20-52 BIOREACTIONS AND BIOPROCESSING

(a) (b) (c)

FIG. 20-37 Three commonly used membrane structures: (a) homogeneous: uniform pore
profile through filter; (b) asymmetric: finer filtering surface faces feed suspension; and (c) com-
posite: two types of materials used . [Reproduced from Bioseparations Science and Engineering,
2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By
Permission of Oxford University Press .]

polyamide, polyether, polycarbonate, polyester, polypropylene, polyethyl- Spiral-wound modules are constructed from flat sheet membranes
ene, regenerated cellulose, poly(vinyl chloride), poly(vinylidene fluoride) separated by spacer screens . The feed solution is fed into one end of the
(PVDF), poly(tetrafluoroethylene) (PTFE), acrylonitrile copolymers, and module and flows through the separator screens along the surface of
polyethersulfones . The inorganic materials used include ceramics, zirco- the membranes . The retentate is then collected on the other end of the
nium oxide, borosilicate glass, stainless steel, and silver . module . The permeate spirals radially inward, eventually to be collected
Cross-flow filtration membranes are available in a variety of configura- through a central tube . The flow in these systems tends to be turbulent .
tions . The membranes are housed in a physical unit called a module . The The main disadvantage of spiral-wound modules is that they are suscep-
membrane module must satisfy a number of mechanical, hydrodynamic, tible to fouling by particulates because of the narrow and irregular flow
and economic requirements, the most important of which are the following through the spacer screens .
[Zeman and Zydney, Microfiltration and Ultrafiltration, Dekker, New York, One effective rotating-type module is the rotating cylinder device origi-
1996, p . 327]: nally proposed by Hallstrom and Lopez-Leiva [Desalination 24: 273 (1978)] .
Mechanical: It must obtain effective (physical) separation of the feed and The feed flows into a thin annular region between two concentric cylinders
permeate streams; provide the necessary physical support for the mem- (Fig . 20-38b) . Either cylinder or both can be porous and can have a mem-
brane (including the ability of the module to endure the required pressure brane bound to the surface . With the membrane on the inner cylinder,
drops and any backflushing) . permeate is collected in the central chamber . When the membrane is on
Hydrodynamic: It must minimize pressure drops through the module (to the outer cylinder, permeate is collected in a separate annular permeate
reduce pumping costs); optimize solute mass transfer (reduce concentra- region adjacent to the outer porous cylinder . The inner cylinder is rotated
tion polarization); minimize particulate plugging or fouling; avoid dead at high speed (typically > 3000 rpm) to induce the formation of Taylor vor-
spots ( for sanitary design); allow for or promote turbulence ( for sanitary tices, which effectively mix the fluid near the surface of the membrane and
design and to reduce boundary layer thickness) . thereby increase the mass-transfer coefficient for the solutes/particles
Economic: It must maximize the membrane packing density (ratio of in the feed . The vortices also help reduce the thickness of the cake or gel
membrane area to module volume); minimize manufacturing costs; permit layer at the membrane surface . One advantage of this system is that the rate
easy access for cleaning and/or membrane replacement; provide sufficient of mass transfer is determined almost entirely by the rate of rotation of the
chemical resistance and operational lifetime; incorporate modularity of inner cylinder, which effectively decouples the mass-transfer characteristics
design for easy scale-up, staging, or cascading . from the feed flow rate . As a result, the rotating devices can be operated at
Several of these criteria are in mutual opposition: for example, modules very low flow rates and with minimal pumping costs . Since, however, the
with high membrane packing density tend to be highly susceptible to plug- energy requirements for rotating the device are usually very high, capital
ging with particulates . Therefore, the choice of a particular module involves costs are high, and scale-up is very difficult, these systems have been limited
balancing these criteria to arrive at the most economic system for each par- to small-scale operation .
ticular application . Scale-Up and Design Experimental testing is generally needed in
There are five types of module configuration for cross-flow filtration the scale-up and design of cross-flow filtration systems . Cross-flow filtra-
(Fig . 20-38): hollow fiber, tubular, flat plate, spiral wound, and rotating . tion modules are available from manufacturers for carrying out laboratory
The tubular module has the same general configuration as the hollow fiber or pilot plant tests . The size of a plant unit can be determined by a direct
module, but the tubes have much larger diameters than the fibers . Some key scale-up of the filtration area based on the feed or permeate flow rate . For
characteristics of these modules are compared in Table 20-18 . this scale-up, however, it is important that the following variables be kept
Hollow fiber modules consist of an array of narrow-bore, self-supporting constant [Datar and Rosen, in Rehm and Reed (eds .), Biotechnology, vol . 3,
fibers that generally have an asymmetric membrane structure . The dense Bioprocessing, VCH, Weinheim, Germany, 1993, p . 486]:
skin layer is usually on the lumen side of the fiber, but it can be placed on the • Inlet and outlet pressures
outside . The feed flows through the lumen of the fibers when the dense skin • Cross-flow (or tangential) velocity
layer is on the lumen side, and the flow is typically laminar, although hol- • Flow channel sizes (height and width)
low fiber systems by some manufacturers can be operated in turbulent flow . • Feed stream properties—test slurries should be representative of the
Since the hollow fibers are self-supporting, they can be cleaned by back- actual process streams
flushing, that is, by reversing the direction of the permeate flow . One disad- • Membrane type and configuration—test data from one design cannot
vantage of hollow fiber modules is that the entire module usually needs to directly be used to design another geometry
be replaced upon the rupture of even a single fiber . Also, to avoid plugging Maintaining the inlet and outlet pressures, the cross-flow velocity, and
of the small-diameter fibers, feed streams generally need to be prefiltered . the flow channel size means that the length of the flow path is constant
Except for some inorganic membranes, tubular membranes are not self- as well . It is also important to make an assessment of the rate of fouling of
supporting and are usually cast in place within a porous support tube made of the membranes . Tests should be performed for several cycles of operation
fiberglass, ceramic, plastic, or stainless steel . The feed flows through the inside to estimate the fouling rate and to determine how fouling can be kept to a
of the tubes, while the permeate flows radially outward across the membrane minimum .
and support tube . Inorganic membranes are usually constructed in a honey- There are four basic modes of operation of cross-flow filtration
comb monolith unit in which the membranes are arranged in a parallel array . (Fig . 20-39): batch concentration, diafiltration, purification, and complete
Tubular systems are typically operated in the turbulent flow regime . Because recycle . In the batch concentration mode, the retained stream containing
of the large diameter of the tubes in these systems, the pumping costs are rela- the product suspended particles or dissolved macromolecules is reduced
tively high compared to other module configurations operated in turbulent in volume . In diafiltration, the volume of the retained stream is kept con-
flow . In addition, large-diameter tubes decrease the filter surface area to mod- stant by the continuous addition of water or buffer, which results in the
ule volume ratio, leading to bigger equipment footprints . The chief advantages removal of low-molecular-weight solutes into the permeate . Diafiltration is
of tubular modules are that they are highly resistant to plugging by particu- commonly used when salt removal or exchange is desired . In the purifica-
lates, and they can be cleaned by backflushing . The diameter of the tubes can tion mode, a low-molecular-weight product passes into the permeate and
be selected to avoid prefiltration of the feed . is thus separated from higher-molecular-weight impurities; or the product
Most flat plate systems are in a rectangular configuration . The flow chan- can be retained and impurities removed in the permeate . In the complete
nels can be open or can have separator screens to improve the mass-transfer recycle mode, both the retained stream and the permeate are returned to
characteristics . Units with open channels are usually operated in laminar the feed tank . Systems may be operated in complete recycle at start-up to
flow, while those with separator screens are often operated with flow in the reach steady state, saturate the membranes, test for leaks and blockage, and
turbulent regime . One disadvantage of these modules is that backflushing adjust the feed rate .
is often impractical because the membranes are effectively supported only Diafiltration is similar to dialysis in that salts are exchanged or removed
on one side . in both modes . Dialysis is a laboratory method that involves placing the
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-53

Membrane
surface

Solvents and
microsolutes

Retained
macrosolutes
Feed Rotating
Hollow fiber cartridge
inlet seal
Membrane
Retained Annular
macrosolutes gap
Separator
screen
Solvents and Membrane
microsolutes Permeate support
Plate-and-frame device collection
tube
Retained Rotating
macrosolutes membrane
Retentate
exit
Rotating
seal
Membrane
Solvents and
microsolutes
Separator Permeate exit
Spiral cartridge screen
(a) (b)
FIG. 20-38 Schematic representations of filter modules . (a) Hollow fiber, plate, and spiral-wound membrane modules . (b) A rotating cylinder mod-
ule . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides
(2015) . By Permission of Oxford University Press .]

feed in a bag made with an ultrafiltration membrane . The bag is placed in a We operate the system in the diafiltration mode, which means that the volume of
large volume of the new solvent desired, and the new solvent is mixed until the solution being desalted is maintained constant . A material balance on the salt in
the system comes to equilibrium or near equilibrium . Complete equilib- the retained volume gives
rium takes 3 to 5 h, while 90 percent equilibrium can be reached in 2 to
3 h (Scopes, Protein Purification, Springer-Verlag, 1982, p . 16) . For a complete V
dc s
= −Qc s (1 − rs )
change of solvent, it is necessary to do the dialysis at least twice . It is fre- dt
quently the job of the engineer to scale up a dialysis done in the laboratory
to a diafiltration in the pilot plant or plant . For scaling and planning, it is where V = volume of solution being desalted
customary to perform material balances on solids mass and liquid volume cs = concentration of salt in volume V
around any component of the operation . rs = fraction of salt rejected by the membrane
Q = filtration rate (volume/time)
Example 20-3 Diafiltration Mode in Cross-Flow Filtration It is desired For this case there is no rejection of salt by the membrane, so that rs = 0 . Integrating the
to use a cross-flow filtration system to desalt 1000 L of a protein solution containing differential equation with the initial time equal to 0 gives
NaCl . The system is capable of operating at a transmembrane flux of 30 L m−2 h−1 .
A membrane is used that will allow complete passage of the salt but no passage of
the protein . To remove 99 .99 percent of the salt, determine the time required and the cs Qt
volume of water required using a cross-flow filtration unit with a membrane area of ln =−
c so V
100 m2 .

TABLE 20-18 Comparison of Key Characteristics of Cross-Flow Membrane Modules*

Channel spacing Packing density Particulate
Module type (cm) (m2/m3) Energy costs plugging Ease of cleaning
Hollow fiber 0 .02–0 .25 1200 Low High Fair
Tubular 1 .0–2 .5 60 High Low Excellent
Flat plate 0 .03–0 .25 300 Moderate Moderate Good
Spiral wound 0 .03–0 .1 600 Low Very high Poor to fair
Rotating 0 .05–0 .1 10 Very high Moderate Fair
*Zeman and Zydney, Microfiltration and Ultrafiltration, Dekker, New York, 1996 . p . 331 .
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and
Demetri P . Petrides (2015) . By Permission of Oxford University Press .
20-54 BIOREACTIONS AND BIOPROCESSING

Buffer or Buffer or
water water

Retentate

CF CF CF CF

Feed
Permeate Permeate with low
molecular weight solutes
(a) Batch concentration (b) Diafiltration

Buffer or Buffer or
water water

CF CF C

Product in
permeate
(c) Purification (d) Complete recycle
FIG. 20-39 The four basic modes of operation of cross-flow filtration (CF = cross-flow filter): before the arrow, start of the operation; after the arrow,
end of the operation . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri
P . Petrides (2015) . By Permission of Oxford University Press .]

where cso = initial salt concentration . Solving for t, we obtain Cross-flow filtration systems may be designed to operate in the batch
or the continuous mode (Fig . 20-41) . In a batch system, feed is pumped
cs through the filtration module and then back to the feed tank . In a variation
V ln
c so (1000 L)(ln 0 .0001) of this mode (called semibatch) for diafiltration, fluid is continuously added
t =− =− = 3 .07 h to the feed tank to keep the feed volume constant (Fig . 20-39b) .
Q  L  2
 30 m 2 h  (100 m ) In the continuous mode of operation, also sometimes called the feed-
and-bleed mode, or retentate bleed mode, feed is added to a recirculation
loop by the feed pump, and concentrate exiting in the retained stream is
L withdrawn from the system so that the concentration factor (i .e ., concen-
Volume of water required = (3 .07 h)  30 2  (100 m 2 ) = 9210 L
 m h tration in the retentate divided by the concentration in the feed) is at the
desired value . When steady state has been achieved, the concentrate will
Thus, a relatively large volume of wastewater will be generated by this process .
be at its maximum concentration, which means that the filtration flux will
be at a minimum throughout the run [Tutunjian, in Moo-Young (ed .), Com-
prehensive Biotechnology, vol . 2, Principles of Biotechnology, Pergamon, 1985,
In designing a diafiltration process, a decision must be made about the p . 411] . It is generally more economical to use a multistage system in a con-
concentration of retained product at which to operate . As this concentration tinuous process (Fig . 20-42) . As more stages are added, the average filtra-
is increased, the filtration flux will decrease according to Eq . (20-24), and the tion flux approaches that for a batch system, and thus the total filtration
total volume of filtrate will decrease for the removal of a given percentage area decreases as more stages are added . This is illustrated in Table 20-19,
of a low-molecular-weight solute . This leads to an optimum concentration where batch ultrafiltration operation is compared with continuous opera-
to minimize the time required, which can be determined mathematically if tion using one, two, three, and five stages .
the relation between filtrate flux and concentration in the bulk fluid (cb) is Because of the economic advantages of continuous operation and
known [Tutunjian, in Moo-Young (ed .), Comprehensive Biotechnology, vol . 2, reduced tankage, this scheme is preferable to batch operation for most large-
Principles of Biotechnology, Pergamon, 1985, p . 411] . scale ultrafiltration operations (Henry et al ., in Perry’s Chemical Engineers’
The basic components in the design of a cross-flow filtration system are
shown in Fig . 20-40 . A pump flows the feed through the filtration module to
give a permeate and a retentate or retained stream . The pump needs to be Back-pressure
sized to provide the desired flow velocity and pressure . The transmembrane valve
pressure is controlled by a back-pressure valve on the retentate stream exit- Retentate Po
ing from the filtration module . Thus the transmembrane pressure drop is
estimated by
CF
1 module
∆pTM = ( pi + po ) − p p (20-37) Permeate
2

where pi and po are the retentate pressures in and out of the module, respec- Feed Pi
tively, and pp is the pressure of the outlet permeate . In designing a cross-flow Pump
filtration system, it is important to minimize the occurrence of gas-liquid
interfaces, since bioproduct denaturation, especially of proteins, can occur FIG. 20-40 Basic components of a cross-flow filtration system . [Reproduced from
at these interfaces in the presence of mechanical shear and turbulent flow Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott
[Virkar et al ., Biotechnol. Bioeng . 23: 425 (1981)] . R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-55

Retentate

Process CF CF
tank

Feed
Pump Permeate Feed Recirculation Permeate
pump pump
(a) (b)
FIG. 20-41 Comparison of (a) batch and (b) single-stage continuous ( feed-and-bleed) cross-flow filtration systems.
[Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and
Demetri P. Petrides (2015). By Permission of Oxford University Press.]

Stage 1 Stage 2 Stage 3

Retentate

CF CF CF

Feed
Feed Recirculation Recirculation Recirculation
pump pump Permeate pump Permeate pump Permeate
FIG. 20-42 Multistage cross- flow filtration system using the retentate bleed mode. [Reproduced from Bioseparations
Science and Engineering, 2d ed., by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By
Permission of Oxford University Press.]

Handbook, 7th ed ., McGraw-Hill, New York, 1997, pp . 22–56) . Another filtration resistance from the filtrate that is reliable or reproducible . Further,
advantage of continuous operation is that it permits the minimization of there is no opportunity to add filter aids to form a cake, since this is usually
the residence time of the product in the cross-flow filtration unit, which is the last step in a clean process .
important for products that are sensitive to heat or shear . Dairy and food Sizing of the filter is almost always based on the maximum allowable
proteins are generally processed continuously, while most pharmaceutical load in the solution to be filtered as well as the desired sterility assur-
and biological products are processed using batch operation [Tutunjian, in ance limit . The sterility assurance limit is the calculated probability
Moo-Young (ed .), Comprehensive Biotechnology, vol . 2, Principles of Biotech- of a single unit of product containing a single microorganism . This is
nology, Pergamon, 1985, p . 411] . expected to be a maximum of 10−4 for aseptic processes, and one usu-
Sterile Filter Design Sterile filtration is an important application of ally designs in an extra order of magnitude at least. One determines the
conventional filters in the pharmaceutical industry, where many products maximum load on the filter by knowing the bioburden or viral load speci-
are sold in sterile form . It is sometimes desirable that other bioseparation fication of the purified pool from the preceding step, or by knowing the
processing steps besides the final one be carried out under sterile condi- acceptable microbial or viral levels in the raw materials that make up that
tions, which requires the filtration of both liquids and air . pool. The retentive requirement for the filter is then divided by the reten-
Sterile Liquid Filtration In typical sterile liquid filtration applications, tion capability (per area) of the filter to be used. The resulting filter area
there is no appreciable filter load, no cake buildup, and no contribution to is the minimum required to meet the sterility goal of fewer than one unit
dose contaminated with one organism per million doses manufactured.
The next step is to ensure that the filter is large enough to allow reasonable
TABLE 20-19 Comparison of Batch and Continuous Ultrafiltration processing times. Typically, one does not want to filter for more than one
Systems Using a Model of Flux as a Function of Bulk Concentration* shift (8 h). At 24 h, questions will be raised about “grow-through” (where
retained microbes will have colonized the filter and “grown through” to
System† Flux (L m−2 h−1) Total area (m2)
the other side). The filtration time for the filter size is determined by
Batch 33.1 (average) 136 finding the clean water flow rate for the filter, either through laboratory
Continuous experiment or through the vendor’s published data, and adjusting for the
One-stage 8.1 555 viscosity of the solution to be filtered relative to water (a correction for the
Two-stage 31.1 243 effect of temperature on viscosity may be necessary). Once the flow rate
8.1 is known, the time to filter a given size batch can be determined. The area
Three-stage 38.7 194
23.4
of the filter can be increased to reduce the filtration time, but cannot be
8.1 decreased without jeopardizing the sterility assurance.
Five-stage 44.7 165 For the ultimate filtration step, in which the product is rendered sterile, a
35.6 depth filter can be added upstream of the membrane filter. This protects the
26.4 sterile filter from fouling during use. This is a critical consideration in high-
17.3 concentration process streams. It is very risky to replace a sterile filter once
8.1 in place, with all the connections downstream sterilized. Depth filters with
∗Tutunjian, in M. Moo-Young (ed.), Comprehensive Biotechnology, vol. 2, Principles of high “dirt loads” can keep particulate matter from reaching the membrane
surface, keeping it clean for the removal of microbes. Two sterile filters are
Biotechnology, Pergamon, Oxford, UK, 1985, p. 411.
†
System design for 10× concentration factor and feed rate of 5000 L/h. sometimes used in series, for added sterility assurance. This gives very high
Flux from Jm = 20 ln (30/cb). probability that dosage units will not be contaminated, but the arrangement
Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G. sometimes becomes difficult to manage in other respects. For instance,
Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission multiple sterility assurance tests must be performed, a requirement that
of Oxford University Press. adds manipulations and increases chances for breaches in sterility.
20-56 BIOREACTIONS AND BIOPROCESSING

Filters used for the purpose of sterilizing pharmaceutical products must Testing bubble points in solvents is feasible, and often performed, but is
be tested to ensure that they have been properly installed and have no considered a destructive form of measurement .
defects that could allow sterility to be compromised . The two most com-
mon tests for a sterile filter for an aqueous product are the bubble point and SEDIMENTATION
the pressure hold or diffusion test . In both tests, a wetted, sterilized filter
is pressurized with air or nitrogen . In the bubble point test, this pressure Sedimentation is the movement of particles or macromolecules in an iner-
is increased until the lowest pressure is found at which a gas passes freely tial field . Its applications in separation technology are widespread . Extremes
through the filter . It is assumed that this is the pressure required to over- of applications range from the settling due to gravity of tons of solid waste
come the surface tension and viscosity of water in the largest membrane and bacteria in wastewater treatment plants to the high speed centrifuga-
pore (where the forces are the weakest) . As the pore size of the membrane tion of a few microliters of blood to determine packed blood cell volume in
decreases, the bubble point increases . This test will also detect a sliced or the clinical laboratory . Accelerations range from 1g in flocculation tanks to
unseated O-ring or other seal failure . The pressure hold test pressurizes a 100,000g in ultracentrifuges for measuring the sedimentation rates of mac-
wetted filter to a pressure lower than the bubble point and measures either romolecules . In bioprocessing, the most frequent applications of sedimen-
the total flow of gas through the membrane (by diffusion) or the rate of pres- tation include the clarification of broths and lysates, the collection of cells
sure decay . Both should be low, and both should meet the manufacturer’s and inclusion bodies, and the separation of fluids having different densities .
specifications for an intact, properly installed sterile filter . Unit operations in sedimentation include settling tanks and tubular
Compatibility and extractable tests need to be performed on each filter centrifuges for batch processing, continuous centrifuges such as disk cen-
type . Compatibility means that the filter does not adsorb an active ingredi- trifuges, and less frequently used unit operations such as field-flow frac-
ent or any other type of active excipient, such as a preservative . Such adsorp- tionators and inclined settlers . Bench-scale centrifuges that accommodate
tion to the filter medium can change the strength (or biological potency) of small samples can be found in most research laboratories and are frequently
the solution after the filtration step . Extractables and leachable chemicals, applied to the processing of bench-scale cell cultures and enzyme prepara-
such as monomers from the membrane or support medium, filter coatings tions . Certain high-speed ultracentrifuges are used as analytical tools for the
or mold release agents, and storage solutions, such as ethanol, must also estimation of molecular weights and diffusion coefficients .
be tested for in pilot studies with the product . These chemicals would be The common types of production centrifuges are shown in Fig . 20-43, and
considered impurities in the final product and should be removed or pre- a comparison of the advantages and disadvantages of the different centri-
vented from leaching . Compatibility and extractables should be tested in a fuge designs is given in Table 20-20 [Bell et al ., in Fiechter (ed .), Advances
sample of the product solution . When this is not possible, because product in Biochemical Engineering/Biotechnology, vol . 26, Springer-Verlag, Berlin,
or excipients would interfere with the analysis of the impurities, a placebo 1983, p . 1] . The principles of sedimentation and more details about equip-
or similar solution should be used . ment for centrifugation are discussed in the Centrifuges subsection . Here,
Air Sterile Filtration Air filters are also sized based on the antici- additional information on this topic that is specific for a bioproduct sepa-
pated flow rate . Airflows are specified with one or more of the following ration process is given, including the range of particle sizes and densities
criteria: pressure differential between areas, air changes per hour, or linear that are encountered, the concept of the sedimentation coefficient and its
flow rate . Pressure differentials are usually measured between rooms, or application, and the concept of equivalent time, which has been found to be
between rooms and corridors . These are specified when it is important for useful for scale-up .
air to flow from a “cleaner” area to a “dirtier” area when doors, windows, Fluid Dynamics During the sedimentation of a particle in the pres-
or pass-throughs are opened between them . Pressure differentials are ence of a centrifugal field, the velocity of the particle reaches a constant
typically 50 to 150 Pa [= 0 .5 to 1 .5 cm of water, or 0 .4 to 1 .1 mm Hg (torr)] . value when all the forces are balanced . For a centrifugal field of accelera-
Air changes per hour are often specified for rooms based on hazardous tion w2R, where w is the centrifuge’s angular velocity and R is the distance of
materials or clean operations the room is designed for . Air changeover is the particle from the center of rotation, at constant particle velocity v for a
the airflow rate divided by the volume of the room . Air changeovers are particle of radius ap and density ρ in a fluid of density ρ0 and viscosity m, an
different from makeup rate, which is the fraction of fresh air injected in the analysis using the equation of motion gives (Harrison et al ., Bioseparations
room’s air handler . Science and Engineering, Oxford University Press, Oxford, UK, 2015, p . 187)
Clean rooms often have very high changeover rates but low makeup rates .
Rooms handling solvents or hazardous materials have high makeup rates (a 2 a 2p (ρ−ρo )ω 2 R
v = (20-38)
100 percent makeup rate is called once-through air) but low air cleanliness 9µ
requirements . The changeover rate varies with the hazard of the material or
the cleanliness requirement for the room . Linear flow rate is usually speci- Creeping flow conditions are usually satisfied in sedimentation . The Reyn-
fied in hoods and smaller spaces . Linear flow rates of 5 to 20 ft/s (= 2 .4 to olds number for sedimenting spherical particles is
6 .1 m/s) are found in laminar flow hoods, in aseptic areas where pharma-
ceuticals are sterile-processed, and in clean rooms where microelectronics 2a p v ρ
Re = (20-39)
are manufactured . Linear flow rates are established to prevent airborne 9µ
microbes and particulate matter from settling in or on the materials being
processed . Creeping flow occurs at Reynolds numbers less than about 0 .1 (Bird et al .,
The air quality of a room or area is typically classified according to the Transport Phenomena, Wiley, Hoboken, N .J ., 2002, p . 59) .
number of airborne particles that can be measured in 1 m3 . Increasingly, Table 20-21 shows the sedimentation velocities for important biopar-
rooms are being classified according to the International Organization for ticles and biomolecules calculated from Eq . (20-38) with ρ0 = 1.0 g/cm3,
Standardization (ISO), which has published a series of documents num- µ = 0.01 g cm−1 s−1 (poise), and representative values of r and ap. The sedi-
bered 14644 (volumes 1 to 5) . These standards classify clean room space as mentation velocity and Reynolds number results shown in Table 20-21 for
1 (cleanest) through 9 (dirtiest), with each classification allowing approxi- yeast cells and bacterial cells at gravitation acceleration can be multiplied
mately one log more particles in a cubic meter of air than the previous . by a centrifuge’s centrifugal acceleration to give the corresponding values
For biotechnology and pharmaceuticals, the most commonly used grades for operation in the centrifuge; for example, at a dimensionless acceleration
are ISO Class 5 (similar to the military standard Class 100) to ISO Class of 10,000, the Reynolds number for yeast cells is 0.07, which means that the
8 (similar to the military standard class 100,000) . ISO Class 5 is used for flow is still creeping.
aseptic processing where products may be briefly exposed to the environ- It is clear from Table 20-21 that gravitational sedimentation is too slow
ment, and ISO Class 8 is used for areas containing primarily closed opera- to be practical for bacteria, and conventional centrifugation is too slow for
tions . High-efficiency particulate air (HEPA) filters are routinely tested for protein macromolecules. In the case of true particles, flocculation is often
integrity by spraying a polyalphaolefin (PAO) aerosol on the feed side of used to increase the Stokes radius ap, while ultracentrifugation is used in
the filters . The typical specification for a PAO test is a 103 reduction in macromolecular separations (Harrison et al., Bioseparations Science and
aerosol particles . Engineering, Oxford University Press, Oxford, UK, 2015, pp. 203, 205).
Vent filters filter air on tanks and drainable equipment . Vent filters typi- When particle density and solvent density are equal, the sedimenta-
cally follow the same specification as the liquid filters used in the process . tion velocity v is zero, and the process is called isopycnic or equilibrium
Vent filters are also sized based on the maximum airflow rate anticipated . sedimentation. This fact is exploited in the determination of molecular
The maximum flow rate is often encountered when the tanks are being filled densities and in the separation of living cells. A density gradient or a
or cleaned . Vent filters must allow flow in both directions . Integrity testing density shelf is employed in such cases. Densities of representative cells,
of vent filters is an evolving technology . These filters are typically hydropho- organelles, and biomolecules measured by this method are given in
bic, so aqueous-based bubble points and pressure holds are not appropri- Table 20-22.
ate . The PAO tests will not work because vent filters are typically membrane An example of a density shelf used for the preparation of cells is the
filters (which are the most compact), while HEPA filters are depth filters . preparation of lymphocytes by sedimentation. The goal of this separation
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-57

Solids

(a) (b) (c)

Solids

(d) (e) (f)

FIG. 20-43 Common types of production centrifuges: (a) tubular bowl; (b) multichamber; (c) disk, nozzle;
(d) disk, intermittent discharge; (e) scroll; and ( f ) basket . Arrows indicate the path of the liquid phase; dashed
lines show where the solids accumulate . [Reproduced from Bioseparations Science and Engineering, 2d ed .,
by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
University Press .]

TABLE 20-20 Comparison of Production Centrifuges* is to remove erythrocytes from a leukocyte population on the basis of a
System Advantages Disadvantages
density shelf . By combining Ficoll, a high-molecular-weight polymer, and
Hypaque, a heavily iodinated benzoic acid derivative, in appropriate pro-
Tubular bowl High centrifugal force Limited solids capacity portions in aqueous buffers, it is possible to achieve a density around 1 .07
Good dewatering Foaming unless special g/cm3 in isotonic solutions . At this density, most white blood cell subpop-
Easy to clean skimming or centripetal
Simple dismantling of bowl pump used ulations will float and nearly all red blood cells will sediment, which is
Recovery of solids difficult in agreement with what would be expected for these cells based on their
densities, shown in Table 20-22 .
Chamber bowl Clarification efficiency remains No solids discharge
constant until sludge space full Cleaning more difficult than
When the concentration of sedimenting particles increases, the sedimen-
Large solids-holding capacity with tubular bowl tation velocity has been found to decrease, a phenomenon known as hin-
Good dewatering Solids recovery difficult dered settling . This effect has been quantified by the following expression
Bowl cooling possible for particles of any shape [Richardson and Zaki, Trans . Inst. Chem. Eng . 32:
Disk centrifuge Solids discharge possible Poor dewatering 36 (1954)]:
Liquid discharge under pressure Difficult to clean
eliminates foaming
Bowl cooling possible
vc = v(1 − ϕ)n (20-40)
Scroll or Continuous solids discharge Low centrifugal force
decanter High feed solids concentration Turbulence created by scroll where vc is the sedimentation velocity of particles in a concentrated sus-
centrifuge pension, v is the velocity of individual particles [Eq . (20-38)], ϕ is the
Basket Solids can be washed well Not suitable for soft biological volume fraction of the particles, and n is a function of only the shape of
centrifuge Good dewatering solids the particle and the Reynolds number . For spherical particles with Re
Large solids holding capacity No solids discharge < 0.2 (usually satisfied during sedimentation), the exponent n has been
Recovery of solids difficult found to be 4 .65 . Equation (20-40) may also be applied to particles of any
*Bell et al ., in Fiechter (ed .), Advances in Biochemical Engineering/Biotechnology, size in a polydisperse system, using the volume fraction for all the par-
vol . 26, Springer-Verlag, Berlin, 1983, p . 1 ticles in the calculation [Richardson and Shabi, Trans. Inst. Chem. Eng.
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . 38: 33 (1960)] .
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission The magnitude of the hindered settling effect for spherical particles as a
of Oxford University Press . function of the particle volume fraction ϕ can be seen in Table 20-23 . Note

TABLE 20-21 Calculated Settling Velocities and Reynolds Number for Example
Bioproducts (Assumes q0 = 1.0 g/cm3 and l = 1.0 cP)
Dimensionless Sedimentation
Bioparticle or Sedimentation Density acceleration velocity v Reynolds
biomolecule radius ap (mm) ρ (g/cm3) (G = ω2R/g) (cm/h) number Re
Yeast cell 2 .5 1 .1 1 0 .5 7 × 10−6
Bacteria cell 0 .5 1 .1 1 0 .02 6 × 10−8
Protein 0 .005 1 .3 104 0 .06 2 × 10−9
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W .
Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford University Press .
20-58 BIOREACTIONS AND BIOPROCESSING

TABLE 20-22 Measured Values of Density of Representative Cells, Example 20-4 Application of the Sedimentation Coefficient In 1974, D.
Organelles, and Biomolecules E. Koppel determined the sedimentation coefficient s20,w for the smaller ribosomes from
Escherichia coli to be 70 S [Koppel, Biochemistry 13: 2712 (1974]. Estimate how long it
Cell, organelle, or biomolecule Density ρ (g/cm3) Ref. would take to completely clarify a suspension of these ribosomes in a high-speed centri-
Escherichia coli 1.09∗ 1 fuge operating at 10,000 rpm with a tube containing the ribosome suspension in which
Bacillus subtilis 1.12 2 the maximum distance of travel of particles radially outward is 1 cm and the initial dis-
tance from the center of rotation to the particles nearest the center of rotation is 4 cm.
Arthrobacter sp. 1.17 3
We can write Eq. (20-41) as
Saccharomyces pombe 1.09 1
Saccharomyces cerevisiae 1.11∗ 1 dR 1
s=
Amoeba proteus 1.02 1 dt ω 2 R
Murine B cells 1.06∗ 1
or
Chinese hamster ovary (CHO) cells 1.06 1
Red blood cells 1.10∗ 4 dR
ω 2 s dt =
White blood cells 1.02∗ 5 R
Peroxisomes 1.26∗ 6 We integrate this equation with the initial condition at t = 0, R = R0 (distance from cen-
Mitochondria 1.20∗ 6 ter of rotation to the particles nearest the center of rotation) to give
Plasma membranes 1.15∗ 6 R
Proteins 1.30∗ 6 ω 2 st = ln
Ro
Ribosomes 1.57∗ 6
DNA 1.68∗ 6 To determine the maximum time required, we evaluate R at the maximum travel of
RNA 2.00∗ 6 the cells measured from the center of rotation (5 cm):

∗Average value.  R 1h
ln   ln(5/4) ×
References  Ro  3600 s
t= = 2 = 8 .1 h
1. Kubitschek, Crit. Rev. Microbiol. 14: 73 (1987). ω2s  rev 2 π rad 1 min  −13
2. Hart and Edwards, Arch. Microbiol. 147: 68 (1987).  10,000 min × rev × 60 s  (70 × 10 s)
3. Illmer, FEMS Microbiol. Lett. 196: 85 (2001).
4. Ponder, J. Biol. Chem. 144: 333 (1942). This should not be an unreasonable amount of time to centrifuge the ribosomes.
5. Barnkob et al., 15th International Conference on Miniaturized Systems for However, since the time varies inversely with the square of the rotation speed, the time
Chemistry and Life Sciences, Oct. 2–6, Seattle, Wash., 2011. can be reduced to 2 h by doubling the speed.
6. Sheeler, Centrifugation in Biology and Medicine, Wiley-Interscience, New York,
1981.
Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G. Equivalent Time To assess the approximate properties of a particle
Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By
Permission of Oxford University Press.
type to be separated, it is sometimes convenient to calculate an “equivalent
time.” To do this, we first define a dimensionless acceleration G, the ratio of
the centrifugal to gravitational acceleration for a particular centrifuge:

that hindered settling can be significant for particle concentrations of a few ω 2R

G≡ (20-43)
percent or greater . g
Sedimentation Coefficient When a body force is applied, velocity where R is usually defined as the radius of the centrifuge bowl. Thus, this
through a viscous medium is usually proportional to the accelerating field dimensionless unit is measured in “g’s”—multiples of the earth’s gravita-
(examples are electric, magnetic, and inertial) . In the case of sedimentation, tional acceleration. A rough approximation of the difficulty of a given sepa-
the resulting constant, a property of both the particle and the medium, is ration by centrifugation is the product of the dimensionless acceleration
the sedimentation coefficient, which is defined as and the time required for the separation. This product is called the equiva-
lent time for the separation and is written as
v
s≡ (20-41) ω2R
ω2R Equivalent time = t (20-44)
g
where v is the steady state velocity of the particle, R is the distance of the Typical values of equivalent time are as follows: 0.3 × 106 s for eukaryotic
particle from the center of rotation, and ω is the angular velocity . Comparing cells, 9 × 106 s for protein precipitates, 18 × 106 s for bacteria, and 1100 × 106 s
this equation with Eq . (20-38), we see that for ribosomes (Belter et al., Bioseparations, Wiley, Hoboken, N.J., 1988, p. 63).
The equivalent time for the centrifugation of cells or biological particles
2 a p 2 (ρ−ρo ) of unknown sedimentation properties may be estimated in a laboratory
s= (20-42)
9µ centrifuge. Samples are centrifuged for various times until a constant vol-
ume of packed cells is reached. The equivalent time Gt is calculated as the
which defines s in terms of only properties of the particle and the medium . product of the G for the particular centrifuge and the time required to reach
This coefficient is usually expressed at 20°C and under conditions (viscosity constant packed cell volume. A centrifuge that has commonly been used for
and density) of pure water as this determination is the Gyro-Tester (Alfa Laval, Inc.).
One approach to scale-up of a centrifugal operation is to assume constant
s20,w (units of S) equivalent time:
(Gt )1 = (Gt )2 (20-45)
The sedimentation coefficient is often expressed in svedberg units, where
10−13 seconds = 1 svedberg unit (S), named after the inventor of the ultracen- where the subscripts refer to centrifuges 1 and 2, respectively.
trifuge, Theodor Svedberg.
Example 20-5 Scale-Up Based on Equivalent Time If bacterial cell debris
has Gt = 54 × 106 s (Belter et al., Bioseparations, Wiley, Hoboken, N.J., 1988, p. 63), how
large must the centrifuge bowl be and what centrifuge speed is needed to effect a full
TABLE 20-23 Effect of Particle Volume Fraction i on the Particle sedimentation in 2 h?
From Eq. (20-44) for Gt, we can estimate the centrifuge speed ω if we know the cen-
Sedimentation Velocity for Spherical Particles
trifuge bowl size and the time of centrifuging. It is reasonable to have a centrifuge that
vc is 10 cm in diameter. Solving Eq. (20-44) for ω by using these values gives
ϕ v 1/2
1/2
 54 × 10 6 s × 9 .81 m 
0.01 0.95  s2 
ω = 
Gtg 
=
0.05 0.79  Rt  0 .05 m × 2(3600) s 
0.10 0.61  
0.20 0.35
rad 1 rev 60 s
= 1213 × × = 11,590 rpm
Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G. s 2 π rad min
Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission
of Oxford University Press. This speed can be achieved in a production tubular bowl centrifuge.
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-59

Another commonly used method to scale up centrifugation is the sigma

analysis method, as discussed under the subsection Sedimentation Centri-
fuges in Sec . 18 . This method uses the velocity of particles at 1g and a sigma
factor that represents the geometry and speed of the centrifuge .

PRECIPITATION
Precipitation, which is the process of coming out of solution as a solid, is an Ab
important method in the purification of proteins that usually comes early in Ag
the purification process . Precipitation is frequently used in the commercial
separation of proteins . The primary advantages of precipitation are that it
is relatively inexpensive, can be carried out with simple equipment, can be
FIG. 20-44 Schematic representation of antibody–antigen (Ab–Ag) interaction .
done continuously, and leads to a form of bioproduct that is often stable in [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison,
long-term storage . Since precipitation is quite tolerant of various impuri- Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
ties, including nucleic acids and lipids, it is used early in many bioseparation University Press .]
processes .
The goal of precipitation is often concentration to reduce volume,
although significant purification can sometimes be achieved . For example, Here, rs and rr are the radius of the protein solute and polymer rod, respec-
all the protein in a stream might be precipitated and redissolved in a smaller tively, V is the partial specific volume of the polymer, cp is the polymer con-
volume, or a fractional precipitation might be carried out to precipitate the centration, and β′ is a constant . Based on this model, we can expect the
protein of interest and leave many of the contaminating proteins in the lowest protein solubility for large proteins .
mother liquor . Molecular size is the predominant factor in a type of precipitation known
In this section, first the focus is upon protein solubility, which is the basis as affinity precipitation . When affinity groups or antibodies to a specific bio-
of separations by precipitation . Then we discuss the basic concepts of parti- molecule (antigen) are added to a solution, the antibody-antigen interaction
cle formation and breakage and the distribution of precipitate particle sizes . can form large multimolecular complexes, as shown in Fig . 20-44 . Such com-
The specific methods that can be used to precipitate proteins are treated plexes are usually insoluble and cause selective precipitation of the antigen .
next . Finally, the methodology to use for the design of precipitation systems The average size of the complex agglomerates is maximized when there is a
is discussed . 1:1 stoichiometric ratio of antibody and antigen . If either is present in great
Protein Solubility The most important factors affecting the solubility excess, only bimolecular complexes will be formed; and there may be no pre-
of proteins are structure and size, protein charge, and the solvent . Explana- cipitation, or low recovery, even if a precipitate is formed .
tions follow for each of these factors . Charge The net charge of a protein has a direct bearing upon the
Structure and Size In the native state, a protein molecule in an aque- protein’s solubility . The solubility of a protein increases as its net charge
ous environment assumes a structure that minimizes the contact of the increases, a result of greater interaction with dipolar water molecules .
hydrophobic amino acid residues with the water solvent molecules and A repulsive reaction between protein molecules of like charge further
maximizes the contact of the polar and charged residues with the water . increases solubility .
The major forces acting to stabilize a protein in its native state are hydro- A simple way to vary the charge on a protein is by changing the pH of the
gen bonding, van der Waals interactions, and solvophobic interactions . In solution . The pH of the solution in which a protein has zero net charge is
aqueous solution, these forces tend to push the hydrophobic residues into called the isoelectric pH or isoelectric point . The isoelectric pH is known as
the interior of the protein and the polar and charged residues to the pro- the pI of the protein . The solubility of a protein is, in general, at its minimum
tein’s surface . For example, one study of 36 globular proteins has shown at the isoelectric point . A typical example is shown in Fig . 20-45 . Nonuni-
that 95 percent of the ionizable groups are solvent accessible [Rashin and form charge distribution, however, results in a dipole moment on the mol-
Honig, J. Mol. Biol . 173: 515 (1984)] . In other studies of 69 proteins, the aver- ecule, which leads to an increase in solubility and a move in the minimum
age solvent-(water-) accessible atomic surface was found to be 57 percent solubility away from the isoelectric point . The effect of the dipole moment is
nonpolar, 25 percent polar, and 19 percent charged [Miller et al ., J. Mol. Biol . discussed further in the following subsection .
196: 641 (1987); Janin et al ., J. Mol. Biol . 204: 155 (1988)] . Thus, in spite of the The net charge of a protein is determined by the following factors: the
forces operating to force hydrophobic residues to the protein’s interior, the total number of ionizable residues, the accessibility of the ionizable resi-
surface of proteins usually contains a significant fraction of nonpolar atoms . dues to the solvent, the dissociation constants (or pKa values) of the ioniz-
The forces acting on a protein lead to the achievement of a minimum able groups, and the pH of the solution [Rothstein, in Harrison (ed .), Protein
Gibbs free energy . For a protein in its native configuration, the net Gibbs free Purification Process Engineering, Dekker, New York, 1994, p . 115] . Besides
energy is on the order of only 10 to 20 kcal/mol . This is a relatively small net the chemical makeup of the ionizable groups, factors that can influence the
free energy, which means that the native structure is only marginally stable pKa values are the chemical nature of the neighboring groups (e .g ., induc-
and can be destabilized by relatively small environmental changes [Privalov, tive effects), the temperature, the chemical nature of the solvent as partially
Annu. Rev. Biophys. Biophys. Chem. 18: 47 (1989)] . reflected by its dielectric constant, and the ionic strength of the solvent .
Water molecules bind to the surface of the protein molecule because of Solvent The solvent affects the solubility of proteins primarily through
association of charged and polar groups and immobilization by nonpolar two parameters, hydrophobicity and ionic strength . The first parameter has
groups . For example, a study of the hydration of human serum albumin been well studied through observations of single-phase solutions of water
found two layers of water around the protein [Van Oss and Good, J. Protein and primary alcohols . Although these solutions can cause protein dena-
Chem . 7: 179 (1988)] . In the layer next to the protein, the water molecules turation at room temperature, denaturation can be avoided at sufficiently
are almost totally oriented, with the hydrogen atoms adjacent to and facing low temperatures . Studies of primary alcohols have shown that denaturing
the albumin surface, while the oxygen atoms face away from the protein efficiency is as follows:
surface . In the second layer of water molecules, most of the water mol-
ecules (70 percent) are nonoriented . These hydration layers are thought methanol < ethanol < propanol < butanol
to promote solubility of the protein by maintaining a distance between the
surfaces of protein molecules . This led to the conclusion that the alcohols with longer alkyl chains are
The size of a protein becomes important with respect to solubility when binding more effectively to nonpolar groups on the protein, weakening
the protein is excluded from part of the solvent . This can happen when non- intraprotein hydrophobic interactions and thus leading to denaturation
ionic polymers that are added to the solution result in steric exclusion of [Bull and Breese, Biopolymers 17: 2121 (1978)] . It is thought that when the
protein molecules from the volume of solution occupied by the polymer . temperature is low, the primary alcohols compete for the water of hydration
Juckes [Biochim. Biophys. Acta 229: 535 (1971)] developed a model for this on the protein and cause the protein molecules to approach more closely, so
phenomenon based on the protein molecule being in the form of a solid that van der Waals interactions lead to aggregation .
sphere and the polymer molecule in the form of a rod, which gave the The ionic strength of the solvent can have both solubilizing and
following equation for S, the solubility of the protein: precipitating effects . The solubilizing effects are referred to as salting in,
while the precipitating actions are called salting out . The salting-in effect
ln S = β ′ − K ′c p (20-46) has been observed for several proteins, including the class of proteins
named euglobulins. These proteins are insoluble in the absence of salt at
where their isoelectric points, but become soluble when salt is added . In contrast,
3 members of another class of proteins, the albumins, are very soluble in water
V  rs + rr  as well as in high concentrations of salt . It is believed that the solubilities
K′= (20-47)
2 .303  rr  of the euglobulins and the albumins differ because the euglobulins have a
20-60 BIOREACTIONS AND BIOPROCESSING

1.0 salting-in effect observed with proteins with high dipole moments . Also it
can be seen that the salting-in term increases more than the salting-out
term as the dielectric constant decreases . The dielectric constant decreases
as the polarity of the solvent decreases . Therefore, the salting-in effect tends
to predominate in relatively nonpolar solvents, while the salting-out effect
is more dominant in aqueous solvents .
0.5 At high ionic strength, the salting-out effect becomes predominant and
can be described empirically by the Cohn equation [Cohn, Physiol. Rev. 5:
349 (1925)]:

ln S = β − K s′I (20-52)

where S is the solubility of the protein and K s′ is a salting-out constant char-

0
acteristic of the specific protein and salt that is independent of tempera-
Iog S (g/L)

ture and pH above the isoelectric point . The constant b, the hypothetical
solubility of the protein at zero ionic strength, depends on only temperature
and pH for a given protein and is a minimum at the isoelectric point . It is
interesting that the Cohn equation is identical in form to the equation that
describes the precipitation of proteins by the addition of nonionic polymers,
–0.5
Eq . (20-46) . In addition, the Kirkwood equation for the solubility of dipolar
ions [Eq . (20-48)] can be arranged to give

ln S p = ln So − ( K s − K i ) I (20-53)

–1.0 which is also identical in form to the Cohn equation, with

β = ln So (20-54)

K s′ = K s − K i (20-55)
3 4 5 6 7
pH Both salting in and salting out are illustrated in Fig . 20-46 for hemoglobin
with ammonium sulfate or sodium sulfate being added . From zero ionic
strength, the solubility of the protein increases to a maximum as salt is
+6 +1 –0.5 added and then continuously decreases as even more salt is added .
Charge Z
FIG. 20-45 The solubility S of insulin in 0 .1 N NaCl as a function of pH . The charge Z is
Example 20-6 Salting Out of a Protein with Ammonium Sulfate
Data were obtained on the precipitation of a protein by the addition of ammonium
the average protonic charge per 12,000 g of insulin at the pH values indicated . Insulin’s
sulfate . The initial concentration of the protein was 15 g/L . At ammonium sulfate con-
pI corresponds to a zero net charge where its solubility is near minimum . (Data from
centrations of 0 .5 and 1 .0 M, the concentrations of the protein remaining in the mother
Tanford, Physical Chemistry of Macromolecules, Wiley, New York, 1961 .)
liquor at equilibrium were 13 .5 and 5 .0 g/L, respectively . From this information, esti-
mate the ammonium sulfate concentration to give 95 percent recovery of the protein
as precipitate .
much higher dipole moment than the albumins [Oncley, in Cohn and Edsall We can use the Cohn equation [Eq . (20-52)], to solve this problem if we can
(eds .), Proteins, Amino Acids, and Peptides, Reinhold, New York, 1943, p . 543] . determine the constants in the equation . Since ionic strength is directly proportional
A theoretical treatment of the interactions between ions and dipoles to concentration c for a given salt [Eq . (20-49)], we can rewrite the Cohn equation as
developed in 1943 by Kirkwood [Cohn and Edsall (eds .), Proteins, Amino
Acids, and Peptides, Reinhold, New York, 1943, p . 276] accounts for salting-in ln S = β − K s′′c
effects by considering the solute size, solute shape, solute dipole moment,
solvent dielectric constant, solution ionic strength, and temperature . One Substituting the experimental data into this equation gives
of Kirkwood’s models was for a spherical dipolar ion with a point dipole
moment u located at the center of the sphere . The equation derived to ln(13 .5) = β − 0 .5 K s′′
describe the interactions is as follows: ln(5 .0) = β − 1 .0 K s′′
 Sp 
ln   = K i I − K s I (20-48) Solving these equations for the constants yields
 So 
β = 3 .60
where Sp is the solubility of the dipolar ion at ionic strength I, S0 is the solu- K s′′= 1 .99 M −1
bility of the dipolar ion in the absence of salt, Ki is the salting-in constant,
and Ks is the salting-out constant . Ionic strength is defined by For 95 percent recovery, the protein solubility in the mother liquor at equilibrium is 5
percent of the initial protein concentration . At this solubility, from the Cohn equation
1
I = ∑ ci zi2 (20-49)
2 i β − ln S 3 .60 − ln(0 .05 × 15)
c= = = 1 .95 M
K s′′ 1 .99
where ci is the molar concentration of any ion and zi is its charge . The
salting-in and salting-out constants can be related to other variables as Precipitate Formation Phenomena By studying the phenomena of
follows: precipitate formation, we can maximize control over the characteristics of
2 the final protein precipitate . Important characteristics of protein precipi-
 u tates are the particle size distribution, density, and mechanical strength .
Ki ∝  (20-50)
 εT  Normally, it is desired to avoid having a large fraction of particles in the small
size range . For proteins, particle sizes near or below 1 µm are considered
Ve to be small . Protein precipitates that consist largely of particles with small
Ks ∝ (20-51)
εT particle sizes can be difficult to filter or centrifuge . Low particle densities
also can lead to filtration or centrifugation problems and can give excessive
where ε is the dielectric constant of the solvent, T is absolute temperature, bulk volumes of the final dried precipitate . Particles with low mechanical
and Ve is the excluded volume of the dipolar ion . Equations (20-48) and strength can give problems with excessive attrition when the dry particles
(20-50) confirm the observed strong relationship between the solubility of are moved . Low strength can also be interpreted as gel formation, which
dipolar proteins and the size of their dipole moment u, as well as the greater leads to major problems in filtration and centrifugation .
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-61

1.2

0.8

0.4

0 (NH4)2SO4

Iog S/S0 –0.4 Na2SO4

–0.8

–1.2

–1.6

–2.0
0 1 2 3 4 5 6
Ionic strength (M)
FIG. 20-46 The effect of (NH4)2SO4 and Na2SO4 on the solubility of hemoglobin: S0 is the solubility in
pure water, and S is the solubility in the salt solution . [Data from Tanford, Physical Chemistry of Macro-
molecules, Wiley, 1961, p . 244 .]

Precipitates form by a series of steps that occur in sequence, which are up to the maximum level of supersaturation, or supersaturation limit, which
the following: (1) initial mixing, (2) nucleation, (3) growth governed by is illustrated in Fig . 20-47 . Note that the rate of nucleation increases to a very
diffusion, and (4) growth governed by fluid motion . There is often some high value at the supersaturation limit .
overlap between these steps . The final size of the precipitate particles High supersaturations generally have negative consequences in carrying
during step 4 is subject to the limits imposed by particle breakage during out precipitations . When the supersaturation is high, the precipitate tends
mixing . The completion of the growth by fluid motion step can be followed to be in the form of a colloid, a gel, or a highly solvated precipitate . To obtain
by an “aging” step, where the particles are mixed until reaching a stable size . precipitate particles having desirable characteristics, the supersaturation
Initial Mixing Initial mixing is the mixing required to achieve homo- should be kept relatively low .
geneity after the addition of a component to cause precipitation . It is Growth Governed by Diffusion The growth of precipitate particles is
important to bring precipitant and product molecules into collision as limited by diffusion immediately after nucleation and until the particles
soon as possible . This is a problem in micromixing, which is also impor- grow to a limiting particle size defined by the fluid motion, which generally
tant in fermentation . This subject was studied by the Russian statisti- ranges from 0 .1 to 10 mm for high and low shear fields, respectively [Ives,
cian Kolmogoroff in the form of the homogeneous isotropic turbulence in Ives (ed .), Scientific Basis of Flocculation, Sijthoff and Noordhoff, Alphen
model, which assumes that mixing between randomly dispersed eddies aan den Rijn, 1978, p . 37] . In a dispersion of particles of uniform size that
is instantaneous and the mixing within eddies is diffusion-limited . It is are growing as dissolved solute diffuses to the particles, the initial rate of
therefore important to know the mean length of eddies, also known as the decrease of particle number concentration N can be described by a second-
Kolmogoroff length, here designated le . It can be calculated from [Bell et al ., order rate equation that was first derived by Smoluchowski [Z. Phys. Chem .
in Fiechter (ed .), Advances in Biochemical Engineering/Biotechnology, vol . 92: 129 (1917)]:
26, Springer-Verlag, 1983, p . 1] .
1/4 dN
 ρν3  − = KAN 2 (20-59)
le =  (20-56) dt
 P /V 
where ρ is the liquid density, ν is the liquid kinematic viscosity, and P/V is
the agitator power input per unit volume of liquid .
It is necessary to mix until all molecules have diffused across all eddies .
This time can be estimated from the Einstein diffusion relationship
δ2
Rate of nucleation

t= (20-57)
2
where d is the diffusion distance and  is the diffusion coefficient for the
molecule being mixed . For spherical eddies of diameter le, this becomes
l e2
t= (20-58)
8
Thus, precipitation is initiated in a well-stirred tank for a period of time
determined on the basis of isotropic turbulence .
Nucleation Nucleation is the generation of particles of ultramicroscopic A B
size . For particles of a given solute to form, the solution must be supersatu-
rated with respect to the solute . In a supersaturated solution, the concen- Supersaturation
tration of the solute in solution is greater than the normal equilibrium FIG. 20-47 Nucleation rate as a function of degree of supersaturation . The normal
solubility of the solute . The difference between the actual concentration in equilibrium solubility is at A, and the supersaturation limit is at B . [Reproduced from
solution and the equilibrium solubility is called the degree of supersatura- Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott
tion, or just supersaturation . The rate of nucleation increases exponentially R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford University Press .]
20-62 BIOREACTIONS AND BIOPROCESSING

Here N is the number of mono-sized particles at any given time t . The con- Growth Governed by Fluid Motion Growth of particles is governed by
stant KA is determined by the diffusivity  and diameter Lmol of the mol- fluid motion after the particles have reached a critical size, typically 1 µm in
ecules that are adding to the particles as follows: diameter [Bell et al ., in Fiechter (ed .), Advances in Biochemical Engineering/
Biotechnology, vol . 26, Springer-Verlag, 1983, p . 1] . In this growth regime,
K A = 8 π Lmol (20-60) particles tend to grow by colliding and then sticking together . This is a floc-
culation process, which is enhanced when electrostatic repulsion between
Integrating Eq . (20-59) gives particles is reduced in comparison to the attractive van der Waals force .
This can be accomplished by raising the ionic strength and lowering the
1 temperature, to reduce the thickness of the electrical double layer, or Debye
N= (20-61)
K A t + 1/N o length, around particles .
For particles of uniform size in a suspension, the initial rate of decrease
For convenience, No is taken as the initial number concentration of dissolved of particle number concentration N due to collisions can be described by a
solute molecules . The Stokes-Einstein equation can be used to estimate the second-order rate equation
diameter of globular proteins, which can be modeled as spheres
dN 2 3
kT − = αL γ N 2 (20-66)
Lmol = B (20-62) dt 3
3 πµ
which was also derived by Smoluchowski [Z. Phys. Chem . 92: 129 (1917)] .
Here a is the collision effectiveness factor ( fraction of collisions that result
where kB is the Boltzmann constant, T is the absolute temperature, and µ is in permanent aggregates), L is the diameter of the particles, and γ is the
the liquid viscosity . shear rate (velocity gradient) . Assuming that the volume fraction of the
Equation (20-61) can be rewritten as
particles (ϕ = πL3N/6) is constant during particle growth governed by fluid
N 1 motion and integrating yields
= (20-63)
N o 1 + N o K At N
= exp (−4 αφγ t /π) (20-67)
With M as the molecular weight of particles at time t and Mo as the molecu- No
lar weight of the solute,
where N0 is now the particle number concentration at the time [t = 0 in
M No Eq . (20-67)] at which particle growth starts to be governed by fluid motion . For
= (20-64) turbulent flow, the average shear rate γ can be estimated by the following equa-
Mo N tion, developed by Camp and Stein [ J. Boston Soc. Civ. Engrs. 30: 219 (1943)]:
so that 1/2
 P /V 
γ = (20-68)
M = M o (1 + K A N o t ) (20-65)  µ 
This equation has been verified experimentally by measuring the molecular where P/V is power dissipated per unit volume and µ is the viscosity of the
weight of precipitating a-casein . The data plotted in Fig . 20-48 indicate good liquid, respectively . Information about the determination of the collision
agreement with Eq . (20-65) after an initial lag time . effectiveness factor a has been given by Bell [Bell et al ., in Fiechter (ed .),
Advances in Biochemical Engineering/Biotechnology, vol . 26, Springer-Verlag,
1983, p . 1] .
Time (s) [•] Precipitate Breakage When precipitate particles grow large enough by
0 5 10 colliding and sticking together, they become susceptible to breakage dur-
ing collisions . The rate of precipitate breakage has been shown to depend
9 on the shear rate and particle concentration . In a study of soy protein pre-
cipitate particles, for example, particle breakup dominated at sizes greater
than 16 µm, and breakup became negligible at low particle volume fractions
8 (<0 .0002) [Brown and Glatz, Chem. Eng. Sci . 42: 1831 (1987)] .
1.0 0.25 0.5
A model that has successfully described the breakup of protein precipi-
tates is the displacement model, which depicts the rate of aggregate size
7 change as a function of displacement from an equilibrium aggregate diam-
eter Le [Twineham et al ., Chem. Eng. Sci . 39: 509 (1984)]:
Molecular weight × 10–8

6 dL
= k ( Le − L)n (20-69)
dt
5 where the rate constant k would be expected to depend on the volume frac-
tion of particles ϕ and the shear rate γ . This model with n = 1 ( first-order) fits
data well for the mean diameter of soy protein particles at constant shear
4 and various particle volume fractions (Fig . 20-49) . The equilibrium diameter
Le has been shown to depend on the shear rate . For isoelectric soy protein
precipitate in laminar Couette shear flow,
3
L e ,v ∝ γ −0 .14 2000 s −1 ≤ γ ≤ 80,000 s −1 (20-70)
2
and for casein precipitated by salting out in a continuous stirred-tank reac-
tor [Twineham et al ., Chem. Eng. Sci . 39: 509 (1984)],
1
Le ,v ∝ γ −0 .21 12 s −1 ≤ γ ≤ 154 s −1 (20-71)
0
0 1 2 3 where Le ,v is the equilibrium particle size at the volume mean of the particle
size distribution . It is remarkable that both correlations are relatively simi-
Time (s) [o, ] lar, given the varation in protein precipitation type, protein type, and shear
FIG. 20-48 Molecular weight–time plots for three concentrations of a3-casein (con-
rate range . Particle breakage is also discussed later in the analysis of the
centrations as indicated on graph in kilograms per cubic meter), aggregating in the particle size distribution in a continuous flow stirred-tank reactor .
presence of 0 .008 M CaCl2 . Molecular weight was determined from light-scattering Precipitate Aging As indicated in Fig . 20-49, protein precipitate parti-
and turbidity measurements . (Data from Bell et al ., Adv. Biochem. Eng . 26: 20, Springer- cles reach a stable size after a certain length of time in a shear field . The time
Verlag, 1983, p . 20 .) period for reaching this stable size is called the aging time . The strength of
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-63

18 protein particles has been correlated with the product of the mean shear
Symbol φ rate and the aging time, γt, which is known as the Camp number. As indi-
cated in Fig. 20-50 for soy protein particles, the mean particle size becomes
17 0.00214 approximately constant after reaching a Camp number of 105. Aging of pre-
0.00329 cipitates helps the particles withstand processing in pumps and centrifuge
0.00675 feed zones without further size reduction.
0.01220 Particle Size Distribution in a Continuous Flow Stirred-Tank
16
Reactor The equations describing precipitate formation and breakage in
the preceding subsection are convenient to use when the size of the pre-
cipitate is changing with time. In addition to being applicable to the batch
15 reactor, these equations can be used to describe precipitate growth in the
tubular reactor (see the Design of Precipitation Systems subsection). How-
ever, for one important type of reactor—the continuous flow stirred-tank
Mean diameter (µm)

14 reactor (CSTR)—operation is at steady state. Because precipitated protein

leaves the CSTR at the same time as the feed protein solution enters, the
product stream leaving the CSTR will have a range of particle sizes, which
13 does not vary with time for operation at steady state. The distribution of
particle sizes can be characterized by performing a population balance for
precipitate particles in the CSTR (Randolph and Larson, Theory of Particu-
12 late Processes, Academic Press, Cambridge, Mass., 1971, p. 49). Before writ-
ing the population balance, it is first necessary to define the population
density distribution function n(L) and the linear growth rate G:
11
dN
n ( L) = (20-72)
dL
10
dL
G= (20-73)
dt
9
Thus, if we know n(L), we can take the integral of n(L)dL to find the total
number of particles bracketed by two given particle diameters, a minimum
and a maximum, for example.
8
0 30 60 90 120 150 180 210 240 270 300 Making a balance on the number ∆N of precipitate particles within a
given size range ∆L and taking the limit as ∆L → 0, we obtain for constant
Time (min)
reactor volume V
FIG. 20-49 Volume mean aggregate diameter as a function of time for soy precipi-
tate particles exposed to shear rate of 1340 s−1 at different particle volume fractions ϕ. d (nG ) n
Lines are drawn for the displacement model. Points are experimental data. [Data from + +B=0 (20-74)
Brown and Glatz, Chem. Eng. Sci. 42: 1831 (1987).] dL τ

1.0

0.9
Ratio of final to initial mean diameters, Lv /Lo,v
——

0.8

0.7

0.6

0.5

0.4

0.3
104 2 3 4 5 6 7 8 9 105 2 3 4 5 6 7 8
Aging parameter, γ–t
FIG. 20-50 The effect of aging on the change in the volume mean diameter of soy protein precipitate
particles exposed to capillary shear. Average capillary rate of shear, 1.7 × 104 s−1; average time of exposure
in capillary, 0.065 s; protein concentration, 30 kg m−3; initial mean diameter L0,v (µm) prior to exposure
to capillary shear: ○, 53.5; ∆, 23.4; ◊, 19 .5; ⊗, 15 .2; ☐, 10 .2; •, 8 .8 . [Data from Bell and Dunnill, Biotechnol.
Bioeng. 24: 1271 (1982) .]
20-64 BIOREACTIONS AND BIOPROCESSING

Example 20-7 Dependence of Population Density on Particle Size and

Primary particle Residence Time in a CSTR For the precipitation of protein in a CSTR, determine
the dependence of the population density distribution n on the particle diameter L and
residence time τ . Consider both particle growth by aggregation and particle breakup .
Use the expressions for the linear growth rate and particle breakup [Eqs . (20-75) and
(20-76)] in the population balance equation [Eq . (20-74)] . This gives

L ≥ L0 d ( K o Ln) n
+ + K 1nL3 = 0
dL τ

Aggregate Differentiating and rearranging, we have

 1  
1+ 
dn   K o τ  K 1 L2 
= − dL +
n  L Ko 
 
 

FIG. 20-51 Schematic drawing of the growth of a precipitate aggregate by collisions of With lower limits of integration of n0 (the population density function at L = L0) and
the aggregate with small primary particles: L0 is the critical diameter above which par- critical diameter L0, we obtain
ticles grow by colliding with and sticking to small primary particles. [Reproduced from
 1 
Bioseparations Science and Engineering, 2d ed., by Roger G. Harrison, Paul W. Todd, Scott n  Lo  
1+
K o τ   −K 
R. Rudge, and Demetri P. Petrides (2015). By Permission of Oxford University Press.] =  exp  1 ( L3 − L3o ) 
no  L   3Ko 

where τ is the mean residence time (= V/Q) and B is the volumetric breakage Methods of Precipitation Methods that have been developed to
or death rate of particles in the size range L to ∆L, such that the net rate of precipitate proteins are based on a knowledge of the solubility of proteins .
disappearance of particles due to breakage per volume is B ∆L. Based on the previous discussion of protein solubility, the most obvious
The growth of protein precipitate particles in a CSTR has been success- methods that emerge are pH adjustment to the isoelectric point of the pro-
fully described by Petenate and Glatz [Biotechnol. Bioeng . 25: 3059 (1983)], tein (called isoelectric precipitation), addition of organic solvents, salting
who used the following expression for the linear growth rate: out, and addition of nonionic polymers .
Isoelectric precipitation is based on the fact that the solubility of a given
protein is generally at a minimum at the isoelectric point (pI) of the pro-
dL  A 
G= =   γ φ1 L = K o L (20-75) tein, which is the pH at zero charge (see Fig . 20-45) . This is a convenient
dt  4 π  method to use when fractionating a protein mixture . For this situation the
pH should be adjusted above the highest pI or below the lowest pI of all
where A is a constant and f1 is the volume fraction of submicrometer-sized the proteins present . The pH is then changed to the nearest pI, where pre-
“primary” particles . In this model, aggregates above a critical diameter L0 cipitate is allowed to form and is then removed . This procedure is repeated
grow because small primary particles collide with aggregates and stick (see for as many proteins as one desires to precipitate . Local extremes of pH
Fig . 20-51) . Primary particles typically have diameters of about 0.2 µm for should be avoided during pH adjustment to minimize protein denaturation
protein precipitation in a CSTR [Petenate and Glatz, Biotechnol. Bioeng . 25: (Harrison et al ., Bioseparations Science and Engineering, Oxford University
3059 (1983)] . Press, Oxford, UK, 2015, p . 32) . There are two advantages of isoelectric pre-
The larger a precipitate aggregate becomes, the more susceptible it is to cipitation when acids are added to cause precipitation: mineral acids are
being broken up by the local shear stresses encountered in turbulent flow . cheap, and several acids (e .g ., phosphoric, hydrochloric, sulfuric) are accept-
For the conditions under which local shear stresses are dominant, the fol- able in protein food products . This method, however, will not work for all
lowing equation for the volumetric breakage or death rate B has been suc- proteins; for example, gelatin, which is a very hydrophilic protein, does not
cessfully used by Petenate and Glatz to describe the breakup of protein precipitate at its isoelectric point in solvents having low ionic strength [Bell
aggregates: et al ., in Fiechter (ed .), Advances in Biochemical Engineering/Biotechnology,
vol . 26, Springer-Verlag, 1983, p . 1] .
B = K µγ 2nL3 = K1nL3 (20-76) Several organic solvents have been used to precipitate proteins, including
alcohols, acetone, and ether . Alcohols, however, have been the most widely
where K is a constant and µ is the viscosity . used in industry . One of the most important processes utilizing alcohol to
A useful parameter to know for the precipitation in a CSTR is the mass- precipitate proteins is the Cohn process to purify therapeutic proteins from
averaged particle size . For spherical particles, the total mass concentration human plasma [Strong, in Kirk and Othmer (eds .), Encyclopedia of Chemi-
MT in terms of the mass-averaged diameter Lm is cal Technology, Interactive Encyclopedia System, 1948, p . 566] . This process
uses ethanol at temperatures below 0°C to minimize denaturation by the
organic solvent . The variables that are manipulated in the Cohn process are
 L3  pH, ionic strength, and ethanol concentration . Ionic strength is kept low,
MT = ρ p π  m  N (20-77)
 6  which leads to a salting-in effect (see the Protein Solubility subsection) . This
salting-in effect is enhanced when ethanol is added . Cohn’s method has
where rp is the density of individual particles and N is the total particle num- been used to obtain albumin, plasminogen, prothrombin, isoagglutinins,
ber concentration . From n(L), we can determine MT and N for spherical par- and γ-globulin starting with blood plasma .
ticles equal to or greater than Lo in size as follows: In the salting out of proteins, salt is dissolved in the solution containing
the proteins . The protein solubility decreases as the salt ionic strength rises
ρpπ ∞
according to the Cohn equation [Eq . (20-52)] . The most important consid-
MT L ≥ Lo =
6 ∫ Lo
L3n ( L) dL (20-78) eration in salting out is the type of salt used . Salts with multiply charged
anions such as sulfate, phosphate, and citrate are the most effective; for
the cation, monovalent ions should be used (Scopes, Protein Purification,
∞
N = ∫ n ( L) dL (20-79) Springer-Verlag, 1982, p . 47) . Following the Hofmeister or lyotropic series,
L ≥ Lo
Lo the salting-out ability of the common multiply charged anions is citrate2− >
phosphate3− > sulfate2−; for the common monovalent cations, the order is
Combining Eqs . (20-77), (20-78), and (20-79) gives the mass-averaged NH4+ > K+ > Na+ .
particle diameter for L ≥ L0: The salt that has the most desirable properties in general for precipitating
1/3 proteins is ammonium sulfate . The solubility of this salt is very high (approx-
 ∞ L3n ( L)dL 
_
Lm
∫L
=  o∞  (20-80)
imately 4 M in pure water) and varies very little in the range of 0 to 30°C . The
density of a saturated solution of ammonium sulfate is 1 .235 g cm−3, which is
L ≥ Lo  
 ∫Lo n ( L ) dL  enough below the density of protein aggregates (approximately 1.29 g cm−3)
to allow centrifugation. Another advantage of using ammonium sulfate is
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-65

that protein precipitates are often very stable for years in 2 to 3 M salt; in
Mean velocity gradient (s–1)
fact, many commercial enzymes are normally sold in ammonium sulfate
solution at high molarity . Furthermore, proteolysis and bacterial action are 100 200 300 400 500
prevented in concentrated ammonium sulfate solutions . The only disadvan-
tage of ammonium sulfate is that it cannot be used above pH 8 because of
the buffering action of ammonia . Sodium citrate is very soluble and is a good
alternative to ammonium sulfate when the precipitation must be performed 20
above pH 8 (Scopes, Protein Purification, Springer-Verlag, 1982, p . 48) .
Several nonionic polymers have been used to precipitate proteins,
including dextran, poly(vinyl pyrrolidone), poly(propylene glycol), and

Particle size, d90 (µm)

poly(ethylene glycol) (PEG) . Of these polymers, by far the most extensively
15
studied is PEG . Several guidelines have been developed for the use of PEG
as a protein precipitant . Solutions of PEG up to 20 percent w/v can be used
without viscosity becoming a problem . PEGs with molecular weights above
4000 have been found to be the most effective (Scopes, Protein Purification,
Springer-Verlag, 1982, p . 59) . Protein destabilization in PEG solutions does 10
not occur until the temperature is significantly higher than room tempera-
ture (>40°C) [Rothstein, in Harrison (ed .), Protein Purification Process Engi-
neering, Dekker, New York, 1994, p . 115] .
Design of Precipitation Systems Once a precipitating agent has
been chosen for the protein of interest, it may be necessary to design a 5
large-scale precipitation process . The safest procedure is to base the design
on a laboratory or pilot plant system that has given acceptable results .
Important considerations in obtaining the best possible plant design are
the following: the type of precipitation reactor, processing conditions ( flow
rates, concentrations, etc .), and assumptions used to scale up to the plant 0
scale . There are three basic types of precipitation reactor: the batch reactor, 0 100 200 300 400 500 600
the continuous stirred-tank reactor (CSTR), and the tubular reactor . These Power per unit volume (W m–3)
types are discussed and compared .
The batch reactor is the simplest of the three types and is often the one FIG. 20-52 The relationship between aggregate size and power per unit volume
for the preparation of isoelectric soy protein precipitate in batch-stirred tanks: d90 is
that is tried first at small scale . Batch precipitation is carried out by slowly the particle size for which all particles that are larger account for 90 wt% of the total .
adding the precipitating agent to a protein solution that is being mixed . Protein concentration is 30 kg m−3. Vessel volumes: △, 0.27 L; ○, 0.67 L; □, 200 L. [Data
Addition of the precipitating agent continues until the desired level of super- from Bell et al., Adv. Biochem. Eng. 26: 40, Springer-Verlag, 1983, p. 40.]
saturation is reached with respect to the protein being precipitated . At this
point nucleation begins, and precipitation proceeds through the steps of
particle growth and aggregation . Mixing continues until the precipitation is precipitant is especially important . This rate should be kept low enough to
complete . The mixing in a batch reactor is generally turbulent . Protein par- avoid high supersaturations that lead to colloidal, highly solvated precipi-
ticles precipitated in a batch reactor tend to be more compact and regular in tates . The concentration of the precipitant being added is also important,
shape than those precipitated in a tubular reactor, apparently because of the with lower concentrations leading to lower supersaturations [Rothstein, in
different shear profiles existing in the two reactors and the length of time the Harrison (ed .), Protein Purification Process Engineering, Dekker, New York,
particles are exposed to this shear [Bell and Dunnill, Biotechnol. Bioeng. 24: 1994, p . 115] .
2319 (1982)] . The shear field in a tubular reactor is essentially homogeneous; The key parameter for scale-up of precipitation is mixing . One recom-
by contrast, in the batch reactor the precipitate particles are exposed to a mended approach to scale up mixing is to first consider using geometric
very wide range of shears and to much longer times of exposure than in the similarity and constant power per unit volume (P/V) (Oldshue, Fluid Mixing
tubular reactor, resulting in improved precipitate mechanical stability . Technology, McGraw-Hill, New York, 1983, p . 198) . For geometric similarity,
In a tubular reactor, precipitation takes place in volume elements that all important dimensions are similar and have a common constant ratio . As
approach plug flow as they move through the tube . Thus, the distance– seen in Fig . 20-52, this method gave reasonable agreement for the d90 par-
particle size distribution history of the particles in a volume element ticle size (the particle size for which all particles that are larger account for
moving through a tubular reactor is comparable to the time–particle size 90 wt% of the total; d90 is also referred to as D10 in describing particle size
distribution history of a stationary volume element in a batch reactor . In distributions) as a function of P/V for the batch isoelectric precipitation of
the tubular reactor, the feed protein solution and the precipitating agent soy protein for vessels ranging in size from 0 .27 to 200 L . If the precipitate is
are contacted in a zone of efficient mixing at the reactor inlet . Good mixing susceptible to shear breakage, however, the assumption of constant P/V for
can be accomplished, for example, by flowing the protein solution through scale-up may not be satisfactory . The impeller tip speed, which determines
a convergent nozzle to a biplanar grid and then introducing the precipitat- the maximum shear rate, rises when P/V is held constant upon scale-up of
ing agent just downstream of the biplanar grid [Virkar et al ., Biotechnol. the reactor volume, as seen in Table 20-24 . These results assume turbulent
Bioeng. 24: 871 (1982)] . The flow pattern in the reactor can be turbulent, flow, where the power number is constant, so that
a property that can be promoted by wire meshes at intervals along the
reactor . In comparison to either the batch reactor or the CSTR, the tubular P ∝ N i3 di5 (20-81)
reactor has the advantages of short fluid residence times, an absence of
moving mechanical parts, uniformity of flow conditions throughout the Here, Ni is the impeller rotation rate, and di is the impeller diameter . It is
reactor, a simple and inexpensive design, and a relatively small holdup of also assumed that the maximum shear rate is proportional to the impel-
fluid . For particles that grow relatively slowly, however, the length of the ler tip velocity (= pNi di ) and that the volume is proportional to di3 . From
tubular reactor can be excessive . Table 20-24 we see that even for a volume scale-up of 10,000 for constant
In the CSTR, fresh protein feed contacts a mixed slurry containing P/V, the tip speed increases by less than a factor of 3 . A volumetric scale-up
precipitate aggregates . The mixing conditions in a CSTR are similar to those factor of 10,000 is typical in scaling up from the lab to the plant .
in a batch reactor . Upon entering the CSTR, fresh protein feed nucleates, the
nucleate particles grow by diffusion, and the submicrometer-size “primary”
particles collide with and adhere to growing aggregates . The degree of TABLE 20-24 Scale-Up of Turbulent Agitation, Assuming Constant P/V
supersaturation can be more easily controlled than in the batch or tubular (Tip speed)large
reactor, which means that the formation of precipitates with undesirable Volume scale-up factor (Tip speed)small
properties is less likely .
A few general statements can be made regarding the processing condi- 10 1.3
tions in precipitation systems . Flows are normally turbulent; flow must 100 1.7
be high enough to avoid inadequate mixing and high supersaturation but 1000 2.2
low enough to avoid excessive particle breakage, leading to particles that 10,000 2.8
are smaller than desirable . For both the batch and tubular reactors, the Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G.
flow regime can be changed from turbulent to laminar during the particle Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission
growth phase to avoid excessive particle breakage . The rate of addition of of Oxford University Press.
20-66 BIOREACTIONS AND BIOPROCESSING

LIQUID CHROMATOGRAPHY AND ADSORPTION Ions do not partition well in hydrophobic phases . It is common, therefore,
to choose a counterion for reversed-phase chromatography . For biological
As discussed in the introduction, adsorption is typically used for the removal mixtures, the counterion is nearly always a strong acid anion, such as tri-
of insolubles such as cells or cell debris as well as for the initial isolation of fluoroacetate, acetate, or chloride . The counterion has a strong effect on the
the product, while chromatography is often used for the final purification separation, by partitioning along with the co-ion of interest . The counterion
steps . Different types of chromatography are used to achieve the high purity can be used to make solutes more or less hydrophobic and will not affect all
required, as is the case for bioproducts such as monoclonal antibodies (see solutes equally .
the subsection Bioprocess Design and Economics for a discussion of the Hydrophobic Interaction Chromatography Hydrophobic interaction
purification of monoclonal antibodies) . Detailed information about the the- chromatography is often used for protein separations . It employs deriva-
ory and equipment for liquid chromatography and adsorption can be found tized polymer resins, with phenyl, butyl, or octyl ligand groups, typically .
in Sec . 16 . In the current section, chromatography and adsorption topics Proteins adhere to the hydrophobic surface under high-salt conditions
are discussed that have particular relevance to bioproducts: (1) adsorbents and redissolve into the mobile phase as the salt concentration is reduced .
that are commonly used to purify biomolecules; (2) the length of unused Hydrophobic interaction chromatography differs from reversed-phase in
bed (LUB) method for scaling up adsorption of biomolecules that obey the that the mobile phase is kept aqueous (polar), and the salt concentration is
Langmuir isotherm, which has often been used to correlate equilibrium used to effect the partitioning to the surface . The mechanism of partition
adsorption data for proteins; (3) agitated-bed adsorption for processing is related to precipitation as opposed to two-phase partitioning (reversed-
cell suspensions in which it is desired to selectively remove a bioproduct phase) or ionic interaction (ion exchange) . This is due to the use of salts that
that has been secreted by the cells; (4) membrane chromatography, which strip proteins of their solvation water . As the concentration of “lyophilic”
has applications for macromolecules such as proteins and DNA; and (5) the salts increases, the probability increases that the protein will aggregate, or
scale-up of the chromatography of proteins . nucleate on the surface of the resin [Roettger et al ., Biotechnol. Prog . 5: 79
Adsorbents for Purification of Bioproducts Many adsorbent res- (1989)] . Hydrophobic interaction chromatography is sensitive to pH, salt
ins have been developed for adsorptive and chromatographic separation used, buffer type, and temperature . Each of these must be carefully con-
of bioproducts . There are two basic resin materials: polymer and silica . trolled to achieve reproducible separation but also to represent an oppor-
Many types of ligand chemistries can be conjugated to either resin material . tunity for increased selectivity .
Typically, however, silica resins most commonly have hydrophobic coatings Affinity Chromatography and Adsorption Affinity chromatography
and are used for reversed-phase chromatography . Polymer resins are more and adsorption take advantage of biological interactions to effect binding
often used in aqueous applications and are conjugated with ion exchange, of specific solutes . Antibodies, antigens, or dyes are conjugated to poly-
hydrophobic interaction, or affinity-type ligands . More detailed information mer resins for the purpose of binding specific solutes from a mixture . For
about polymer and silica adsorbents can be found in the subsection Adsor- instance, an antibody could be raised against a target protein . The antibody
bents and Ion Exchangers in Sec . 16 . would be conjugated to a resin, usually via cyanogen bromide activation .
Surface area and particle size are very important properties of adsor- The antibody then captures the solute out of the mixture, and the impurities
bent resins . The resin provides the surface area for the adsorption, which flow through the column . The solute can be recovered by changing the pH,
is generally 100 to 1500 m2/g . The surface area on the outer surface of a increasing the salt concentration, or adding a displacer, that is, a molecule
10-µm-diameter solid sphere is 1 .7 m2/g, so it follows that most of the sur- that also has some affinity for the antibody, or other binding agent (affinity
face area is in the internal porosity of the particle . Since this surface area ligand) on the stationary phase .
is accessed by molecular diffusion (one class of resins, called perfusion res- Other examples of specific interactions that can be used to isolate pro-
ins, allows convection through some macropores), the path length for this
teins are enzyme–ligand, enzyme–cofactor, and receptor–agonist (antago-
diffusion is important . This path length is the radius of the resin particle .
nist) . In each case, one member of any of these pairs may be immobilized
Therefore, both the diameter of the particle and its internal surface area are
to a resin to isolate the desired partner . Affinity chromatography is often
important for the resin performance . Resins have large ranges of surface
coupled with cloning techniques to synthesize the target molecule with
areas and particle sizes, and they do not necessarily co-vary .
an “epitope” or recognition sequence that can be captured by the affinity
The various adsorbents used for adsorption and chromatography of
biomolecules are discussed below based on the type of operation that is ligand . Affinity chromatography is used from small-scale research (e .g .,
carried out . high-throughput screening) to large-scale purification .
Ion Exchange Adsorption and Chromatography A comprehensive dis- An example of a type of affinity chromatography that has had a major
cussion of ion exchange adsorbents is given in the subsection Adsorbents impact on the purification of monoclonal antibodies is the use of immobi-
and Ion Exchangers in Sec . 16 . lized protein A in process scale purification [Vunnum, in U . Gottschalk (ed .),
Reversed-Phase Chromatography Reversed-phase chromatography Process Scale Purification of Antibodies, Wiley, Hoboken, N .J ., 2009, p . 79] .
employs a hydrophobic phase bonded to the surface of the resin . Typically, Protein A is often used as the first chromatography step in the purification,
reversed-phase resins are silica based . Reversed phase is so named because which gives a large purification factor and results in simplifying the purifi-
the partitioning of solutes between the mobile phase and stationary phase cation process . Protein A has affinity for the constant region of Ig-G anti-
is opposite to that observed with bare silica . In other words, hydrophobic bodies, so it is generally applicable to the capture and purification of most
solutes bind in higher proportions in reversed phase, while hydrophilic monoclonal antibodies . It is a polypeptide with a molecular mass of 54 kDa
solutes bind in higher proportion in “normal phase .” Solutes are typically that is found in the cell wall of Staphylococcus aureus. Recombinant protein
introduced into reversed-phase columns in water, or with minimal amounts A that lacks the membrane-binding region is used in affinity purification .
of organic solvent, so that most solutes partition to the stationary phase . One of the drawbacks of the protein A and other affinity methods is that
The organic content of the mobile phase is slowly increased, typically as a low levels of the affinity ligand or its fragments can co-elute with the target
percent of acetonitrile, methanol, or isopropanol, thereby decreasing the molecule . In particular, when an affinity step is used as the first step in the
polarity of the mobile phase . purification process, it is exposed to significant levels of extracellular prote-
Hydrophobic phases that are bonded to silica are typically aliphatic octyl ases that liberate fragments of the ligand into the column effluent . However,
(C8), octyldecyl (C18), phenyl (C6 aromatic), and methyl (C1) . The different trace amounts of affinity ligands and others such as high-molecular-weight
chain lengths and densities (called coverage in the commercial literature) of aggregates of the antibody and host cell protein can be removed in one or
the different bonded phases obviously lead to more or less hydrophobicity . two subsequent chromatography steps . More information about the process
The entire surface of the silica cannot be fully covered with a monolayer scale purification of monoclonal antibodies using protein A is found in the
of the desired phase, however, because of steric effects . Bare silica remains Bioprocess Design and Economics subsection .
exposed, and this bare silica can participate in the separation by interact- Immobilized Metal Affinity Chromatography Some proteins have
ing with hydrophilic molecules, or hydrophilic domains of large molecules, high affinities for specific metals . This affinity may be either structural, as
thereby altering the binding . The strategy employed for covering this in the case of metalloproteins, which require metal centers for their biologi-
exposed silica surface depends almost as much on the specifics of the sepa- cal activity, or based on the content of specific amino acid residues, such as
ration achieved as it does on the chain length of the bonded phase . Polymer- histidine, tryptophan, and cysteine, which have increased affinity for transi-
ized phases represent an attempt to cover the surface by polymerizing the tion metals such as nickel and copper . Techniques have been developed to
alkyl chains together at their point of attachment to the silica . End-capped immobilize metal ions with spacer arms onto polymer resins . These resins
resins use a short chain length group, such as methyl or ethyl, to cover the are referred to as IMAC resins, and they are used to purify proteins that have
unreacted surface sites . Resins that do not utilize either method are left so one of the two characteristics mentioned above . Genetic engineering has
intentionally to take advantage of the “mixed-mode” separation that may also been used to enable IMAC to be performed, with cloned target biomole-
result . The separation characteristics of these resins are difficult to repro- cules being fused to polyhistidine tags [Arnold, Bio/Technology 9: 152 (1991)] .
duce precisely, leading to considerable variation in separation between dif- Size Exclusion Chromatography Sometimes referred to as gel filtration,
ferent manufacturers’ resins, and within one manufacturer’s resin from one size exclusion chromatography (SEC) separates solutes on the basis of their
lot of material to another . size . There is no derivatization of the polymer gel, and there is no binding
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-67

between the solutes and the resin . Molecules larger than the largest pores to the train of mixed columns. In the use of this model for the recovery of
in the gel (the exclusion limit) cannot enter the gel and are eluted first . an antibiotic, equilibrium was modeled by the Freundlich isotherm written
Smaller molecules enter the gel to varying extents, depending on their size in the form
and shape, and thus are retarded on their passage through the bed . In gen-
eral, SEC resins are hydrophilic polymer gels with a broad distribution of ci∗,n = bqia,n (20-85)
pore sizes . The pore size is dependent on the degree of polymerization of
the gel . Size exclusion chromatography is a useful technique, especially for where a and b are constants.
changing buffers or for removing small molecules from protein solutions . Equations (20-82) to (20-85) constitute a set of mathematical relation-
Because of the lack of binding between the solute and the resin, however, ships that govern the performance of each column in the train. These simul-
the capacity of this technique is low . Size exclusion effects may be in action taneous equations can be solved by the Runge-Kutta numerical method to
in all the above-described techniques, since all resins are macroporous . predict the effluent and adsorbent concentrations as a function of time.
Agitated-Bed Adsorption Agitated-bed adsorption processes have Excellent agreement between predicted and experimental adsorption data
been developed to allow removal of a product secreted by the cells without for the recovery of the antibiotic novobiocin in a three-stage train has been
first having to remove the cells . In this type of process, cell culture broth obtained using this method (Belter et al., Bioseparations, Wiley, Hoboken,
is passed through a series of agitated columns containing an adsorbent, as N.J., 1988, p. 63).
shown in Fig . 20-53 . Each column has screens at the inlet and outlet that are The scale-up of a series of agitated columns containing adsorbent
designed to retain the adsorbent within the column but allow the broth to (Fig. 20-53), operated in the periodic countercurrent mode, follows directly
pass through . When the concentration of the product in the effluent of the from the mathematical model presented for this process [Eqs. (20-82)
last column in the series reaches a certain value, the flow is stopped, and the to (20-85)] (Belter et al., Bioseparations, Wiley, Hoboken, N.J., 1988, p. 63).
lead column is taken out of the train . Periodic countercurrent operation is Besides the increases in the flow rate and volume that occur upon scale-up,
obtained by advancing each of the remaining columns in the train, placing the mixing patterns may also change (Harrison et al., Bioseparations Science
a regenerated column of adsorbent in the last position, and restarting the and Engineering, Oxford University Press, Oxford, UK, 2015, p. 305).
feed flow . The lead column taken out of the train is washed with the adsor- Length of Unused Bed (LUB) Method for Scale-Up of Fixed-Bed
bent agitated to remove the broth solids, and the product is eluted from the Adsorption This method allows scale-up of fixed-bed adsorption based on
adsorbent, usually in the fixed-bed mode . data from laboratory columns, keeping the particle size and superficial veloc-
This process has advantages over filtration, in that there is no filter aid to ity constant (Thomas and Crittenden, Adsorption Technology and Design,
dispose of and it is not necessary to wash a filter cake containing the cells, Butterworth-Heinemann, Waltham, Mass., 1998, p. 165). In discussing the
so losses of the product are often less . The equipment for this process is less LUB method of scale-up, it is necessary to define the break-point time tb and
expensive and easier to maintain than that used for centrifugation . The dis- the ideal adsorption time t* on a breakthrough curve, which are indicated in
advantage is that expensive solid adsorbents are more easily fouled by the Fig. 20-54. The break-point time is usually taken at the relative concentration
dirtier feed stream and so require harsher or more expensive regeneration ci/ci0 = 0.05 or 0.10, where ci0 is the feed concentration (McCabe et al., Unit
procedures and more frequent replacement compared to adsorbents uti- Operations of Chemical Engineering, McGraw-Hill, New York, 1993, p. 819).
lized with streams with fewer impurities . Also, resin attrition can be an issue . Since only the fluid last exiting the column has this concentration, the aver-
A useful mathematical model for this process has been developed [Belter age fraction of the solute removed from the start of feeding to the break-point
et al ., Biotechnol. Bioeng. 15: 533 (1973)] . The continuity equation for the nth time is usually 0.99 or higher. The ideal adsorption time is the time for break-
column in the train can be written for separand i as through that would occur if the solute were in perfect equilibrium with the
bed of adsorbent, which would give a vertical breakthrough curve. For a sym-
Rate of separand in − rate of separand out = rate of accumulation of separand metrical breakthrough curve, the ideal adsorption time is the time at which
ci/ci0 = 0.5. At the ideal adsorption time for a bed initially free of the solute to
dci ,n dq
Qci ,n −1 − Qci ,n = VL + V R i ,n (20-82) be adsorbed, based on a unit area of bed cross section,
dt dt
vci 0t ∗ = Lρb qi ,sat (20-86)
where Q = volumetric flow rate
ci,n−1, ci,n = separand concentration in feed to and effluent from column n, where L = bed length
respectively qi,sat = average adsorbent phase concentration of solute i in equilib-
VL = liquid volume in column rium with feed concentration ci0, based on the adsorbent weight
VR = volume of adsorbent in column (weight of solute i per weight of adsorbent)
qi,n = separand concentration in adsorbent phase of column n aver- v = superficial, or linear, velocity ( flow rate divided by the cross-
aged over an adsorbent particle sectional area)
t = time ci0 = concentration of solute i in feed
ρb = bulk density of adsorbent
The rate of mass transfer of separand to the adsorbent phase is described by
a linear driving force expression The ideal adsorption time is therefore given by
dqi ,n Lρb qi ,sat
= K a (ci ,n − ci∗,n ) (20-83) t∗ = (20-87)
dt vci 0
where ci∗,n is the separand concentration in the bulk liquid when it is at equi- The amount of the solute adsorbed at the break point can be determined by
librium with qi,n, and Ka is an overall mass-transfer coefficient that can be integrating the breakthrough curve up to time tb, as indicated in Fig. 20-55.
correlated to experimental data as The width of the breakthrough curve defines the width of the mass-transfer
zone in the bed.
K a = Ae − B ( qi /qi ,sat ) + De − E ( qi /qi ,sat ) (20-84)

Here A, B, D, and E are constants, and qi,sat is the adsorbent phase concentra- 1
tion which is in equilibrium with the separand concentration ci0 in the feed
ci /ci0

0.5
Q
ci,3 0
1 2 3 tb t*
Q Q Q Time t
VL VL VL
ci,0 ci,1 ci,2 FIG. 20-54 Breakthrough curve for a fixed-bed adsorber, showing the break-point
time tb, chosen at 5 percent of the solute feed concentration, and the ideal adsorption
FIG. 20-53 Train of agitated-bed adsorption columns for the processing of cell time t*. The ratio of mobile phase concentration of solute to solute concentration in
culture broth. [Reproduced from Bioseparations Science and Engineering, 2d ed., feed to the adsorber is ci/ci0. [Reproduced from Bioseparations Science and Engineering,
by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). 2d ed., by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides
By Permission of Oxford University Press.] (2015). By Permission of Oxford University Press.]
20-68 BIOREACTIONS AND BIOPROCESSING

1 1.0

ci /ci0

0
tb
Time t
FIG. 20-55 Integration of the breakthrough curve for a fixed-bed adsorber. The area

ci/ci0
of integration to the left of a vertical line at time t is proportional to the amount of 0.5
solute adsorbed up to that time. The ratio of mobile phase concentration of solute to
solute concentration in feed to the adsorber is ci/ci0. [Reproduced from Bioseparations
Science and Engineering, 2d ed., by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and
Demetri P. Petrides (2015). By Permission of Oxford University Press.]

For adsorption where the equilibrium isotherm is favorable, which is true

for the Langmuir isotherm that can often be used for protein adsorption,
the concentration profile in the mass-transfer zone takes on a character-
istic shape that does not change as the zone propagates through the bed
[Ruthven, in Kroschwitz (ed .), Kirk-Othmer Encyclopedia of Chemical Tech-
nology, vol . 1, 4th ed ., Wiley-Interscience, Hoboken, N .J ., 1991, p . 493] . At the 0
break-point time, the adsorbent between the inlet of the bed and the begin- 0 10 20 30 40 50
ning of the mass-transfer zone is completely saturated (in equilibrium with t (h)
the solute in the feed) . The adsorbent in the mass-transfer zone goes from
being completely saturated to being almost free of solute, and the adsorbent FIG. 20-56 Breakthrough curve for the fixed-bed adsorption of a pharmaceutical
product in Example 20-8. [Reproduced from Bioseparations Science and Engineering, 2d
could be assumed to be on the average about half-saturated . This would be ed., by Roger G. Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015).
equivalent to half the adsorbent in the mass-transfer zone being saturated By Permission of Oxford University Press.]
and the other half being unused . The scale-up principle is that the length of
the unused bed in the mass-transfer zone does not change as the bed length
is changed [Collins, AIChE Symp. 74(63): 31 (1967)] . We use the LUB method, which involves graphical integrations of the breakthrough
The length of unused bed (LUB) can be determined directly from the curve to determine the amount of solute adsorbed at various times. The breakthrough
breakthrough curve obtained experimentally . If qi,b is the average adsorbent curve with concentration in dimensionless form is therefore plotted, with the curve
phase concentration of solute i at the break-point time, then the fraction extended to ci /ci0 = 1.0, assuming that the curve is symmetric about ci /ci0 = 0.5 (Fig. 20-56).
of bed capacity utilized at the break-point time is qi,b/qi,sat . Therefore, the By doing graphical integrations of this curve, we find the following:
unused fraction of the bed is 1 − qi,b/qi,sat. From Eq. (20-86),
Total solute adsorbed at saturation (ci/ci0 = 1.00)
qi ,b t b 44 .0
= (20-88)  ci 
qi ,sat t ∗ = Qci 0 ∫  1 − c
0 i0 
dt = Qci 0 (31 .5 h)

so that LUB can be written as Total solute adsorbed at break point (ci/ci0 = 0.05)
21 .5
 ci 
 qi ,t 
LUB =  1 − b  L =  1 − b∗  L
t
(20-89)
= Qci 0 ∫  1 − c  dt = Qci 0 (21 .2 h)
i0 
 qi ,sat   t  0

where Q is the volumetric flow rate. Thus, t∗ and tb are 31.5 and 21.2 h, respectively, for
where tb and t∗ are stoichiometric times determined by integration of the this column. From Eq. (20-89)
breakthrough curve:
LUB Qc (21 .2 h)
∞ = 1 − i0 = 1 − 0 .673 = 0 .327
 c  L Qci 0 (31 .5 h)
t = ∫  1 − i  dt
∗
(20-90)
LUB = 0 .327(15 cm) = 4 .90 cm
0  c i0 

t
b
 c  The adsorbent loading at saturation per total bed volume is
tb = ∫  1 − i  dt (20-91)
0
ci 0  cm 3 U
400 × 0 .00075 3 × 31 .5 h
h cm U
In scale-up calculations, the length of column required can easily be ρb qi ,sat = 2 2 = 0 .0321 3
3 .14 × 5 cm × 15 cm cm
found by adding the LUB to the length calculated by assuming local equilib-
rium, with a shock wave concentration front [Ruthven, in Kroschwitz (ed.), 4
Kirk-Othmer Encyclopedia of Chemical Technology, vol. 1, 4th ed., Wiley- For the column 30 cm high, the LUB does not change, so that
Interscience, Hoboken, N.J., 1991, p. 493].
LUB 4 .90
= = 0 .163
Example 20-8 Scale-Up of the Fixed-Bed Adsorption of a Pharmaceu- L 30
tical Product The breakthrough data given in the table below were obtained for The superficial velocity stays the same and is
the adsorption of a pharmaceutical product in a laboratory column (5-cm diameter ×
15-cm high) at a feed flow rate of 400 mL/h and feed concentration of 0.75 U/L, where
U is units of biological activity of the pharmaceutical product. It is desired to scale up cm3
400 cm
the process to operate in a column 30 cm high. What break-point time can be expected v= h = 20 .4
in the 30-cm-high column? 3 .14 × 5 2 cm2 h
4
t (h) ci (U/L) From Eq. (20-87),
20.5 0.01
U
26.7 0.20
Lρb qi,sat 30 cm × 32 .1
32.0 0.39 ∗
t = = L = 62 .9 h
vci 0 cm U
36.0 0.53 20 .4 × 0 .75
h L
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-69

Therefore, from Eq . (20-89), dispersion due to mixing volumes upstream and downstream of the mem-
brane for flow distribution .
LUB 
tb = t ∗  1 −  = (62 .9 h)(1 − 0 .163) = 52 .6 h
Membranes for chromatography have been fabricated and are commer-
 L  cially available as thin sheets, flat disks, hollow fibers, and monolithic rods,
disks, and tubes [Avramescu et al ., in A . Pabby et al . (eds .), Handbook of Mem-
Membrane Chromatography The use of filters and membranes brane Separations, CRC Press, Boca Raton, Fla ., 2009, p . 25] . Hollow fibers
to adsorb substances from a process stream has a long history . Indeed, are usually bundled within a shell using potting material . For large-scale
the classic elementary school experiment, in which India ink is sepa- production, thin membrane sheets are usually stacked or spiral-wound .
rated into a rainbow of colors as it ascends a wetted strip of filter paper, Various configurations of membrane systems are illustrated in Fig . 20-57 .
is a form of membrane chromatography, in which the pigment chemi- The volumetric binding capacities are lower than in conventional chroma-
cals have differential binding affinities for the cellulose fibers that form tography, and the depths are not typically high .
the filter material . However, the use of membranes in chromatography The case typically made for membrane chromatography is that the dis-
received renewed interest with the development of microporous adsorp- posable nature of the filters and the large volumes of water required to oper-
tive membranes, in which the surfaces of the pores are chemically ate a fixed column make membrane chromatography a viable alternative to
derivatized with an interactive ligand . Membrane chromatography was column chromatography for polishing applications for removal of contami-
first performed in 1988 by [Brandt et al ., Bio/Technology 6: 779 (1988)] nants such as viruses, DNA, and endotoxins [Fraud et al ., BioPharm. Int. 23:
using hollow fiber membranes functionalized with either the protein 44 (2010)] . However, in analyses such as these, the chromatography column
gelatin for isolating fibronectin from human plasma or protein A for is sized at 100 times the size of the membrane; in other words, the chroma-
purifying an IgG antibody . Compared to the equivalent chromatography tography column will be used at 1 percent of its dynamic binding capacity,
column packed with 100-µm adsorbent particles, the membrane chro- where the membrane will be used at 100 percent dynamic binding capac-
matography column volume and fluid residence time were less by more ity . Furthermore, the dynamic binding capacity of the chromatography
than 3 orders of magnitude . resin typically exceeds that of the membrane configuration [70 mg/mL for
The major difference between membrane and packed-bed chromatogra- Sepharose Q-HP (GE Life Sciences Technical Literature, available at www .
phy is that in membrane chromatography the transport of solutes to their gelifesciences .com) versus 50 mg/mL for the Sartobind Q (Sartorius tech-
binding sites takes place predominantly by convection through the pores nical literature, available at www .sartorius .com), for example] . The actual
(rather than by diffusion in the stagnant fluid inside the adsorbent parti- ligand density in the two formats is approximately 50 mM for the Sartobind
cles for packed-bed chromatography), thereby reducing both process time Q versus 140 mM for the Sepharose Q-HP . This demonstrates two related
and liquid volume . Much higher flow rates can be used in membrane chro- points . First, the ligand density achievable in membrane chromatography
matography than in packed beds, and the binding efficiency is relatively is much lower (three times in this comparison) because the surface area
independent of the feed flow rate over a wide range [Ghosh, J. Chromatogr. for adsorption is lower than in resin chromatography . Second, because this
A 952: 13 (2002)] . Although they have a shorter flow path, membrane internal surface area in resin chromatography is less accessible than for
chromatography systems have generally been found to achieve resolu- membrane chromatography, the proportion of dynamic binding capacity
tions comparable to those with packed beds of adsorbent [Yamamoto and available to a large molecule such as a protein is much smaller for resin than
Sano, J. Chromatogr. 597: 173 (1992); Gerstner et al ., J. Chromatogr . 596: for membranes .
173 (1992)] . However, membrane chromatography systems typically are Another perceived limitation of using membranes as chromatogra-
used for adsorption of contaminants in a process stream, such as DNA or phy media is the cleanliness of the feed stream required . For the specific
endotoxin, or low levels of host cell proteins, rather than for the binding application of stripping extremely low-concentration contaminants from
and elution of a product molecule and its closely related variants, such as a concentrated antibody solution as the last step (with the cleanest feed
those caused by oxidation or amino acid misincorporation . For example, stream) in the process, a high solids load could potentially foul the mem-
in an antibody purification process, a final step may be the reduction of brane surface and blind pores . Ultimately, only the largest pores will be
DNA, endotoxin, and certain host cell proteins . In this case, the monoclo- left open, diffusion path lengths will increase, and flow will deteriorate .
nal antibody need not bind to the chromatography media (resin or mem- This issue is easily overcome by adding a prefilter upstream of the mem-
brane); therefore, a pH may be chosen where the antibody does not bind, brane, just as is frequently done for conventional chromatography . The
while the contaminants do . This is called the flow-through mode, or frontal contaminant load on the membrane itself is addressed by sizing the mem-
chromatography . brane for the anticipated load, again, just as one would with conventional
Another application where membrane chromatography is particularly chromatography . For example, in the polishing of antibody solutions using
suitable is in the purification of large proteins (molecular weight > 250,000) anion exchange, many contaminants, including viruses, nucleic acids, and
[Ghosh, J. Chromatogr. A 952: 13 (2002)] . Proteins this large diffuse slowly endotoxins, tend to be negatively charged at neutral pH, while antibodies
into the pores of chromatography adsorbent media, but this is not a limi- are positively charged at this pH and flow through without binding [Zhou
tation in membrane chromatography since diffusion paths are very short . et al ., BioPharm Int. 20: 26 (2007)] .
In membrane chromatography, there is dispersion in the system due to Scale-up of membrane chromatography is typically achieved by
the flow found in open pores, and the diffusion of molecules in the radial increasing the diameter of the filter, conserving the interstitial velocity
direction across the parabolic flow profile and also in the axial direction . from scale to scale . However, it is possible to scale up by maintaining
There is some dispersion due to the kinetics of adsorption and desorption, the ratio of membrane thickness to interstitial velocity constant [Etzel
which are a feature of conventional chromatography as well . Finally, there is and Riordan, in Shukla et al . (eds .), Process Scale Bioseparations for the

Hollow fiber
Polymer rod
Spiral wound

Membrane stack

FIG. 20-57 Various configurations used in membrane chromatography systems . The arrows represent the direc-
tion of flow, and the membrane cross-sectional area is shaded in gray . [Reproduced from Bioseparations Science and
Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
of Oxford University Press .]
20-70 BIOREACTIONS AND BIOPROCESSING

Biopharmaceutical Industry, Taylor and Francis, 2007, p . 278], and when V = column volume
this is done, the longer or thicker membranes would have to be able V0 = column void volume
to perform with higher pressure drops and higher velocities than their u = interstitial fluid velocity = superficial velocity v/column void
scaled-down counterparts . fraction ε
dp = particle diameter
Example 20-9 Comparison of Time for Diffusion Mass Transfer in Con-
ventional Chromatography and Membrane Chromatography Estimate This equation has been found to be valid over a wide range of experimen-
the time for diffusion mass transfer in conventional chromatography and in mem- tal conditions (Yamamoto et al ., ibid .) . To remove the volume terms from
brane chromatography . Assume a typical particle diameter for an adsorbent used in the expression for resolution, we make the definitions
preparative chromatography of 100 µm, pore diffusion coefficient  p of 3 × 10−7 cm2
s−1 of a representative protein bovine hemoglobin in size exclusion media at 20°C Q ε uAc
Q= = (20-93)
[Athalye, J. Chromatog . 589: 71 (1992)], superficial velocity of 600 cm/h in the pores in V V
membrane chromatography possible in membrane chromatography with no compro-
mise in efficiency [Zhou et al ., BioPharm Int . 20: 26 (2007)], void fraction of 0 .6 for the G = Vg (20-94)
membranes, and a membrane pore size of 1 µm . Assume the fluid has the properties
of water . where Q is the inlet flow rate, ε is the column void fraction, and A is the
The time required for diffusion can be calculated using Einstein’s diffusion equation column cross-sectional area . These substitutions lead to
1/2
δ2  
t= ε
2 p Rs ∝  2
(20-95)
 G (1 − ε )Qd p
For conventional chromatography, the diffusion length is 50 µm (one-half the particle
diameter) . The type of convective flow regime in the pores in membrane chromatog- Thus, for scale-up with constant resolution from scale 1 to scale 2 for the
raphy can be determined by calculating the Reynolds number in a cylindrical pore: same product and the same column void fraction, the scale-up equation is
cm G1 Q 1d p2 1 = G2 Q 2 d p2 2 (20-96)
600
h × 1 .0 g × 1 cm × 1 h
d m   r 1 µm ×
v
 ε 0 .6 cm3 10 4 µm 3600 s Thus, as the particle size increases on scale-up, the flow rate relative to the
Re = = = 0 .003
m g column volume must decrease and/or the gradient slope must decrease to
0 .01
cm ⋅ s maintain constant resolution, which seems correct intuitively .
where dm is the diameter of a pore in the membrane, v is the fluid superficial velocity, ε In practice, the gradient and the stationary phase size and chemistry are
is the void fraction, and ρ and µ are the density and viscosity of the fluid, respectively . not changed upon scale-up . This is so because it is easy to develop lab-scale
Thus, this is laminar, creeping flow since Re is less than 0 .1, and the velocity in the processes that use the same resin and same gradient that can be used at
pore has a parabolic profile . Since the protein is being adsorbed on the surface of the the commercial process scale . Therefore, in practice, only the ratio between
pore, there will be a radial concentration gradient from the surface to the center of column volume and flow rate needs to be addressed:
the pore, and the maximum diffusion length will therefore be 0 .5 µm (one-half the
pore diameter) . Q1 Q2
Therefore, for the conventional chromatography the diffusion time from the Ein- = (20-97)
stein diffusion equation is V1 V2

 1 cm 
2 Scale-up from a well-designed process development or preparatory column
(50 μm)2 ×  4 is reasonably straightforward . When the bed height can be maintained on
 10 µm 
t= = 42 s scale-up, the mobile phase linear velocity remains the same, and the column
cm 2 is simply scaled by diameter . Scaling from a column 1 cm in diameter to a
2 × 3 .0 × 10 −7
s column having a diameter of 10 cm would constitute a 100-fold increase in
For membrane chromatography, assuming the diffusion coefficient in the pores of the scale . This is the most conservative way to scale up a column, but is not a
membrane is the same as in pores of the conventional chromatography adsorbent, necessary constraint .
2
 1 cm  Example 20-10 Scale-Up of Protein Chromatography A column 20 cm
(0 .5 µm)2 ×  4
 10 μm  long, with an internal diameter of 5 cm, gives sufficient purification to merit scale-up .
t= = 0 .004 s The column produces 3 .2 g of purified protein per cycle, and a cycle takes 6 h, from
cm2
2 × 3 .0 × 10 −7 equilibration through regeneration . To design a throughput of 10 g/h, what are the new
s column’s dimensions if the superficial velocity is held constant and it is assumed that
These calculations show that the diffusion time is four orders of magnitude less for the gradient and particle size are not changed on scale-up?
membrane chromatography compared to conventional chromatography . A similar Equation (20-97) is applicable since the gradient and particle size remain the same .
result was obtained by Klein (Affinity Membranes, Wiley, Hoboken, N .J ., 1991, p . 135) for According to this equation, the residence time is the same at small and large scales,
the diffusion time of IgG antibody in membrane chromatography with hollow fibers and therefore the cycle time is not changed . Thus, the scaled-up column must produce
compared to a fast-flow gel matrix, both using immobilized protein A . 6 h/cycle ×10 g/h = 60 g/cycle . Since the flow rate is proportional to the throughput of
In summary, membrane chromatography is a technology that is growing in impor- protein,
tance in bioseparation science . There is a place for low-capacity, high-throughput Q2 60 g/cycle
adsorption media for the stripping of low-level contaminants from a process stream . = = 18 .75
The membranes are available with virtually all the same chemistries as conventional Q1 3 .2 g/cycle
chromatography . They can be made inexpensively and fit well with disposable tech-
nologies currently in favor in the industry . From Eq . (20-97),
Q1 Q
= 2
Scale-Up of the Chromatography of Proteins Chromatography A1 L1 A2 L2
scale-up algorithms typically account for changes in bed height and diam-
eter, linear and volumetric flow rate, and particle size . A general approach which leads to
to scale-up is based on keeping the resolution constant . Yamamoto et al . v1 v 2
[ J. Chromatog . 409: 101 (1987); Ion-Exchange Chromatography of Proteins, =
L1 L2
Dekker, 1988, p . 263] have developed the following proportionality for
resolution Rs of proteins in linear gradient elution ion exchange chromatog- Since the superficial velocity is constant,
raphy and hydrophobic interaction chromatography:
L1 = L2
1/2
m L
Rs ∝   (20-92) Therefore,
 g (V - V0 )ud p2 
2
π  2  L2
D
where m = diffusion coefficient of protein in solution Q2 V2  2  D 
2

L = column length = = 2 =  2  = 18 .75

 D1 
π  1  L1
Q1 V1 D
g = slope of gradient (change in concentration of gradient per
 2
volume of gradient)
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-71

where D1 and D2 are the column diameters for columns 2 and 1, respectively . Since New product candidates
D1 = 5.0 cm, we obtain

D2 = (18 .75)0 .5 D1 = 21 .6 cm
Feasibility Level 1
It is not always necessary to scale according to bed diameter . The flow rate may be
normalized against the volume of the empty column, to give units of time−1. This nor- Evaluation of product
malized flow rate is held constant from one scale to another, as shown in Eq. (20-97). opportunities
Considerable research and practice indicate that this technique is equally effective
Level 2
and less restrictive compared to holding linear velocity constant. In this case, column
bed height may be increased or decreased, depending on the requirements of pressure Development stage
drop and mechanical seals. A shallower bed gives a lower linear velocity and a wider Setting development Level 3
diameter. A deeper bed gives a higher linear velocity (and higher pressure drop) and a
proportionally narrower bed diameter. Therefore, bed height may need to be scaled on objectives,
the basis of pressure drop constraints. preparation of budgets
Level 4
BIOPROCESS DESIGN AND ECONOMICS
Market Level 5
Given a product and a desired annual production rate (process through-
entry
put), bioprocess design endeavors to answer the following and other related
questions: What are the required amounts of raw materials and utilities
needed for a single batch? What is the total amount of resources consumed
per year? What is the required size of process equipment and supporting
utilities? Can the product be produced in an existing facility, or is a new Commercial products
plant required? What is the total capital investment? What is the manu-
FIG. 20-58 Types of design estimates during the life cycle of a product [Frohlich,
facturing cost? What is the optimum batch size? How long does a single in Biopharmaceutical Process Economics and Optimization (conference proceedings),
batch take? How much product can be generated per year? Which process Sept. 30–Oct. 1, 1999, International Business Communications, Westborough, Mass].
steps or resources constitute scheduling and throughput bottlenecks? What
changes can increase throughput? What is the environmental impact of
the process (i.e., amount and type of waste materials)? Which design is the engineers for order-of-magnitude estimates. The table lists the capital
“best” among several plausible alternatives? investment for five large-scale facilities built to manufacture therapeutic
Definitions and Background Process design is the conceptual work monoclonal antibodies using cell culture (by growing mammalian cells in
done prior to building, expanding, or retrofitting a process plant. It consists of stirred-tank bioreactors). Column 2 displays the number of the production
two main activities, process synthesis and process analysis. Process synthesis bioreactors, the working volume of each, and the total working volume. For
is the selection and arrangement of a set of unit operations (process steps) instance, a Genentech facility includes six production bioreactors, each hav-
capable of producing the desired product at an acceptable cost and quality. ing a working volume of 15 m3. Column 4 displays the total capital invest-
Process analysis is the evaluation and comparison of different process syn- ment, and column 5 displays the ratio of the total capital investment to the
thesis solutions. In general, a synthesis step is usually followed by an analysis total production bioreactor volume. The ratio ranges from 3.3 to 6.2 with an
step, and the results of analysis determine the subsequent synthesis step. average value of $4.6 million per cubic meter of bioreactor volume. Based
Process design and project economic evaluation require integration of on the data of Table 20-26, an engineer may conclude with some confidence
knowledge from many different scientific and engineering disciplines and that the capital investment for a new 100 m3 (total production bioreactor
are carried out at various levels of detail. Table 20-25 presents a common volume) cell culture facility would be in the range of $330 million to $620
classification of design and cost estimates and typical engineering costs for million and most likely around $460 million. Please note, however, that
a $50 million capital investment for a new plant. advances in technology (e.g., cell lines that generate higher product titers
Figure 20-58 presents the need for design estimates of various types dur- and the use of single-use systems) and other factors may render such data
ing the life cycle of product development and commercialization. The trap- obsolete and reduce the accuracy of order-of-magnitude estimates. As a
ezoidal shape of the diagram represents the drastic reduction in product result, cost estimates are progressively refined as new product candidates
candidates as we move from feasibility studies to commercialization. In move through the development life cycle shown in Fig. 20-58.
fact, the chances of commercialization at the research stage for a new prod- Most engineers employed by operating companies usually perform level 2
uct are only about 1 to 3 percent, at the development stage they are about and 3 studies. Such studies take weeks or months to complete using appro-
10 to 25 percent, and at the pilot plant stage they are about 40 to 60 percent priate computer aids. The main objective of such studies is to evaluate
(Douglas, Conceptual Design of Chemical Processes, McGraw-Hill, New York, alternatives and pinpoint the most cost-sensitive areas—the economic “hot
1988, p. 4). spots”—of a complex process. The results of such analyses are used to plan
Order-of-magnitude estimates are usually practiced by experienced engi- future research and development and to generate project budgets.
neers who have worked on similar projects in the past. They take minutes Level 4 and 5 studies are usually performed by the engineering and con-
or hours to complete, but the error in the estimate can be as high as 50 per- struction companies hired to build new plants for promising new products
cent. Table 20-26 presents a sample of data typically used by experienced that are at an advanced stage of development. Such estimates are beyond
the scope of this section. Instead, the focus of the material in the rest of
this section will be on level 1, 2, and 3 studies. Also note that opportunities
TABLE 20-25 Types of Design Estimates, Their Cost and Accuracy for creative process design work are usually limited to preliminary studies.
for a $50 Million Project* By the time detailed engineering work has been initiated, a process is more
Level Type of estimate Error (%) Cost ($1000)
1 Order-of-magnitude estimate ≤ 50
(ratio estimate) based on similar previous TABLE 20-26 Capital Investments for Cell Culture Facilities
projects Bioreactor Completion Investment $ million
2 Project planning estimate ≤ 30 30–100 capacity (m3) year ($ millions) per m3
(budget estimation) based on knowledge of
major equipment items Boehringer Ingelheim 6 × 15 = 90 2003 296 3.3
3 Preliminary engineering (scope ≤ 25 250–750 (Germany)
estimate) based on sufficient data to permit Lonza Biologics 3 × 20 = 60 2004 207 3.4
the estimate to be budgeted (Portsmouth, N.H.)
4 Detailed engineering (capital approval stage) ≤ 15 1250–2000 Genentech 6 × 15 = 90 2005 450 5.0
based on almost complete process data (Oceanside, Calif.)
5 Procurement and construction (contractor’s ≤ 10 3500–7000 Bristol Myers Squibb 6 × 20 = 120 2009 750 6.2
estimate) based on complete engineering (Devens, Mass.)
drawings, specifications, and site surveys Roche Pharmaceuticals 6 × 12.5 = 75 2009 375 5.0
(Switzerland)
∗Douglas, Conceptual Design of Chemical Processes, McGraw-Hill, New York, 1988, p. 7.
Reproduced from Bioseparations Science and Engineering, 2d. ed., by Roger G. Har- Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G.
rison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission of Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission
Oxford University Press. of Oxford University Press.
20-72 BIOREACTIONS AND BIOPROCESSING

than 80 percent fixed . Furthermore, the vast majority of important deci- minute-by-minute time dependency of events and the animation of the
sions for capital expenditures and product commercialization are based process . Material balances, equipment sizing, and cost analysis tasks are
on results of preliminary process design and cost analysis . This explains usually out of the scope of such models .
why it is so important for a new engineer to master the skills of preliminary The benefits from the use of process simulators depend on the type of
process design and cost estimation . product, stage of development, and size of the investment . For commodity
Environmental impact assessment is an activity closely related to process biological products, such as biofuels, minimization of capital and operating
design and cost estimation . Biochemical plants generate a wide range costs is the primary benefit . For high-value biopharmaceuticals, systematic
of liquid, solid, and gaseous waste streams that require treatment prior process development that shortens the time to commercialization is the
to discharge . The cost associated with waste treatment and disposal has primary motivation . Figure 20-59 shows a pictorial representation of the
skyrocketed in recent years due to increasingly strict environmental regula- benefits from the use of such tools at the various stages of the commercial-
tions . This cost can be reduced through minimization of waste generation ization process .
at the source . However, generation of waste from a chemical or biochemical When product and process ideas are first conceived, process modeling
process is dependent on the process design and the manner in which the tools are used for project screening, selection, and strategic planning based
process is operated . Thus, reducing waste in an industrial process requires on preliminary economic analyses . During this phase, the company’s pro-
intimate knowledge of the process technology . In contrast, waste treatment cess development groups are looking into the various options available for
is essentially an add-on at the end of the process . In addition, minimization synthesizing, purifying, characterizing, and formulating the final product .
of waste generation must be considered by process engineers at the early The process undergoes constant changes during development . Typically, a
stages of process development . Once a process has undergone significant large number of scientists and engineers are involved in the improvement
development, it is difficult and costly to make major changes . Furthermore, and optimization of individual processing steps . The use of process simu-
regulatory constraints that are unique to the pharmaceutical industry lators at this stage can introduce a common language of communication
restrict process modifications after clinical efficacy of the drug has been and facilitate team interaction . A computer model of the entire process can
established . These are only some of the reasons why process synthesis and provide a common reference and evaluation framework to facilitate process
analysis must be initiated at the early stages of product development . development . The impact of process changes can be readily evaluated and
Synthesis of Bioseparation Processes Rules of thumb, or heuris- documented in a systematic way . Once a reliable model is available, it can
tics, were given in the introduction for use in developing flowsheets for the be used to pinpoint the cost-sensitive areas of a complex process . These are
recovery and purification of biological products, and examples were given of usually steps of high capital and operating cost or low yield and production
the application of these rules of thumb to the placement of various biosepa- throughput . The findings from such analyses can focus further lab and pilot
ration unit operations . plant studies to optimize those portions of the process . The ability to experi-
Process Analysis The flowsheets created during process synthesis ment on the computer with alternative process setups and operating con-
must be analyzed and compared on the basis of capital investment, manu- ditions reduces the costly and time-consuming laboratory and pilot plant
facturing cost, environmental impact, and other criteria to decide which effort . A simulator can also evaluate the environmental impact of a process .
ideas to consider further . Methodologies for estimating capital investment For instance, material balances calculated for the projected large-scale
and manufacturing cost are presented later in the Process Economics sub- manufacturing reveal environmental hot-spots . These are usually process
section . In both cases, estimation is based on the results of material and steps that utilize organic solvents and other regulated materials with high
energy balances and equipment sizing . These calculations are typically disposal costs . Environmental issues not addressed during process develop-
done using spreadsheets or process simulators . These tools allow the pro- ment may lead to serious drawbacks during manufacturing .
cess design team to characterize a processing scenario, and then quickly With process development near completion at the pilot plant level, simu-
and accurately redo the entire series of calculations for a different set of lation tools are used to systematically design and optimize the process for
assumptions and other input data . commercial production . Availability of a good computer model can greatly
Spreadsheets Spreadsheet applications, such as Microsoft Excel, have facilitate the transfer of a new process from the pilot plant to the large-scale
become as easy to use as word processors and graphics packages . In its sim- facility . If a new facility needs to be built, process simulators can size pro-
plest form, a spreadsheet is an electronic piece of paper with empty boxes, cess equipment and supporting utilities and estimate the required capital
known as cells . The user can enter data in those cells, perform calculations,
and generate results . Results from spreadsheets can be easily plotted in a
variety of graphs .
Process Simulators and Their Benefits Process simulators are soft-
ware applications that enable the user to readily represent and analyze Idea generation
integrated processes . They have been in use in the petrochemical industries Project screening, strategic planning
since the early 1960s . Established simulators for those industries include
Aspen Plus and Aspen HYSYS from Aspen Technology, Inc . (Burlington,
Mass .), ChemCAD from Chemstations, Inc . (Houston, Tex .), and PRO/II
from SimSci-Esscor, Inc . (Lake Forest, Calif .) .
The simulators mentioned above have been designed to model primarily
continuous processes and their transient behavior . Most biological prod- Process development
ucts, however, are produced in batch and semicontinuous mode (Koro- Evaluation of alternatives
vessi and Linningerr, Batch Processes, Taylor and Francis, Oxford, UK, 2006;
Common language of communication
Heinzle et al ., Development of Sustainable Bioprocesses, Wiley, Hoboken, N .J .,
2006) . Such processes are best modeled with batch process simulators that
account for time dependency and sequencing of events . The first simulator
designed specifically for batch processes was called Batches ( from Batch
Process Technologies in West Lafayette, Ind .) . It was commercialized in the
mid-1980s . All its operation models are dynamic, and simulation always From R&D to manufacturing
involves integration of differential equations over a period of time . In the
mid-1990s, Aspen Technology (Burlington, Mass .) introduced Batch Plus Facility design, tech transfer, and process fitting
(now called Aspen Batch Process Developer), a recipe-driven simulator that
targeted batch pharmaceutical processes . Around the same time, Intelli-
gen, Inc . (Scotch Plains, N .J .) introduced SuperPro Designer [Petrides et al .,
Pharm. Eng. 22: 56 (2002); Toumi et al ., Pharm. Eng . 20, March/April 2010] .
SuperPro Designer is a flowsheet-driven simulator that handles material Manufacturing
and energy balances, equipment sizing and costing, economic evaluation,
environmental impact assessment, process scheduling, and debottleneck- Ongoing optimization, debottlenecking,
ing of batch and continuous processes . cycle time analysis, and reduction
Discrete-event simulators have also found applications in the biopro- Production planning and scheduling
cessing industries . Established tools of this type include ProModel from
ProModel Corporation (Orem, Utah), Arena and Witness from Rockwell
Automation, Inc . (Milwaukee, Wis .), Extend from Imagine That, Inc . (San FIG. 20-59 Benefits of using process simulators . [Reproduced from Bioseparations
Jose, Calif .), and FlexSim from FlexSim Software Products, Inc . (Orem, Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and
Utah) . The focus of models developed with such tools is usually on the Demetri P . Petrides (2015) . By Permission of Oxford University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-73

investment . In transferring production to existing manufacturing sites (tech- validation of the model is necessary . In its simplest form, a review of the
nology transfer), process simulators can be used to evaluate the various sites results by an experienced engineer can play the role of validation .
from a capacity and cost point of view and to select the most appropriate Process Economics The preliminary economic evaluation of a
one . project for manufacturing a biological product usually involves estima-
In large-scale manufacturing, simulation tools are mainly used for on- tion of capital investment, estimation of operating costs, and analysis of
going process optimization and debottlenecking studies . Furthermore, profitability . For biopharmaceuticals, another figure worth considering
tools that are equipped with batch process scheduling capabilities can be is the average cost of new drug development, which is in the range of
used to generate production schedules on an ongoing basis in a way that $500 million to $1 billion [DiMasi et al ., J. Health Econ . 22: 151 (2003);
does not violate constraints related to the limited availability of equipment, Gilbert et al ., The Business and Medicine Report 21: Nov . 2003] . Much of
labor resources, utilities, inventories of materials, etc . this figure represents research and development (R&D) spending for
Using a Biochemical Process Simulator The minimum requirements all unsuccessful products . In other words, the actual average develop-
for a biochemical process simulator are the ability to handle batch as well ment cost per successful drug may be $100 to $200 million, but because
as continuous processes and the ability to model the unit operations that more than 90 percent of new projects never reach commercialization,
are specific to bioprocessing . Because SuperPro Designer ( from Intelligen, the average overall R&D cost skyrockets . This order-of-magnitude cost
Inc .) has the ability to satisfy these requirements, we will use it to illustrate increase reinforces the need for effective process design tools and meth-
the role of such tools in bioprocess design . A functional evaluation version odologies that assist engineers and scientists in efficiently evaluating
of SuperPro Designer and additional information on bioprocess simulation and eliminating nonpromising project ideas at the very early stages of
can be obtained at the website www .intelligen .com . Tutorial videos on the product and process development .
use of SuperPro Designer can be viewed at www .intelligen .com/videos . Capital Cost Estimation The capital investment for a new plant includes
To model an integrated process using a simulator, the user starts by devel- three main items: direct fixed capital (DFC), working capital, and start-up
oping a flowsheet that represents the overall process . For instance, the flow- and validation cost . The DFC for small to medium biotechnology facilities
sheet of a hypothetical process can be drawn on the main window of SuperPro is usually in the range of $50 to $200 million, whereas for large facilities it
Designer . The flowsheet is developed by putting together the required unit is in the range of $250 to $750 million . For preliminary design purposes, the
operations (sometimes referred to as unit procedures, as explained later in this various items of DFC are estimated based on the total equipment purchase
section) and joining them with material flow streams . Next, the user initial- cost (PC) using several multipliers sometimes called Lang factors . Table 20-27
izes the flowsheet by registering (selecting from the component database) the provides ranges and average values for the multipliers and a skeleton for the
various materials used in the process and specifying operating conditions and calculations . Detailed definitions of the various cost items and additional
performance parameters for the various operations . information can be found in traditional process design textbooks and the
Most biochemical processes operate in batch or semicontinuous mode . technical literature [Turton et al ., Analysis, Synthesis, and Design of Chemical
This is in contrast to continuous operation, which is typical in the petro- Processes, 5th ed ., Pearson, Upper Saddle River, N .J ., 2018; Douglas, Concep-
chemical and other industries that handle large throughputs . In continu- tual Design of Chemical Processes, McGraw-Hill, New York, 1988; Peters and
ous operations, a piece of equipment performs the same action all the time, Timmerhaus, Plant Design and Economics for Chemical Engineers, 5th ed .,
which is consistent with the notion of unit operations . In batch process- McGraw-Hill, New York, 2002; Ulrich et al ., A Guide to Chemical Process Design
ing, however, a piece of equipment goes through a cycle of operations . For and Economics, Wiley, Hoboken, N .J ., 1984; Valle-Riestra et al ., Project Evalua-
instance, a typical chromatography cycle includes equilibration, loading, tion in the Chemical Process Industries, McGraw-Hill, New York, 1983; Garrett,
washing, elution, and regeneration . In SuperPro Designer, the set of opera- Chemical Engineering Economics, Van Nostrand Reinhold, New York, 1989;
tions that comprise a processing step is called a unit procedure (versus a unit Seider et al ., Process Design Principles—Synthesis, Analysis, and Evaluation,
operation) . Each unit procedure contains individual tasks (e .g ., equilibration, Wiley, Hoboken, N .J ., 1999; Towler and Sinnott, Chemical Engineering Design;
loading) called operations . A unit procedure is represented on the screen Principles, Practice and Economics of Plant and Process Design, Butterworth-
with a single equipment icon . In essence, a unit procedure is the recipe of a Heinemann (Elsevier), Oxford, U .K ., 2008] .
processing step that describes the sequence of actions required to complete Notice the wide range of multiplier values in Table 20-27 for estimating
that step . The significance of the unit procedure is that it enables the user to the cost of buildings . Plants for commodity biochemicals, such as ethanol
describe and model the various activities of batch processing steps in detail . and citric acid, fall on the low end of the range . Conversely, biopharmaceu-
For every operation within a unit procedure, SuperPro includes a math- tical facilities with their expensive heating, ventilation, and air condition-
ematical model that performs material and energy balance calculations . ing (HVAC) requirements fall on the high end . The average value of 0 .45
Based on the material balances, SuperPro performs equipment-sizing cal- corresponds to relatively large plants that produce medium- to high-value
culations . If multiple operations within a unit procedure dictate different products (e .g ., industrial enzymes) .
sizes for a certain piece of equipment, the software reconciles the different For more accurate estimation of building costs, it is necessary to estimate
demands and selects an equipment size that is appropriate for all opera- the process area required based on the footprint of the equipment and the
tions . In other words, the equipment is sized to ensure that it will not be space required around the equipment for safe and efficient operation and
overfilled during any operation but is no larger than necessary (to minimize
capital costs) . In addition, the software checks to ensure that the vessel con-
tents will not fall below a user-specified minimum volume (e .g ., a minimum
impeller volume) for applicable operations . TABLE 20-27 Fixed Capital Cost Estimation
Before any simulation calculations can be done, the user must ini- Range of multiplier
tialize the various operations by specifying operating conditions and Cost item Average multiplier values
performance parameters through appropriate dialog windows . After ini-
Total plant direct cost (TPDC)
tialization of the operations, the simulator performs material and energy Equipment purchase cost (PC)
balances for the entire process and estimates the required sizes of equip- Installation 0 .50 × PC 0 .2–1 .5
ment . Optionally, the simulator may be used to carry out cost analysis and Process piping 0 .40 × PC 0 .3–0 .6
economic evaluation calculations . The fundamentals of process econom- Instrumentation 0 .35 × PC 0 .2–0 .6
ics are described in the next subsection . Insulation 0 .03 × PC 0 .01–0 .05
Other tasks that can be handled by process simulators include pro- Electrical 0 .15 × PC 0 .1–0 .2
cess scheduling, environmental impact assessment, debottlenecking, Buildings 0 .45 × PC 0 .1–3 .0
and throughput analysis . Issues of process scheduling and environmental Yard improvement 0 .15 × PC 0 .05–0 .2
impact assessment are addressed in the Illustrative Examples subsection . Auxiliary facilities 0 .50 × PC 0 .2–1 .0
Throughput analysis and debottlenecking is the analysis of the capacity and
Total plant indirect cost (TPIC)
time utilization of equipment and resources (e .g ., utilities, labor, raw mate-
Engineering 0 .25 × TPDC 0 .2–0 .3
rials) . The objective is to identify opportunities for increasing throughput Construction 0 .3–0 .4
0 .35 × TPDC
with the minimum possible capital investment (see Illustrative Examples for
additional information on the subject) . Total plant cost (TPC) TPDC + TPIC
Having developed a good model using a process simulator or a spread- Contractor’s fee 0 .05 × TPC 0 .03–0 .08
sheet, the user may conduct virtual experiments with alternative process Contingency 0 .10 × TPC 0 .07–0 .15
setups and operating conditions . This may potentially reduce costly and Direct fixed capital (DFC) TPC + contractor’s fee and
time-consuming laboratory and pilot plant effort . One must be aware, how- contingency
ever, that the GIGO (garbage in, garbage out) principle applies to all com- Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
puter models . More specifically, if some assumptions and input data are Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
incorrect, the outcome of the simulation will not be reliable . Consequently, of Oxford University Press .
20-74 BIOREACTIONS AND BIOPROCESSING

TABLE 20-28 Building Cost Estimation (Year 2012 Prices)*

240
Air circulation rates
Space function Unit cost ($/m2) (volume changes/h)
Process area† 220

Hardware cost ($1000)

ISO Grade 8 3000–3750 10–20
ISO Grade 7 3750–5200 30–70 200
ISO Grade 6 6700–9000 70–160
ISO Grade 5 9000–12,000 200–600
Mechanical room (utilities) 450–900 180
Laboratory 1500–3000
Office 750–900 160
∗Frolich, in Biopharmaceutical Process Economics and Optimization (conference pro-
ceedings), International Business Communications, Sept . 30–Oct . 1, 1999 . 140
†
The ISO Grade refers to ISO 14644, “Cleanrooms and associated controlled environ-
ments .” An earlier system uses a “class number” that refers to the maximum number
of particles 0 .5 μm or larger per cubic foot . Classes 100,000, 10,000, 1000, and 100 are 120
similar to ISO Grades 8, 7, 6, and 5, respectively .
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission 100
of Oxford University Press . 0 10 20 30 40 50 60
Membrane area (m2)
FIG. 20-61 Purchase cost of membrane filtration systems (2012 prices) . [Reproduced
maintenance . Then the building cost is estimated by multiplying the area of from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul
the various sections (e .g ., process, laboratory, office) of a plant by an appro- W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
priate unit cost provided in Table 20-28 . This table, which was developed University Press .]
by DPS Biometrics (Framingham, Mass .), also provides information on air
circulation rates for the various process areas, which determine the sizing
and power requirements of HVAC systems . Note that equipment purchase cost is a strong function of industrial
Table 20-27 indicates a wide range in the equipment installation cost application and plant location . The data of Figs . 20-60 to 20-63 are appli-
multipliers . Using multipliers that are specific to individual equipment cable to biopharmaceutical facilities in developed countries . The cost of
items leads to the most accurate estimates . In general, equipment delivered membrane filtration systems used in the food, biofuel, and water purifi-
mounted on skids has a lower installation cost . cation industries is more than an order of magnitude lower compared to
For preliminary cost estimates, Table 20-27 clearly shows that the fixed the biopharmaceutical industry . The much larger equipment scale and the
capital investment of a plant is a multiple (usually 3 to 10 times) of its equip- less stringent equipment specifications relative to biopharmaceuticals
ment purchase cost . The low end of the range applies to large-scale facilities are responsible for the large difference in cost . The same trend applies to
that produce biofuels and commodity biochemicals . The high end applies to the cost of chromatography columns, storage tanks, reactor vessels, and
biopharmaceutical facilities . most other equipment items . A good source of cost data for equipment
The equipment purchase cost can be estimated from vendor quotations, used in the biofuel and biomaterial industries is available from the U .S .
published data, company data compiled from earlier projects, and the use of Department of Energy [DOE/NETL-2002/1169 (2002) . Process Equipment
process simulators that are equipped with appropriate costing capabilities . Cost Estimation . Available at http://www .osti .gov/bridge/purl .cover .
Vendor quotations are time-consuming to obtain and are, therefore, usu- jsp?purl=/797810-Hmz80B/native/] . Additional sources for bioprocessing
ally avoided for preliminary cost estimates . Instead, engineers tend to rely equipment cost data are available in the literature [Kalk and Langlykke,
on the other three sources . Figures 20-60 to 20-63 provide equipment cost in Demain and Solomon (eds .), Manual of Industrial Microbiology and Bio-
data for disk-stack centrifuges, membrane filtration systems, chromatogra- technology, American Society for Microbiology, Washington D .C ., 1986,
phy columns, and vertical agitated tanks that meet the specifications of the p . 363; Reisman, Economic Analysis of Fermentation Processes, CRC Press,
biopharmaceutical industry . The cost of the membrane filtration systems Boca Raton, FL, 1988, p . 55] .
includes the cost of the skid, tank, pumps, and automation hardware and Often, cost data for one or two discrete equipment sizes are available, but
software . The tanks are appropriate for buffer preparation . They include the cost for a different-size piece of equipment must be estimated . In such
a low-power agitator, but no heating/cooling jacket . The data represent cases, the scaling law (expressed by the following equation) can be used:
average values from several vendors . a
 Size2 
Cost 2 = Cost1  (20-98)
 Size1 

The mathematical form of the scaling law explains why cost-versus-size

data graphed on logarithmic coordinates tend to fall on a straight line . The
500 value of the exponent a in Eq . (20-98) ranges between 0 .5 and 1 .0, with an
average value for vessels of around 0 .6 (this explains why the scaling law
Purchase cost ($1000)

is also known as the “0 .6 rule”) . According to this rule, when the size of a
vessel doubles, its cost will increase by a factor of (2/1)0.6, or approximately
52 percent . This result is often referred to as the economy of scale . In using
the scaling law, it is important to make sure that the piece of equipment
whose cost is being estimated has a size that does not exceed the maximum
available size for that type of equipment .
The prices of equipment change with time owing to inflation and other
market conditions . That change in price is captured by the Chemical
Engineering Plant Cost Index (CE Index) that is published monthly by
Chemical Engineering magazine . The index I is used to update equipment
cost data according to the following equation:

50 I 
Cost 2 = Cost1  2  (20-99)
10 100 1000  I1 
Sigma factor (1000 m2)
Another factor that affects equipment purchase cost is the material of
FIG. 20-60 Purchase cost of disk-stack centrifuges versus Σ factor (2012 prices) . construction . For instance, a tank made of stainless steel costs approxi-
[Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, mately 2 .5 to 3 times as much as a carbon-steel tank of the same size . A
Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford tank made of titanium costs around 15 times the cost of a carbon-steel
University Press .] tank of the same size . Other factors that affect equipment cost include the
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-75

10,000

1000

Purchase cost ($1000)

100

10

1
1 10 100 1000 10,000
Column volume (L)
FIG. 20-62 Purchase costs of chromatography columns made of acrylic tube and stainless-steel bed sup-
ports (2012 prices) . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison,
Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford University Press .]

finishing of the metal surface and the instrumentation that is provided with (e .g ., in dollars per kilogram) . The unit cost and selling prices of bioprod-
the equipment . ucts are inversely proportional to market size (Harrison et al ., Biosepara-
Working capital accounts for cash that must be available for invest- tions Science and Engineering, Oxford University Press, Oxford, UK, 2015,
ments in ongoing expenses and consumable materials . These expenses may p . 2) . Low-molecular-weight commodity biochemicals and biofuels that are
include raw materials for 1 to 2 months, labor for 2 to 3 months, utilities for produced in large quantities cost around $1 to $5/kg to make . Citric acid,
1 month, waste treatment/disposal for 1 month, and other miscellaneous whose production is analyzed later in this section, is a product of this type .
expenses . The required amount of working capital for a process is usually 10 Specialty biochemicals that are used as food supplements (e .g ., vitamins)
to 20 percent of the DFC . and flavoring agents have a manufacturing cost of $5 to $100/kg . The manu-
Start-up and validation costs can also represent a significant capital facturing cost of therapeutic proteins produced in large quantities is in the
investment for a biopharmaceutical plant . A value of 20 to 30 percent of DFC range of $1/g to $1000/g . Human serum albumin (HSA) which is extracted
is quite common . from blood plasma and has an annual production volume of more than
Operating Cost Estimation The operating cost to run a biochemical 500 metric tons lies close to the low end . The manufacturing cost of thera-
plant is the sum of all ongoing expenses including raw materials, labor, con- peutic proteins with annual production volume ranging from a few hun-
sumables, utilities, waste disposal, and facility overhead . Dividing the annual dreds of kilograms to a few metric tons is in the range of $50 to $1000/g . The
operating cost by the annual production rate yields the unit production cost insulin and monoclonal antibody processes analyzed later in this section

500
Purchase cost ($1000)

50
100 1000 10,000 100,000
Volume (L)
FIG. 20-63 Purchase cost of agitated tanks made of stainless steel (2012 prices) . [Reproduced from Biosep-
arations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri
P . Petrides (2015) . By Permission of Oxford University Press .]
20-76 BIOREACTIONS AND BIOPROCESSING

TABLE 20-29 Operating Cost Items and Ranges list of commonly used raw materials in the biochemical industries . Note that
Cost item Type of cost Range of values (% of total)
the price of a raw material can vary widely depending on its required purity .
This can be clearly seen in the case of water . Water for injection (WFI), for
Raw materials Direct 10–80 instance, costs 100 to 500 times as much as city water . Raw materials for bio-
Labor Direct 10–50 pharmaceuticals are typically required to be USP/NF grade . Prices for a wide
Consumables Direct 1–50
range of chemicals are available online at www .spectrumchemicals .com .
Lab/QC/QA Direct 1–50
Waste disposal Direct 1–20
Labor is estimated based on the total number of operators, which in turn
Utilities Direct 1–30 is calculated by summing the operator requirements of the various opera-
Facility overhead Indirect 10–70 tions as a function of time . As will become clear in the examples discussed
Miscellaneous Indirect 0–20 later, the labor requirement in a batch manufacturing facility varies with
time . In a single-product facility, the number of operators in each shift must
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . be based on maximum demand during that shift . In multiproduct facili-
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission ties, each product line can employ a certain number of dedicated operators
of Oxford University Press .
and rely on floating operators during periods of peak demand . In general,
smaller facilities tend to utilize a larger number of operators per processing
represent products of this type . The manufacturing cost of interferons, step because these plants are less automated . For instance, a small biotech
erythropoietin (EPO), and other therapeutic proteins with very low annual company may utilize two or three operators to set up a fermenter, whereas
production volume ( from hundreds of grams to a few kilograms) is more in a large, highly automated fermentation facility, a single operator may
than $10,000/g [Jagschies, BioPharm Int . 21: 72 (2008)] . handle the setup of six different fermenters remotely from the control room .
Table 20-29 displays the various types of operating cost, their direct or In general, a typical biotech company that deals with high-value products
indirect nature, and ranges for their values relative to the total operat- will allocate at least one operator to each processing step (centrifugation,
ing cost . Sometimes cost items are categorized as either fixed or variable . membrane filtration, chromatography, etc .) during its operation . The setup
Fixed costs are incurred regardless of the volume of product output . The of a step may require multiple operators for a short period . The annual
clearest case of a fixed cost is depreciation, which is part of the equipment- cost of an operator (including salary and benefits) varies widely around
dependent cost . The clearest case of a variable cost would be the cost of raw the globe . It is in the range of $4000 to $10,000 in developing nations and
materials . Most other costs have a fixed component and a variable compo- can exceed $50,000 in developed countries (Pollak et al ., Contract Pharma,
nent . It is obvious from the wide range of values in Table 20-29 that industry Jan ./Feb . 2012) .
averages cannot predict the operating cost of a process; a certain level of Consumables are items that may be used up, fouled, or otherwise dam-
detailed calculations is required . aged during processing, such as membranes, chromatography resins, and
The raw materials cost includes the cost of all fermentation media, activated carbon . These items must be replaced periodically . As the exam-
recovery chemicals, and cleaning materials . For commodity biochemicals, ples later in this section will illustrate, the high unit cost of chromatography
such as ethanol, the cost of fermentation media is the main component . For resins and their frequent replacement can make them a major compo-
high-value products, the solutions used for product recovery and equipment nent of the manufacturing cost . The unit cost of typical ion exchange and
cleaning can be a major part of the raw materials cost . Table 20-30 provides a hydrophobic interaction chromatography resins used for the purification
of proteins is in the range of $500 to $2000 per liter of resin . The unit cost
of protein A affinity resins that are commonly used for the purification of
TABLE 20-30 Common Bioprocessing Raw Materials (Year 2012
monoclonal antibodies is in the range of $5000 to $15,000 per liter of resin .
Prices)
The replacement frequency of such resins is in the range of 50 to 200 cycles
Raw material Comments Price ($/kg) of usage (the high-end resins have a longer useful life) . In contrast, the unit
cost of polymeric chromatography resins used for the purification of small
C Source
biomolecules (e .g ., amino acids) is substantially lower (under $100 per liter
Glucose Solution 70% w/v 0 .30–0 .40
of resin), and their life is longer (1000 to 2000 h of operation) . Likewise, the
Corn syrup 95% Dextrose equivalent 0 .40–0 .50
unit cost of silica-based resins used for water demineralization is around
Molasses 50% Fermentable sugars 0 .12–0 .20 $0 .5 per liter, and their life is in the range of 2000 to 6000 h of operation (the
Soybean oil Refined 1 .10–1 .30 life strongly depends on the composition of the treated materials) .
Corn oil Refined 1 .30–1 .40 Regarding membrane filtration operations, the unit cost of MF/UF mem-
Ethanol USP tax-free 0 .80–0 .90 branes used in the biopharmaceutical industry (in the form of hollow fiber
Methanol Gulf Coast 0 .40–0 .45 cartridges or cassettes) is in the range of $300 to $800/m2 . Such membranes
n-Alkanes 0 .75–0 .90 typically handle 10 to 50 filtration cycles before disposal . The unit cost of
N Source related membranes used in industrial biotechnology (e .g ., for production
Ammonia Anhydrous, fertilizer grade 0 .30–0 .60 of industrial enzymes) is considerably lower (under $200/m2), and the
Soybean flour 44% protein 0 .45–0 .50 expected life is more than 2000 h of operation . The cost of membranes used
Cottonseed flour 62% protein 0 .50–0 .60 for large-scale water purification is under $50/m2, and their useful life is at
Casein 13 .5% w/w total N 10 .00–12 .00
least 6000 h of operation . In general, ceramic membranes cost more than
polymeric ones, but they last longer .
Ammonium sulfate Technical 0 .17–0 .25
The cost of disposable bags or containers, also known as single-use sys-
Ammonium nitrate Fertilizer grade 33 .5% N, bulk 0 .20–0 .30
tems, is part of the consumables cost as well . Disposable bags have become
Urea 46% N, agricultural grade 0 .55–0 .65
popular in biopharmaceutical manufacturing because they eliminate the
Yeast Brewers, debittered 1 .25–1 .40 need for cleaning and sterilization in place (Papavasileiou et al ., BioPharm
Whey Dried, 4 .5% w/w N 1 .25–1 .40 Int. Suppl., Nov . 2008, p . 16) . Other advantages of single-use systems include
Salts increased processing flexibility and shorter validation, start-up, and com-
KH2PO4 USP, granular 1 .65–1 .85 mercialization times . Table 20-31 provides information on disposable bags
K2SO4 Granular, purified 2 .80–3 .00 used for the preparation and storage of buffer solutions and fermentation
Na2HPO4 1 .40–1 .80
MgSO4 ⋅ 7H2O 0 .45–0 .55
ZnSO4 ⋅ 7H2O Agricultural grade, powder 0 .65–0 .75 TABLE 20-31 Disposable Bags for Preparation
Other and Storage of Solutions (Year 2012 Prices)
Process water 0 .0001–0 .001 Volume (L) Bags for storage ($) Bags for mixing ($)
RO water 0 .005–0 .01
Water for injection 0 .02–0 .5 50 310 600
H3PO4 (85% w/w) Food grade 3 .5 – 4 .5 100 340 690
NaOH 0 .2–0 .5 200 360 820
HCl (37% w/w) 0 .7–0 .8 500 460 930
H2SO4 (98% w/w) 0 .15–0 .25 1000 650 1180

Reproduced from Bioseparations Science and Engineering, 2d ed . by Roger G . Reproduced from Bioseparations Science and Engineering, 2d ed .,
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P .
of Oxford University Press . Petrides (2015) . By Permission of Oxford University Press .
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-77

media . Bags with mixing capability are required for solution preparation . Profitability Analysis Estimates of capital investment, operating cost,
Similar bags are used for inoculum preparation in rocking and stirred-tank and revenues of a project provide the information needed to assess its prof-
bioreactors . Bags for stirred-tank bioreactors are available with working vol- itability and attractiveness from an investment point of view . There are
umes of up to 2000 L . A number of biopharmaceuticals are produced exclu- various measures for assessing profitability . The simplest ones include gross
sively in single-use systems . Note that large disposables bags (larger than margin, return on investment (ROI), and payback time, and they are calcu-
50 L) utilize appropriate supporting skids . lated by using the following equations:
Laboratory, QC, and QA activities include off-line analysis, quality control
(QC), and quality assurance (QA) costs . Chemical and biochemical analysis Gross profit
Gross margin = (20-100)
and physical property characterization, from raw materials to final product, Revenues
are a vital part of biochemical operations . The laboratory/QC/QA cost is
usually 10 to 20 percent of the operating labor cost . However, for certain Net profit per year
Return on investment (ROI) = × 100 % (20-101)
biopharmaceuticals that require a large number of very expensive assays, Total investment
this cost can be as high as the operating labor . For such cases, it is impor-
tant to account for the number and frequency of the various assays in detail, Total investment
Payback time (yr) = (20-102)
since changes in lot size that can reduce the frequency of analysis can have Net profit per year
a major impact on profit margins .
The treatment of wastewater and the disposal of solid and hazardous where gross profit is equal to annual revenues minus the annual operating
materials are other important operating costs . The amount and composi- cost, and net profit is equal to gross profit minus income taxes plus depre-
tion of the various waste streams are derived from the material balances . ciation . All variables are averaged over the lifetime of a project .
Multiplying the amount of each waste stream by the appropriate unit cost Other measures that are more involved, such as the net present value
yields the cost of treatment and disposal . Treatment of low biological oxy- (NPV) and internal rate of return (IRR), consider the cash flows of a project
gen demand (BOD) wastewater (< 1000 mg/L) by a municipal wastewater over its evaluation life and the time value of money . Detailed definitions for
treatment facility usually costs $0 .2 to $0 .5/m3 . This is not a major expense NPV and IRR can be found in the literature [Peters and Timmerhaus, Plant
for most biotech facilities that deal with high-value products . However, Design and Economics for Chemical Engineers, 4th ed ., McGraw-Hill, New York,
disposal of contaminated solvents (typically generated by chromatogra- 1991, p . 327; Towler and Sinnott, Chemical Engineering Design; Principles,
phy steps) and other regulated compounds can become a major expense Practice and Economics of Plant and Process Design, Butterworth-Heinemann
because the unit disposal cost can be more than $1/kg . Waste disposal may (Elsevier), Oxford, U .K ., 2008, p . 366] . The examples presented next demon-
also become a problem if an unwanted by-product is generated as part of strate how these measures facilitate the decision-making process .
the recovery chemistry of a process (see the citric acid example) . Disposal Illustrative Examples In this section, SuperPro Designer is used to
of single-use systems via incineration costs $100 to $200 per metric ton of illustrate the analysis and evaluation of the production of three biological
material . products . The first example analyzes the production of citric acid, a com-
Utilities costs include the cost of heating and cooling agents as well modity organic acid heavily used in the beverage industry . The second deals
as electricity . The amounts are calculated as part of the material and with the bacterial production of recombinant human insulin, the first com-
energy balances . Aerobic fermenters are major consumers of electricity, mercial product of modern biotechnology . The third example focuses on
but downstream processing equipment generally does not consume much the production of monoclonal antibodies (mAbs) from mammalian cells
electricity . In terms of unit cost, electricity costs $0 .05 to $0 .15/kWh . The cultured in stirred-tank bioreactors . The generation of the flowsheets for
cost of heat removal using cooling water is in the range of $0 .002 to $0 .01 the production of all three products was based on information available in
per 1000 kcal of heat removed . The cost of cooling using chilled water and the patent and technical literature combined with engineering judgment
refrigerants is in the range of $0 .05 to $0 .1 per 1000 kcal of heat removed . and experience with other biological products . These examples are used to
The cost of producing steam for use as a heating medium is around $5 draw general conclusions on the manufacturing cost of biological products .
to $15/1000 kg depending on pressure (low, medium, high), type of fuel The computer files for these examples are available as part of the evaluation
used for its generation, and scale of production . The cost of clean steam version of SuperPro Designer at the website www .intelligen .com . The flow-
(generated utilizing highly purified water) is around $50 to $100/1000 kg sheets and charts of the examples are available in color format in the readme
(depending on the scale of production and level of water purity) . Clean files of the corresponding SuperPro Designer examples . Additional examples
steam is used in biopharmaceutical facilities for sterilizing equipment as and pertinent publications are available at www .intelligen .com/literature .
part of equipment cleaning (e .g ., “steam-in-place” or SIP operations) . Note Citric Acid Production A number of organic acids are produced via fer-
that manufacturers often classify purified water used for buffer prepara- mentation . Of these, citric acid is produced in the largest amount [> 1,800,000
tion and equipment cleaning as a utility and not as a raw material, thus metric tons (t) per year] . Citric acid is marketed as citric acid monohydrate
increasing the cost contribution of utilities . The insulin example, pre- or as anhydrous citric acid . The majority of citric acid (> 60 percent) is used
sented later in this section, describes a methodology for the systematic in the food and beverage industries to preserve and enhance flavor . In the
sizing of systems that supply purified water . chemical industries (which represent 25 to 30 percent of total utilization),
Facility overhead costs account for the depreciation of the fixed capital the uses of citric acid include the treatment of textiles, softening of water,
investment, maintenance costs for equipment, insurance, local (property) and manufacturing of paper . In the pharmaceutical industry (10 percent of
taxes, and possibly other overhead-type expenses . For preliminary cost esti- total utilization), citrate is used as a buffer and formulation excipient, iron
mates, the entire fixed capital investment is usually depreciated linearly citrate is used as a source of iron, and citric acid is used as a preservative for
over a 10-year period . In the real world, the U .S . government allows corpora- stored blood, tablets, and ointments, and in cosmetic preparations (Crueger
tions to depreciate equipment in 5 to 7 years and buildings in 25 to 30 years . and Crueger, Biotechnology—A Textbook of Industrial Microbiology, 2d ed .,
The value of land cannot be depreciated . The annual maintenance cost Sinauer, Sunderland, Mass ., 1989, p . 134) . Citric acid is increasingly being
can be estimated as a percentage of the equipment’s purchase cost (usually used in the detergent industry as a replacement for polyphosphates .
10 percent) or as a percentage of the overall fixed capital investment (usu- Citric acid was first recovered in 1869 in England from calcium citrate,
ally 3 to 5 percent) . Insurance rates depend to a considerable extent on the which was obtained from lemon juice . Its production by filamentous fungi
maintenance of a safe plant in good repair condition . A value for insurance has been known since 1893 . The first production via surface culture fer-
in the range of 0 .5 to 1 percent of DFC is appropriate for most bioprocessing mentation was initiated in 1923 . Production using stirred-tank fermenters
facilities . The processing of flammable, explosive, or highly toxic materials began in the 1930s, and presently this is the preferred method for large-scale
usually results in higher insurance rates . The local (property) tax is usually 2 manufacturing . The plant considered in this example produces around
to 5 percent of DFC . The factory expense represents overhead cost incurred 18,000 t/yr of crystal citric acid, which represents approximately 1 percent
by the operation of non-process-oriented facilities and organizations, such of the current world demand .
as accounting, payroll, fire protection, security, and cafeteria . A value of 5 to The entire flowsheet is shown in Fig . 20-64 . Molasses, the carbon source
10 percent of DFC is appropriate for these costs . of fermentation, is diluted with water from about 50 percent ferment-
Included in miscellaneous costs are ongoing R&D, process validation, able sugars content to 20 percent in a blending tank (V-101) . Suspended
and other overhead-type expenses that can be ignored in preliminary cost particulate material is then removed by filtration (PFF-101) . Metal ions,
estimates . Other general expenses of a corporation include royalties, adver- particularly iron, are subsequently removed by an ion exchange chroma-
tising, and selling . If any part of the process or any equipment used in the tography column (C-101), and the purified raw material solution is then
process is covered by a patent not assigned to the corporation undertaking heat-sterilized (ST-101) . Nutrients (i .e ., sources of ammonium, phosphorus,
the new project, permission to use the technology covered by the patent magnesium, potassium copper, and zinc) are dissolved in water (V-104) and
must be negotiated, and some form of royalty or license fee is usually heat-sterilized (ST-101) . (It is necessary to sterilize the first three nutrients
required . Advertising and selling covers expenses associated with the activi- separately to avoid a precipitate .) The fermentation cycle is 7 days, and the
ties of the marketing and sales departments . production is handled by seven fermenters that operate in staggered mode .
20-78

C-Source

Reg buffer
139.84 MT/batch Vent-1
S-104
Wash S-108

Water-1a
S-11'
Water-1b
S-109 P-6/ST-101
P-1/V-101 P-5/V-103
P-2/PFF-101 P-3/V-102 Heat sterilization
Mixing P-4/C-101 Storage S-112
Filtration Storage
Ion exchange
347.50 MT/batch

Amm sulfate
Fermentation section
Vent-2

Vent-3
Other salts P-12/AF-101
S-115
P-8/ST-101 27.96 MT/batch Air filtration
S-116
Heat sterilization
P-7/V-104 Vent-4
Mixing Air
S-117
Water-2 S-113 P-10/AF-102 S-114
Air filtration
P-13/V-105
Sulfuric acid P-11/FR-101
P-9/G-101 Storage
Fermentation
Gas compression
Vent-6 456.69 MT/batch
S-118
Water-4 Lime
Vent-5
Biomass Water-3 385.44 MT/batch
S-121
S-119

P-16/RVF-102 S-120
Aq-waste-1
P-17/V-107 Product recovery P-14/RVF-101
P-15/V-106 Biomass removal
Gypsum formation
433.06 MT/batch Product precipitation
Isolation section

S-122
S-123 S-126
562.62 MT/batch Water-6
Water-5
Gypsum Water vapor
S-125
509.37 MT/batch
P-21/RDR-101 Product
S-124 P-20/RVF-104 Air-2
Rotary drying
P-18/RVF-103 Product recovery Aq-waste-2
P-19/CR-101 54.80 MT/batch
Gypsum removal
Continuous crystallization

242.90 MT/batch

FIG. 20-64 Citric acid production flowsheet . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of
Oxford University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-79

Since the plant operates around the clock, one fermentation batch is initi- sulfate) generated . The quantities of these compounds depend on the chem-
ated daily and another one is completed daily . Each fermenter has a vessel istry of the purification process and cannot be reduced without changing
volume of 350 m3 and generates broth of around 315 m3 . A three-step seed the recovery technology . Since this gypsum is contaminated with biomass,
fermenter train (not shown in the flowsheet) supplies inoculum to each pro- it has little or no commercial value . A disposal cost of $50/MT (metric ton)
duction fermenter (FR-101) . A pure culture of the mold Aspergillus niger is was assumed in this example . The large amount of wastewater is also worth
used to inoculate the smallest seed fermenter . When optimum growth of noting .
mycelium is reached, the contents of the seed fermenter are transferred to A list of major equipment items along with their purchase costs (gener-
the next stage fermenter, which is approximately 10 times larger . Similarly, ated by SuperPro Designer) is shown in Table 20-33 . The total equipment
this larger seed fermenter inoculates the production fermenter with about cost for a plant of this capacity is around $10 .4 million . Note that approxi-
10 percent volume of actively growing mycelium broth . Air is supplied by a mately 30 percent of the equipment cost is associated with the seven pro-
compressor (G-101) at a rate that gradually increases from 0 .15 vvm (volume duction fermenters . The fermenters are made of stainless steel to minimize
of air per volume of liquid per minute) to 1 .0 vvm . Cooling water removes leaching of heavy metals that affect product formation . The final item, cost
the heat produced by the exothermic process (2990 kcal/kg of citric acid of unlisted equipment, accounts for the cost of the seed fermenters, pumps,
formed) and maintains the temperature at 28°C . The fermented broth is dis- and other secondary equipment that is not considered explicitly . Table 20-34
charged into the holding tank (V-105), which acts as a buffer tank between displays the various items of the direct fixed capital (DFC) investment . The
the batch upstream section and the continuous downstream section . total DFC for a plant of this capacity is around $43 .6 million or approxi-
Purification starts with the removal of biomass by a rotary vacuum filter mately 4 .2 times the total equipment cost .
(RVF-101) . The clarified fermentation liquor flows to an agitated reactor A summary of the operating costs is shown in Table 20-35 . The raw
vessel (V-106) where approximately 1 part of hydrated lime, Ca(OH)2, for materials cost is the most important, accounting for 40 .9 percent of the
every 2 parts of liquor is slowly added to precipitate calcium citrate . The overall operating cost . This is quite common for commodity biochemicals .
lime solution must be very low in magnesium content to minimize losses Molasses is the most expensive raw material, accounting for 67 percent of
due to generation of relatively soluble magnesium citrate . Calcium citrate is the raw materials cost . The purification chemicals, sulfuric acid and cal-
separated by a second rotary vacuum filter (RVF-102), and the citrate-free cium hydroxide, account for 18 .7 percent and 9 .3 percent of the overall raw
filtrate (Aq-Waste-1) is sent to a wastewater collection tank . The calcium
citrate cake is sent to another agitated reactor vessel (V-107), where it is
acidified with dilute sulfuric acid to form a precipitate of calcium sulfate
(gypsum) . A third filter (RVF-103) removes the precipitated gypsum and TABLE 20-33 Equipment Specification and Purchase Costs for
yields an impure citric acid solution in the filtrate . Careful control of the pH Citric Acid Production (Year 2012 Prices)
and temperature of each precipitation step is important for maximizing the Quantity Name Description Unit cost ($) Cost ($)
yield of citric acid . The resulting solution is concentrated and crystallized
using a continuous evaporator/crystallizer (CR-101) . The crystals formed 1 AF-101 Air filter 100,000 100,000
are separated by rotary vacuum filtration (RVF-104) and dried in a rotary Rated throughput = 7.7 m3/s
dryer (RDR-101) . If the final product is required in high purity, then treat- 1 AF-102 Air filter 37,000 37,000
ment with activated carbon may precede crystallization to remove colo- Rated throughput = 3.0 m3/s
rants . In addition, ion exchange is sometimes used to remove metal ions 1 C-101 Chromatography column 150,000 150,000
and other ionic species . Column volume = 4.66 m3
An equipment occupancy chart for consecutive batches can be developed 1 CR-101 Crystallizer 542,000 542,000
using SuperPro Designer (Harrison et al ., Bioseparations Science and Engi- Vessel volume = 130 m3
neering, Oxford University Press, Oxford, UK, 2015, p . 475) . The process batch 7 FR-101 Fermenter 436,000 3,052,000
time is approximately 200 h (or 8 .3 days) . This is the time elapsed from the Vessel volume = 350 m3
preparation of raw materials to the final product of a single batch (excluding 1 G-101 Centrifugal compressor 1,560,000 1,560,000
the time required for inoculum preparation) . The duration of each fermenta-
Compressor power = 1,430 kW
tion batch is 160 h (6 .7 days) . The availability of seven production fermenters
1 PFF-101 Plate & frame filter 145,000 145,000
operating in staggered mode (out of phase) enables the plant to initiate a
new batch every 24 h . The upstream portion of the process (i .e ., raw mate- Filter area = 335 m2
rial preparation and fermentation) operates in batch mode . The downstream 1 RDR-101 Rotary dryer 475,000 475,000
section (product recovery and purification) operates continuously . Drying area = 85 m2
A summary of the overall material balances (expressed as input-output 1 RVF-101 Rotary vacuum filter 154,000 154,000
component balance) is given in Table 20-32 . “CA crystal” stands for crys- Filter area = 46 m2
talline citric acid and represents the final product . Glucose represents the 1 RVF-102 Rotary vacuum filter 214,000 214,000
fermentable carbohydrates in molasses (50 percent w/w) . Note the large Filter area = 83 m2
amounts of Ca(OH)2 and sulfuric acid consumed and gypsum (calcium 1 RVF-103 Rotary vacuum filter 195,000 195,000
Filter area = 71 m2
1 RVF-104 Rotary vacuum filter 137,000 137,000
TABLE 20-32 Overall Material Balances for Citric Acid (CA)
Filter area = 35 m2
Production (kg/yr)
1 ST-101 Sterilizer 308,000 308,000
Component In Out Out – In Rated throughput = 34 m3/h
Ammonium sulfate 278,000 26,000 −252,000 1 V-101 Blending tank 503,000 503,000
Biomass 0 2,014,000 2,014,000 Vessel volume = 300 m3
CA crystal 0 18,250,000 18,250,000 1 V-102 Blending tank 503,000 503,000
Ca(OH)2 11,717,000 558,000 −11,159,000 Vessel volume = 300 m3
Calcium citrate 0 623,000 623,000 1 V-103 Blending tank 503,000 503,000
CO2 0 3,861,000 3,861,000 Vessel volume = 300 m3
Citric acid 0 657,000 657,000 1 V-104 Blending tank 139,000 139,000
Glucose 22,934,000 275,000 −22,659,000 Vessel volume = 35 m3
Gypsum 0 19,972,000 19,972,000 2 V-105 Flat bottom tank 198,000 396,000
Impurities 459,000 459,000 0 Vessel volume = 350 m3
Nutrients 1,894,000 383,000 −1,511,000 1 V-106 Neutralizer 126,000 126,000
Oxygen 65,171,000 57,994,000 −7,177,000 Vessel volume = 42 m3
NaOH 185,000 185,000 0 1 V-107 Neutralizer 94,000 94,000
Sulfuric acid 14,979,000 576,000 −14,403,000 Vessel volume = 15 m3
Water 308,879,000 320,663,000 11,784,000 Unlisted Eequipment 1,037,000
Total 426,496,000 426,496,000 0 Total 10,370,000
Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G. Reproduced from Bioseparations Science and Engineering, 2d ed., by Roger G.
Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission Harrison, Paul W. Todd, Scott R. Rudge, and Demetri P. Petrides (2015). By Permission
of Oxford University Press. of Oxford University Press.
20-80 BIOREACTIONS AND BIOPROCESSING

TABLE 20-34 Fixed Capital Estimate Summary for Citric Acid eliminates the consumption of Ca(OH)2 and H2SO4 and the generation of the
Production (Year 2012 Prices, $) unwanted CaSO4 . Butanol has been used as an extractant, as has tributyl
phosphate . Ion pair extraction by means of secondary or tertiary amines
Total plant direct cost (TPDC) dissolved in a water-immiscible solvent (e .g ., octyl alcohol) provides an
Equipment purchase cost 10,370,000 alternative route . Developments in electrodialysis membranes could lead to
Installation 3,726,000 recovering citric acid directly from the fermentation broth by this technique
Process piping 3,111,000 (Blanch and Clark, Biochemical Engineering, Dekker, New York, 1997, p . 611) .
Instrumentation 2,074,000 Human Insulin Production Insulin facilitates the metabolism of car-
Insulation 311,000 bohydrates and is essential for the supply of energy to the cells of the body .
Electrical 1,037,000 Impaired insulin production leads to the disease diabetes mellitus, which
Buildings 2,074,000 is a significant cause of death and health problems in industrialized and
Yard improvement 1,556,000 developing countries [Barfoed, Chem. Eng. Prog . 83: 49 (1987); World Health
Auxiliary facilities 1,037,000 Organization, fact sheet no . 310, 2011] .
TPDC 25,295,000 Human insulin is a polypeptide consisting of 51 amino acids arranged
Total plant indirect cost (TPIC) in two chains: chain A with 21 amino acids and chain B consisting of 30
Engineering 5,059,000 amino acids . The A and B chains are connected by two disulfide bonds .
Construction 7,589,000 Human insulin has a molecular weight of 5808 and an isoelectric point of
TPIC 12,648,000 5 .4 . Human insulin can be produced by four different methods:
Total plant cost (TPC = TPDC + TPIC) 37,943,000 • Extraction from human pancreas
Contractor’s fee 1,897,000
• Chemical synthesis via individual amino acids
Contingency 3,794,000
• Conversion of pork insulin, or “semisynthesis”
• Fermentation of genetically engineered microorganisms
Direct fixed capital (DFC) 43,634,000 Extraction from the human pancreas cannot be practiced because the
TPC + contractor’s fee and contingency availability of raw material is so limited . Total synthesis, while technically
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
feasible, is not economically viable because the yield is very low . Production
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission based on pork insulin, also known as semisynthesis, transforms the porcine
of Oxford University Press . insulin molecule into an exact replica of the human insulin molecule by sub-
stituting a single amino acid, threonine, for alanine in the G-30 position .
This technology has been developed and implemented by Novo Nordisk A/S
materials cost, respectively . The following prices were assumed: $0 .15/kg of (Denmark) . However, this option is also quite expensive because it requires
molasses, $0 .013/kg of 10 percent w/w H2SO4 solution, $0 .08/kg of Ca(OH)2, the collection and processing of large amounts of porcine pancreases . In
and $0 .1/m3 of process water . The facility-dependent cost is the second addition, the supply is limited by the availability of porcine pancreas .
most important, accounting for 28 .7 percent of the overall cost . Depre- At least three alternative technologies have been developed for produc-
ciation of the fixed capital investment and maintenance of the facility are ing human insulin based on fermentation and utilizing recombinant DNA
the main contributors to this cost . Utilities are the third largest expense, technology [Ladisch and Kohlmann, Biotechnol. Prog . 8: 469 (1992)] . The
accounting for 14 .8 percent of the overall cost . Electricity and cooling water first successful technique of biosynthetic human insulin (BHI) production
utilized by the fermenters are the main contributors to this cost . Labor lies based on recombinant DNA technology was the two-chain method . This
in the fourth position, and the environmental cost (waste treatment/dis- technique was developed by Genentech, Inc . (South San Francisco) and
posal) is fifth . Disposal unit costs of $1/m3 and $50/MT (metric ton) were scaled up by Eli Lilly and Company (Indianapolis, Ind .) . Each insulin chain
assumed for liquid and solid (gypsum) waste streams, respectively . The dis- is produced as a β-galactosidase fusion protein in Escherichia coli, form-
posal of gypsum accounts for 85 percent of the overall environmental cost . ing inclusion bodies . The two peptide chains are recovered from the inclu-
The overall unit production cost is approximately $1 .4/kg, which is roughly sion bodies (IBs), purified, and combined to yield human insulin . Later, the
equal to the early 2012 selling price of citric acid [Saleem, http://www .prlog . β-galactosidase operon was replaced with the tryptophan (Trp) operon,
org/11293569-china-citric-acid-price-trend-outlook-2011 .html (2011)] . This resulting in a substantial yield increase .
can be explained by noting the excess citric acid production capacity around The so-called intracellular method of making proinsulin eliminates the
the world (which keeps profit margins low), and the fact that most operating need for the separate fermentation and purification trains required by
citric acid plants are rather old and partially depreciated . If depreciation is the two-chain method . Intact proinsulin is produced instead . The proin-
ignored, the facility-dependent cost is reduced by more than 80 percent, and sulin route has been commercialized by Eli Lilly [Kehoe, in Sikdar et al .,
the overall unit cost drops to around $1/kg . (eds .), Frontiers in Bioprocessing, CRC Press, Boca Raton, Fla ., 1989, p . 45] .
Based on the preliminary evaluation of this project idea, one should not Figure 20-65 shows the key transformation steps . The E. coli cells overpro-
recommend investing in new citric acid production capacity unless there duce Trp-LE´-Met-proinsulin (Trp-LE´-Met is a 121-amino acid peptide
is a combination of favorable conditions . Obviously, availability of inexpen- signal sequence; proinsulin, with 82 amino acids, is a precursor to insulin)
sive equipment (e .g ., by acquiring an existing facility) and raw materials in the form of inclusion bodies, which are recovered and solubilized . Proin-
(e .g ., by locating the plant near a source of low-cost molasses) are the most sulin is released by cleaving the methionine linker using cyanogen bromide
important factors . Development or adoption of a superior technology may (CNBr) . The proinsulin chain is subjected to a folding process to allow inter-
also change the attractiveness of citric acid production . Such a technology molecular disulfide bonds to form; and the C peptide, which connects the
is actually available; it utilizes extraction for citric acid recovery [Roberts, A and B chains in proinsulin, is then cleaved with enzymes to yield human
in McKetta and Cunningham (eds .), Encyclopedia of Chemical Processing insulin . A number of chromatography and membrane filtration steps are
and Design, vol . 8, Dekker, New York, 1979, p . 324] . Recovery by extraction required to purify the product .
A second (extracellular) method of producing proinsulin was devel-
oped by Novo Nordisk A/S . It is based on yeast cells that secrete insulin as
a single-chain insulin precursor [Barfoed, Chem. Eng. Prog. 83: 49 (1987);
TABLE 20-35 Operating Cost Summary for Citric Acid Production World Health Organization, fact sheet no . 310, 2011] . Secretion simplifies
(Year 2012 Prices) product isolation and purification . The precursor contains the correct
Citric acid Annual Proportion disulfide bridges and is therefore identical to those of insulin . It is converted
Cost item crystals ($/kg) cost ($/yr) of total (%) to human insulin by transpeptidation in organic solvent in the presence of
a threonine ester and trypsin followed by deesterification . Another advan-
Raw materials 0 .57 10,310,000 40 .92 tage of the secreted proinsulin technology is that by employing a continuous
Facility-dependent 0 .40 7,223,000 28 .67 bioreactor–cell separator loop, it is possible to reuse the cells .
Labor 0 .12 2,102,000 8 .34 In this example, we analyze a process based on the intracellular proinsu-
Consumables 0 .00 15,000 0 .06 lin method . The economics of this process were studied previously by Datar
Lab/QC/QA 0 .01 210,000 0 .83 and Rosen, in Asenjo (ed .), Separation Processes in Biotechnology, Dekker,
Waste treatment and disposal 0 .09 1,611,000 6 .39 New York, 1990, p . 741, and by Petrides et al ., Biotechnol. Bioeng. 5: 529 (1995) .
Utilities 0 .21 3,724,000 14 .78 The annual world demand for insulin and insulin analogs was over
Total 1 .40 22,195,000 100 .00 50,000 kg in 2010 and was growing at an annual rate of around 20 percent
[Nair, BioPharm Int . 24: 11 (2011)] . The plant analyzed in this example has
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
a capacity of around 1800 kg of purified biosynthetic human insulin (BHI)
of Oxford University Press . per year . This is a relatively large plant for producing polypeptide-based
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-81

Biomass The IBs are recovered in the heavy phase (with a yield of 98 percent), while
Cell harvesting most of the cell debris particles remain in the light phase . This is possible
because the density (1 .3 g/cm3) and size (diameter about 1 mm) of the IBs
Cell disruption
are significantly greater than those of the cell debris particles . The IB sludge,
Inclusion bodies which contains approximately 20 percent solids w/w, is washed with WFI
IB recovery containing 0 .66 percent w/w Triton-X100 detergent (the volume of solution
IB dissolution is twice the volume of inclusion body sludge) and recentrifuged (P-14) using
the same centrifuges as before (DS-101) . The detergent solution facilitates
Trp-LE’-Met-Proinsulin purification (dissociation of debris and soluble proteins from inclusion bod-
CNBr cleavage ies) . The exit temperature is maintained at 10°C . The slurry volume at the
end of the primary recovery section is around 1440 L .
Proinsulin (unfolded) The inclusion body suspension is transferred to a glass-lined reaction
tank (V-101) and is mixed with urea and 2-mercaptoethanol to final con-
Oxidative sulfitolysis centrations of 300 g/L (5 M) and 40 g/L, respectively . Urea is a chaotropic
agent that dissolves the denatured protein in the inclusion bodies, and
Proinsulin-SSO3 2-mercaptoethanol is a reductant that reduces disulfide bonds . A reaction
Folding, S-S bond formation time of 8 h is required to reach a solubilization yield of 95 percent . The inclu-
sion bodies are composed of 80 percent w/w Trp-LE´-Met-proinsulin, with
Proinsulin (refolded) the remainder being other (contaminant) proteins . At the end of the solu-
bilization reaction, a diafiltration unit (DF-101) is used to replace urea and
Enzymatic conversion 2-mercaptoethanol with WFI and to concentrate the solution . This opera-
tion is performed in 6 h with a recovery yield of 98 percent . All remaining
Insulin (crude) fine particles (biomass, debris, and inclusion bodies) are removed by means
of a polishing dead-end filter (DE-101) . This polishing filter protects the
Purification chromatographic units that are used further downstream . The solution vol-
ume at this point is around 2200 L .
Purified human insulin
The fusion protein is cleaved with CNBr (cyanogen bromide) into the signal
FIG. 20-65 Human insulin from proinsulin fusion protein . [Reproduced from Biosep- sequence Trp-LE´-Met, which contains 121 amino acids, and the denatured
arations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . proinsulin (82 amino acids) in a glass-lined reactor (V-102) . The reaction is
Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford University Press .] carried out in a 70 percent formic acid solution containing 30-fold molar
excess CNBr (stoichiometrically, 1 mol of CNBr is required per mole of Trp-
LE´-Met-proinsulin) . The reaction takes 12 h at 20°C and reaches a yield of
95 percent . The mass of the released proinsulin is approximately 30 percent
biopharmaceuticals . The plant operates around the clock, 330 days a year . of the mass of Trp-LE´-Met-proinsulin . A small amount of cyanide gas is
A new batch is initiated every 48 h, resulting in 160 batches per year . The formed as a by-product of the cleavage reaction . Detailed information on
fermentation broth volume per batch is approximately 37 .5 m3 . CNBr cleavage is available in the patent literature (Di Marchi, U .S . Patent
The entire flowsheet for the production of BHI is shown in Fig . 20-66 . It 4,451,396, 1984) . The formic acid, unreacted CNBr, and generated cyanide
is divided into four sections: fermentation, primary recovery, reactions, and gas are removed by applying vacuum and raising the temperature to around
final purification . Note: A “section” in SuperPro is simply a set of unit proce- 35°C (the boiling point of CNBr) . This operation is carried out in a rotary
dures (processing steps) . If you open the file “insulin .spf ” in SuperPro, you vacuum evaporator (CSP-101) and takes 1 h . Since cyanide gas is toxic, all
will see that all the unit procedures within a given section have their own air exhausted from the vessels is scrubbed with a solution of hypochlorite,
distinctive color (blue, green, purple, and black for fermentation, primary which is prepared and maintained in situ (Bobbit and Manetta, U .S . Patent
recovery, reactions, and final purification, respectively) . 4,923,967, 1990) .
Fermentation media are prepared in a stainless-steel tank (BT-101) and Sulfitolysis of the denatured proinsulin takes place in a glass-lined reac-
sterilized in a continuous heat sterilizer (ST-101) . The axial compressor tor (V-103) under alkaline conditions (pH 9–11) . This operation is designed
(G-101) and the absolute filter (AF-101) provide sterile air and ammonia to unfold proinsulin, break any disulfide bonds, and add SO3 moieties
to the fermenter at an average rate of 0 .5 vvm . A two-step seed fermenter to all sulfur residues on the cysteines . The product of interest is human
train (not shown in the flowsheet) is used to inoculate the 50-m3 production proinsulin(S—SO3—)6 (protein–S–sulfonate) . The sulfitolysis step is neces-
fermenter (FR-101) with transformed E. coli cells . These cells are used to sary for two reasons: (1) The proinsulin probably is not folded in the correct
produce the Trp-LE´-Met-proinsulin precursor of insulin, which is retained configuration when expressed in E. coli as part of a fusion protein, and (2)
in the cellular biomass . The fermentation time in the production fermen- the cyanogen bromide treatment tends to break existing disulfide bonds .
ter is about 18 h, and the fermentation temperature is 37°C . The final con- The final sulfitolysis mixture contains 50 percent w/w guanidine HCl (6 M),
centration of E. coli in the production fermenter is about 30 g/L (dry cell 0 .35 percent ammonium bicarbonate (NH4HCO3), 3 percent Na2SO3, and
weight) . The Trp operon is turned on when the E. coli fermentation runs out 1 .5 percent Na2S4O6 (Di Marchi, U .S . Patent 4,451,396, 1984) . A reaction time
of tryptophan . The chimeric protein Trp-LE´-Met-proinsulin accumulates of 12 h is required to reach a yield of 95 percent . The presence of the dena-
intracellularly as insoluble aggregates (inclusion bodies), and this decreases turing reagent (guanidine HCl) prevents refolding and cross-folding of the
the rate at which the protein is degraded by proteolytic enzymes . In the base same protein molecule onto itself or two separate protein molecules onto
case, it was assumed that the inclusion bodies (IBs) constitute 20 percent of each other . Urea may also be used as a denaturing reagent . Upon completion
total dry cell mass . At the end of fermentation, the broth is cooled down to of the sulfitolysis reaction, the sulfitolysis solution is exchanged with WFI
10°C to minimize cell lysis . After completion of each processing step in the to a final guanidine HCl concentration of 20 percent w/w . This procedure,
fermentation section (and subsequent sections), the equipment is washed P-21, utilizes the DF-101 diafilter that also handles buffer exchange after IB
to prepare for the next batch of product . solubilization . The human proinsulin(S—SO3—)6 is then chromatographi-
After the end of fermentation, the broth is transferred into a surge tank cally purified by means of three ion exchange columns (C-101) operating in
(VT-101), which isolates the upstream section from the downstream section parallel and each running four cycles per batch . Each column has a diam-
of the plant . Three disk-stack centrifuges (DS-101) operating in parallel are eter of 140 cm and a bed height of 25 cm . A cation exchange resin is used
used for cell harvesting . Note that a single unit procedure icon in the Super- (SP Sepharose Fast Flow from GE Healthcare Biosciences) operating at pH
Pro model may represent multiple equipment items operating in parallel . 4 .0 . The eluant solution contains 69 .5 percent w/w WFI, 29 percent urea,
During centrifugation, the broth is concentrated from 37,000 L to 9157 L, and and 1 .5 percent NaCl . Urea, a denaturing agent, is used to prevent incorrect
most of the extracellular impurities are removed . The cell recovery yield is refolding and cross-folding of proinsulin(S—SO3—)6 . The following operat-
98 percent . The cell sludge is diluted with an equal volume of buffer solution ing assumptions were made: (1) The column is equilibrated for 30 min prior
(buffer composition: 96 .4 percent w/w water for injection (WFI), 0 .7 percent to loading . (2) The total resin binding capacity is 20 mg/mL . (3) The eluant
EDTA, and 2 .9 percent Tris-base) in a blending tank (BT-102) . The buffer volume is equal to 5 column volumes (CVs) . (4) The total volume of the solu-
facilitates the separation of the cell debris particles from inclusion bodies . tions for column wash, regeneration, and storage is 15 CVs . (5) The protein
Next, a high-pressure homogenizer (HG-101) is used to break the cells and of interest is recovered in 1 .5 CVs of eluant buffer with a recovery yield of
release the inclusion bodies . The exit temperature is maintained at around 90 percent .
10°C . The same centrifuges as before (DS-101) are then used for inclusion A refolding operation catalyzes the removal of the SO3 moiety and then
body recovery (P-12) . The reuse of these centrifuges can be seen by not- allows disulfide bond formation and correct refolding of the proinsulin to
ing that procedures P-9 and P-12 have the same equipment name, DS-101 . its native form . It takes place in a reaction tank (V-104) . This process step
20-82

Media EDTA solution Primary recovery section Triton-X-100

Fermentation section P-6/AF-102
Air filtration 9.30 MT/batch
2.85 MT/batch
Water Liq Waste 2
P-2/ST-101
Heat sterilization Liq waste 1
P-1/BT-101
Mixing S-142
38.01 MT/batch P-11/HG-101 S-118
Ammonia P-10/BT-102 Homogenization P-12/DS-101
P-5/AF-101 P-9/DS-101 Blending/Storage Centrifugation P-13/BT-103
Air filtration P-7/FR-101 Centrifugation Blending/Storage
P-8/VT-101
Air Fermentation Storage
S-143
P-4/G-101
CNBr/HCOOH
Centrifugal compression 4.34 MT/batch

S-109 MrEtOH/Urea
11.04 MT/batch
WFI-1
Liq waste 3
9.55 MT/batch
S-122
WFI-2 S-127
Guan HCI S-114
P-16/DF-101 P-14/DS-101
P-17/DE-101 Centrifugation
P-19/CSP-101 Diafiltration P-15/V-101
Dead-End Filtration
Rotary evaporator P-18/V-102 Liq Waste 4 S-11' IB solubilization
P-21/DF-101 CNBr cleavage
P-20/V-103 Liq Waste 5
Diafiltration
Sulfitolysis
Liq Waste 6 S-125 S-158
12.55 MT/batch 8.37 MT/batch
S-146 S-159
S-147 S-160
Reactions section Enzymes
WFI-4
S-131
S-135 S-148
WFI-3 S-131 S-149
S-136 S-154
S-123
S-129
S-152
S-119
P-27/DF-102 Liq Waste 10 P-28/C-103
S-144 S-Sepharose Liq Waste 11
Diafiltration
P-24/DF-102 Liq Waste 8 P-25/C-102 P-26/V-105 S-138
18.16 MT/batch
P-22/C-101 P-23/V-104 Diafiltration HIC column Enzyme conversion
S-132
S-Sepharose Refolding 4.54 MT/batch
Liq Waste7 Liq Waste 9
S-161
55.40 MT/batch
89.27 MT/batch EtOH S-165 1.79 MT/batch
Find purification section
S-173

S-126 WFI-6
S-174 S-166
S-162 S-175 S-167
S-137 WFI-5
11.50 kg/batch

Product
P-36/FDR-101 S-157 P-33/DF-105
Freeze drying Diafiltration P-31/C-105 P-31/DF-104
P-34/V-106 P-30/C-104 P-29/DF-103
Gel filtration Diafiltration RP-HPLC Diafiltration
Liq Waste 17 P-35/BCF-101 Crystallization
Basket centrifugation Liq Waste 16 Liq Waste 12
Liq Waste 14
1.94 MT/batch Liq Waste 15 Liq Waste 13
1.61 MT/batch 2.62 MT/batch
40.40 MT/batch 13.16 MT/batch

FIG. 20-66 Insulin production flowsheet . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-83

involves treatment with mercaptoethanol (MrEtOH), a reductant that facili- recovered in 0 .5 CV of eluant buffer with a recovery yield of 90 percent . The
tates the disulfide interchange reaction . It is added at a ratio of 1 .5 mol of mobile phase is a solution of acetic acid .
mercaptoethanol to 1 mol of SO3 . Dilution to a proinsulin(S—SO3—)6 con- Next, a diafilter (DF-105) is used to concentrate the purified insulin solu-
centration of less than 1 g/L is required to prevent cross-folding of proinsu- tion by a factor of 10 . The solution volume at this point is around 180 L,
lin molecules . The reaction is carried out at 8°C for 12 h and reaches a yield which contains approximately 12 .8 kg of insulin . This material is pumped
of 85 percent . After completion of the refolding step, the refolding reagents into a jacketed and agitated tank (V-106) . Ammonium acetate and zinc
are replaced with WFI, and the protein solution is concentrated using a chloride are added to the protein solution until each reaches a final con-
diafiltration unit (DF-102), which has a product recovery yield of 95 percent centration of 0 .02 M [Datar and Rosen, in Asenjo (ed .), Separation Processes
(5 percent of the protein denatures) . The volume of the solution at this point in Biotechnology, Dekker, New York, 1990, p . 741] . The pH is then adjusted to
is around 4500 L . Next, the human proinsulin is chromatographically puri- between 5 .4 and 6 .2 . The crystallization is carried out at 5°C for 12 h . Insulin
fied in a hydrophobic interaction chromatography (HIC) column (C-102) . The crystallizes with zinc with the following stoichiometry: insulin6-Zn2 . Step
following operating assumptions were made: (1) The column is equilibrated recovery on insulin is around 90 percent .
for 30 min prior to loading . (2) The total resin binding capacity is 20 mg/mL . The crystals are recovered with a basket centrifuge (BCF-101) with a yield
(3) The eluant volume is equal to 6 CVs . (4) The total volume of the solu- of 95 percent . Finally, the crystals are freeze-dried (FDR-101) . The purity of
tions for column wash, regeneration, and storage is 15 CVs . (5) The protein the crystallized end product is between 99 .5 and 99 .9 percent as measured
of interest is recovered in 1 CV of eluant buffer with a recovery yield of by analytical high-performance liquid chromatography (HPLC) . Approxi-
90 percent . (6) The material of a batch is handled in three cycles . mately 11 .5 kg of product is recovered per batch . The overall recovery yield
The removal of the C-peptide from human proinsulin is carried out is around 32 percent .
enzymatically (using trypsin and carboxypeptidase B) in a reaction vessel Table 20-36 displays the material requirements in kilograms per year, per
(V-105) . Trypsin cleaves at the carboxy-terminus of internal lysine and argi- batch, and per kilogram of main product (MP = purified insulin crystals) .
nine residues, and carboxypeptidase B removes terminal amino acids . The The solutions of H3PO4 (5 percent w/w) and NaOH (0 .5 M) are used for
amount of trypsin used is rate-limiting and allows intact human insulin to equipment cleaning . WFI is used for preparing all the buffers utilized in
be formed . Carboxypeptidase is added to a final concentration of 4 mg/L, product purification as well as all the cleaning solutions . Note the large
while trypsin is added to a final concentration of 1 mg/L . The reaction takes amounts of formic acid, urea, guanidine hydrochloride, acetic acid, and ace-
place at 30°C for 4 h and reaches a conversion yield of 95 percent . The vol- tonitrile required per kilogram of final product . All these materials end up
ume of the solution at this point is around 4300 L . in waste streams .
A purification sequence based on multimodal chromatography, which In the base case, this waste is treated and discarded . However, opportuni-
exploits differences in molecular charge, size, and hydrophobicity, is used ties may exist for recycling some chemicals for in-process use and recover-
to isolate biosynthetic human insulin . A description of all the purification ing others for off-site use . For instance, formic acid (HCOOH), acetonitrile,
steps follows . and urea are good candidates for recycling and recovery . Formic acid is used
The enzymatic conversion solution is exchanged with WFI and concen- in large quantities (11 tons per batch) in the CNBr cleavage step (V-102),
trated by a factor of 4 in a diafilter (DF-102) . An ion exchange column (C-103) and it is removed by means of a rotary vacuum evaporator (CSP-101), along
is used to purify the insulin solution . The following operating assumptions with small quantities of CNBr, H2O, and urea . The recovered formic acid can
were made: (1) The column is equilibrated for 30 min prior to loading . (2) be readily purified by distillation and recycled in the process . Around 2 .2
The total resin binding capacity is 20 mg/mL . (3) The eluant volume is equal metric tons per batch of urea is used for the dissolution of inclusion bodies
to 8 CVs, and the eluant is an 11 .5 percent w/w solution of NaCl in WFI . (V-101), and 17 metric tons per batch is used in the first chromatography
(4) The total volume of the solutions for column wash, regeneration, and step (C-101) to purify proinsulin(S—SO3)6 before its refolding . Approxi-
storage is 14 CVs . (5) The protein of interest is recovered in 1 .5 CV of eluant mately 90 percent of the urea appears in just two waste streams (Liq Waste
buffer with a recovery yield of 95 percent . (6) The material from each batch 4 and 8) . It is unlikely that these urea-containing streams can be purified
is handled in four cycles . The solution volume at this point is around 1780 L .
Next, the ion exchange eluant solution is exchanged with WFI in a diafil-
ter (DF-103) and is concentrated by a factor of 2 .0 . A recovery yield of 98 TABLE 20-36 Raw Material Requirements for Human Insulin
percent was assumed for this step (2 percent denatures) . Production: 1 Batch = 11.5 kg Main Product (MP = Purified Insulin
The purification of the insulin solution proceeds with a reversed-phase Crystals)
high-performance liquid chromatography (RP-HPLC) step (C-104) . Detailed Requirement
information on the use of RP-HPLC for insulin purification is available in the
literature . Analytical studies with a variety of reversed-phase systems have Raw material kg/year kg/batch kg/kg MP
shown that an acidic mobile phase can provide excellent resolution of insu-
Glucose 782,200 4,889 425 .1
lin from structurally similar insulinlike components . Minor modifications
Salts 71,400 446 38 .8
in the insulin molecule, resulting in monodesamido formation at the 21st
Water 9,715,000 60,719 5,279 .9
amino acid of the A chain, or derivatization of amines via carbamylation
or formylation results in insulin derivatives having significantly increased H3PO4 (5% w/w) 3,979,000 24,869 2,162 .5
retention . Derivatives of this nature are typical of the kind of insulinlike NaOH (0 .5 M) 3,842,000 24,013 2,088 .0
components that are found in the charge stream going into the reversed- WFI 51,890,000 324,313 28,201 .1
phase purification . The use of an acidic mobile phase results in the elution Ammonia 81,600 510 44 .3
of all the derivatives after the insulin peak, while the use of mildly alkaline EDTA 10,420 65 5 .7
pH results in derivatives eluted on either side of the parent insulin peak . TRIS base 43,160 270 23 .5
An ideal pH for insulin purification is in the region of 3 .0 to 4 .0, since this Triton-X-100 3,035 19 1 .6
pH range is far enough below the isoelectric pH of 5 .4 to provide for good MrEtOH 98,660 617 53 .6
insulin solubility . An eluant buffer with an acetic acid concentration of Urea 3,054,000 19,088 1,659 .8
0 .25 M meets these operational criteria because it is compatible with the CNBr 15,270 95 8 .3
chromatography and provides good insulin solubility . A 90 percent insulin Formic acid 1,752,000 10,950 952 .2
yield was assumed in the RP-HPLC step with the following operating condi- Guanidine-HCl 805,600 5,035 437 .8
tions: (1) the column is equilibrated for 30 min prior to loading; (2) the total Na2O6S4 24,160 151 13 .1
resin binding capacity is 15 mg/mL; (3) the column height is 25 cm; (4) the NH4HCO3 5,551 35 3 .0
eluant volume is 6 CVs and its composition is 25 percent w/w acetonitrile, Sodium sulfite 48,320 302 26 .3
1 .5 percent w/w acetic acid, and 73 .5 percent w/w WFI; (5) the total volume Sodium chloride 775,500 4,847 421 .5
of the solutions for column wash, equilibration, regeneration, and storage Acetic acid 975,700 6,098 530 .3
is 6 CVs; and (6) the protein of interest is recovered in 1 CV of eluant buffer Sodium hydroxide 137,200 858 74 .6
with a recovery yield of 90 percent . Enzymes 3 0 0 .0
The RP-HPLC buffer is exchanged with WFI and concentrated by a factor
Acetonitrile 764,700 4,779 415 .6
of 2 .0 in a diafilter (DF-104) that has a product recovery yield of 98 percent
Ammonium acetate acetate 181 1 0 .1
(2 percent denatures) . Purification is completed by a gel filtration chroma-
tography column (C-105) . The following operating assumptions were made: Zinc chloride 320 2 0 .2
(1) The column is equilibrated for 30 min prior to loading . (2) The sample Total 78,874,980 492,943 42,866 .8
volume is equal to 5 percent of the column volume . (3) The eluant volume Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
is equal to 4 CVs . (4) The total volume of the solutions for column wash, Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
depyrogenation, stripping, and storage is 6 CVs . (5) The protein of interest is of Oxford University Press .
20-84 BIOREACTIONS AND BIOPROCESSING

economically for in-process recycling . However, these solutions can be Utilities Waste disposal
concentrated, neutralized, and shipped off site for further processing and 0.2% 7.3%
utilization as a nitrogen fertilizer .
Approximately 4 .8 metric tons (MT) per batch of acetonitrile is used in Lab/QC/QA
the reversed-phase HPLC column (C-104), and most of it ends up in the Raw materials
1.8% 27.3%
waste stream of the column (Liq Waste 13) along with 6 .8 MT of water,
1 .85 MT of acetic acid, and small amounts of NaCl and other impurities . It is
unlikely that acetonitrile can be recovered economically to meet the high-
purity specifications for a step so close to the end of the purification train .
Consumables
However, there may be a market for off-site use .
The scheduling and equipment utilization for consecutive batches can 31.9%
be determined using SuperPro Designer (Harrison et al ., Bioseparations Sci-
ence and Engineering, Oxford University Press, Oxford, UK, 2015, p . 489) . The
batch time is approximately 11 days . This is the time required to go from the Facility-dependent
preparation of raw materials to final product for a single batch (excluding
inoculum preparation) . However, since most of the equipment items are uti- 24.1%
lized for much shorter periods within a batch, a new batch is initiated every Labor
48 h . The equipment with the least idle time between consecutive batches 7.3%
is the time (or scheduling) bottleneck (V-104 in this case) that determines the
maximum number of batches per year . Its occupancy time (approximately FIG. 20-67 Breakdown of manufacturing cost for human insulin production .
45 h) is the minimum possible time between consecutive batches . The pro- [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison,
duction line operates around the clock and processes 160 batches per year . Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
University Press .]
Process scheduling is closely related to the determination of the annual
capacity of a batch process . The last part of this example discusses how
changes in scheduling and installation of additional equipment can be used
to increase process throughput and reduce manufacturing cost . associated with fermentation . The other 92 .7 percent is associated with the
Another characteristic of batch processing is the variable demand for recovery and purification sections . Most of the cost is associated with the
resources (e .g ., labor, utilities, and raw materials) as a function of time . For reactions section because of the large amounts of expensive chemicals and
example, SuperPro Designer can be used for tracking the use of WFI and consumables required for purification .
for sizing the equipment for WFI needed to produce insulin by this process Finally, for each raw material used in the process, Table 20-38 displays
(Harrison et al ., Bioseparations Science and Engineering, Oxford University the price, annual cost, and contribution to the overall raw materials cost .
Press, Oxford, UK, 2015, p . 490) . WFI, urea, and H3PO4 (5 percent w/w) are the top three contributors to the
The results of the economic evaluation are shown in Table 20-37 . The raw materials cost . The H3PO4 and NaOH solutions are used for equipment
detailed tables for these calculations are available as part of the evaluation cleaning .
version of SuperPro Designer . For a plant of this capacity, the total capital Other assumptions for the economic evaluation include the following: (1)
investment is $178 million . The unit production cost is $61 per gram of puri- A new manufacturing facility will be built and dedicated to production of
fied insulin crystals . Assuming a selling price of $100/g, the project yields 1800 kg/yr of purified insulin . (2) The entire direct fixed capital is depreci-
an after-tax internal rate of return (IRR) of 35 percent and a net present ated linearly over a period of 10 yr . (3) The project life-time is 15 yr . (4) The
value (NPV) of $250 million (assuming a discount interest rate of 7 percent) . unit cost of membranes is $800/m2 and they are replaced every 50 cycles . (5)
Based on these results, this project represents a very attractive investment . The average unit cost of chromatography resins is $1500/L . (6) The waste
However, if amortization of up-front R&D costs is considered in the eco- disposal cost is $5/m3 for low BOD streams and $150/m3 for streams con-
nomic evaluation, the numbers change drastically . For instance, a modest taining significant amounts of solvents and other regulated chemicals .
amount of $100 million for up-front R&D cost amortized over a period of In the base case, a new batch is initiated every 48 h . Most of the equip-
10 yr reduces the IRR to 16 .8 percent and the NPV to $153 million . ment items, however, are utilized for less than 24 h per batch . If the market
The operating cost is broken down in Fig . 20-67 . The cost of consumables conditions are favorable, this provides the opportunity for increasing plant
is the most important, accounting for 31 .9 percent of the overall manu- throughput without major capital expenditures . A realistic improvement is
facturing cost . Consumables represent the expense of periodically replac- to initiate a batch every 24 h . This will require a new fermenter of the same
ing the resins of the chromatography columns and the membranes of the size as the original fermenter, whose operation will be staggered relative
membrane filters . The cost of raw materials is the second most important, to the existing unit so that one fermenter is ready for harvesting every day .
accounting for 27 .3 percent of the overall cost . The facility-dependent cost is Such a production change will also require additional equipment of the fol-
third, accounting for 24 .1 percent of the overall cost . This cost item accounts lowing types: (1) disk-stack centrifuges to reduce the occupancy of DS-101
for the depreciation and maintenance of the facility and other overhead to less than 24 h, (2) two reaction vessels to reduce the occupancy of V-103
expenses . Labor and Lab/QC/QA account for 9 .1 percent . The treatment and V-104, and (3) membrane filters to reduce the occupancy of DF-104 and
and disposal of waste materials accounts for 7 .3 percent of the total cost . As DF-105 .
mentioned in the material balance section, recycling and reuse of some of The additional capital investment for such a change is around $55 million .
the waste materials may reduce this cost . This additional investment will allow the plant’s capacity to be doubled, and
The percentage of the operating cost associated with each flowsheet
section is displayed in Fig . 20-68 . Only 7 .3 percent of the overall cost is Fermentation
7.3%

TABLE 20-37 Key Economic Evaluation Results for Final purification Primary recovery
Human Insulin Production 23.4% 8.0%
Direct fixed capital $145 million
Total capital investment $178 million
Plant throughput 1803 kg/yr
Manufacturing cost $110 million/yr
Unit production cost $61/g
Selling price $100/g
Reactions
Revenues $180 million/yr
61.3%
Gross profit $70 million/yr
Taxes (40%) $28 million/yr
IRR (after taxes) 35%
NPV ( for 7% discount interest rate) $250 million
FIG. 20-68 Cost distribution per flowsheet section for human insulin production .
Reproduced from Bioseparations Science and Engineering, 2d ed ., [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison,
by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
Petrides (2015) . By Permission of Oxford University Press . University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-85

TABLE 20-38 Cost of Raw Materials for Human Insulin Production to 2 .2-L roller bottles (RBR-101; shake flasks are frequently used instead
(Year 2012 Prices) of roller bottles for suspension cultures), then to 100-L and subsequently
to 200-L rocking bioreactors that utilize disposable bags (BBS-101 and
Bulk material Unit cost ($/kg) Annual cost ($) Percent BBS-102) . Sterilized media are fed at the appropriate amount in all four
Glucose 0 .60 469,300 1 .56 initial steps (3 .6, 11 .4, 43 .6, 175 .4 kg/batch, respectively) . The broth is then
Salts 1 .00 71,400 0 .24 moved to the first stirred seed bioreactor (DSBR-101), which utilizes 1000-L
Water 0 .05 485,800 1 .61 disposable bags . The second seed bioreactor (SBR-102) is a 5000-L stainless-
H3PO4 (5% w/w) 1 .00 3,979,000 13 .19 steel vessel . For the two seed bioreactors, the media powder is dissolved in
NaOH (0 .5 M) 0 .50 1,921,200 6 .37 WFI in two prep tanks (MP-101 and MP-102) and then sterilized/fed to the
WFI 0 .10 5,188,500 17 .20 reactors through 0 .2-µm dead-end filters (DE-101 and DE-102) . In the cell
Ammonia 0 .70 57,100 0 .19
culture section, serum-free low-protein media powder is dissolved in WFI
EDTA 18 .50 192,700 0 .64
in a stainless-steel tank (MP-103) . The solution is sterilized using a 0 .2-µm
dead-end polishing filter (DE-103) . A 20,000-L stainless stirred-tank bioreac-
TRIS base 6 .00 259,000 0 .86
tor (BR-101) is used to grow the cells, which produce the therapeutic mono-
Triton-X-100 1 .50 4,600 0 .02
clonal antibody (mAb) . The production bioreactor operates in fed-batch
MrEtOH 3 .00 296,000 0 .98 mode . High media concentrations are inhibitory to the cells, so one-half of
Urea 1 .52 4,642,200 15 .38 the media is added at the start of the process and the rest is fed at a variable
CNBr 11 .00 168,000 0 .56 rate during fermentation . The concentration of dry media powder in the ini-
Formic acid 1 .60 2,802,400 9 .29 tial feed solution is 17 g/L . The cell culture time is 12 days . The volume of
Guanidine-HCl 2 .15 1,732,000 5 .74 broth generated per bioreactor batch is approximately 15,000 L, which con-
Na2O6S4 0 .60 14,500 0 .05 tains roughly 30 .5 kg of product (the product titer is approximately 2 g/L) .
NH4HCO3 1 .00 5,600 0 .02 Between the downstream unit procedures there are 0 .2-µm dead-end fil-
Sodium sulfite 0 .40 19,300 0 .06 ters to ensure sterility . The generated biomass and other suspended com-
Sodium chloride 1 .23 954,000 3 .16 pounds are removed using a disk-stack centrifuge (DS-101) . During this
Acetic acid 2 .50 2,439,400 8 .08 step, roughly 5 percent of mAb is lost in the solids waste stream . The bulk
Sodium hydroxide 3 .50 480,300 1 .59 of the contaminant proteins are removed using a protein A affinity chroma-
Enzymes 500,000 1,691,100 5 .60 tography column (C-101) which processes a batch of material in four cycles .
Acetonitrile 3 .00 2,294,200 7 .60 The following operating assumptions were made for each chromatography
Ammonium acetate 15 .00 2,700 0 .01 cycle: (1) Resin binding capacity is 15 g of product per liter of resin . (2) The
Zinc chloride 12 .00 3,800 0 .01 eluant or elution buffer is a 0 .6 percent w/w solution of acetic acid, and its
Total 30,174,100 100 .00 volume is equal to 5 column volumes (CVs) . (3) The product is recovered
in 2 CVs of eluant with a recovery yield of 90 percent . (4) The total volume
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . of the solution required for column equilibration, wash and regeneration is
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission 14 CVs . The entire procedure takes approximately 27 h and requires a resin
of Oxford University Press . volume of 502 liters . The protein solution is then concentrated fivefold and
diafiltered with two volumes of buffer (in P-21/DF-101) . This step takes
approximately 8 .4 h and requires a membrane of 21 m2 . The product yield
the new unit production cost will be around $55/g . The reduction in the unit is 97 percent . The concentrated protein solution is then chemically treated
production cost is rather small because the majority of the manufacturing for 1 .5 h with Polysorbate 80 to inactivate viruses (in P-22/V-111) . The ion
cost is associated with consumables and raw materials that scale approxi- exchange (IEX) chromatography step (P-24/C-102) that follows processes
mately linearly with production . one batch of material in three cycles . The following operating assumptions
Therapeutic Monoclonal Antibody Production Monoclonal antibod- were made for each cycle: (1) The resin’s binding capacity is 40 g of product
ies are large protein molecules that consist of two main regions, the Fab per liter of resin . (2) A gradient elution step is used with a sodium chloride
( fragment antigen binding) region and the Fc ( fragment crystallizable) concentration ranging from 0 .0 to 0 .1 M and a volume of 5 CVs . (3) The prod-
region . Monoclonal antibodies (mAbs) are the fastest-growing segment in uct is recovered in 2 CVs of eluent buffer with a mAb yield of 90 percent .
the biopharmaceutical industry . Currently mAbs are used to treat various (4) The total volume of the solutions required for column equilibration,
types of cancer, rheumatoid arthritis, psoriasis, severe asthma, macular wash, regeneration, and rinse is 16 CVs . The step takes approximately 22 .3 h
degeneration, multiple sclerosis, and other diseases . More than 20 mAbs and requires a resin volume of 210 L . Ammonium sulfate is then added to
and Fc fusion proteins are approved for sale in the United States and Europe, the IEX eluate (in P-25/V-109) to a concentration of 0 .75 M . This increases
and approximately 200 mAbs are in clinical trials for a wide variety of indi- the ionic strength of the eluate in preparation for the hydrophobic interac-
cations [Walsh, Trends Biotechnol . 23: 553 (2005); Pavlou and Belsey, Eur. J. tion chromatography (HIC, P-26/C-103) step that follows . Like the IEX step
Pharm. Biopharm . 59: 389 (2005)] . The market size for mAbs in 2010 was in which preceded it, the HIC step processes one batch of material in three
excess of $35 billion [Rader, Contract Pharma, July/Aug . 2012] . cycles . The following operating assumptions were made for each cycle of
The high-dose demand for several mAbs translates into annual produc- the HIC step: (1) The resin binding capacity is 40 g of product per liter of
tion requirements for purified product in the metric ton range . This exam- resin . (2) The eluant is a sodium chloride (4 percent w/w) sodium dihydrogen
ple illustrates the analysis of a large-scale mAb process . Again, the modeling phosphate (0 .3 percent w/w) solution, and its volume is equal to 5 CVs .
and calculations are performed with SuperPro Designer . (3) The product is recovered in 2 CVs of eluant buffer with a recovery yield of
The flowsheet of the overall process is displayed in Fig . 20-69 . The gen- 90 percent . (4) The total volume of the solutions required for column equili-
eration of the flowsheet was based on information available in the patent bration, wash, and regeneration is 12 CVs . The step takes approximately
and technical literature combined with the authors’ engineering judgment 22 h and requires a resin volume of 190 L . A viral filtration step (DE-105)
and experience with such processes [Kelley, Biotechnol. Prog. 23: 995 (2007)] . follows . It is a dead-end type of filter with a pore size of 0 .02 µm . Finally,
The process in this example produces 1544 kg/yr of purified mAb . The flow the HIC elution buffer is exchanged for the product bulk storage (PBS) buffer
diagram of Fig . 20-69 is a simplified representation of the actual process and concentrated 1 .5-fold (in DF-102) . This step takes approximately 20 h
because it lacks all the buffer preparation and holding activities . Such pro- and requires a membrane area of 10 m2 . The approximately 800 L of final
cesses require 20 to 30 buffer solutions for product purification . These solu- protein solution is stored in twenty 50-L disposable storage bags (DCS-101) .
tions are prepared in mixing tanks and then stored in holding tanks located The overall yield of the downstream operations is 63 .2 percent, and 19 .3 kg
close to the purification train . The tanks required for buffer preparation of mAb is produced per batch .
and holding add to the capital investment of the facility, while the required The equipment occupancy chart of the process for consecutive batches
labor adds to the manufacturing cost . The model files for this example that can be drawn using SuperPro Designer (Harrison et al ., Bioseparations Sci-
are part of the evaluation version of SuperPro Designer (available at www ence and Engineering, Oxford University Press, Oxford, UK, 2015, p . 498) . The
.intelligen .com) include the tanks for buffer preparation and holding . In schedule represents a plant that has a single production train . The batch
addition, the capital and operating costs associated with buffer prepara- time is approximately 50 days . This is the time required from the start of
tion activities were considered in the cost analysis results presented in this inoculum preparation to the final product purification of a single batch .
example . Additional information on this example in the form of two tutorial A new batch is initiated every 2 weeks (14 days) . The production bioreac-
videos is available at www .intelligen .com/videos . tor (BR-101) is the time (scheduling) bottleneck . On an annual basis, the
The upstream part is split into two sections: the Inoculum Preparation plant processes 20 batches and produces approximately 386 kg of purified
section and the Bioreaction section . The inoculum is initially prepared in mAb . The analysis revealed that under these conditions the downstream
225-mL T-flasks (TFR-101) . Next, the material from the T-flasks is moved train is underutilized, and the cycle time of the process—the time between
20-86

ProtA-Equil.

ProtA-Wash
S-001
S.003
Inoculum Prep Primary recovery
ProtA-Elut
Protein-A
ProtA-Reg.
3.62 L/batch 15.07 L/batch

P-1/TFR-101 P-2/RBR-101 S-004

T-Flask (225 mL) Roller Bottle (2.2 L) P-16/DE-108
P-14/V-101 P-15/DS-101 P-19/DE-109
Surge Tank Centrifugation Polishing Filter P-18/C-101
P-17/V-103 Polishing Fitler S-038
S-006 S-035 Centrifugation Protein A Column ProtA-Waste
S-008 S-009
S-033 46.84 m3/batch 3.86 m3/batch
Pool Tank

58.79 L/batch
S-043
S-005 S-041 Chemical virus inactivation
P-3/BBS-101 P-4/BBS-102
Bag Bioreactor (100 L) Bag Bioreactor (200 L) S-2 S-110
S-011 234.66 L/batch S-045
S-103 S-102
S-101
S-044
P-20/V-107 P-22/V-111
S-9 P-21/DF-101 P-23/DE-110
Virus Inactivation
Storage S-042 Diafiltration Polishing Fiter
S-012 S-048
771.84 L/batch
772.02 L/batch
P-6/MP-101 P-7/DE-101
Sterile Filtration P-5/DSBR-101
Media Prep Bag Bioreactor S-12
S-10 944.44 L/batch HIC-Equil
IEX-Equiil.
IEX-Wash Amm. Sulfate
S-018 HIC-Wash HIC Chrom
IEX-WFI IEX Chrom HIC-El

S-019 IEX-EI
IEX-Strip HIC-Reg.
P-10/DE-102 P-8/SBR-102 P-27/DE-106
P-9/MP-101 IEX-Rinse
Sterile Filtration Seed Bioreactor (5 m3) P-25/V-109 Dead-End Filtration
Media Prep IEX Pool Tank 1.35 m3/batch
S-052
S-5 P-24/C-102 P-26/C-103
S-025 S-6 IEX-Waste HIC HIC-Waste 1.15 m3/batch
3.81 m3/batch IEX
3 Chromatography 9.80 m3/batch
Chromatography 12.67 m /batch
Cell Culture
S-8
Viral exclusion Final filtration
S-024 S-058
S-057

P-12/MP-103 P-13/DE-103
S-025b Media Prep Sterile Filtration 766.28 L/batch
S-105
S-104
15.31 m3/batch P-28/V-108 Final Product
P-29/DE-105 P-30/V-110 P-32/DE-107 P-33/DCS-101
HIC Pool Tank Storage Freeze in 50 L
Viral Exclusion P-31/DF-102 Final Polishing
S-059 Plastic Bags
P-11/BR-101 Filtration Diafiltration S-060 Filtration
S-024b Production Bioreactor (20 m3) 766.28 L/batch
P-35/DE-104
P-34/MP-104 Sterile Filtration
Media Prep S-1

FIG. 20-69 Monoclonal antibody production flowsheet (L = liters) . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides
(2015) . By Permission of Oxford University Press .]
 DOWNSTREAM PROCESSING: SEPARATION AND PURIFICATION 20-87

TABLE 20-39 Raw Material Requirements for mAb Production: TABLE 20-41 Breakdown of the Manufacturing Cost for mAb
1 Batch = 19.3 kg Main Product (MP= Purified mAb) Production
Material kg/yr kg/batch kg/kg MP Cost item $ million/yr $/g %
Inoc . media solution 18,630 232 .88 12 .066 Raw materials 16 .67 10 .8 12 .86
WFI 8,430,000 105,375 .00 5,459 .845 Facility-dependent 60 .54 39 .2 46 .71
Serum-free media 37,340 466 .75 24 .184 Labor 18 .89 12 .2 14 .58
H3PO4 (5% w/w) 2,277,000 28,462 .50 1,474 .741 Consumables 23 .58 15 .3 18 .19
NaOH (0 .5 M) 2,054,000 25,675 .00 1,330 .311 Lab/QC/QA 4 .45 2 .9 3 .43
NaOH (0 .1 M) 7,815,000 97,687 .50 5,061 .528 Miscellaneous 5 .47 3 .5 4 .23
Amm . sulfate 12,160 152 .00 7 .876 Total 129.60 83.9 100.00
Polysorbate 80 6 0 .08 0 .004
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
Protein A equil . buffer 1,967,000 24,587 .50 1,273 .964 Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
Protein A elution buffer 800,200 10,002 .50 518 .264 of Oxford University Press .
Protein A reg . buffer 480,400 6,005 .00 311 .140
NaCl (1 M) 184,200 2,302 .50 119 .301
IEX elution buffer 16,130 201 .63 10 .447 chromatography resins, membrane filters, and disposable bags that need to
IEX equil . buffer 664,900 8,311 .25 430 .635 be replaced on a regular basis . Labor and raw materials costs come third
HIC elution buffer 239,200 2,990 .00 154 .922
and fourth, accounting for 14 .6 percent and 12 .9 percent of the total cost,
respectively . The miscellaneous cost item (4 .2 percent of total) accounts for
HIC equil . buffer 449,600 5,620 .00 291 .192
heating/cooling utilities, electricity, and environmental costs . The cost of
Concentrated PBS 14,370 179 .63 9 .307
WFI, commonly classified as a utility cost in industry, is accounted for in
EtOH (10% w/w) 363,000 4,537 .50 235 .104
the cost of raw materials in this example . In terms of cost distribution per
Total 25,823,136 322,789.20 16,724.83 section, 62 percent of the cost is associated with the upstream section and
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . 38 percent with the downstream .
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission The economic evaluation relies on the following key assumptions: (1)
of Oxford University Press . A new manufacturing facility will be built and dedicated to production of
1544 kg/yr of mAb . (2) The entire direct fixed capital is depreciated linearly
over a period of 12 yr . (3) The project lifetime is 16 yr . (4) The unit cost of WFI
is $0 .15/L . (5) The cost of the serum-free media (in powder form) is $300/kg .
consecutive batches—is relatively long . The cycle time of the process can (6) All the chemicals used are high-purity grade . (7) The unit cost of mem-
be reduced and the plant throughput increased by installing multiple bio- branes is $400/m2 . (8) The unit cost of chromatography resins is $6000/L,
reactor trains that operate in staggered mode (out of phase) and feed the $1200/L, and $2050/L for columns C-101, C-102, and C-103, respectively .
same purification train . For a case of four bioreactor trains feeding the same (9) The chromatography resins are replaced every 60, 50, and 50 cycles for
purification train, the new cycle time is 3 .5 days, which is one-fourth of the columns C-101, C-102, and C-103, respectively .
original . Under these conditions, the plant processes 80 batches per year After a model of the entire process has been developed on the computer,
and produces 1544 kg/yr of mAb . Some biopharmaceutical companies have tools such as SuperPro Designer can be used to ask and readily answer
installed more than four bioreactor trains per purification train in order to “what if ?” questions and to carry out sensitivity analysis with respect to key
achieve cycle times as low as 2 days . design variables . In this example, the impact of product titer (varied from 1
The material requirements of the process are summarized in Table 20-39 . to 10 g/L) and bioreactor size (10,000 and 20,000 L) on unit production cost
Note the large amount of WFI utilized per batch . The majority of WFI is is evaluated . Figure 20-70 displays the results of the analysis . All points cor-
consumed for cleaning and buffer preparation . respond to four production bioreactors feeding a single purification train .
Key economic evaluation results generated using the built-in cost func- For low product titers, the bioreactor volume has a considerable effect on
tions of SuperPro Designer are displayed in Table 20-40 . The total capital the unit production cost . For instance, for a product titer of 1 g/L, going
investment ( for the case with the four bioreactor trains) is around $477 from 10,000 to 20,000 L of production bioreactor volume reduces the unit
million . The total annual operating cost is $130 million, resulting in a unit cost from $230/g to $162/g . On the other hand, for high product titers (e .g .,
production cost of around $84/g (1544 kg of purified mAb is produced annu- around 5 g/L), the impact of bioreactor scale is not as important . This can be
ally) . Assuming a selling price of $200/g, the project yields an after-tax inter- explained by the fact that at high product titers, the majority of the manu-
nal rate of return (IRR) of 24 .3 percent and a net present value (NPV) of $560 facturing cost is associated with the purification train . It is therefore wise
million (assuming a discount interest rate of 7 percent) . to shift R&D efforts from cell culture to product purification as the product
A breakdown of the operating cost contributors is presented in titer in the bioreactor increases . A key assumption underlying the sensitivity
Table 20-41 . The facility-dependent cost is the most important item, analysis is that the composition and cost of the cell culture media are inde-
accounting for 46 .7 percent of the manufacturing cost or $39 .2/g of final pendent of product titer .
product . This is common for high-value products that are produced in small
quantities in expensive facilities . Depreciation of the fixed capital invest-
ment and maintenance of the facility are the main contributors to this cost . 250
Consumables are the second most important operating cost, accounting for
18 .2 percent of the total or $15 .3/g of final product . Consumables include
200
Production cost ($/g)

150
TABLE 20-40 Key Economic Evaluation Results for mAb 4 × 10 kL
Production
4 × 20 kL
Direct fixed capital $365 million 100
Total capital investment $477 million
Plant throughput 1544 kg of mAb/yr
50
Manufacturing cost $130 million/yr
Unit production cost $84/g of mAb
Selling price $200/g of mAb 0
Revenues $309 million/yr 0 1 2 3 4 5 6 7 8 9 10 11
Gross profit $179 million/yr Product titer (g/L)
IRR (after taxes) 24 .3%
NPV ( for 7% discount interest rate) $560 million FIG. 20-70 The mAb production cost as a function of product titer and production
bioreactor volume (L = liters) . The arrow indicates the product titer for this example .
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison,
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
of Oxford University Press . University Press .]
20-88 BIOREACTIONS AND BIOPROCESSING

PRODUCT QUALITY ATTRIBUTE CONTROL

INTRODUCTION In the case of chemical or synthetic organic pharmaceutical manufactur-

ing, target species PQA and undesired by-products are, for the most part,
This section focuses on the implementation of product quality attribute well characterized, and their mode of action is understood . Sample collec-
control (PQAC) for biopharmaceutical product manufacturing . PQAC of tion and analysis methods have been well established, so that when they are
these products is mainly concerned with controlling aspects of the prod- combined with process optimization and control strategies, a robust PQAC
uct that may impact the function of the product, such as patient outcomes . program can be implemented . The biopharmaceutical industry is begin-
The majority of the discussion is focused on the upstream portion of the ning to overcome technical hurdles, such as reliable automated sampling
manufacturing process . Process perturbations and environmental condi- and integrated analytical analysis, that have historically hampered PQAC
tions at this stage, where the macromolecules are actually created in the implementation .
host cells, can have the greatest impact on critical product quality attributes Defining CPQA for a biopharmaceutical is significantly more challenging
(CPQAs) . Application of PQAC in the downstream purification process, compared to defined chemical entities . Biopharmaceutical quality attri-
where unwanted species such as aggregates or charge variants are cleared butes include parameters that impact safety and efficacy in the patient that
from the product, is also discussed . may only be known by their effects in clinical studies or through limited in
Compared to chemical and small molecule pharmaceutical manufactur- vitro bioassays . Biopharmaceuticals are large, complex molecules whose
ing, PQAC in the biopharmaceutical industry has lagged by several decades . structures are related to CPQA, but may also vary during the course of an
Application of PQAC to biopharmaceutical processing is challenging due upstream bioprocess . If biopharmaceutical lots having a distribution of prod-
to the complexity of the product and its mode of action, the cellular pro- uct attributes with unique biological responses are used in clinical develop-
cesses creating the product, unique issues with representative sampling and ment, then the commercial process may be required to match that CPQA
analysis, and the rigorous regulatory process they are subject to . Other bio- distribution, further challenging a deterministic PQAC approach . In the case
based processes, such as the manufacture of biofuel, can use many of the of biosimilars, target PQA are defined by matching the innovator’s molecule .
tools discussed here, but often enjoy a simplified PQAC approach due to the These targets are determined by off-line analysis of attributes such as gly-
relative simplicity of the molecules produced and the governing regulatory cosylation, deamidation, and aggregates . The control of these parameters
requirements . to match the innovator molecule is often a priority in the development and
Biopharmaceutical processes are routinely controlled for tempera- production of a biosimilar molecule and can benefit from a PQAC approach .
ture, pressure, pH, dissolved gases, and other critical process parameters On-line sensors exist for temperature, pH, and dissolved gases, and in situ
(CPPs) . Unlike in other industries, the linkage between the controlled cell mass and medium composition can be determined with dielectric and
process variables and product quality is not always clear . Without direct, Raman spectroscopy, respectively . However, there are no near-term on-line
near real-time CPQA measurement, implementation of PQAC is very solutions to directly measure product quality attributes . Therefore, a key ele-
difficult . As a result, upstream biopharmaceutical processes have often ment of any PQAC scheme is a means to automatically collect and analyze rep-
been operated as a black box . This is an open-loop or recipe-based con- resentative bioreactor samples . Figure 20-71 shows an example PQAC process
trol scheme with respect to CPQA . The commercial process is simply that employs a combination of online and at-line data collection approaches .
operated within a set of process parameter targets established during Biopharmaceutical PQAC schemes continue to become more sophisti-
development with the expectation that, by operating within this design cated and refined . Many companies have committed to it in their develop-
space, the process will consistently achieve required product quality tar- ment programs, and several regulatory agencies have voiced their support
gets . Biopharmaceutical manufacturing processes operate this way due for PQAC strategies . With increased understanding of the PQA themselves,
to process and product complexities, coupled with limited analytics and improved consistency in sample collection and analysis, and increased
bioreactor sampling capabilities . regulatory support, companies will continue to establish PQAC programs .

Control algorithms Predictive model At-line data

Process levers
Media addition Percent Viable cell volume,
Adjustment of
algorithm to hit galactosylation as a galactose, glucose,
galactose and
target galactosylation function of input and percent
glucose feed rates
profile parameters galactosylation

On-line data
Dissolved oxygen, pH,
capacitance, Raman

Protein A
UPLC/LCMS Analyzer
Cell removal sample
Cell-free Glycosylation profile
preparation
Bioreactor sample

Aseptically Whole-broth bioprocess analyzers

collected • Viable cell volume
whole-broth • Galactose concentration
sample • Glucose concentration

FIG. 20-71 PQAC application in which a target mAb galactosylation level is maintained by controlling galactose feed concentration .
 PRODUCT QUALITY ATTRIBUTE CONTROL 20-89

REGULATORY GUIDANCE AND INDUSTRY TRENDS On-line, in-line, and automated sampling and analysis tools have been
developed to facilitate development of a data-driven QbD strategy focused
PQAC concepts are driving significant changes in regulatory guidance and on process operations and PQA . Time course data of metabolites, glycosyl-
approaches to consistently achieve CPQA . Historically biopharmaceutical ation patterns, and other PQA can be generated using these tools . Examples
manufacturing has been performed as a series of batch unit operations . include Raman, near IR, and dielectric spectroscopy (capacitance) probes
Product quality and release testing were performed only after process com- installed in bioreactor probe ports . These data along with chromatography
pletion . This approach led to the development of process validation, where and mass spectroscopy data analysis can be used with metabolite profil-
a process is developed and locked in . Consistency was then demonstrated ing technologies to predict cell culture productivity endpoints . Small-scale
over a relatively small number of runs, using the locked process . Once vali- bioreactor design of experiments (DOE) studies are also performed to gain
dated, manufacturers strove to minimize changes in an effort to maintain a better understanding of factors that affect PQA . These data streams can
consistent process results . As a result, biopharmaceuticals have often been inform predictive modeling approaches, which can lead to process control
produced using a validated process that uses the same operational parame- strategies focused on real-time PQA optimization in bioreactors .
ters from run to run on the basis that the manufactured product retains the Downstream protein purification unit operations may also benefit from
same quality attributes . Having operated this way for decades, this has been a cohesive QbD strategy since they can also impact product attributes .
an accepted biopharmaceutical regulatory compliance approach, although Downstream purification often consists of a series of distinct unit opera-
advances in technology and regulatory guidance are leading to changes . tions . These downstream technologies may include host cell removal (depth
The FDA and other regulatory agencies have recognized the importance filtration), primary purification (protein A bind and elute), viral load reduc-
of shifting toward a PQAC approach . As a result, several guidance docu- tion through inactivation (low pH) and removal (nanofiltration), and pol-
ments addressing process analytical technology (PAT) and quality by design ishing steps (ion exchange resins or membrane adsorbers) . Additionally, a
(QbD) have been published . These include Guidance for Industry, PAT—A tangential flow filtration (TFF) step is typically included to perform a buffer
Framework for Innovative Pharmaceutical Development, Manufacturing and exchange and then concentrate the protein to achieve the required antibody
Quality Assurance, FDA, 2004; ICH Q8, Guidance for Industry, Q8 Pharma- drug substance formulation . PQAs affected by these steps may include con-
ceutical Development, Rev . 2, FDA, Nov . 2009; ICH Q9, Guidance for Industry, taminants such as charge variants, host cell proteins, reduced variants, pro-
Q9 Quality Risk Management, FDA, 2006; and ICH Q10, Guidance for Indus- tein aggregation, viral load reduction, and product purity .
try, Q10 Pharmaceutical Quality System, FDA, 2009 . Together these guidance In contrast to upstream operations, downstream operations are of shorter
documents outline a system in which a design space where the process can duration, and the relationship between CPP and process step objectives, such
be successfully operated is characterized, real-time bioreactor monitoring as aggregate removal, is more readily characterized and measured compared
using online and at-line technologies is implemented, and the knowledge to cell culture processes, making them ideal candidates for a QbD approach .
gained is used to actively control product quality . This fuller understanding Downstream processing is also more likely to be a standardized sequence of
may lead to a PQAC scheme and possibly a real-time release testing (RTRT) unit operations, often using a defined platform process . Upstream processes
program, although regulatory agencies will continue to require rigorous require customized approaches from product to product since cell lines,
validation of these processes . medium formulations, variable biomolecular constructs, and variable CPQA
are driven by the therapeutic application . By using platform downstream chro-
QUALITY BY DESIGN (QbD) matography processes, for example, the CPPs that dictate the quality outcomes
may be readily defined and characterized by a set of bench-scale experiments .
The FDA’s Office of Biotechnology Products states that the basic concept of Process parameters to be evaluated may include pH, conductivity, tempera-
QbD is to “provide guidance on pharmaceutical development to facilitate ture, buffer gradients, stability, peak collection criteria, resin lifetime, and
design of products and processes that maximizes the product’s efficacy and process run times . Applying PAT tools during the development phase enables
safety profile while enhancing product manufacturability .” Furthermore, it rapid generation of robust data sets as part of DOE studies, leading to well-
is stated that “knowledge gained during the pharmaceutical development characterized process design space and a QbD-enabled commercial process .
program is critical for enhanced understanding of product quality and pro- PQAC Implementation The biotech industry is moving toward incor-
vides a basis for risk management and increased regulatory flexibility .” QbD porating PQAC into fed-batch, perfusion, and fully continuous processes .
is a working methodology, focused on systematically evaluating, under- Accomplishing this requires knowledge of CPQAs and their relation to
standing, and refining manufacturing processes with the goal of delivering CPPs, a means to rapidly analyze CPQAs and process behavior and direct
an optimized product with an understanding of the relationship of PCQAs the results into models and control systems . For example, companies are
and CPPs . This can lead to the establishment of a design space and appro- developing automated mAb glycan profile control beginning at bench and
priate control strategy . pilot plant scales . As PQAC is demonstrated at those scales, it will progress
QbD is a proactive scientific approach to “build” quality into a product into clinical manufacturing to support biopharmaceutical product devel-
rather than “test” quality into a product . Its objective is to identify process opment and regulatory approval . The time required for these programs to
parameters that impact product attributes (desired and undesired) so that reach the approval stage and begin commercial manufacturing may mean
controls can be implemented so that the desired product is manufactured . it will be many years before biopharmaceuticals are manufactured using
The objectives include providing increased understanding of the pro- PQAC . Given the faster path to commercialization for veterinary products,
cess operating space, including linking critical process parameters to the coupled with cost control drivers, the first commercial biomolecule manu-
required target product quality parameters . This knowledge allows greater factured using a PQAC scheme may be a veterinary product .
flexibility in accommodating process changes, while also establishing con- Understanding of biopharmaceutical CPQAs and their impact on mol-
trol ranges for a PQAC approach . QbD strategies can include better under- ecule function is rapidly increasing . For example, Ghaderi studied which
standing of manufacturing process inputs, including raw material testing, mAb glycosylation patterns to target and which ones to avoid [Ghaderi,
process parameters and PAT integration, and outputs, including product Zhang, Hurtado-Ziola, and Varki, Biotech. and Genetic Eng. Rev . 28: 147–176
identification, efficacy, performance, and safety testing . (2012)] . Once CPQAs are identified, supporting PQAC technology develop-
Protein therapeutics are inherently diverse, and variants may exist within ment can advance . A key requirement of PQAC is the ability to measure
each bioreactor production lot . A formulated drug product may be com- CPQAs and process behavior in near real time . Upstream and downstream
prised of proteins having somewhat different attributes associated with sample collection and preparation systems are maturing, making it possible
both the active molecule and other components present in the final drug to routinely and automatically collect whole-broth samples and analyze
formulation . This variability can be due to process variations in upstream them for viable cell density and volume, medium and metabolite concen-
and downstream processing unit operations, or it may be the result of ana- trations (glucose, lactate etc .), and dissolved gases . Samples can also be
lytical variability in particular bioassays . Although CPQAs may be linked to prepared using automated cell removal, which is required to perform ana-
manufacturing CPPs, the CPQAs themselves, including efficacy, immunoge- lytical protein A chromatography preprocessing, followed by HPLC/UPLC
nicity, and other safety-related attributes, are uniquely associated with the or LCMS analysis to measure PQA such as glycan profiles .
protein product . Since CPP variations impact protein CQPAs, biomanufac- To complete a PQAC process, CPQA and other process analytical results
turing processes are ideal opportunities to apply QbD principles . can be fed into predictive models that drive control schemes to maintain
Attributes most commonly affected in the bioreactor include glyco- CPQA at a desired level . Significant improvements in such predictive model
sylation (sugar residues), charge variants (oxidation, deamidation), and development have been reported [Van Impe, Vercammen, and Derlinden,
protein molecular weight (reduction, aggregation) . Batch and fed-batch bio- Procedia Food Science 1: 965–971 (2001)] .
reactor processes are dynamic wherein cell state is changing as a function
of cell age, feeding strategy, medium composition, metabolite buildup, and BIOPROCESS ANALYSIS TECHNOLOGIES—GENERAL
biopharmaceutical production . Since these may be CPPs impacting CPQAs,
implementing a QbD strategy incorporating process and analytical moni- Consistently delivering quality biopharmaceuticals typically requires one or
toring and at the bioreactor can lead to better understanding of the process more PAT tools to monitor and control processes in real time . The increas-
parameters that correlate with CPQAs . ingly common use of bioprocess automation to achieve tighter control of
20-90 BIOREACTIONS AND BIOPROCESSING

product quality is driving the development and adoption of PAT for process BIOPROCESS ANALYSIS TECHNOLOGIES—UPSTREAM
monitoring and integration with feedback control of both CPP and CPQA . TECHNOLOGIES
Significant progress has been made in the development of analytical
methods for measuring CPQA . These analytical methods, however, are often In cell culture processes, small variations in process conditions or medium
off-line analyses performed days or even weeks after sample collection . This formulations can cause wide variations in the quality and quantity of the
analytical time delay makes PQAC difficult to implement because control product produced . For example, Le analyzed over 200 commercial cell
adjustments to a bioprocess in any PQAC scheme cannot be implemented culture runs with nominally identical process conditions [Le, Kabbur,
before the measurement of the CPQA is completed . The challenge to be Pollastrini, Sun, Mills, Johnson, and Hu, J. of Biotech . 162: 210–223 (2012)] .
addressed to allow implementation of any PQAC scheme therefore is to inte- They found that minor variations in the seed culture resulted in large run-
grate and complete sample collection and analysis within a short enough to-run variations in cellular metabolism related to lactate production and
time frame to make a PQAC scheme reasonable . Once CPQA results can be consumption in the production bioreactor (Fig . 20-73) . Conditions that led
rapidly and automatically generated, the data can be fed into process mod- to the accumulation of lactate resulted in lower product titers, whereas
els and control algorithms, which may lead to development of automated conditions that favored cellular metabolism switching from lactate produc-
CPP and attribute control of biopharmaceutical manufacturing production . tion to consumption yielded the highest product titers . Using multivariate
What follows is a brief overview of currently adopted bioprocess PAT analysis, it was shown that lactate could be used as a process performance
technologies . Additional details can be found in Al-Rubeai, Animal Cell Cul- indicator . If lactate levels could also be tied to CPP, it could form the basis
ture, Springer, New York, 2015, pp . 185–222 . of a real-time control strategy to maximize production . In this example, PAT
Online versus At-Line Technologies Bioreactor sterility require- to monitor lactate levels and control bioreactor conditions based on these
ments have traditionally favored online PAT for upstream PAC, which does measurements could be implemented to reduce process variability .
not require bioreactor sampling, over at-line PAT which requires sampling . Established PAT Online technologies that provide continuous mea-
This is so because automated PQAC schemes relying on sophisticated at- surements include pH, dissolved oxygen, temperature, feed, and gas flow
line technology require consistent and reliable sample collection, which rates . These are established, well-understood technologies, and they may
presents several hurdles for any automated sampling system . First, bioreac- come in disposable or single-use compatible formats . These parameters are
tor sterility must be maintained during sampling . In the case of mamma- typically independently controlled to a target set point . Modulation of these
lian cell culture, the smallest microbial contamination can result in product parameters often affects the cell state globally and is therefore less desirable
loss worth several million dollars . Second, once a sample is removed from as a means of independent, targeted product attribute controls compared to
the bioreactor, the system must consistently route viscous, high-solids- controls that specifically target quality attributes .
content samples to multiple instruments for weeks, or in the case of per- Off-line measurements, which require manual bioreactor sampling, are
fusion cell culture for months, consistently and without clogging . There needed to monitor other process parameters such as extracellular metabo-
are now automated aseptic sampling technologies commercially available lite and medium concentrations (glucose, lactate, glutamine, glutamate,
that have demonstrated the ability to collect representative samples while ammonium), electrolyte concentrations (sodium, potassium, calcium),
maintaining bioreactor sterility, allowing traditionally off-line analytical osmolality, cell density, carbon dioxide, and product concentration . Cell
techniques to be applied in a pseudo-online, or at-line fashion (Pepper, culture medium composition (amino acids, trace metals) measurements
Modernizing Biopharma Sampling, Contract Pharma, Sept . 2014) . Autosam- are less frequently employed, although automated sampling is beginning to
pling systems can collect whole-broth samples directly from the bioreactor make them more commonplace .
and route those samples to multiple analytical destinations . Whole-broth Raman and NIR Spectroscopy Raman and near-infrared (NIR)
or cell-retained samples are routed to bioanalyzers to measure cell density, spectroscopic technologies have been used to continuously monitor and
viability, protein titer, metabolites, salts, and dissolved gases . Samples ana- control bioreactor processes, although not as often as other instruments .
lyzed by chromatographic or mass spectrometer methods must be cell-free . These technologies measure bioreactor broth composition via a probe
Autosampling systems provide a mechanism for removing the cells, result- inserted into the bioreactor which introduces a light source . The incident
ing in a cell-free sample . Prior to analysis, these cell-free samples can be pro- light from the probe interacts with the bioreactor contents within the mea-
cessed in a liquid handler employing methods such as protein A processing, surement volume . The light is absorbed or scattered in a specific manner
solid-phase extraction, dilution, digestion, or filtration . Cell-free, processed according to the specific mixture of chemical species present in the mea-
samples, analyzed with the full capability of HPLC/UPLC and LCMS types of surement volume of the probe . This light interaction can be detected and
systems, can generate the full spectrum of PQA data such as glycosylation used to generate a spectrum consisting of various peaks quantifying the
profiles, amino acid concentrations, and levels of deamination and aggrega- light interaction as a function of wavelength . This spectrum is a convolu-
tion (Fig . 20-72) . tion of the response of all components that interact with the incident light .
Automated downstream PAC requires higher sampling frequencies Multivariate analysis (MVA) or other models can be developed to interro-
compared to bioreactors to enable real-time monitoring and control, driv- gate the spectra to identify and quantify the various species that responded
ing adoption of rapid at-line online PAT . Automated sampling technology to the light source .
is being applied to downstream processes to integrate traditionally off- These spectroscopic technologies can measure many different analytes
line analysis, such as UPLC systems . Contamination control in automated simultaneously and are available in single-use compatible formats . Due
downstream sampling systems is generally simpler than in upstream to the complexity of the bioreactor contents, interpreting the spectrum
applications, since the processing buffers are not growth-promoting, and requires analysis and modeling approaches such as partial least squares
bioburden reduction can be achieved through chemical sanitization, or (PLS) . MVA and PLS model development requires generating spectra from
presterilized closed processing solutions . Additionally, the streams being known solutions having representative concentrations of all species present
sampled are relatively low-viscosity and free of suspended solids, making during production . Some of the data are used to generate the model, with
sample preparation less challenging . the remainder used to validate it . Controlling mAb glycation levels using
For both upstream and downstream processes, the selection of which Raman spectroscopy integrated with a PLS model was demonstrated [Berry,
PAT to apply depends on which PQA must be controlled, knowledge of how Biotech. Prog. 32: 224–234 (2016)] .
these PQAs are related to process conditions, and which PAT measurements Predictive modeling approaches have also been used to generate empiri-
are needed to monitor relevant process conditions . cal “fingerprints” of the bioreactor contents from the raw spectral data and

Bioreactor Sampling device Cell removal Sample Automated

system processing analysis
Sample
Bioreactor Sampling device Whole-broth analyzer Automated
routing
analysis
Whole-broth analyzer
Automated
Bioreactor Sampling device Whole-broth analyzer analysis

FIG. 20-72 An automated sampling and analysis system capable of collecting samples from multiple bioreactors and sending those
samples to multiple analytical destinations .
 PRODUCT QUALITY ATTRIBUTE CONTROL 20-91

instruments has enabled many off-line metabolite assays to be used

15 at-line . These data can be used for near real-time bioreactor monitor-
ing and control in conjunction with cell density and other measure-
ments . This provides insight into cell metabolic states over the course
High lactate runs
Lactate (g/L)

10 of a bioreactor process, which is known to be a key determiner of prod-

uct quality attributes [Zhao, Eng. Life Sci . 15: 459–468 (2015)] . Which
specific extracellular metabolites are measured depends on the level of
understanding of their effect on cellular metabolism and product attri-
5 butes . Commonly measured metabolites include major inputs and out-
puts of central metabolism . Some examples include ammonia, lactate,
Low lactate runs glucose, galactose, amino acids, and nucleosides . Currently available
0 analyzers include various commercial bioanalyzers, amino acid analyz-
0 50 100 150 200 250 ers, as well as LC/UPLC instruments . These have all been implemented
Time (h) at-line using automated sampling systems . Some of these techniques,
such as LC-based analysis, require automated cell removal as part of the
FIG. 20-73 Example of cell culture variability . Lactate concentrations are shown sampling system prior to analysis .
for nominally identical commercial-scale cell culture productions runs . Runs where Product Attributes A major goal of bioreactor PAT has been to enable
lactate continued to increase over time had lower product titers, whereas runs where consistent biopharmaceutical production meeting CPA targets . Direct
lactate levels stayed low or decreased had the highest titers . [Reprinted from Le, K ., measurement of these attributes and incorporation into a PAC scheme has
Pollastrini, S ., Mills, J ., and Hu, W ., “Multivariate Analysis of Cell Culture Bioprocess
Data—Lactate Consumption as a Process Indicator,” J. Biotech. 162: 214 (2012) . Repro-
benefited from the same robust automation platforms that enable at-line
duced with permission of Elsevier . © 2012 Elsevier B .V . All rights reserved .] extracellular metabolite measurements . Many product attribute measure-
ments have been deployed as at-line instruments to constantly monitor
bioreactor CPA . Many product attribute measurements require cell removal
and initial product purification, typically using preparative pro-A chroma-
to correlate them with specific culture states . More information on the tography for mAbs and mAb variants, prior to analysis . These methods use
generation and use of spectral fingerprints can be found in Lourenço, Anal. sophisticated automated sampling and liquid handling systems to prepro-
Bioanal. Chem. 404: 1211–1237 (2012) . This empirical bioreactor state infor- cess and direct samples to a wide range of analytical instruments . Product
mation (a state estimator) can be used in a model predictive control (MPC) attribute-related measurements include product concentration, glycosyl-
strategy . For both PLS and MPC, significant amounts of data are required to ation patterns, high- and low-molecular-weight species, aggregates, and
build, validate, and successfully apply these methods to bioreactors . charge variants . The specific product attributes being measured depend
Dielectric Spectroscopy Dielectric (or capacitance) spectroscopy (DS) greatly on understanding the impact of the product attributes on the safety
technology is often used to measure viable cell mass, which can be corre- and efficacy of the product .
lated to viable cell density for a range of host cells including CHO, microbial,
yeast, and insect cells . DS relies on the fact that the cellular membrane of
live cells is impermeable to ions . When exposed to an alternating electric BIOPROCESS ANALYSIS TECHNOLOGIES—DOWNSTREAM
field, the membranes become polarized, resulting in a capacitance related TECHNOLOGIES
to the dielectric properties . Capacitance is proportional to the biomass,
dependent on cell size, cell type, suspended versus adherent cells, tempera- Since downstream processes involves a series of unit operations, the
ture, and physiological cell state . Cell debris, gas bubbles, and most medium application of PAT depends on the step being monitored . Typical down-
components are not detected by DS . In most cases nonviable cells are also stream unit operations include solids-liquid separation/clarification, protein
not detected by DS, although there are cases where cells which are classified separation/purification, viral inactivation/clearance, product concentra-
as nonviable do produce a capacitance signal (typically depending on the tion/formulation, and sterile filtration . Each one has a specific role in meet-
specific mode of cell death resulting in the nonviable cells) . A method for ing requirements for purity, impurities removal, and preparing the product
detecting and correcting for the phenomenon has been published [Downey, stream for the next step in the production process .
Biotech. Prog . 30: 479–487 (2014)] . Online PAT Online PAT tools routinely employed in downstream
The measurement frequency used depends on the host cell and is deter- processes include pH, conductivity, UV, turbidity, flow rates, pressure, and
mined by comparing medium with and without cells present . Frequency- temperature . Most of these can be a CPP for chromatography steps . For
scanning DS instruments are available that may obviate the need for this example, pressure may be a CPP for viral filtration, and pH for viral inac-
step . To determine biomass or cell density from capacitance, the data are tivation . PAT is used to monitor and document process performance and
correlated to cell density determined by other instruments such as off-line is integrated with process controllers to modulate process parameters .
image analysis cell counters . Changing bioreactor conditions that affect Chromatographic protein purification steps often use UV absorbance, often
cell size, hence cell volume, will affect the instrument accuracy, unless at 280 nm and/or 300 nm, to measure relative protein concentrations and
accounted for in the correlations . as an indicator for product peak collection start and stop times, which can
DS advantages include directly measuring viable cell mass, being a simple affect purity and yield . Conductivity and pH are routinely used to monitor
continuous online technology, and available in single-use compatible for- buffer composition during chromatographic purification and are impor-
mats . Although measurement accuracy may be impacted in the declining tant parameters that can affect protein stability, binding affinity, and the
culture phase as viability drops, and initial functional calibration for the separation of product from contaminants . Turbidity can be used to measure
specific application is required, the simplicity of this PAT tool has led to its suspended solids in clarified liquid streams as part of bioreactor harvest
wide application . Potential DS applications can include measuring specific operations . It has been used to monitor the separation efficiency of spent
cellular phenomena such as cell size, growth rates, physiological cell state, cell removal from the product stream by centrifuges and to monitor poten-
adherent cell density, and ion polarization [Justice, Biotech. Adv. 29: 391–401 tial depth filtration breakthrough .
(2011)] . These advanced measurements will likely require data-driven, pre- At-Line PAT Application of new technologies may lead to PAT for pro-
dictive modeling approaches similar to spectroscopic techniques . cess related impurities (host cell proteins, DNA, antifoam, leached protein A)
Soft Sensors In addition to considering individual technologies for and product-related attributes (size, aggregates, acid/base variants, high/
measuring specific aspects of a process, multiple measurements can be low molecular-weight species, and product concentrations) . Automated
combined to predict the value of a process variable of interest . The pro- sampling technologies are enabling offline measurements to be performed
cess of using easily measured secondary variables to infer a primary mea- at-line . One example is controlling product loading onto chromatography
surement is known as a soft sensor . Robust application requires a data columns using at-line analytical chromatography instruments to moni-
set to be generated, which can be used to train a predictive model, and tor product concentration in the feed stream . Additionally, impurities such
a second set of data for verification . Soft sensor approaches are typically as aggregate levels can be monitored and controlled to optimize step yield
employed when direct in-process measurement of the parameter of inter- while meeting product specifications . Because downstream samples are
est is impractical, or where multiple physical measurements have to be cleaner, automated systems can be simplified . However, the need for rapid
analyzed together by application of a multivariate method to predict spe- analysis is increased due to short step cycle times .
cific outcomes, such as product quality and titer [Gunther, Comput . Chem.
Eng . 33: 88–96 (2009)] . PRODUCT ATTRIBUTE CONTROL
Extracellular Metabolites While many extracellular metabolites
may be measured using online spectroscopic technologies, development Variations in raw materials, process parameters, and cell state can signifi-
of robust autosampling systems combined with automated analytical cantly affect bioprocess outcomes . The complexity of raw materials makes
20-92 BIOREACTIONS AND BIOPROCESSING

these variations very difficult or expensive to detect, and they may only
manifest themselves when used in the process . This is driving an increas- Set point
ing emphasis on product attribute control (PAC) in the face of raw material
variability .
Open-Loop Attribute Control Bioreactor PAC can be implemented MPC
using open-loop or closed-loop control, although some process parameters,
such as temperature, will use closed-loop control independent of any PAC In-process Forecast Control
measurements Controller
scheme . Open-loop control is the most commonly used PAC method, pri- model element
marily due to an inadequate linkage between CPQA and CPP . In open-loop
control, process parameters, cell lines, and medium are developed to ensure
the process is robust against disturbances, and is intended to be consis-
tently operated within an established design space . This process “recipe” Process
is established using laboratory scale DOE that can vary from product to
product, even with similar host cell lines . Because open-loop control has
been the de facto standard for PAC, much of the required technology is FIG. 20-74 Schematic representation of a model-predictive controller . The control-
commercially available, making it a simple, quick, low-effort way to imple- ler uses in-process measurements of the controlled variables and the culture state as
ment a recipe-controlled process . This approach is limited by an inability to inputs to a predictive model . The predictive model forecasts are used to determine
respond to unforeseen, or more extreme, process disturbances beyond what the appropriate control element manipulations needed to achieve the desired process
the process was designed and tested for . In extreme cases, this may lead to outcome .
process failures . To increase process robustness against such disturbances,
additional raw material vigilance and development may be required to
minimize process variances, sometimes even after the process has reached control handles to achieve a given change in a product attribute are often
commercial scale . not fully characterized . In addition to the fundamental metabolic-based
Closed-Loop Attribute Control Increasingly biopharmaceutical studies used to identify control handles for many important biopharma-
manufacturers are looking toward closed-loop PQAC as both a development ceutical PQA, empirically based methods can also be employed to identify
and a commercial strategy to meet increasing regulatory expectations, and the relationship of process inputs to a given output . Application of machine
to compete in the biosimilar market segment . The goal of closed-loop con- learning methods in conjunction with metabolic studies may also be used to
trol is a process that is more robust and has less product variability . Closed- identify suitable control handles from cell culture data . An example applica-
loop control actively controls the process trajectory toward a desired CPQA tion of this type of methodology, where bioreactor data were mined to deter-
endpoint by manipulating CPP . This requires an understanding of which mine candidate CPP for titer, is presented here in Charaniya, J. Biotechnol.
and to what degree CPP affects CPQA . An additional complexity is that the 147: 186–197 (2010) .
degree to which a controlled parameter affects a CPQA can vary during the A process for identifying potential control handles and development
course of a batch or fed-batch process . of control schemes has been described for IgG1 mAb glycan distribution
To be a candidate for closed-loop PAC, an attribute must be able to be [St . Amand, Biotech. Bioeng. 111: 1957–1970 (2014)] . In that example a
measured in process; one or more controllable process parameters must DOE was executed followed by an analysis of variation statistical analysis
exist that can positively and/or negatively manipulate the attribute (ideally (ANOVA) to determine main and interaction effects of manganese, galac-
independently of others), and detailed knowledge of how the manipulated tose, and ammonia levels on glycan distribution . The results were used to
parameters affect the attribute must be obtained . There is some debate as perform a controllability analysis in which controllable modes—those that
to whether closed-loop PAC is more suited as a development tool used to have a significant impact on glycan distributions—were identified and
arrive at a recipe-driven commercial process where PAC can predictably ranked according to their potential to effect changes in glycan distributions .
achieve the target range, or if it can be used to routinely control commer- This formed the basis of a controllability analysis which identified both
cial processes . It is likely that there will be a range of situations, from those input modes (i .e ., suitable control parameters) and output modes, which
where PAC is dependent on a few well-characterized variables so closed- together identify what biological processes are controllable to achieve
loop control can be robustly and efficiently implemented, to more complex a desired outcome . The full description of this analysis is provided in the
cases where either it is not feasible or the process variability is low and the supplemental information provided by Amad .
impact on critical PQA low, so active control does not provide a clear ben- Last, MPC requires a model to predict the change of a specific output in
efit or advantage . It can be expected that as monitoring and development response to a set of measured process inputs . Data-driven predictive mod-
tools advance, so will the implementation of closed-loop PQAC for routine els can use experimental data to derive these relationships in the form of
biopharmaceutical production . either fundamental or empirical relationships . Models used for MPC must
Implementation of Product Attribute Control for Bio-Based incorporate causal and dynamic predictions of how a given manipulated
Processes In many bioreactions the effect resulting from adjustments variable will affect the desired output variable . Experimental techniques
to a process parameter on a product attribute is both complex and depen- commonly employed in the field of systems identification can be applied
dent on the cellular metabolic state due to the highly interactive cellular to generate such relationships . See the following reference for more
processes . For example, Yang [Biotech. Prog . 16: 751–759 (2000)] found information on applying system identification techniques to processes:
that the state of ammonia accumulation (a major metabolic by-product Isermann, Identification of Dynamic Systems: An Introduction with Applica-
of amino acid metabolism) in CHO cultures also had a significant effect tions, 1st ed ., Springer-Verlag, Berlin, 2011 . Downey described the practi-
on terminal sialylation of glycans, an important mAb PQA . Because of cal application of a simple system identification experiment to develop a
this complexity and state dependence, control schemes in cell culture model predictive controller of galactosylation of a mAb produced in CHO
processes often involve some form of model-predictive control (MPC) to cell culture process [Downey, Biotech. Prog. 33: 1647-1661 (2017)] . In this
account for multivariate interactions when anticipating the effects of example, the causal, dynamic relationship between galactose feeds and
manipulating control parameters . A simplified MPC scheme is shown in galactosylation of a mAb were determined by subjecting a culture grown
Fig . 20-74 . in a pseudo-steady-state culture to step increases in galactose concentra-
Although MPC is used in many applications (see Sec . 8, Process Control tion, and recording the resulting dynamic response of mAb galactosyl-
and Instrumentation), there are some unique considerations for biologics . ation over time .
Any feedback control scheme needs in-process measurements of the con- Generation of MPC models typically requires large amounts of experi-
trolled variable . For multivariate, state-dependent bioreactor systems, a mental data that span the expected parameter ranges in order to provide
number of in-process measurements are needed to quantify the state of reliable, robust control . The relationship between inputs and outputs in
the system, or to use as model inputs, to enable closed-loop MPC . Process cell culture processes can be highly nonlinear, multivariate systems and
measurements must be taken often enough to allow the controller to adjust therefore difficult to construct an accurate model from limited data . For
to dynamic changes in the process outputs . For bioreactor processes, the MPC, however, a sufficiently accurate model may be obtained by approxi-
required sampling frequency is dependent on the system dynamics includ- mating a system using linear methods . Zupke described the application of
ing host cell metabolic rates and the rate of change of the manipulated and MPC to control high-mannose species generated in a cell culture process
controlled variables . In many cases, manual sampling may be insufficient, [Zupke, Biotech. Prog. 31: 1433–1441 (2015)] . The functional, causal rela-
and at-line PAT is needed . tionship between mannose feeds and high-mannose residues on a mAb
One or more control handles that can be manipulated to achieve desired were determined using small-scale batch experiments . A model predic-
changes in the product attributes of interest must also be identified . Unlike tive controller was constructed from this fundamental relationship using a
processes in more established industries, in cell culture processes the receding-horizon method .
 EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS 20-93

EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS

HIGH CELL DENSITY CULTURES (HCDCs) the economical generation of biological products . Thus, high cell density
cultures are ideal for yielding high volumetric production rates; but signifi-
Introduction High cell density cultures are used to achieve higher bio- cant challenges exist .
reactor efficiencies . A high cell density culture is characterized by having A number of growth considerations need to be taken into account for high
at least 10 times the normal cell density of a common batch reactor . For cell density mammalian cultures . Oxygenation, CO2 removal, and media
yeasts, archaea, and bacteria, this means biomass concentrations greater design are critical for maximizing cell count . Adjustment of substrate feed
than 100 g/L . Applying high-density culturing to mammalian cells can rate (or perfusion rate if a perfusion system is used) is important to meet
result in cell concentrations in excess of 107 cells/mL to 108 cells/mL . This the metabolic demand of increasing cell density for fed-batch and continu-
approach is more economical since less growth medium and volume are ous cultures . Also, controlled nutrient feeding can contribute to consistent
required, enhancing cell productivity per volume, which improves reactor product qualities . In addition to growth considerations, there are a num-
utilization and efficiency . These high-density cultures are also amenable to ber of limitations to high cell density cultures . Mass-transfer and mixing
continuous cultivation, resulting in reduced size and capital costs . However, demands become more prevalent for high cell density cultures . Increasing
high-density cell reactors are difficult to design and challenging to main- oxygen demand at high cell density requires higher oxygen delivery rates to
tain; therefore, specific strategies and issues associated with the high cell the bioreactor . In addition, a significant amount of dissolved CO2 accumu-
density approach are discussed . lates at high cell density . Base addition, often used to combat high dissolved
Microbial Reactors A variety of reactor types are used for high-density CO2 (which can lower pH), can lead to cultures with high osmolarity, result-
microbial systems including stirred-tank reactors, dialysis membrane reac- ing in the loss of cell growth and even cell death . Nonuniform mixing can
tors, cyclone reactors, gas-lift reactors, and shaken ceramic flasks . Cur- promote heterogeneity in concentrations, and cell aggregation can result in
rently, the stirred-tank reactor operating under fed-batch conditions is the segregation of cells within different mixing zones . Physical space is another
most widely used approach for large-scale high-density cell production . limitation to production in high cell density culture processes . Theoretically,
Stirred-tank reactors are commonly used, but for HCDC, internal or exter- the maximum cell density for mammalian cell cultures is approximately
nal cell retention methods are needed so that substrate can be refreshed . 109 cells/mL . Typical values on the order of 107 or 108 cells/mL occupy about
Internal retention has advantages because pumping cells out of the reac- 1 to 10 percent of the total volume area achieved .
tor causes stress, and filters or membranes can be used to hold back the While currently most high-density animal cultures are obtained through
cells . The dialysis membrane reactor also uses internal cell retention and is feed batch cultures, cell retention is one particular approach in reactor
advantageous because it allows for continuous removal of toxic and inhibi- design to increase cell density . Most membrane systems can be challenging
tory products without added stress on the cells . Dialysis fermentation can with respect to handling very high cell density cultures . Cell immobiliza-
be achieved using two-vessels linked with a dialysis device or using one ves- tion and encapsulation work are feasible for higher densities (on the order of
sel with two chambers . The disadvantage to dialysis is the potential loss of 108 cells mL−1), but the intensive labor and high preparatory cost make this
valuable nutrients while filtering inhibitory ones . approach impractical for industrial production . Many cell lines are adapted
An important consideration for growing high-density cultures is the need to grow in suspension as single-cell or small aggregates while certain attach-
to maintain sufficient levels of nutrients to satisfy the increased number ment systems still require the presence of solid surfaces .
of cells while not inhibiting cells with too much substrate . Fed-batch and Media supplementation or fortification is essential for optimizing the
dialysis fermentation techniques are able to overcome these restrictions . number of viable cells attainable in HCDC, and the media exchange rate
Continuous or intermittent feeding is recommended to meet the demands directly affects production levels and yield . In many cases, complete media
of these cultures . For fed-batch reactors, mathematical models and indi- exchange can occur periodically to maintain a cell density of 107 cells/mL
rect and direct feedback control systems are used to optimize the feed rate while media exchange rates should be increased in order to maintain a den-
and nutrient concentrations . Direct control may not be sufficient because sity of 108 cells/mL . However, high perfusion rates can pose challenges to the
of the lack of sensors for fast and reliable measurement and the complex- efficiency of cell retention systems . Additionally, minimizing media perfu-
ity of the cellular dynamics . Bioactivity patterns such as changes in pH or sion rates is important to minimize processing costs . The optimal solution
dissolved oxygen concentration are instead used as indirect measures to is often to develop special concentrated or fortified medium for fed-batch
overcome these limitations . Noninvasive optical measurement methods are and perfusion reactors .
also attractive for industrial use . There are multiple and varied challenges Process control is also required for consistent production . In high-density
encountered during HCDC of microbial cultures including solubility of perfusion reactors, cell densities generally increase from 0 .5 × 105 cells mL−1 to
solid and gaseous substrates in aqueous media . For example, low oxygen 107 to 108 cells/mL . Initially during cell accumulation, media can be provided
solubility and inhibition of substrates with respect to growth can be pre- in order to maximize cell growth . During the stationary or production phase,
vented by controlling the supply of the carbon or oxygen source and through cell densities vary less and should be controlled at set points through media
metabolic engineering . Other issues include instability, volatility, and deg- feeding and environmental control . Optimal conditions for high cell density
radation of products; accumulation of products or metabolic by-products also require proper pH, temperature, and dissolved oxygen control . Addition-
to a growth-inhibiting level; high evolution rates of CO2, heat generation, ally, nutrient and metabolic by-product concentrations can be optimized
and viscosity of the medium . For example, acetate production can become through media exchange as needed to optimize reactor performance .
inhibitory in fed-batch, microbial cultures . One approach is to remove Another mixing issue is dissolved CO2 accumulation . The CO2 produced
the inhibitory acetate through integration of fermentation and separa- can be incorporated as bicarbonate, and CO2 can be stripped from the
tion processes . Another physical limitation for HCDC is the reduced mix- medium and transferred to the gas phase by increased sparging and mass
ing efficiency that occurs with increasing fermenter size . To address issues transfer . However, CO2 stripping may proceed at a lower rate than oxygen
involving substrates, a controlled feeding strategy can be applied in concert delivery, creating a gradual increase in dissolved CO2 . See the Carbon
with control of pH and other process parameters . Limiting essential nutri- Dioxide Stripping subsection for more detailed discussion of CO2 control
ents may reduce growth but also limit production of inhibitor by-products in bioreactors .
such as acetate . The fed-batch approach is now widely used to obtain high However, base addition can engender other issues including high osmo-
cell density with dialysis fermentation as an alternative . Nevertheless, due to larity and additional mixing issues . When sufficient time is not allowed for
the vast array of microorganisms, a number of unknowns remain for eluci- thorough mixing and homogenization of the initial base, the continual addi-
dating the optimal conditions and bioprocesses for HCDC . Further studies tion of base can cause locally elevated pH values, which can then cause cell
are required to evaluate microbial physiologies in order to develop optimal lysis . The resulting cell lysis can lead to viscous regions with high pH . Since
processes . The development of online sensor systems and monitoring and bulk pH is not affected, incorporating additional base amplifies the issues
control algorithms will ensure further enhancement of HCDC processes . and leads to formation of “snowball” structures of lysed cells . To resolve this,
Additionally, intracellular regulatory mechanisms remain to be investigated placement of a base addition line close to the impeller system can provide
and overcome, using metabolic engineering and synthetic biology . As a sufficient mixing for homogenization . Thus, correctly tuning agitation, base
result of these advances, highly productive industrial HCDC bioprocesses feed location and proper adjustment of the base pump feed rate can yield
will continue to be in demand and have a significant impact on the biotech- robust pH control .
nology industry . Lack of adequate reactor performance and control can lead to hetero-
Mammalian Bioprocessing High-yielding mammalian processes can geneities in the bioreactor . Heterogeneities are generally a result of poor
lower process operating costs and increase biomanufacturing efficiencies . mixing as reactor sizes are scaled up and cell densities increase . These
For mammalian cell cultures, high-density culture is employed to maximize inconsistencies can result in not only nutrient and pH gradients, but also
20-94 BIOREACTIONS AND BIOPROCESSING

oxygen-starved dead zones which facilitate cell lysis and alter product qual-
ity and yield . Poor reactor performance can also result in cell aggregates
and clumping at high densities . Aggregation can cause segregation of cells,
where larger aggregates migrate to lower agitation zones, leading to lower
productivity and reduced process efficiency . Heterogeneity in the cellu-
lar environment can lower cell viability, density, and productivity which

Pervaperator
will reduce the overall performance . Proper mixing, media feeding and Substrate
exchanges, and control in suspension cell cultures will provide more consis- Condenser Product
tent quality and product from mammalian HCDC in biotechnology .
Fermenter
INTEGRATED CELL CULTURE AND PRODUCT RECOVERY
The biotechnology industry has leveraged human ingenuity to manipulate
and modify organisms to produce pharmaceuticals, antibodies, biofuels, Effluent
metabolites, and other high-value biochemicals . Since the early 1990s, the
industry has witnessed significant growth in production, commercial-
ization, and scale . However, this growth is limited by costly and separate
upstream cell culture/fermentation and downstream product recovery Compressor
steps . Product compounds are typically produced by batch, fed-batch, or
continuous fermentation, followed by filtration or sedimentation to sepa- FIG. 20-75 Schematic of integrated fermentation-pervaporation process .
rate cells from spent medium . The media are then subjected to a series of
costly separation steps for secreted products . Unfortunately, this distinct
separation phase can be inefficient and costly due to energy requirements,
supplies, material transport, and equipment maintenance [Schügerl, Karl, is often used to separate two or more immiscible liquids such as an organic
Biotechnology Advances 18: 581–599 (2000)] . liquid-phase solution from an aqueous liquid-phase fermentation solution,
In recent years, methodologies have been implemented to eliminate the which are in direct contact unlike perstraction which separates the phases
product recovery bottleneck by integrating the separation and fermentation using a membrane . To perform this extraction, however, the extraction sol-
processes . Efforts to enhance production efficiency include techniques that vent must be nontoxic to the culture and have a high partition coefficient for
incorporate the cell culture and product recovery steps into a single or a few the component of interest . Economic studies show that ethanol extraction
steps in order to rapidly develop higher-quality products at lower cost and from a continuous fermentation is more cost-effective than conventional
with improved bioprocess performance . Some of these methods include separation methods for ethanol recovery . Other applications involving
pervaporation, perstraction, extraction, adsorption, gas stripping, precipi- extraction include separations of antibiotics such as cephalosporins or
tation, and crystallization (Table 20-42) . These methods of in situ product other beta-lactam precursors . Extraction from culture, however, does not
recovery are also desirable since product compounds such as ethanol, buta- always remove the accumulated organic acids or metabolites efficiently .
nol, and others can be toxic or inhibit growth at the elevated concentrations Furthermore, toxicity of the solvent, especially when organics are used, can
present in a fermentation culture . In situ recovery allows for increased sub- be an issue, which can be reduced by either immobilizing the cells in culture
strate consumption, higher cell densities, elevated product yields, and more or using protective agents such as propylene glycol .
efficient and economical integrated operations . To minimize the toxicity of an organic, two-phase aqueous extraction can
Pervaporation One method for bypassing the product recovery bot- be implemented in which both immiscible phases are water-based solu-
tleneck is to include a pervaporation process (Fig . 20-75) . During pervapo- tions . The two phases are distinct based on the presence of two distinct
ration a pressure differential is applied across a highly selective membrane dissolved polymers, such as polyethylene glycol and dextran in the two
which separates volatiles from the biologics recycle stream . Pervaporation aqueous phases, or include a polymer in one phase and a salt solution in
is usually carried out with the feed stream set at atmospheric pressure while the other . These aqueous two-phase extraction systems can be especially
the permeate stream is subject to vacuum . The difference in partial pres- advantageous for recovering proteins or other sensitive biologics since the
sure between the components in the feed stream causes the species with a compounds will not denature in either of the two liquid phases . Ionic liq-
higher vapor pressure to evaporate, thereby causing a concentration gradi- uids, in which salts are present in a liquid state, represent another potential
ent within the membrane, which becomes the driving force for mass trans- extraction solvent that eliminates organics, although their capabilities in
fer . Permeate from the pervaporation process is condensed and subjected to integrated recovery systems are not demonstrated .
a cold trap for further separation and improved process selectivity . Gas Stripping and Adsorption Gas stripping is a separation process
Perstraction Perstraction is similar to pervaporation in that a selectively whereby a vapor stream is used to physically remove one or more compo-
permeable membrane separates the fermentation stream from the separation nents from a liquid stream . Adsorption is the adhesion of a substance known
stream . However, perstraction membranes are hollow tubes that allow the as an adsorbate to a surface known as an adsorbent . Gas stripping and adsorp-
flow of two liquid streams on separates sides with a solute material transferred tion have been applied to remove volatile metabolites from the vapor phase of
between them . This process works by circulating fermentation broth through fermentation processes . Recovery by stripping the fermentation culture can
the membranes and back to the vessel, allowing only the component of inter- be achieved by sparging with an inert gas (Fig . 20-76) . An inert gas bubbling
est to seep through into the harvesting stream . This harvesting stream will throughout the broth forces volatile components into the gas phase .
include a solvent, often organic, that solubilizes one of the metabolites of inter- The gas-phase mixture is then sent through either a condenser or an
est from the culture media to be purified . The target metabolite mixes and dis- adsorption process, which removes target compounds and recycles the
solves into solvent with a high solubility for the component of interest and is inert stream . Studies have been conducted to test for the recovery of ethanol
subsequently purified . One advantage of perstraction is that toxic products and butanol produced by fermentation . Adsorbent media such as activated
from a fermentation can be eliminated; however, the presence of an in-stream carbon, polystyrene, charcoal, and divinylbenzene-styrene beads have also
membrane separation can be costly and subject to clogging and fouling . been tested . In situ adsorption may be used in an effort to remove ammonia
Extraction Extraction refers to the separation technique of extracting and other inhibitory side products as well as to improve the production of
a compound from a mixture . In situ liquid-liquid extraction can also be used antibiotics by fermentation . However, in situ adsorption can be adversely
to recover volatile components from a fermentation process . This method affected by clogging due to proteins and cells . Once again, immobilizing

TABLE 20-42 Benefits and Limitations of the Proposed Processes

Stripping Adsorption Extraction Pervaporation Perstraction
Capacity Moderate Low High Moderate Low
Selectivity Low Low High Moderate High
Fouling Low High Moderate Low Low
Operational simplicity High Low Low High High
When one is designing a process, the demand, selectivity, and process maintenance should be taken into account
in order to maximize overall process effectiveness . From Groot, W . J ., van der Lans, R . G . J . M ., and Luyben, K . Ch . A .
M ., “Technologies for Butanol Recovery Integrated with Fermentations,” Process Biochemistry, 27: 64 (1991) .
 EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS 20-95

Fermenter gas exchange resins can become clogged with suspended cells, these fermenta-
tions are also often carried out after immobilizing cells .

ADVANCED BIOREACTOR CONTROL TECHNIQUES

Condenser
For industry, control of bioreactors can help solve many of the prob-
lems associated with inconsistent product quality and variable product
output . By maintaining bioreactor operation at optimal operating con-
Gas ditions throughout the fermentation life cycle, much of the uncertainty
that comes from bioreactor performance variability can be minimized .
Excess gas Eliminating uncertainty and maintaining product quality and consis-
Fermenter tency through optimal control of biomanufacturing operations are two of
the major objectives of the highly regulated biopharmaceutical industry .
The abilities to satisfy regulatory requirements, meet production goals,
and speed up process development times are all vital to the success of
an industrial biotechnology operation . For this reason, advanced process
Condensate control methodologies will become increasingly important in biotechnol-
receiver ogy . Some of the advanced process control methods that can be used in
FIG. 20-76 Recovery by gas stripping and condensation .
bioreactors are summarized in Table 20-43 . Advanced control of bioreac-
tors is aimed at optimizing processes and controlling levels of three main
parameters: biomass, substrate, and product output . To achieve these
goals, researchers use a variety of both traditional control techniques
cells in a matrix inside the reactor is a technique that can be used to reduce and more advanced control techniques, such as model-predictive control,
adsorbent clogging . adaptive control, and habituating control .
Precipitation with a Base Precipitation with an appropriate base is a Model-Predictive Control As shown in Fig . 20-77a, MPC functions by
simple and gentle method for recovering citric or lactic acid produced dur- tuning control parameters based on a process model . Through the use of the
ing fermentation . Usually a weak base such as calcium carbonate is added to model, the experimenter determines the optimal operating conditions and
the broth; in solution the calcium carbonate reacts with organic acids and optimal controller settings for the system . The controller can then be tuned
makes calcium salts which precipitate out of solution due to a low aqueous to desired settings to maintain these operating conditions . The model can
solubility . The precipitate is then removed from the broth, and the organic be either an ad hoc form based on an off-line mathematical model or an
acids are liberated by one of the following methods: treatment with a strong adaptively determined linear dynamic on-line model . In the off-line model,
acid, solvent extraction, or electrodialysis . theoretical laws and correlations are used to create a model . The use of
Electrodialysis Electrodialysis (ED) is conducted by applying an elec- theoretical predictions to create a static model is easier as no real-time data
trical potential across a semipermeable membrane allowing for ions to flow must be collected and the model only needs to be derived once . However,
to their respective poles . ED is an effective tool which takes advantage of the the downside is that this type of model can be inconsistent with the actual
ionic character of target metabolites in order to separate and purify them . process when predictions do not match experimental output . Using an on-
Studies have shown that this process can be as effective as precipitation line model helps to eliminate this problem . For these models, the model is
without leading to significant waste products . The downside to electrodialy- based on collected process data, and it can be altered based on the actual
sis is that cell-free broth may be required to preserve the bipolar ED mem- conditions and behaviors of the bioreactor system . The easiest models to
branes; therefore, a cell separation method consisting of immobilization or create are linear models, and these are generated in both static and dynamic
filtration must be employed before ED is performed . varieties . Linearization techniques are used for creating linear models for
Crystallization Crystallization is a phase separation and purification control . Local linearization is used when the perturbations around a steady-
technique wherein crystals are precipitated from a liquid solution . Purifica- state operating condition are small . For example, for a single-variable sys-
tion by crystallization is appropriate primarily with cell-free reaction mix- tem with x as the system variable, the change with time around the steady
tures; thus a process may be established with immobilized cultures . In situ state xs can be linearized with approximation as
crystallization applied to purification and separation of antibiotics can be  ∂g 
performed where the antibiotic crystallizes in a fermenter . For instance, this  g ( x − x s ) ≈ g ( x s ) + ∂ x  ( x − x s ) (20-103)
phenomenon can be seen when producing tetracycline with S. aureofaciens . xs

Ion Exchange Ion exchange is a method used to exchange ions

between two electrolytes or between an electrolytic solution and a sub- Similarly, in nonlinear systems with multiple variables such as a single-
input, single-output system, with s as the system variable, c as the controlled
stance . This method can be applied to the separation of compounds from
variable, and a model given below, the change with time around the steady-
a fermentation culture . Specifically, the ionic character of metabolites
state values ( s s , c s ) can be linearized with approximation as
is exploited when subjecting a fermentation broth to a recovery process
involving ion exchange . This process requires use of a specialized powerful
ionic exchange resin, which will remove organic acids or other metabolites  ds ∂g   ∂g 
from a broth and can be used in situ . In situ studies on the incorporation of  dt = g ( s , c ) ≈ g ( s s , c s ) + ∂ s  ( s − s s ) + ∂c  (c − c s ) (20-104)
ss cs
ion exchange revealed increased process productivity . However, because ion

TABLE 20-43 Comparison of Advanced Bioreactor Control Techniques

Method Salient features Pros Cons
Model-predictive • Predicts corrective • Provides systematic changes • Computational complexity
control control action based based on operating conditions • Often prone to providing
on a mathematical • Offers effective control for ambiguous solution
model multivariable systems
Adaptive • Controls substrate • Gives real-time information on • Presents higher complexity than
control concentration as well state of biomass conventional PID controllers
as production rate • Eliminates errors quickly with • Requires advanced expertise for
fewer fluctuations running and maintenance
Habituating • Provides manipulation • Minimizes the input cost of • Offers limited systematic
control of more inputs than affecting control in nonlinear mechanism for creating additional
the controlled outputs bioreactor models controlled outputs
• Provides efficient performance • Requires complex control efforts
in closed loops by utilizing all for introduction of outputs
available inputs
20-96 BIOREACTIONS AND BIOPROCESSING

Manipulated feed Manipulated feed

(substrate (substrate
concentration) concentration)
Nonmanipulated Nonmanipulated
feed feed

Output Output

Sampling Sampling

Sensors Sensors

Model Adaptive Control

predictive Predictive control algorithm
Optimization
control model Software Software
sensors sensors
(a) (b)
FIG. 20-77 Control systems: (a) model predictive control, (b) adaptive control .

Local linearization can often be applied to provide model flexibility . or this control method can operate to alter the stability conditions of a reac-
Moreover, custom linearization techniques are also used for special non- tor from outside to inside of normal and optimized operating conditions .
linearities . Different factors are considered before choosing a linearization Adaptive control is commonly used to control either the substrate concen-
technique . These factors include, but are not limited to, the specific nature tration in the bioreactor or the production rate of biomass or product . These
of the control problem, the time and cost incurred, and the quality of pro- control schemes can also provide valuable indirect information about the
cess model at hand . Once identified, these simplified process models are state of the biomass within the reactor, thus helping to monitor growth and
easier to work with and for many processes provide an adequate level of rate of generation of a desired product .
control . However, this does not necessarily provide a very robust control Habituation Control This approach is particularly useful for continu-
scheme . Alternative approaches can use mathematical techniques to trans- ous processes; it is a more complex method in which multiple manipulated
form nonlinear systems into more simplified linear systems that enhance inputs are used to control a single output . A major advantage of this control
controller design and application . The application of a nonlinear model is is that the manipulation of several inputs in conjunction with minimizing
more challenging but may increase the validity of the controller . Optimiz- deviations of inputs from steady-state values allows for tighter control of
ing control is typically applied in disturbance situations and in changing the output . For example, one of the commonly controlled parameters in a
operating conditions . For example, model predictive control can be applied bioreactor operation is biomass concentration, which in turn is dependent
to optimize conditions such as biomass and substrate concentration or on the substrate concentration . Substrate concentration in the bioreactor
metabolite and protein production rates . The presence of many distur- can be tightly controlled by using both the dilution rate and the feed sub-
bances will cause the control to break down by disrupting the steady-state strate concentration as manipulated variables . Controlling the substrate
optimum on which control relies . One way to increase control robustness concentration also enables control of other growth parameters such as
is to combine linearization techniques with adaptive control . The adaptive the formation rates of the desired products . Habituating control offers the
control, as described in the next subsection, helps to minimize some of the opportunity to control substrate concentration or biomass by a concerted
effects of the disturbances to allow for tighter control . and coordinated manipulation of both the substrate feed concentration
Adaptive Control As shown in Fig . 20-77b, adaptive control provides and the dilution rate, which increases the rate of substrate into and out of
process control for systems with large perturbations from a steady-state the bioreactor . Furthermore, proper tuning of two parameters rather than a
operating condition by adjusting the control parameters to maintain per- single one can facilitate tight regulation over the control objective and ulti-
formance objectives . This control technique can be applied to any dynamic mately improved process performance and product consistency [Carrondo
process exhibiting parameter variations as well as to those processes which et al ., Biotechnology J . 7: 1522–1529 (2012)] .
have variable parameters due to constantly changing physical conditions,
such as fouling in heat exchangers . It operates as a self-tuning controller ADVANCED SEPARATION APPROACHES*
that works with or without the necessity of a process model . This control
scheme relies upon measurements of certain indicator process variables . Electrophoresis Electrophoresis is a separation technique often
Because this method takes into account real-time data and tunes the con- applied to the analysis of biological or other polymeric samples . This tech-
troller based on these data, adaptive control schemes are often quite robust nique is among the most powerful for estimating purity because of its sim-
in maintaining functionality and accuracy even with high levels of uncer- plicity, speed, and high resolution, and also because there is only a small
tainty . This makes adaptive control a good option for processes with con- probability that any of the components being analyzed will be lost during
stantly changing parameters, and it can also be helped by the presence of the process of analysis . Electrophoresis has frequent application to analysis
a model, even though the control provided may not be the tightest . There of proteins and DNA fragment mixtures . The high resolution of electropho-
are three main types of adaptive controllers: scheduled adaptive control, resis has made it an important tool in the advancement of biotechnology .
model-reference adaptive control (MRAC), and self-tuning adaptive control . Variations of this methodology are being used for DNA sequencing and fin-
Scheduled adaptive control works on a set of predetermined process control gerprinting; for isolating active biological factors associated with diseases
data and uses preprogrammed measures to achieve the control . However, such as cystic fibrosis, sickle cell anemia, myelomas, and leukemia; and
MRAC exercises simulation of a predetermined process reference model for establishing immunological reactions between individual compounds .
with current process data to calculate the difference between actual process Electrophoresis is an effective analytical tool because the electric field does
output and the model simulation output . The goal of MRAC is to reduce the not affect a molecule’s structure, and it is highly sensitive to small differ-
difference between actual process output and the model simulation output ences in molecular charge, size, and sometimes shape . The electrophoresis
by changing process parameters . By contrast, to achieve control, the self- systems with the highest capacity are free-flow electrophoresis, density gra-
tuning adaptive control allows a dynamic model-building process based on dient electrophoresis, recycling free-flow isoelectric focusing, and rotating
current process data as opposed to using a fixed predetermined model . This isoelectric focusing . The principles of operation of these are discussed next .
makes self-tuning control the most flexible adaptive control, but also most
prone to errors as a poor model-building process could potentially lead to ∗Portions of the Advanced Separation Approaches section have been reproduced
a highly unstable system . Adaptive control is effective when used both in from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul
start-up and disturbance situations . Thus, adaptive control can operate to W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission of Oxford
bring a reactor from start-up to stability within normal operating conditions, University Press .
 EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS 20-97

Free-Flow Electrophoresis In free-flow electrophoresis, a sample to be the concentration of each separand . The result is a series of non-Gaussian
separated is continuously injected into a thin rectangular channel through peaks, and yield may be low since separand distributed in the tails of the
which a carrier fluid flows, with an electric field that is perpendicular to the parabolic profiles might not be collected .
flow (Fig . 20-78a) . This results in differential deflection of charged solutes and In the example shown in Fig . 20-78a, the flow is from bottom to top, but
particles, leading to separation by differences in electrophoretic mobility . Appli- downward and horizontal configurations exist . Carrier fluid exits in thin
cation of the electric field is initiated before sample loading and is achieved via outgoing channels at the opposite end of the chamber . The vertical flow is
two electrode compartments separated from the flow chamber by semiperme- laminar, resulting in a parabolic velocity distribution in the yz plane using
able membranes at the two vertical edges of the thin chamber . Each electrode the axes specified by the diagram . Thus
compartment contains a single metal wire electrode and an electrolyte solution
consisting of a nontoxic buffer and dissolved KCl or NaCl at an ionic strength  y2 
higher than that of the carrier buffer . To maximize field strength (hence sepa- v z ( y ) = v max  1 − 2  (20-105)
rand migration velocity) and minimize current density (hence heating and nat-  b 
ural convection), the carrier buffer must have the lowest ionic strength possible
that is compatible with separand solubility and stability (separands could be where y is the distance from the center of the chamber in the y direction and b is
living cells) . An upward-flowing buffer stream becomes warmer as it rises and the chamber half-thickness . This velocity profile means that separands near
is more stable against natural convection than is a downward-flowing stream . the front and back chamber walls move vertically the most slowly, spend
A thermostated water jacket can be in contact with the front and back planes, more time in the electric field, and are therefore carried farther horizontally
which are usually made of glass or polycarbonate . in the field than their identical counterparts flowing faster at the center of
In free-flow electrophoresis, sample solution is injected into one of the the chamber . The time required for a volume element to reach the outlet is
in-flowing channels or directly into a sample port in the chamber . There are
limits on the concentration of sample based on solution density . A sample L
that is too dense will not rise with upward flow, and a sample that is too light t( y) = (20-106)
vz ( y )
will rise to the top of the chamber before significant separation can occur;
vice versa for downward flow . Ideally the carrier buffer and sample solution
should have the same density and very similar conductivities . Collection of where L is the distance from the inlet to the outlet .
separands occurs at the top (or receiving end) of the chamber at collection The electric field also causes laminar electro-osmotic flow at the cham-
ports (“outlet fractions” in the diagram), which are connected by fine tubing ber walls in the x direction (toward the negative electrode) with the result
to a collection vessel . The process is continuous, so carrier buffer is flowing
out of outlets continuously during a separation . Separands exit according to 1  3y2 
v x ( y ) = U o E  2 − 1 (20-107)
their mobility . Separation may be quantified by identifying and measuring 2  b 

Veo(y)
Outlet
fractions
L M H
Collection port

Buffer Electrode
– + + membrane
Vz(y) = Vmax Cooling jacket
Separated
Electrode separands
Vz(y) = 0
membrane Density gradient
Z
Electrode membrane
Y
Sample zone
X +b
0 Electrode
membrane Floor
–b
Sample inlet
–

Buffer Sample
(a) Carrier fluid in (b) + Gradient + Floor

Focusing Neutral
cell screens screens
Loading ports
Cooling
Anolyte tube Catholyte

+ –

+ –
Cooling
tube
Catholyte
pH zones
Collection Rotation Anion exchange
pH zones membrane
reservoirs Collection
(Thermostated) Pump Circulation lines ports
Anolyte
(c) (d)

FIG. 20-78 Schematic drawings of four methods of preparative electrokinetic separation . (a) free-flow electrophoresis (FFE), (b) density-gradient electrophoresis (DGE),
(c) recycling free-flow isoelectric focusing (RIEF), and (d) rotating isoelectric focusing (Rotofor) . In (a), separand concentration trajectories are denoted for low- (L), medium-
(M), and high- (H) mobility separands . [Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P .
Petrides (2015) . By Permission of Oxford University Press .]
20-98 BIOREACTIONS AND BIOPROCESSING

where Uo = electro-osmotic mobility of the chamber walls due to their zeta in high resolution and purity, but the pH range must be selected to include
(surface) potential , E = applied electric field strength (voltage per length) given the isoelectric pH of the separand of interest .
by E = I/keA, and ke = solution electrical conductivity in siemens per meter . The mixing effects of natural convection due to temperature and concen-
Separand molecules near the wall will migrate more slowly toward their tration gradients can be counteracted by rotation of the separation vessel at
complementary electrode than will those at the center of the chamber . This a carefully chosen velocity, resulting in a constantly changing gravity vector,
relationship between laminar flow ( y-z profile) and electro-osmotic flow hence rotating isoelectric focusing (Fig . 20-78d) . The focusing chamber is
( y-x profile) leads to an exact balancing of the two migrations for certain filled with ampholyte solution with the sample mixed in using several of the
separands and gives a flat profile of the separand concentration across the loading ports . After filling without bubbles, the filling ports are sealed, and
thin chamber, as shown by the letter M on the diagram in Fig . 20-78a . And rotation is begun . The ampholytes and sample molecules will focus simul-
L and H correspond to lower and higher mobility separands, respectively . taneously . Cooling water is flowed through the central cooling tube, rota-
Separands are resolved on the basis of distance migrated X in the x direction tion is initiated, then power is switched on . After focusing, the solution in
according to the following equation: each compartment is rapidly drained through the collection ports without
mixing .
 1  3y2  
X = U el t ( y ) + U o  2 − 1 t ( y )  E (20-108)
 2  b   MAGNETIC BIOSEPARATIONS

Density Gradient Electrophoresis (DGE) DGE takes advantage of Magnetic capture is a powerful, highly selective method for collecting sepa-
differences in the electrophoretic mobility and buoyancy of particles . In rands present in low abundance in a highly mixed suspension or solution .
this type of electrophoresis, separands move through a density gradient in a Once a separand has reacted with magnetic beads by extensive mixing,
static liquid (Fig . 20-78b) . Separation of particulate separands is slowed due resulting in affinity adsorption, the beads are captured and held by a mag-
to increased upward buoyant force . In density gradient electrophoresis, car- netic force so that all other components can be rinsed away, much as in a
rier buffer is loaded (typically) from the bottom of a cylindrical column by centrifugation wash step . The general principle is not unlike that of affinity
a density gradient maker with low density followed by high density . Sample chromatography, in which ligand molecules are affixed, usually by covalent
is loaded in a solution having higher density than the bottom of the den- reactions, to polymer microspheres (or nanoparticles) that contain iron
sity gradient, and it is followed by a very dense fluid that forms a “floor” oxide ( from 3 to 80 percent by mass) . Examples of ligand molecules include
under the fluids above it in the column . The electric field is applied after nucleic acids, glycoconjugates, avidin, and most often antibodies . Magnetic
loading the sample . Applying an electric field to a column of suitable geom- reagents have been manufactured for several purposes, and their sizes now
etry results in negligible electro-osmotic flow relative to the electrophoretic range from some 20 nm up to 10 µm in diameter . While most are spherical,
velocity of the separands . However, Joule heating of the fluid is inevitable some may be irregular in shape, especially if the iron oxide content is high .
and is counteracted by thermostating the column with a water jacket . The Physical Principles of Magnetic Separations It is desirable that
density gradient counteracts the thermal convection due to Joule heating . much of the iron oxide be in the form of magnetite, Fe3O4, which provides
The separands should be allowed to migrate upward to ensure adequate unpaired electrons whose orbits can be oriented in a magnetic field, result-
separation . For collection of separands, the contents of the column are ing in magnetization . This condition is referred to as paramagnetism, in
either drained slowly out the bottom or pumped out the top of the column, which the material is magnetic if, and only if, it is in a magnetic field . This is
and fractions are collected into individual containers as in preparative chro- important because materials that retain their magnetism after the magnetic
matography . The resolution of DGE is based on the number and size of frac- field is removed are likely to aggregate and become difficult to handle . Para-
tions collected . Purity can be very high depending on fraction volume . magnetic materials do not exhibit remanent magnetization as ferromag-
The forces acting on a particle in density gradient electrophoresis are the netic materials do, such as iron, cobalt, and permanent-magnet ceramics .
electric field, the gravitational field, and the buoyant force . Balancing these When a magnetizing field H (measured in amperes per meter, A m−1)
forces leads to a particle velocity for particles migrating upward toward the returns to zero, ferromagnets are still magnetic, but paramagnetic materi-
anode (positively charged electrode) as als are not . The magnetization M is proportional to H so that

2
M = χH (20-110)
2 a p g (ρ−ρo )
v = U el E − (20-109)
9 µ where the constant of proportionality is the volume magnetic susceptibility
χ . Magnetic flux density B is usually expressed in SI units of tesla (T, 1 T =
where Uel is the electrophoretic mobility of the charged particle, E is the field 1 N A−1 m−1) and is related to the applied field by the magnetic permeability,
strength or voltage gradient (voltage V per length L between the electrodes), which in free space is µo = 4p × 10−7 T m A−1, a universal constant that can
and the second term is the sedimentation velocity for spheres of density r be expressed in the more convenient units of newtons per squared ampere
and radius ap in a fluid of density ro and viscosity µ under the influence (N A−2):
of gravity . This velocity will not be constant as a function of the distance
migrated in the z direction (upward) because ro changes with z . B = µ0 H (20-111)
Recycling Free-Flow and Rotating Isoelectric Focusing In recy-
cling isoelectric focusing (RIEF, Fig . 20-78c), electrophoretic separation In the presence of a large magnet having magnetizing field H (A m−1),
takes place as separands, in a solution containing ampholytes, flow through a small particle of volume V and susceptibility ∆χ above that of the sur-
compartments in a focusing cell across which an electric field is applied . rounding fluid will experience magnetization M = ∆χ H (A m−1) and a force
Separands and ampholytes pass repeatedly from compartment to compart- Fm (in newtons) that is proportional to the gradient of the flux density pro-
ment and back to a reservoir associated with each compartment by elec- duced by the external magnet ∇B (T m−1) at the instantaneous location of
trophoresis until they become neutral at their isoelectric pH in one of the the particle, given by Moeser et al ., AIChE J . 50: 2835 (2004),
compartments . At this point, electrophoretic migration ceases, and steady
state has been reached . Then each of the reservoirs is drained for sample Fm = V∆χH ⋅ ∇B (20-112)
collection . The number of fractions equals the number of reservoirs . It is
necessary to adjust flows to be the same in and out of all the reservoirs; flow but since B = µoH, the force can be expressed in terms of the strength of the
balancing is very important . permanent magnet:
The electric field is applied across the compartmented cell only, and the
initial current is typically several milliamperes . While the electric current B
Fm = V ∆χ ⋅∇B (20-113)
and pumps are on, fluid is circulated out of the reservoirs through the com- µo
partments of the electrophoresis cell and back to the reservoirs . Eventually
the fluid in each compartment is at a different pH, resulting in a pH gradient The features of this relationship are used to understand the equipment
across the separation cell . Each separand migrates to the compartment in required to perform magnetic separations using magnetic reagents . The
the cell that is at its isoelectric pH, where Uel = 0, and therefore is found in the inducing field H plays the role of magnetizing the paramagnetic reagent
corresponding reservoir having that pH . At the end of the focusing period, by orienting unpaired electron spins so that the reagent will be attracted
the fluid contained in each reservoir is drained into a separate container, its toward the magnetizing source by the gradient ∇B (see Fig . 20-79a) .
pH is measured, and the contents are assayed for separands . The relatively Equipment and Reagents The volume of available commercial
small number of fractions will typically result in “single-fraction peaks” in magnetic particles is a deliberate variable that may vary by more than
the plot of concentration versus pH . A broad range of pHs, for example 2 106-fold (4 .0-µm versus 40-nm diameter, for example) . The quality of mag-
to 10, will result in low resolution and purity; each reservoir could contain netic material incorporated into the bead determines ∆χ, which is the rela-
several separands . A narrow range of pHs, for example, 4 .2 to 4 .8, will result tive susceptibility of the average bead material, that is, its susceptibility
 EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS 20-99

Magnetic nanoparticles

Polymer coat

Reactive linker
Bo
Biospecific
ligand (Ab)
(b)
M = B/µo

∇B Sample mixture
(a)
Buffer stream

Sample mixture

Steel beads

Magnetic particles
on the surface of Magnetic
steel beads
(c) (d) Nonmagnetic

FIG. 20-79 Magnetic separations . (a) The role of magnetic flux density B and gradient ∇B in applying ponderomotive
(motion-causing) force to a magnetically susceptible particle (represented as a circle with an arrow passing through it) .
The arrow through the particle represents the polarization of the magnetization M. (b) The general configuration of a
magnetic reagent nanoparticle or microparticle showing susceptible core, nonsusceptible material coat, linker reac-
tive molecule, and external ligand for reacting with separands with high specificity . (c) Operation of a column-based
(equilibrium) separator showing packed column of steel beads or susceptible metal wool to which magnetic reagents
adsorb and from which they are desorbed for batch collection by removing the magnetic field . (d) Operation of a flow-
based (rate) cylindrical separator having a core and inlet and outlet splitters showing removal of magnetic reagents
from an inner to an outer parallel annular flow for continuous collection of separands . [Reproduced from Bioseparations
Science and Engineering, 2d ed ., by Roger G . Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By
Permission of Oxford University Press .]

minus that of the surrounding fluid . Parenthetically, a solution containing formation are utilized by the different magnets used in the different devices:
a high concentration of paramagnetic solutes (ions) could diminish or even (1) The simpler “tube and magnet” types expose the suspension of magnetic
reverse this value . There are four features of magnetic reagents to consider reagent (beads, ferrofluid) to one pole of a dipole magnet, which forms flux
besides bead diameter including the manner in which beads are manufac- lines as shown in Fig . 20-79a . (2) When steel spheres or wires are placed in
tured, the magnetically susceptible material, the structural bead material a column served by dipole magnets, flux lines enter the spheres essentially
(nonsusceptible), the conjugation site on the bead material, and the biospe- perpendicular to the sphere surface, creating intensely converging field lines
cific ligand (see Fig . 20-79b) . Table 20-44 is a list of each of these choices, and and therefore a high flux gradient leading to the term high-gradient magnetic
the listed components can be combined in all possible permutations . separation (HGMS) as in Fig . 20-79c . (3) Three forms of Halbach magnet
Magnetic separation devices fall into three broad categories: tube and assemblies have been used and are available for larger-volume separations .
magnet equilibrium separators, high-gradient batch column separators, and A Halbach array is a special arrangement of multiple permanent magnets
flowing continuous separators . Different methods of magnetic field gradient that augments the magnetic field on one side of the array while canceling the
field to near zero on the other side . A Halbach quadrupole assembly forms a
central bore within which the channel of the quadrupole magnetic separator
TABLE 20-44 Examples of Susceptible Materials, Nonsusceptible
(Fig . 20-79d) is held .
Magnetic Separation Applications Applications include bench-
Materials, Reactive Linkers, and Biospecific Ligands Used in
scale protein and nucleic acid purification, protein and nucleic acid diag-
Magnetic Reagents
nostics (“bead-based assays”), and magnetic cell sorting . Bench-scale
Susceptible Nonsusceptible Reactive Biospecific purification is typically achieved using biospecific affinity ligands in much
materials materials linkers ligands the same way as in affinity chromatography . Magnetic purification can also
Magnetite (Fe3O4) Polystyrene Carboxyl Biotin
be used as a concentration step . A typical procedure consists of coupling the
biospecific ligand to the surface of magnetic beads . Unreacted antibody (or
Maghemite (γ-Fe2O3) Silica Amino Biotin (or streptavidin)
other ligand) is washed away by magnetically aggregating the paramagnetic
Alginate Epoxy Antibodies particles . Magnetic separation has unique advantages in cell purification;
Agarose Tosyl Glycoconjugates it may be used to remove, select, or enrich specific subsets in cell mixtures
Complementary nucleic for application to regenerative medicine and rare-cell analysis . In the case of
acids selected cells to be retained, chemical and enzymatic methods are available
Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G . for the removal of magnetic reagents from cells after separation .
Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission There are many ready-made magnetic reagents precoated with biospe-
of Oxford University Press . cific ligands available commercially . If desired reagents are unavailable,
20-100 BIOREACTIONS AND BIOPROCESSING

on the individual components’ interaction with the stationary phase relative

to their interactions with an eluent, which is responsible for eluting the indi-
vidual products from the mixture . In SMB, the column is set up so that the
two products come out at different outlets due to their differing interactions
with the stationary solid and eluent or desorbent liquid phase . Assuming
the mixture only has two components, their velocities can be determined
experimentally or predicted theoretically . The component with stronger
1 2 3 4 interactions with the column will have net movement in one direction, and
the component with weaker interactions will have net movement in the
opposite direction . Assuming the input is located at the middle of the col-
umn and the column is long enough, the separated components can be col-
lected at opposite ends . To keep the movement near continuous, each bed
is split into subunits (Fig . 20-80) . For small-scale separations, 4 to 8 subunits
are used; for large-scale, up to 24 . Instead of moving the bed, the desired
and undesired product exit stages are continuously adjusted to create the
simulation of a moving bed (Fig . 20-81) .
FIG. 20-80 Simulated moving-bed column illustration . [Li et al . Continuous Process- Advantages of SMB relative to traditional elution chromatography include
ing in Pharmaceutical Manufacturing 1: 1–27 (2014) .] the fact that, because the process is only space-dependent, it can be run
continuously . SMB is applicable at any scale; the only adjustments involve
sizes of columns and equipment . Although SMB is usually conducted in the
liquid phase, it can also be carried out in the gas phase, allowing for more
efficient transfer . SMB also has lower solvent consumption and more effi-
they can be synthesized by the user . Vendors of magnetic beads provide cient use of the solid adsorbent bed compared to traditional chromatogra-
chemically activated surfaces and instructions for the conjugation of phy, as it can be run on overload . SMB also requires fewer theoretical plates
user-supplied biospecific ligands . It is easy to remove unreacted biospe- to achieve the same purity as with standard elution chromatography . When
cific ligand and (if used) linker compounds by holding the beads in place traditional separation techniques are too harsh, SMB can be used instead
with a magnet while the reactant solution is decanted and the beads of distillation, etc . The disadvantage to SMB, however, is that it is primarily
are rinsed with final buffer before resuspending . To purify a separand, used to split a mixture into two; so if there are more than two components
beads are added to the mixed sample solution and held—often at 4°C to be separated, several units must be used [Juza et al ., Trends in Biotechnol .
to preserve specificity—with gentle agitation while separand molecules 18: 108–118 (2000)] .
or cells collide with the bead surfaces and react . Adequate agitation is Theory A moving bed can be formulated to include salt gradient in ion
required for this reaction, since magnetic beads can be considered “non- exchange mode in which two initial sections have a higher salt concentra-
Brownian” and all reagents must be transported to them by diffusion and tion, and the next two sections have a lower salt concentration, thus creating
mixing, a salt gradient [Li et al ., Continuous Processing in Pharmaceutical Manufac-
turing 1: 1–27 (2014)] . In ion exchange chromatography used for protein
purification, the gradient allows for desorption of adsorbed proteins in the
first two sections, and a higher adsorption of proteins in the two farther
SIMULATED MOVING-BED CHROMATOGRAPHY downstream sections . The gradient can be created by introducing lower salt
Typically, downstream separation and purification of proteins relies on dif- concentration at the feed and higher salt concentration at the desorbent
ferent physical properties, such as size, charge, hydrophobicity, and specific portion . Equation (20-114) represents the mass balance over volume of the
binding capability . Simulated moving-bed (SMB) chromatography is a semi- bed k for proteins and salt . Equation (20-115) represents the mass balance in
continuous or continuous alternative to conventional batch elution chro- the particles for proteins and salt . These balances are described according
matography . Traditional chromatography is a time-dependent process: it to the linear driving force approximation .
involves collecting the components at different times from the same outlet . Mass balance over a volume element of the bed k for proteins and salt is
SMB, on the other hand, is space-dependent and accumulates the separate
components simultaneously from different locations . ∂C ik ∂ 2 C ik uk ∂C ik 1 − ε B
In moving-bed chromatography, the feed is provided and products are = DLk − − kPik (qik∗ − qik ) (20-114)
recovered simultaneously and continuously, but there are practical chal- ∂t ∂Z 2 εB ∂Z εB
lenges to implementing this approach . Alternatively, in a simulated moving
bed, the feed inlet, the eluent, or desorbent inlet and two output product Mass balance in the particles for proteins and salt is
positions are moved continuously, giving the impression of a moving bed
(Figs . 20-80 and 20-81) . When a mixture passes over a solid column, the ∂qik
components in a mixture react differently with the stationary phase and = kPik (qik∗ − qik ) (20-115)
therefore move at different rates . They are also removed differently based ∂t

Feed Extract Raffinate Feed

1 1 1

4 2 4 2 4 2

3 3 3
Raffinate Eluent Eluent Extract
(a) (b) (c)
FIG. 20-81 Simulated moving-bed operation: (a) continuous rapid circulation of flow in one direction; (b) and (c) are examples of configurations that create
the simulated movement by changing the feed valves . [Li et al . Continuous Processing in Pharmaceutical Manufacturing 1: 1–27 (2014) .]
 EMERGING BIOPHARMACEUTICAL AND BIOPROCESS TECHNOLOGIES AND TRENDS 20-101

EXPANDED-BED CHROMATOGRAPHY
Out
Proteins are involved in a number of industries such as pharmaceuticals,
food processing, textiles, leather goods, detergents, and even paper . Effec-
tive protein separation and purification is essential for economic produc-
tion of quality product . A three-step separation and purification method is
typically employed . These three steps include capture, intermediate separa-
tion, and final purification [Li et al ., Continuous Processing in Pharmaceuti-
cal Manufacturing 1: 1–27 (2014)] .
The expanded bed utilizes particles in a fluidized environment, whereas
classic column chromatography uses a solid phase in a packed bed .
Expanded-bed chromatography utilizes a particle size gradient which Zone 3 L3
allows for local zones of separation (Figs . 20-82 and 20-83) . Following cap- Product 3
ture, the target proteins can be eluted through an elution solution sepa-
rately from contaminants and impurities in the original entities . Strides are
being made by using expanded-bed adsorption for primary recovery of pro-
teins from a variety of sources including, but not limited to, E. coli homog- Zone 2 L2
enate, yeast fermentation, mammalian cell culture, milk, and animal tissue Product 2
extracts [Hjorth, Trends Biotechnol. 15: 230–235 (1997)] .
Adsorbents When an adsorbent is chosen, there are a number of
trade-offs to consider . The current trend to use dense solid-core materi- Zone 1 L1
als as adsorbents has the benefits of allowing much higher feed velocities
while not being sensitive to ionic strength . Low-density adsorbents have Product 3
been shown to have a higher adsorption capacity in pilot-scale experiments, Feed
but that capacity has been shown to decrease in the presence of high ionic
strength . An effective adsorbent will have a high-density matrix and a salt- FIG. 20-83 Three-zone model for protein adsorption kinetics in expanded beds .
tolerant ligand . The use of a high-density matrix minimizes the consump- Zone 1 is characterized by its location at the bottom of the column, large adsorbent
tion of dilution buffer . The use of a salt-tolerant ligand is preferred because particle size, small bed voidage, and significant liquid axial diffusion . Zone 2 is char-
there is a lack of sensitivity to ionic strength and salt concentration does not acterized by its location at the middle of the column, smaller adsorbent particle size, a
decrease through dilution [Asghari and Jahanshahi, J. Chromatogr . A, 1257: larger bed voidage, and a weaker liquid axial dispersion than in zone 1 . Zone 3 is char-
89–97 (2012)] . A number of adsorbents have been utilized for research and acterized by its location at the top of the column, the smallest adsorbent particle size,
high bed voidage, and smallest liquid axial dispersion . [Li et al . Continuous Processing
commercial purposes, as shown in Table 20-45 . in Pharmaceutical Manufacturing 1: 1–27 (2014) .]
Modeling of the Expanded Bed Modeling an expanded bed is
more complex than modeling a fixed bed . A three-zone model takes into
account the variations in adsorption behavior throughout the bed . The
variations are due to different adsorbent particle sizes, bed voidage, and
liquid axial dispersion coefficients at the bottom, middle, and top zones of
the columns (zones 1 to 3 in Fig . 20-83) and are described by Eqs . (20-116) Equations (20-116) through (20-118) correspond to the material balance
through (20-118) . for the bulk liquid phase in each zone, the mass balance for the adsorbent
phase in each zone, and the pore diffusion for the adsorbent in each zone,
respectively .

∂2 C k uk ∂C k ∂C k 1 - e Bk 3
DLk - - - k fk [C k − (c k )r = Rk ] = 0 (20-116)
∂ Z 2 e Bk ∂ Z ∂t e Bk Rk

∂q k ∂2 qk 3 TABLE 20-45 Recent Commercial and Research Adsorbent Use in

(1 − ε Bk ) = DSk + (1 − ε Bk ) k fk [C k − (c k )r = Rk ] (20-117) Expanded-Bed Chromatography*
∂t ∂Z 2 Rk
Core Year
Adsorbent components developed Use
∂c ∂q  ∂ 2 c 2 ∂c k 
ε p k + k = ε p D p  2k + (20-118)
∂t ∂t  ∂r r ∂r  β-Cyclodextrin polymer Tungsten carbide 2010 Research
Agarose Tungsten carbide 2010 Research
6% Cross-linked agarose Crystalline quartz 2010 Commercial
6% Cross-linked agarose Crystalline quartz 2010 Commercial
Cellulose Tungsten carbide 2011 Research
6% Cross-linked agarose Crystalline quartz 2011 Commercial
Hydrogel-filled Zirconium oxide 2011 Commercial
Agarose Nickel (nanoporous) 2012 Research
Agarose Zinc (nanoporous) 2012 Research
6% Cross-linked agarose Crystalline quartz 2012 Commercial
6% Cross-linked agarose cover Crystalline quartz 2012 Commercial
with polyvinyl pyrrolidone
3% Agarose Tungsten carbide 2013 Research
Polyacrylamide-based Cryogel Titanium oxide 2013 Research
6% Cross-linked agarose Tungsten carbide 2013 Commercial
Agarose Tungsten carbide 2013 Commercial
Hydrogel-filled Zirconium oxide 2013 Commercial
6% Cross-linked agarose Crystalline quartz 2013 Commercial
Agarose Tungsten carbide 2013 Commercial
Agarose Tungsten carbide 2013 Commercial
(a) (b) (c) (d) Agarose Tungsten carbide 2013 Commercial
FIG. 20-82 Expanded-bed chromatography . (a) Adsorbent is embedded; (b) the col- ∗Adapted from Li, P ., Gomes, P . F ., Loureiro, J . M ., and Rodrigues, A . E ., “Proteins
umn is fluidized and a concentration gradient is formed; (c) the feed is injected, and Separation and Purification by Expanded Bed Adsorption and Simulated Moving Bed
particulates and debris move past the bed and out of the column, while the product Technology,” Continuous Processing in Pharmaceutical Manufacturing 1: 1–27 (2014) .
interacts with the adsorbent; (d) the column is repacked and the flow is reversed to Reproduced from Bioseparations Science and Engineering, 2d ed ., by Roger G .
collect the product . [Li et al . Continuous Processing in Pharmaceutical Manufacturing Harrison, Paul W . Todd, Scott R . Rudge, and Demetri P . Petrides (2015) . By Permission
1: 1–27 (2014) .] of Oxford University Press .
This page intentionally left blank
 Section 21

Solids Processing and Particle Technology

Karl V. Jacob, B.S. Fellow, The Dow Chemical Company; Lecturer, University of Michigan; Fellow,
American Institute of Chemical Engineers (Section Editor, Particle Characterization, Pneumatic Conveying,
Screening)

Greg Mehos, Ph.D., P.E. Senior Project Engineer, Jenike & Johanson, Inc. (Bulk Solids Flow and Hopper
Design)

John W. Carson, Ph.D. President, Jenike & Johanson, Inc., Founding member and past chair
of ASTM Subcommittee D18.24, “Characterization and Handling of Powders and Bulk Solids” (Bulk Solids Flow
and Hopper Design)

Yi Fan, Ph.D. Associate Research Scientist, The Dow Chemical Company (Solids Mixing)

Ben J. Freireich, Ph.D. Technical Director, Particulate Solid Research, Inc. (Solids Mixing, Size
Enlargement)

James F. Koch, M.S. Senior Process Engineering Specialist, The Dow Chemical Company (Size
Reduction, Screening)

Shrikant V. Dhodapkar, Ph.D. Fellow, The Dow Chemical Company (Feeding, Metering, and Dosing);
Fellow, American Institute of Chemical Engineers

Pradeep Jain, M.S. Senior Fellow, The Dow Chemical Company (Feeding, Metering, and Dosing)

PARTICLE CHARACTERIZATION Centrifugal Sedimentation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-12

Particle Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-5 Sieving Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-13
Specification for Particulates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-5 Elutriation Methods and Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-13
Particle Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-5 Differential Electrical Mobility Analysis (DMA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-13
Particle Size Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-5 Surface Area Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Model Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 Particle Size Analysis in the Process Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 At-Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Average Particle Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 On-Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Specific Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 In-Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Particle Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Equivalent Projection Area of a Circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-6 Reference Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-14
Feret’s Diameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-7
Sphericity, Aspect Ratio, and Convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-7 BULK SOLIDS FLOW AND HOPPER DESIGN
Fractal Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-7 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-15
Sampling and Sample Splitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-7 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-15
Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-8 Flow Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-15
Wet Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-8 Flow Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-16
Dry Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-8 Analysis of Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-17
Particle Size Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-8 Solids-Induced Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-17
Laser Diffraction Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-8 Cylinder Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-18
Image Analysis Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-9 Hopper Section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-18
Dynamic Light Scattering Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-10 Funnel Flow Hoppers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-19
Acoustic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-11 Eccentric Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-19
Single-Particle Light Interaction Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-11 Bulk Solids Flow Properties Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-19
Small-Angle X-Ray Scattering Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-11 Cohesive Strength and Internal Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-19
Focused-Beam Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-11 Bulk Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-21
Electrical Sensing Zone Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-11 Wall Friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-22
Gravitational Sedimentation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-12 Permeability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-23
Sedimentation Balance Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-12 Abrasive Wear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-23

The authors gratefully acknowledge the contributions of Wolfgang Witt, Ralf Weinekötter, Douglas Sphar, Erik Gommeran, Richard Snow, Terry Allen, Jim Litster, and Bryan Ennis,
coauthors of the 8th Edition, Sec . 21 .

21-1
21-2 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Bin Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-23 Sifters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62

Mass Flow Hopper Angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-23 Tumbling Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62
Mass Flow Outlet Dimensions to Prevent Arching . . . . . . . . . . . . . . . . . . . . . . . . . 21-24 Probability Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-66
Mass Flow Discharge Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-28 Revolving Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-66
Funnel Flow Outlet Size to Prevent Arching and Ratholing . . . . . . . . . . . . . . . . . 21-28 Paddle-Driven Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-66
Expanded Flow Hopper Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-29
Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-29 SIZE REDUCTION
Useful Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-29
Flow aids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-31 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-67
Inserts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-31 Industrial Uses of Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-67
Air Assist and fluidization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-31 Types of Grinding: Particle Fracture versus Deagglomeration . . . . . . . . . . . . . . . 21-67
Standpipes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-32 Wet versus Dry Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-67
Chutes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-32 Typical Grinding Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-68
Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-32 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-68
Caking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-33 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-68
Process Vessels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-34 Single-Particle Fracture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-68
Energy Required and Scale-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-69
Energy Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-69
SOLIDS MIXING Fine Size Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-69
Principles of Solids Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-35 Breakage Modes and Grindability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-69
Industrial Relevance of Solids Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-35 Grindability Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-70
Mixing Mechanisms: Dispersive and Convective Mixing . . . . . . . . . . . . . . . . . . . . . 21-36 Operational Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-71
Segregation in Solids and Demixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-36 Mill Wear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-71
Percolation Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-36 Safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-71
Agglomeration Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-37 Temperature Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Vibration-Induced Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-37 Hygroscopicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Interstitial Fluid-Driven Segregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-37 Dispersing Agents and Grinding Aids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Other Segregation Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-37 Cryogenic Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Mixture Quality: The Statistical Definition of Homogeneity . . . . . . . . . . . . . . . . . . . 21-37 Size Reduction Combined with Other Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Equipment for Mixing of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-38 Size Reduction Combined with Size Classification . . . . . . . . . . . . . . . . . . . . . . . . . 21-72
Bunker and Silo Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-38 Internal Size Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-73
Pneumatic Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-39 Size Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-73
Tumbling Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-40 Other Systems Involving Size Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-73
Agitated Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-40
Paddle and Plow Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-40 MODELING AND SIMULATION
Ribbon Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-40
Fluidizing Paddle Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-40 OF GRINDING PROCESSES
Screw Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-41 Modeling of Milling Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-73
Sigma-Blade Mixers or Z-Blade Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-41 Batch Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
Impaction Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-41 Grinding Rate Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
High-Shear Mixers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-42 Breakage Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
Other Mixing Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-42 Solution of Batch Mill Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
Blending technology selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-42 Continuous-Mill Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
Residence Time Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-74
Solution for Continuous Milling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-75
PNEUMATIC CONVEYING OF SOLIDS Closed-Circuit Milling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-75
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-43 Data on Behavior of Grinding Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-76
The Gas/Solid System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-43 Grinding Rate Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-76
Zenz or State Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-44 Scale-Up and Control of Grinding Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-76
Saltation Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-44 Scale-Up Based on Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-76
Choking Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-46 Parameters for Scale-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-76
Taxonomy of Pneumatic Conveying Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-46
Pneumatic Conveying System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47
Air Movers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47 CRUSHING AND GRINDING EQUIPMENT:
Fans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47 DRY GRINDING—IMPACT AND ROLLER MILLS
Blowers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47 Jaw Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Compressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47 Design and Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Conveying Pipe and Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-47 Comparison of Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Feeders for Pneumatic Conveying Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-48 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Design and Analysis of Dilute-Phase Conveying Systems . . . . . . . . . . . . . . . . . . . . . 21-49 Gyratory Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Gas Pressure Drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-49 Design and Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-77
Acceleration Pressure Drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-49 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-78
Pressure Drop Due to Lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-50 Control of Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-78
Bend Pressure Drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-50 Impact Breakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Solids Friction Pressure Drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-51 Hammer Crusher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Pressure Drop Due to Solid Gas Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-51 Cage Mill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Calculation Strategy for Dilute-Phase Conveying . . . . . . . . . . . . . . . . . . . . . . . . . . 21-51 Prebreakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Dense-Phase Conveying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-52 Hammer Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Prediction of Dense-Phase Conveying Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-53 Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-79
Secondary Air Injection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-53 Pin Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Dense-Phase System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-54 Universal Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Conveying System Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-55 Hammer Mills with Internal Air Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Roll Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
SCREENING Roll Press . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-56 Roll Ring-Roller Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Wire Mesh/Screen Cloth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-57 Raymond Ring-Roller Mill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-80
Commercial Screen Cloth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-58 Pan Crushers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-81
Screen Cloth Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-59 Design and Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-81
Screen Cloth Materials of Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-59 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-81
Fundamentals of Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-59
Classification Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-60
Grade Efficiency Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-60 CRUSHING AND GRINDING EQUIPMENT:
Screener Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-60 FLUID-ENERGY OR JET MILLS
Screen Blinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-60 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-81
Screener Perfomance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62 Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82
Screener Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62 Spiral Jet Mill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82
Vibrating Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62 Opposed Jet Mill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82
Reciprocating Screens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-62 Other Jet Mill Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82
 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY 21-3

CRUSHING AND GRINDING EQUIPMENT: SIZE ENLARGEMENT

WET/DRY GRINDING—MEDIA MILLS Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-88
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-88
Media selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-82 Binding Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-88
Tumbling Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-83 Wet Agglomeration Rate Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-88
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-83 Compaction Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-93
Multicompartment Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-83 Agglomeration Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-94
Material and Ball Charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-84 Wet Agglomeration Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-94
Dry versus Wet Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-84 Compaction Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-97
Dry Ball Milling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-84 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-99
Wet Ball Milling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-84 Population Balance Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-99
Mill Efficiencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-100
Capacity and Power Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85
Stirred Media Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 FEEDING, METERING, AND DOSING
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-100
Attritors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Volumetric Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-100
Vertical Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Screw Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-101
Horizontal Media Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Vibratory Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-102
Annular Gap Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Belt Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-103
Manufacturers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Apron Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-103
Performance of Bead Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-85 Rotary Vane/Airlock Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-103
Residence Time Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-86 Rotary Plow (Plough) Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-104
Vibratory Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-86 Table Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-104
Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87 Slide Gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-105
Residence Time Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87 Gravimetric Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-105
Hicom Mill . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87 Solids Flowmeters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-106
Planetary Ball Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87 Performance Metrics for Gravimetric Feeders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-106
Disk Attrition Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87 Dosing or Batching Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-109
Dispersers and Emulsifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87
Media Mills and Roll Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87
Dispersion and Colloid Mills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87
Pressure Homogenizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87
Microfluidizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21-87

Nomenclature
Note that many symbols have multiple definitions . Also some symbols have only one set of units associated with them . This means that they are associated
with an empirical equation .
U .S . customary U.S. customary
Symbol Definition SI units units Symbol Definition SI units units

A Pipe area m2 ft2 H Absolute humidity kg water/kg air lb water/lb air

A, B, C Constants for Eq. (21-23) — — H (θ′) Hopper function, Fig. 21-51 dimensionless dimensionless
A0 Cross-sectional area of outlet m2 ft2 h Height between hopper apex m ft
a Distance from the scatter m ft and cylinder-hopper transition
to the detector h Liquid layer thickness mm in
a Acceleration m/s2 ft/s2 ha Asperity height mm in
ak,n Constants for batch grinding eq. dimensionless dimensionless I0 Illuminating intensity W/m2 W/ft2
B Outlet size m ft I (θ) Intensity of scattered light W/m2 W/ft2
B Bend factor, see Table 21-7 dimensionless dimensionless K Permeability m/s ft/s
B Nucleation rate number/m3/s number/ft3/s K, a, b, c, d Stegmaier constants (Table 21-8) dimensionless dimensionless
Bmin Minimum outlet size m ft K0 Permeability at reference density m/s ft/s
∆Bk,u Breakage function, Eq. (21-126) dimensionless dimensionless K Wave number dimensionless dimensionless
C, n Constants, Eq. (21-119) k Janssen coefficient dimensionless dimensionless
Ci Impact crushing resistance ft ⋅ lb/in k Stress concentration factor dimensionless dimensionless
Cw Concentration of water in air kg/m3 lb/ft3 kB Boltzmann constant J/K J/K
CPF Area concentration 1/cm2 1/in2 L Length m ft
c Average slug velocity m/s ft/s L Roll crusher roll length cm in
cD Constant defined by Eq. (21-111) dimensionless dimensionless l Crack length mm in
D Silo cylinder diameter m ft lp Material layer thickness on screen m ft
D Pipe diameter m ft M Sample mass kg lb
D Rotor diameter of rotating mill m ft Mk,r kth moment of qr(x) distribution mm in
DF Critical rathole diameter m ft Mc , Ms Mass flow rate of coarse & feed kg/h lb/h
Dm Media diameter cm in m Geometric coefficient (0 or 1) dimensionless dimensionless
d Wire size mm in m s Solids discharge rate kg/s lb/s
d Roll gap cm in N Total number of particles number number
dp Particle diameter m ft in system
dp Geometric mean particle diameter mm in Nc Critical rotation speed rpm rpm
dp∗ Matsumoto critical diameter m ft n As defined in Eq. (21-38) dimensionless dimensionless
d50 Cut size mm in na Amount of absorbed gas, BET mol/g mol/lb
d3,2 Sauter mean diameter mm in P Pressure Pa lb/in2
E Work done during grinding J ft ⋅ lb P Settled weight g lb
Ei Work index, Eq. (21-123) kWh/ton P Probability dimensionless dimensionless
e Elementary charge C C p Average value of particle property — —
e Coefficient of restitution dimensionless dimensionless p Number of elementary charges — —
fc Unconfined yield stress Pa lb/ft2 Pa, Pb, Pc Probabilities of screen passage dimensionless dimensionless
ff Flow factor dimensionless dimensionless Pt Total pressure Pa lb/in2
g Gravitational constant m/s2 ft/s2 Pwsat Saturation pressure of pure water Pa lb/in2
G Growth rate m3/s ft3/s ∆P Pressure drop Pa psia
Gc Solids flux at choking kg/(m2 ⋅ s) lb/( ft2 ⋅ s) Q Roll crusher capacity cm3/min
G(xi) Grade efficiency of ith fraction dimensionless dimensionless Qi(x) Cumulative distribution dimensionless dimensionless
G (φt) Rathole function, Fig. 21-64 dimensionless dimensionless i = 0 number, i = 1 length
H Silo cylinder height m ft i = 2 area, i = 3 volume
H Solids drop height m ft q Modulus of the scattering vector 1/m 1/ft
21-4 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Nomenclature (Continued )
U .S . customary U.S. customary
Symbol Definition SI units units Symbol Definition SI units units
qi(x) Distribution density dimensionless dimensionless εtap Tapped effective porosity dimensionless dimensionless
i = number, i = 1 length η Hydrodynamic viscosity of liquid Pa∙s P
i = 2 area, i = 3 volume ηk Classifier efficiency dimensionless dimensionless
q, qF , qR Mill flows: total, feed, recycle kg/h lb/h θ Angle of stress rotation degrees degrees
R Radius m ft θ Solids impact angle degrees degrees
R Recycle ratio dimensionless dimensionless θ Three phase contact angle degrees degrees
Reff Effective pore radius mm in θ′ Hopper angle from vertical degrees degrees
RH Hydraulic radius m ft θend Pyramidal hopper end angle degrees degrees
r Radial distance from hopper vertex m ft θside Pyramidal hopper side angle degrees degrees
RH Relative humidity dimensionless dimensionless θv Hopper valley angle degrees degrees
S Distance along chute m ft κ Rizk constant, Eq. (21-85) dimensionless dimensionless
SF Stress factor λF Fluid friction factor dimensionless dimensionless
SI Stress intensity λZ Solids friction factor dimensionless dimensionless
Su Grinding rate function for size u 1/s 1/s m Solids loading ratio/phase ratio dimensionless dimensionless
SV Volume specific surface 1/mm 1/in m Liquid viscosity Pa ⋅ s
s Peripheral roll speed cm/min ft/min ms Solids loading ratio at saltation dimensionless dimensionless
smax Maximum pore saturation dimensionless dimensionless ρ Slurry density kg/m3 lb/ft3
s(θ′) Stress function, Eq. (21-61) dimensionless dimensionless ρb Bulk density kg/m3 lb/ft3
T Temperature °C or K °F or K ρba Bulk density at zero pressure kg/m3 lb/ft3
t Time s sec gradient, see Eq. (21-70)
Ug Superficial gas velocity m/s ft/s ρbO Bulk density at outlet kg/m3 lb/ft3
Up Particle velocity m/s ft/s ρb0 Reference bulk density kg/m3 lb/ft3
Ut Terminal velocity m/s ft/s ρbmin Minimum bulk density kg/m3 lb/ft3
u Collision viscosity m/s ft/s ρbmax Maximum bulk density kg/m3 lb/ft3
us Superficial gas velocity m/s ft/s ρf Fluid density kg/m3 lb/ft3
V Hopper volume m3 ft3 ρg Gas density kg/m3 lb/ft3
V Stream velocity m/s ft/s ρm Media density kg/m3 lb/ft3
V0 Initial solids velocity on a chute m/s ft/s ρp Particle density kg/m3 lb/ft3
V1 Chute impact velocity m/s ft/s ρs Particle density kg/m3 lb/ft3
V2 Chute velocity after impact m/s ft/s σ Standard deviation of sample
Vc Choking velocity m/s ft/min σ Normal stress Pa lbf /ft2
Vd Droplet volume cm3 in3 σ Impact pressure Pa lbf /ft2
vs Saltation velocity m/s ft/min σavg Average stress Pa lbf /ft2
Vt Rotating mill tip speed m/s ft/s σh Horizontal stress Pa lbf /ft2
W Weight undersize g lb σi Local stress at tip of a crack Pa lbf /ft2
Ws Conveying system mass flow rate kg/s lb/s σN Gross applied stress Pa lbf /ft2
w Screen aperture size mm in σss Steady-state normal stress Pa lbf /ft2
w Solid-to-liquid mass ratio dimensionless dimensionless σv Vertical stress Pa lbf /ft2
wk Weight fraction on kth screen dimensionless dimensionless σvht Mean vertical stress Pa lbf /ft2
X Solid moisture content kg water/kg solid lb water/lb solid σw Wall stress Pa lbf /ft2
X Particle size, Eq. (21-119) mm in σx, σy Normal stresses Pa lbf /ft2
XF Feed particle size mm in σY Granule yield strength Pa lbf /ft2
XP Product particle size mm in σ0 Reference stress Pa lbf /ft2
x or xi Particle size mm in σ1 Major consolidation stress Pa lbf /ft2
xSt Stokes diameter mm in σ2 Minor consolidation stress Pa lbf /ft2
Z(x) Electron mobility of particle size x C C σ Abutment stress Pa lbf /ft2
z Depth of solids m ft τ Shear stress Pa lbf /ft2
τss Steady-state shear stress Pa lbf /ft2
Greek letters τwall Wall stress Pa lbf /ft2
τxy, τyx Shear stresses Pa lbf /ft2
α Chute angle degrees degrees φ Kinematic angle of internal friction degrees degrees
α Empirical constant, Eq. (21-51) kg/m3 lb/ft3 φ Sphericity dimensionless dimensionless
β Compressibility degrees degrees φ′ Wall friction angle degrees degrees
β Coalescence kernel number/s number/s ω Mill rotational speed 1/s 1/s
Γ Decay rate 1/s 1/s Ca Capillary number dimensionless dimensionless
γ Liquid surface tension dyn/cm Frp Particle Froude number dimensionless dimensionless
δ Effective angle of friction degrees degrees Frs Solids Froude number dimensionless dimensionless
δ Rizk constant, Eq. (21-85) dimensionless dimensionless Kn Knudsen number dimensionless dimensionless
δ Angle of disc granulator degrees degrees L Loschmidt number 1/mol 1/mol
ε Voidage dimensionless dimensionless Rep Particle Reynolds number dimensionless dimensionless
εc Voidage at choking dimensionless dimensionless Stdef Deformation Stokes number dimensionless
εeff Effective porosity dimensionless dimensionless Stv Viscous Stokes number dimensionless dimensionless
εmin Minimum granule porosity dimensionless dimensionless St∗v Critical viscous Stokes number dimensionless dimensionless
 PARTICLE CHARACTERIZATION

General References: Terence Allen, Particle Size Measurement, vol . 1: Powder Sampling TABLE 21-1 Tabular Presentation of Particle-Size Data
and Particle Size Measurement and vol . 2: Surface Area and Pore Size Determination,
5th ed ., Springer Netherlands, 1997 . P . M . Gy, Sampling of Particulate Materials: Theory 1 2 3 4 5 6 7
and Practice, Elsevier, 1979 . Bart and Sun, “Particle Size Analysis Review,” Anal. Chem . xi, ∆xi, q– 3,i = ∆Q3,i/∆x i
57: 151R (1985) . Miller and Lines, “Critical Reviews in Analytical Chemistry,” 20(2): i mm ∆Q3,i mm 1/mm Q3,i q–∗3,i
75–116 (1988) . Herdan, Small Particles Statistics, Butterworths, London, 1960 . Orr and
DalleValle, Fine Particle Measurement, 2d ed ., Macmillan, New York, 1960 . Kaye, Direct 0 0 .063 0 .0000
Characterization of Fine Particles, Wiley, New York, 1981 . Van de Hulst, Light Scattering 1 0 .090 0 .0010 0 .027 0 .0370 0 .0010 0 .0028
by Small Particles, Wiley, New York, 1957 . K . Leschonski, “Representation and Evaluation 2 0 .125 0 .0009 0 .035 0 .0257 0 .0019 0 .0027
of Particle Size Analysis Data,” Part. Part. Syst. Charact. 1: 89–95 (1984) . Karl Sommer, 3 0 .180 0 .0016 0 .055 0 .0291 0 .0035 0 .0044
Sampling of Powders and Bulk Materials, Springer, 1986 . H . Merkus, Particle Size 4 0 .250 0 .0025 0 .070 0 .0357 0 .0060 0 .0076
Measurements: Fundamentals, Practice, Quality, Springer Netherlands, 2009 . M . Alderliesten, 5 0 .355 0 .0050 0 .105 0 .0476 0 .0110 0 .0143
“Mean Particle Diameters, Part I: Evaluation of Definition Systems,” Part. Part. Syst. 6 0 .500 0 .0110 0 .145 0 .0759 0 .0220 0 .0321
Charact. 7: 233–241 (1990); “Part II: Standardization of Nomenclature,” Part. Part. Syst. 7 0 .710 0 .0180 0 .210 0 .0857 0 .0400 0 .0513
Charact. 8: 237–241 (1991); “Part III: An Empirical Evaluation of Integration and 8 1 .000 0 .0370 0 .290 0 .1276 0 .0770 0 .1080
Summation Methods for Estimating Mean Particle Diameters from Histogram Data,” 9 1 .400 0 .0610 0 .400 0 .1525 0 .1380 0 .1813
Part. Part. Syst. Charact . 19: 373–386 (2002); “Part IV: Empirical Selection of the Proper 10 2 .000 0 .1020 0 .600 0 .1700 0 .2400 0 .2860
Type of Mean Particle Diameter Describing a Product or Material Property,” Part. Part. 11 2 .800 0 .1600 0 .800 0 .2000 0 .4000 0 .4755
Syst. Charact . 21: 179–196 (2004); “Part V: Theoretical Derivation of the Proper Type of 12 4 .000 0 .2100 1 .200 0 .1750 0 .6100 0 .5888
Mean Particle Diameter Describing a Product or Process Property,” Part. Part. Syst. Charact . 13 5 .600 0 .2400 1 .600 0 .1500 0 .8500 0 .7133
22: 233–245 (2005) . ISO 9276, Representation of Results of Particle Size Analysis . H . C . 14 8 .000 0 .1250 2 .400 0 .0521 0 .9750 0 .3505
van de Hulst, Light Scattering by Small Particles, Structure of Matter Series, Dover, 1981 . 15 11 .20 0 .0240 3 .200 0 .0075 0 .9990 0 .0713
Craig F . Bohren and Donald R . Huffman, Absorption and Scattering of Light by Small 16 16 .000 0 .0010 4 .800 0 .0002 1 .0000 0 .0028
Particles, Wiley-Interscience, 2007 . Bruce J . Berne and Robert Pecora, Dynamic Light
Scattering: With Applications to Chemistry, Biology, and Physics, unabridged edition,
Dover, 2000 . J . R . Allegra and S . A . Hawley, “Attenuation of Sound in Suspensions and If Qr (x) is differentiable, the distribution density function qr (x) can be calcu-
Emulsions: Theory and Experiment,” J. Acoust. Soc. America 51: 1545–1564 (1972) . lated as the first derivative of Qr(x), or
xi
dQr ( x )
PARTICLE SIZE qr ( x ) =
dx
or Qr ( x i ) = ∫ q ( x ) dx
x min
r (21-2)
Specification for Particulates The behavior of dispersed matter
is generally described by a large number of parameters, e .g ., the powder’s It is helpful in the graphical representation to identify the distribution type,
bulk density, flowability, and degree of aggregation or agglomeration . Each as shown for the cumulative volume distribution Q3(x) and volume distribu-
parameter might be important for a specific application . In solids processes tion density q3(x) in Fig . 21-1 . If qr(x) displays one maximum only, the distri-
such as comminution, classification, agglomeration, mixing, crystallization, bution is called a monomodal size distribution. If the sample is composed of
or polymerization, or in related material handling steps, particle size plays two or more different-size regimes, then qr(x) shows two or more maxima
an important role . Often it is the dominant quality factor for the suitability and is called a bimodal or multimodal size distribution.
of a specific product in the desired application . PSDs are often plotted on a logarithmic abscissa (Fig . 21-2) . While the Qr(x)
Particle Size As particles are extended three-dimensional objects, values remain the same, care has to be taken for the transformation of the
only a perfect spherical particle allows for a simple definition of the particle distribution density qr(x), as the corresponding areas under the distribution
size x—as the diameter of the sphere . In practice, spherical particles are very density curve must remain constant (in particular the total area remains 1,
rare . So usually equivalent diameters are used, representing the diameter of or 100 percent) independent of the transformation of the abscissa . So the
a sphere that behaves as the real (nonspherical) particle in a specific sizing transformation has to be performed by
experiment . Unfortunately, the measured size now depends on both the ∆Qr ,i
method used for assessing particle size and the subsequent data analysis . qr∗ (ln x i − 1 , ln x i ) = (21-3)
It is unreasonable to expect identical results for the particle size even for ln( x i /x i −1 )
instruments using the same measurement method . This equation also holds if the natural logarithm is replaced by the logarithm
In most applications more than one particle is observed . As each indi- to base 10 .
vidual may have its own particle size, methods for data reduction have been
introduced . These include the particle size distribution, a variety of model
distributions, and moments (or averages) of the distribution . Also note Example 21-1 From Table 21-1 one can calculate, e .g .,
that these methods can be extended to other particle attributes . Examples
∆Q3,11 0.16
include pore size, porosity, surface area, color, and electrostatic charge dis- q3,11 = = = 0.2 µm −1
tributions, to name but a few . ∆x 11 0.8 µm
Particle Size Distribution A particle size distribution (PSD) can be dis- ∗ ∆Q3,11 0.16
q3,11 = q3∗ (ln x 10 , ln x 11 ) = =
played as a table or a diagram . In the simplest case, one can divide the range ln( x11 /x 10 ) ln(2.8 µm/2.0 µm)
of measured particle sizes into size intervals and sort the particles into the 0.16
corresponding size class, as displayed in Table 21-1 (shown for the case of = = 0.4755
ln1.4
volume fractions) .
Typically the fractions ∆Qr,i in the different size classes i are summed and
normalized to 100 percent, resulting in the cumulative distribution Q(x), also
known as the percentage undersize . For a given particle size x, the Q value
represents the percentage of the particles finer than x .
If the quantity measure is “number,” Q0(x) is called a cumulative number
distribution . If it is length, area, volume, or mass, then the corresponding length
[Q1(x)], area [Q2(x)], volume, or mass distributions are formed [Q3(x)]; mass and
volume are related by the specific density ρ . The index r in this notation repre-
sents the quantity measure (ISO 9276-1:1998, Representation of Results—Part 1
Graphical Representation) . The choice of the quantity measured is of decisive
importance for the appearance of the PSD, which changes significantly when
the dimension r is changed . As, e .g ., one 100-mm particle has the same volume
as 1000 10-mm particles or 1 million 1-mm particles, a number distribution
is always dominated by and biased to the fine fractions of the sample while
a volume distribution is dominated by and biased to the coarse fractions .
The normalization of the fraction ∆Qr,i to the size of the corresponding
interval leads to the distribution density qr ,i , or
∆Qr ,i n n
qr ,i =
∆x i
and ∑ ∆Q r ,i = ∑ qr ,i ∆x i = 1 = 100% (21-1)
FIG. 21-1 Histogram q3 ( x ) and Q3(x) plotted with linear abscissa .
i =1 i =1

21-5
21-6 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

where x 1,2 is the weighted average diameter of the area distribution, also
known as Sauter mean diameter. It represents a particle having the same
ratio of surface area to volume as the distribution, and it is also referred to as
a surface-volume average diameter. The Sauter mean is an important average
diameter used in solids handling and other processing applications where
aspects of two-phase flow become important, as it appropriately weights
the contributions of the fine fractions to surface area . For nonspherical par-
ticles, a shape factor has to be considered .

Example 21-2 The Sauter mean diameter and the volume-weighted particle
size and distribution given in Table 21-1 can be calculated by using ISO 9276-2:2014,
Representation of Results of Particle Size Analysis—Part 2: Calculation of Average Particle
Sizes/Diameters and Moments from Particle Size Distributions via Table 21-2 .
The Sauter mean diameter is

M 3,0 1 n ln ( x i /x i −1 )
x 1,2 = M 1,2 = = with M 1,3 = ∑ ∆Q3,i
M 2,0 M −1,3 i =1 x i − x i −1
–∗
FIG. 21-2 Histogram q 3(x) and Q3(x) plotted with a logarithmic abscissa . which yields

1
x 1,2 = = 2.110882
Model Distribution While a PSD with n intervals is represented by 0.473736
2n + 1 numbers, further data reduction can be performed by fitting the size
distribution to a specific mathematical model . The logarithmic normal dis- The volume-weighted average particle size is
tribution or the logarithmic normal probability function is one common
model distribution used for the distribution density, and it is given by 1 n
x 1,3 = M 1,3 = ∑ ∆Q3,i ( x i + x i −1 )
2 i =1
1 −0.5 z 2 1  x 
qr∗ ( z ) = e with z = ln  (21-4)
s  x 50,r 
which yields
2π
1
The PSD can then be expressed by two parameters, namely, the mean size x 1,3 = (7.280590) = 3.640295
2
x50,r and, e .g ., by the dimensionless standard deviation s (ISO 9276-5:2005,
Methods of Calculations Relating to Particle Size Analysis Using Logarithmic
Normal Probability Distribution) . The data reduction can be performed by PARTICLE SHAPE
plotting Qr(x) on logarithmic probability graph paper or using the fitting
methods described in ISO 9276-3:2008, Adjustment of an Experimental Curve For many applications not only the particle size but also the shape are of
to a Reference Model . This method is mainly used for the analysis of powders importance; e .g ., toner powders should be spherical while polishing pow-
obtained by grinding and crushing and has the advantage that the transfor- ders should have sharp edges . Traditionally in microscopic methods of size
mation between PSDs of different dimensions is simple . The transformation analysis, direct measurements are made on enlarged images of the particles
is also log-normal with the same slope s . by using a calibrated scale . While such measurements are always encour-
Other model distributions used are the normal distribution (Laplace- aged to gather a direct sense of the particle shape and size, care should
Gauss), for powders obtained by precipitation, condensation, or natural be taken in terms of drawing general conclusions from limited particle
products (e .g ., pollens); the Gates-Gaudin-Schuhmann distribution (biloga- images . Furthermore, with the strong progress in computing power, instru-
rithmic), for analysis of the extreme values of fine particle distributions ments have become available that acquire the projected area of many particles
(Schuhmann, Am. Inst. Min. Metall. Pet. Eng., Tech . Paper 1189 Min . Tech ., in short times, with a significant reduction in data manipulation times .
1940); or the Rosin-Rammler-Sperling-Bennet distribution for the analysis of Standardization of shape parameters is given in ISO 9276-6:2008 Descriptive
the extreme values of coarse particle distributions, e .g ., in monitoring grind- and Qualitative Representation of Particle Shape and Morphology.
ing operations [Rosin and Rammler, J. Inst. Fuel 7: 29–36 (1933); Bennett, Equivalent Projection Area of a Circle Equivalent projection area of
ibid ., 10: 22–29 (1936)] . a circle is widely used for the evaluation of particle sizes from the projection
Moments Moments represent a PSD by a single value . With the help area A of a nonspherical particle .
of moments, the average particle sizes, volume specific surfaces, and other
mean values of the PSD can be calculated . The general definition of a x EQPC = 2 A / π (21-8)
moment is given by (ISO 9276-2:2014, Calculation of Average Particle Sizes/
Diameters and Moments from Particle Size Distributions)
x max TABLE 21-2 Table for Calculation of Sauter Mean Diameter
M k ,r = ∫
x min
x k qr ( x ) dx (21-5) and Volume Weighted Particle Size

∆Q∗3,i ∆Q∗3,i
where Mk,r is the kth moment of a qr(x) distribution density and k is the I xi, mm ∆Q3,i ln (xi/xi–1) ln (xi/xi–1) ln (xi/xi–1) / (xi + xi–1),
power of x. (xi–xi–1) (xi–xi– 1) mm
Average Particle Sizes A PSD has many average particle sizes . The 0 0 .0630
general equation is given by 1 0 .0900 0 .0010 0 .3567 13 .2102 0 .013210 0 .000153
2 0 .1250 0 .0009 0 .3285 9 .3858 0 .008447 0 .000194
3 0 .1800 0 .0016 0 .3646 6 .6299 0 .010608 0 .000488
x k ,r = k M k ,r (21-6) 4 0 .2500 0 .0025 0 .3285 4 .6929 0 .011732 0 .001075
5 0 .3550 0 .0050 0 .3507 3 .3396 0 .016698 0 .003025
6 0 .5000 0 .0110 0 .3425 2 .3620 0 .025982 0 .009405
Two typically employed average particle sizes are the arithmetic average par- 7 0 .7100 0 .0180 0 .3507 1 .6698 0 .030056 0 .021780
ticle size x k ,0 = M k ,0 [e .g ., for a number distribution (r = 0) obtained by count- 8 1 .0000 0 .0370 0 .3425 1 .1810 0 .043697 0 .063270
ing methods], and the weighted average particle size x 1,r = M 1,r [e .g ., for a 9 1 .4000 0 .0610 0 .3365 0 .8412 0 .051312 0 .146400
volume distribution (r = 3) obtained by sieve analysis], where x 1,r represents 10 2 .0000 0 .1020 0 .3567 0 .5945 0 .060635 0 .346800
the center of gravity on the abscissa of the qr(x) distribution . 11 2 .8000 0 .1600 0 .3365 0 .4206 0 .067294 0 .768000
Specific Surface The specific surface area can be calculated from size 12 4 .0000 0 .2100 0 .3567 0 .2972 0 .062418 1 .428000
distribution data . For spherical particles this can simply be calculated by 13 5 .6000 0 .2400 0 .3365 0 .2103 0 .050471 2 .304000
using moments . The volume specific surface is given by 14 8 .0000 0 .1250 0 .3567 0 .1486 0 .018577 1 .700000
15 11 .2000 0 .0240 0 .3365 0 .1051 0 .002524 0 .460800
M 16 16 .0000 0 .0010 0 .3567 0 .0743 0 .000074 0 .027200
6 6
SV = or SV = = 2,0 = 6 M 1,3 (21-7) ∑0 .473736 7 .280590
x 1,2 M 1,2 M 3,0
 PARTICLE CHARACTERIZATION 21-7

SAMPLING AND SAMPLE SPLITTING

As most of the sizing methods are limited to small sample sizes, an impor-
xF,min tant prerequisite to accurate particle size analysis is proper powder sam-
pling and sample splitting (ISO 14488:2007, Particulate Materials—Sampling
and Sample Splitting for the Determination of Particulate Properties) .
xF,max When one is determining particle size (or any other particle attribute
xF,max such as chemical composition or surface area), it is important to recognize
that the error associated in making such a measurement can be described
by its variance, or
xF,max 90
σ observed
2
= σ actual
2
+ σ 2measurement (21-9)
FIG. 21-3 Definition of Feret diameters .
σ 2measurement = σ sampling
2
+ σ analysis
2
(21-10)

That is, the observed variance in the particle size measurement is due to
Feret’s Diameter Feret’s diameter is determined from the projected area
both the actual physical variance in size and the variance in the measure-
of the particles by using a slide gauge (Fig . 21-3) . In general, it is defined as the
ment . More importantly, the variance in measurement has two contributing
distance between two parallel tangents of the particle at an arbitrary angle .
factors: variance due to sampling, which would include systematic errors in
In practice, the minimum xF,min and maximum Feret diameters xF,max, the mean
the taking, splitting, and preparation of the sample; and variance due to the
Feret diameter x F , and the Feret diameters obtained at 90° to the direction of
actual sample analysis, which would include not only the physical measure-
the minimum and maximum Feret diameters xF,max90 are used . The minimum
ment at hand, but also how the sample is presented to the measuring zone,
Feret diameter is often used as the diameter equivalent to a sieve analysis .
which can be greatly affected by instrument design and sample dispersion
Other diameters used in the literature include Martin’s diameter or the
edges of an enclosing rectangle. Martin’s diameter is a line, parallel to a fixed (discussed later) . Successful characterization of the sample (in this discus-
sion, taken to be measurement of particle size) requires that the errors in
direction, which divides the particle profile into two equal areas .
measurement be much less than actual physical variations in the sample
These diameters offer an extension over volume equivalent diameters to
itself, especially if knowledge of sample deviations is important . In this
account for shape deviations from spherical . As with any other quality mea-
sure of size, many particles must be measured to determine distributions of regard, great negligence is unfortunately often exhibited in sampling efforts .
these particle size diameters . With the advent of high-speed image process- Furthermore, measured deviations in particle size or other properties are
often incorrectly attributed to and reflect upon the measuring device, where
ing, particle size and shape can be determined quickly . For shape character-
in fact they are caused by inattention to proper sampling and sample split-
ization, these devices are able to generate galleries of particle shapes which
can be very helpful in solving process and product problems . Particles can ting . Worse still, such deviations caused by poor sampling may be taken as
true sample deviations, causing undue and frequent process corrections .
be sorted, for example, by fractal dimension, fiber length, or sphericity . The
Powders may be classified as nonsegregating (cohesive) or segregating
engineering challenge is to connect these shapes to product characteristics
or plant processing issues . ( free-flowing) . Representative samples can be more easily taken from cohe-
sive powders, provided that they have been properly mixed . For wet samples
Sphericity, Aspect Ratio, and Convexity Parameters describing the
shape of the particles include the following: a sticky paste should be created and mixed from which the partial sample
The sphericity ψS (0 < ψS ≤1) is defined by the ratio of the perimeter of a is taken .
circle with diameter xEQPC to the perimeter of the corresponding projection In the case of segregating powders, four key rules should be followed,
although some apply or can be equally employed for cohesive materials as
area A . And ψS = 1 represents a sphere .
The aspect ratio ψA (0 < ψA ≤1) is defined by the ratio of the minimum to well . These rules are especially important for in-line and on-line sampling,
the maximum Feret diameter ψA = xFeret min/xFeret max . It gives an indication of the discussed below . Allen (Allen, Particle Size Measurement, Volume 1: Powder
Sampling and Particle Size Measurement and Particle Size Measurement,
elongation of the particle . Some literature also used 1/ψA as the definition
Volume 2: Surface Area and Pore Size Determination, 5th ed ., Springer,
of sphericity .
Netherlands, 1997) suggests the following:
The convexity ψC (0 < ψC ≤ 1) is defined by the ratio of the projection area A
to the convex hull area A + B of the particle, as displayed in Fig . 21-4 . 1 . The particles should be sampled while in motion . Transfer points are
In Fourier techniques the shape characteristic is transformed to a signa- often convenient and relevant for this . Sampling a stagnant bed of segregat-
ing material by, e .g ., thieves disrupts the state of the mixture and may be
ture waveform . Beddow and coworkers (Beddow, Particulate Science and
Technology, Chemical Publishing, New York, 1980) take the particle centroid biased to coarse or fines .
as a reference point . A vector is then rotated about this centroid with the tip 2 . The whole stream of powder should be taken in many short time inter-
vals in preference to part of the stream being taken over the whole time, i .e .,
of the vector touching the periphery . A plot of the magnitude of the vector
a complete slice of the particle stream . Furthermore, any mechanical collec-
versus its angular position is a wave-type function . This waveform is then
tion point should not be allowed to overfill, since this will make the sample
subjected to Fourier analysis . The lower-frequency harmonics constituting
bias toward fines, and coarse material rolls off formed heaps .
the complex wave correspond to the gross external morphology, whereas
3 . The entire sample should be analyzed, splitting down to a smaller sample
the higher frequencies correspond to the texture of the fine particle .
Fractal Dimension This was introduced into fine particle science by if necessary . In many cases, segregation of the sample will not affect the
Kaye and coworkers (Kaye, Direct Characterization of Fine Particles, Wiley, measurement, provided the entire sample is analyzed . There are, however,
New York, 1981), who show that the noneuclidean logic of Mandelbrot can exceptions in that certain techniques may only analyze one surface of the
be applied to describe the ruggedness of a particle profile . A combination final sample . In the case of chemical analysis, an example would be near
infrared spectroscopy operated in reflectance mode as opposed to trans-
of fractal dimension and geometric shape factors such as the aspect ratio
can be used to describe a population of fine particles of various shapes, and mission . Such a technique may still be prone to segregation during the final
these can be related to the functional properties of the particle . analysis .
4 . A minimum sample size exists for a given size distribution, generally
determined by the sample containing a minimum number of coarse par-
ticles representative of the customer application . While many applications
involving fine pharmaceuticals may only require milligrams to establish
a representative sample, other cases such as detergents and coffee might
require kilograms . Details are given in the standard ISO14488:2007, Particu-
late Materials—Sampling and Sample Splitting for the Determination of
Particulate Properties.
A A B In this regard, one should keep in mind that the sample size may also
reflect variation in the degree of mixing in the bed, as opposed to true size
differences . (See also the subsection Solids Mixing: Measuring the Degree
of Mixing .) In fact, larger samples in this case help minimize the impact of
segregation on measurements .
FIG. 21-4 Definition of the convex hull area A + B for the projection area A of a The estimated maximum sampling error on a 60:40 blend of free-flowing
particle . sand using different sampling techniques is given in Table 21-3 .
21-8 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

TABLE 21-3 Reliability of Selected Sampling Method

Estimated maximum
Method sampling error, %
Cone and quartering 22 .7
Scoop sampling 17 .1
Table sampling 7 .0
Chute splitting 3 .4
Spinning riffling 0 .42

The spinning riffler (Fig . 21-5) generates the most representative samples .
In this device a ring of containers rotates under the powder feed . If the pow-
der flows a long time with respect to the period of rotation, each container
will be made up of many small fractions from all parts of the bulk . Many
different configurations are commercially available . Devices with small
numbers of containers (say, 8) can be cascaded n times to get higher split-
ting ratios 1:8n . This usually creates smaller sampling errors than does using FIG. 21-6 Examples of commercial splitting devices . Spinning riffler and standard
splitters with more containers . A splitter simply divides the sample into splitters . (Courtesy of Retsch Corporation.)
two halves, generally pouring the sample into a set of intermeshed chutes .
Figure 21-6 illustrates commercial rifflers and splitters .
For reference materials sampling errors of less than 0 .1 percent are PARTICLE SIZE MEASUREMENT
achievable (S . Röthele and W . Witt, Standards in Laser Diffraction, PARTEC,
5th European Symposium Particle Characterization, Nürnberg, 1992, There are many techniques available to measure the particle size distribu-
pp . 625–642) . tion of powders or droplets . The wide size range, from nanometers to mil-
limeters, of particulate products, however, cannot be analyzed by using only
a single measurement principle . Added to this are the usual constraints
DISPERSION of capital costs versus running costs, speed of operation, degree of skill
Many sizing methods are sensitive to the agglomeration state of the sample . required, and, most important, the end-use requirement .
In some cases, this includes primary particles, possibly with some percent- If the particle size distribution of a powder composed of hard, smooth
age of such particles held together as weak agglomerates by interparticle spheres is measured by any of the techniques, the measured values should
cohesive forces . In other cases, strong aggregates of the primary particles be identical . However, many different size distributions can be defined for
may also exist . Generally, the size of either the primary particles or the any powder made up of nonspherical particles . For example, if a rod-shaped
aggregates is the matter of greatest interest . In some cases, however, it particle is placed on a sieve, then its diameter, not its length, determines
may also be desirable to determine the level of agglomerates in a sample, the size of aperture through which it will pass . If, however, the particle is
requiring that the intensity of dispersion be controlled and variable . Often allowed to settle in a viscous fluid, then the calculated diameter of a sphere
the agglomerates have to be dispersed smoothly without comminution of of the same substance that would have the same falling speed in the same
aggregates or primary particles . This can be done either in gas (dry) or in fluid (i .e ., the Stokes diameter) is taken as the appropriate size parameter of
liquid (wet) by using a suitable dispersion device which is stand-alone or the particle . Since the Stokes diameter for the rod-shaped particle will obvi-
integrated in the particle-sizing instrument . If possible, dry particles should ously differ from the rod diameter, this difference represents added infor-
be measured in gas and wet particles in suspension . mation concerning particle shape . The ratio of the diameters measured by
Wet Dispersion Wet dispersion separates agglomerates down to the two different techniques is called the shape factor.
primary particles by a suitable liquid . Dispersing agents and optional cavita- Historically methods primarily using mechanical, aerodynamic, or hydro-
tion forces induced by ultrasound are often used . Care must be taken that dynamic properties for discrimination and particle sizing have been used,
the particles not be soluble in the liquid, or that they not flocculate . Micros- but today methods based on the interaction of the particles with electro-
copy and zeta potential measurements may be of utility in specifying the magnetic waves (mainly light), ultrasound, or electric fields dominate .
proper dispersing agents and conditions for dispersion . Laser Diffraction Methods Over the past 30 years laser diffraction
Dry Dispersion Dry dispersion uses mechanical forces for the has developed into a leading principle for particle size analysis of all kinds
dispersion . While a simple fall-shaft with impact plates may be sufficient for of aerosols, suspensions, emulsions, and sprays in laboratory and process
the dispersion of coarse particles, say, >300 mm, much higher forces have to environments .
be applied to fine particles . The scattering of unpolarized laser light by a single spherical particle can
In Fig . 21-7 the agglomerates are sucked in by the vacuum generated be mathematically described by
through expansion of compressed gas applied at an injector . They arrive at I0
low speed in the dispersing line, where they are strongly accelerated . This I (θ) = {[ S1 (θ)]2 + [ S 2 (θ)]2 } (21-11)
2k 2a 2
creates three effects for the dispersion, as displayed in Fig . 21-8 .
With suitable parameter settings, agglomerates can be smoothly dis- where I(θ) is the total scattered intensity as function of angle θ with respect
persed down to 0 .1 mm [K . Leschonski, S . Röthele, and U . Menzel, Entwicklung to the forward direction; I0 is the illuminating intensity; k is the wave num-
und Einsatz einer trockenen Dosier-Dispergiereinheit zur Messung von ber 2π/λ; a is the distance from the scatterer to the detector; and S1(θ) and
Partikelgrößenverteilungen in Gas-Feststoff-Freistrahlen aus Laser- S2(θ) are dimensionless, complex functions describing the change and
Beugungsspektren; Part. Charact. 1: 161–166 (1984)] without comminution amplitude in the perpendicular and parallel polarized light . Different algo-
of the primary particles . rithms have been developed to calculate I(θ). The Lorenz-Mie theory is based

Powder feed Aerosol beam

Dispersing line

FIG. 21-7 Dry disperser RODOS with vibratory feeder VIBRI creating a fully dispersed
FIG. 21-5 Spinning riffler sampling device . aerosol beam from dry powder . (Courtesy of Sympatec GmbH .)
 PARTICLE CHARACTERIZATION 21-9

V1 small particle large particle

collision

V2

ro ro
(a)

V1 collision
(a)

Fourier lens detector

working distance f RD
(b)
V1 I(r) r
θ θ
dv ω θ
dx
θ I(θ)
V2 particle ensemble
(c) (b)

FIG. 21-8 Interactions combined for dry dispersion of agglomerates . (a) Particle-to-
particle collisions . (b) Particle-to-wall collisions . (c) Centrifugal forces due to strong intensity
velocity gradients .

on the assumption of spherical, isotropic, and homogenous particles and

that all particles can be described by a common complex refractive index
m = n − iκ. Index m has to be precisely known for the evaluation, which is
difficult in practice, especially for the imaginary part κ, and inapplicable for
mixtures with components having different refractive indices .
The Fraunhofer theory considers only scattering at the contour of the
particle and the near forward direction . No preknowledge of the refractive
index is required, and I(θ) simplifies to
2
I0  J (α sin θ) 
I (θ) = α4  1  (21-12)
2 k 2 a 2  α sin θ 
(c) 31 21 11 ...rj rj+1
with J1 as the Bessel function of the first kind and the dimensionless size
parameter α = πx/λ . This theory does not predict polarization or account FIG. 21-9 (a) Diffraction patterns of laser light in forward direction for two different
for light transmission through the particle . particle sizes . (b) The angular distribution I (θ) is converted by a Fourier lens to a spa-
For a single spherical particle, the diffraction pattern shows a typical ring tial distribution I (r) at the location of the photodetector . (c) Intensity distribution of a
structure . The distance r0 of the first minimum to the center depends on small particle detected by a semicircular photodetector .
the particle size, as shown in Fig . 21-9a . In the particle sizing instrument,
the acquisition of the intensity distribution of the diffracted light is usually
performed with the help of a multielement photodetector . in addition to size, relevant shape information becomes available by the
Diffraction patterns of static nonspherical particles are displayed in method . Currently, mainly instruments creating a 2D image of the 3D par-
Fig . 21-10 . As all diffraction patterns are symmetric to 180°, semicircular ticles are used . Two methods have to be distinguished .
detector elements integrate over 180° and make the detected intensity Static image analysis is characterized by nonmoving particles, e .g ., on a
independent of the orientation of the particle . microscope slide (Fig . 21-11) . The depth of sharpness is well defined, result-
Simultaneous diffraction on more than one particle results in a superpo- ing in a high resolution for small particles . The method is well established
sition of the diffraction patterns of the individual particles, provided that and standardized (ISO 13322-1:2014, Particle Size Analysis—Image Analysis
particles are moving and diffraction between the particles is averaged out . Methods, Part 1: Static Image Analysis Methods), but can handle only small
This simplifies the evaluation, providing a parameter-free and model-inde- amounts of data . The particles are oriented by the base; overlapping par-
pendent mathematical algorithm for the inversion process (M . Heuer and ticles have to be separated by time-consuming software algorithms, and the
K . Leschonski, “Results Obtained with a New Instrument for the Measure- tiny sample size creates a massive sampling problem, resulting in very low
ment of Particle Size Distributions from Diffraction Patterns,” Part. Part. statistical relevance of the data . Commercial systems reduce these effects
Syst. Charact. 2: 7–13, 1985) . by using large or even stepping microscopic slides and the deposition of the
Today the method is standardized (ISO 13320-1:2009, Particle Size Analysis— particles via a dispersing chamber . As all microscopic techniques can be
Laser Diffraction Methods—Part 1: General Principles), and many compa- used, the size range is only defined by the microscope used .
nies offer instruments, usually with the choice of Fraunhofer and/or Mie
theory for the evaluation of the PSD . The size ranges of the instruments
have been expanded by combining low-angle laser light scattering with 90°
or back scattering, the use of different wavelengths, polarization ratio, and
white light scattering, etc . It is now ranging from below 0 .1 mm to about
1 cm . Laser diffraction is currently the fastest method for particle sizing at
highest reproducibility . In combination with dry dispersion it can handle
large amounts of sample, which makes this method well suited for process
applications .
Instruments of this type are available, e .g ., from Malvern Ltd . (Mastersizer),
Sympatec GmbH (HELOS, MYTOS), Horiba (LA, LS series), Beckmann Coulter
(LS 13320), or Micromeritics (Saturn) .
Image Analysis Methods The extreme progress in image capturing
and exceptional increase of the computational power within the last few FIG. 21-10 Calculated diffraction patterns of laser light in forward direction for
years have revolutionized microscopic methods and made image analysis nonspherical particles: square, pentagon, and floccose . All diffraction patterns show
methods very popular for the characterization of particles, especially since, a symmetry to 180° .
21-10 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

FIG. 21-13 Diagram of Leeds and Northrup Ultrafine Particle Size Analyzer (UPA),
FIG. 21-11 Setup of static (left) and dynamic (right) image analysis for particle using fiber optics in a backscatter setup .
characterization .
The measured decay rates Γ are related to the translational diffusion
Dynamic image analysis images a flow of moving particles . This allows coefficients D of spherical particles by
for a larger sample size . The particles show arbitrary orientation, and the
number of overlapping particles is reduced . Several companies offer sys- 4π θ k BT
tems which operate in either reflection or transmission, with wet disper- Γ = Dq 2 with q= sin and D= (21-13)
λ0 2 2 πηx
sion or free fall, with matrix or line-scan cameras . The free-fall systems are
limited to well-flowing bulk materials . Systems with wet dispersion only where q is the modulus of the scattering vector, kB is the Boltzmann con-
allow for smallest samples sizes and slow particles . As visible light is used stant, T is the absolute temperature, and η is the hydrodynamic viscosity
for imaging, the size range is limited to about 1 mm at the fine end . This of the dispersing liquid . The particle size x is then calculated by the Stokes-
type of instruments has been standardized (ISO 13322-2:2006, Particle Size Einstein equation from D at fixed temperature T and η known .
Analysis—Image Analysis Methods, Part 2: Dynamic Methods) . DLS covers a broad range of diluted and concentrated suspension . As the
Common to all available instruments are small particle numbers, which theory is only valid for light being scattered once, any contribution of mul-
result in poor statistics . Thus recent developments have yielded a combina- tiple scattered light leads to erroneous PCS results and misinterpretations .
tion of powerful dry and wet dispersion with high-speed image capturing . So different measures have been taken to minimize the influence of multiple
Particle numbers up to 107 can now be acquired in a few minutes . Size and scattering .
shape analysis is available at low statistical errors [W . Witt, U . Köhler, and The well-established photon correlation spectroscopy (PCS) uses highly
J . List, “Direct Imaging of Very Fast Particles Opens the Application of the diluted suspensions to avoid multiple scattering . The low concentration of
Powerful (Dry) Dispersion for Size and Shape Characterization,” PARTEC particles makes this method sensitive to impurities in the liquid . So usually
2004, Nürnberg] . very pure liquids and a clean-room environment have to be used for the
Dynamic Light Scattering Methods Dynamic light scattering (DLS) is preparation and operation (ISO 13321:1996, Particle Size Analysis—Photon
now used on a routine basis for the analysis of particle sizes in the submi- Correlation Spectroscopy) .
crometer range . It provides an estimation of the average size and its distri- Another technique (Fig . 21-13) utilizes an optical system which minimizes
bution within a measuring time of a few minutes . the optical path into and out of the sample, including the use of backscatter
Submicrometer particles suspended in a liquid are in constant brownian optics, a moving-cell assembly, or setups with the maximum incident beam
motion as a result of the impacts from the molecules of the suspending fluid, intensity located at the interface of the suspension to the optical window
as suggested by W . Ramsay in 1876 and confirmed by A . Einstein and M . (Trainer, Freud, and Weiss, Pittsburgh Conference, Analytical and Applied
Smoluchowski in 1905/06 . Spectroscopy, Symp. Particle Size Analysis, March 1990; ISO 22412:2008, Par-
In the Stokes-Einstein theory of brownian motion, the particle motion at ticle Size Analysis—Dynamic Light Scattering) .
very low concentrations depends on the viscosity of the suspending liquid, Photon cross-correlation spectroscopy (PCCS) uses a novel three-dimensional
the temperature, and the size of the particle . If viscosity and temperature cross-correlation technique which completely suppresses the multiple scat-
are known, the particle size can be evaluated from a measurement of the tered fractions in a special scattering geometry . In this setup two lasers A
particle motion . At low concentrations, this is the hydrodynamic diameter. and B are focused to the same sample volume, creating two sets of scattering
DLS probes this motion optically . The particles are illuminated by a patterns, as shown in Fig . 21-14 . Two intensities are measured at different
coherent light source, typically a laser, creating a diffraction pattern, show- positions but with identical scattering vectors .
ing in Fig . 21-12 as a fine structure from the diffraction between the par-
ticles, i .e ., its near-order . As the particles are moving from impacts of the     
thermal movement of the molecules of the medium, the particle positions q = k A − k1 = kB − k2 (21-14)
change with the time t .
The change of the position of the particles affects the phases and thus the Subsequent cross-correlation of these two signals eliminates any contribu-
fine structure of the diffraction pattern . So the intensity in a certain point tion of multiple scattering . So highly concentrated, opaque suspensions can
of the diffraction pattern fluctuates with time . The fluctuations can be ana- be measured as long as scattered light is observed . High count rates result in
lyzed in the time domain by a correlation function analysis or in the fre- short measuring times . High particle concentrations reduce the sensitivity
quency domain by frequency analysis . Both methods are linked by Fourier of this method to impurities, so standard liquids and laboratory environ-
transformation . ments can be used, which simplifies the application [W . Witt, L . Aberle,
and H . Geers, “Measurement of Particle Size and Stability of Nanoparticles
in Opaque Suspensions and Emulsions with Photon Cross Correlation
Spectroscopy,” Particulate Systems Analysis, Harrogate (UK), 2003] .

FIG. 21-14 Scattering geometry of a PCCS setup . The

 sample volume is illuminated
FIG. 21-12 Particles illuminated by a gaussian-shaped laser beam and its corre- by two incident beams . Identical scattering vectors q and the scattering volumes are
sponding diffraction pattern show a fine structure . used in combination with cross-correlation to eliminate multiple scattering .
 PARTICLE CHARACTERIZATION 21-11

RF generator RF detector particle, (2) the measurement of the amount of light extinction caused by the
x << λ particle presence, (3) the measurement of the residence time during motion
entrainment through a defined distance, or (4) particle velocity .
Many commercial instruments are available, which vary in optical design,
light source type, and means, and how the particles are presented to the light .
Instruments using light scattering cover a size range of particles of 50 nm
to about 10 mm (liquid-borne) or 20 mm (gas-borne), while instruments
using light extinction mainly address liquid-borne particles from 1 mm to
the millimeter size range . The size range capability of any single instrument
is typically 50:1 . International standards are as follows: ISO 13323-1:2000,
Determination of Particle Size Distribution—Single-Particle Light Interaction
Methods, Part 1: Light Interaction Considerations; ISO 21501-2:2007, Determi-
λ x >> λ
scattering nation of Particle Size Distribution—Single Particle Light-Interaction Methods,
measuring zone Part 2: Light-Scattering Liquid-Borne Particle Counter ; ISO 21501-3:2007, Part 3:
Light-Extinction Liquid-Borne Particle Counter ; ISO 21501-4:2007, Part 4:
Light-Scattering Airborne Particle Counter for Clean Spaces .
FIG. 21-15 Setup of an ultrasonic attenuation system for particle size analysis . Instruments using the residence time, such as the aerodynamic particle
sizers, or the particle velocity, as used by the phase Doppler particle analyzers,
measure the particle size primarily based on the aerodynamic diameter .
Acoustic Methods Ultrasonic attenuation spectroscopy is a method well Small-Angle X-Ray Scattering Method Small-angle X-ray scattering
suited to measuring the PSD of colloids, dispersions, slurries, and emul- can be used in a size range of about 1 to 300 nm . Its advantage is that the scat-
sions (Fig . 21-15) . The basic concept is to measure the frequency-dependent tering mainly results from the differences in the electron density between the
attenuation or velocity of the ultrasound as it passes through the sample . particles and their surroundings . As internal crystallites of external agglom-
The attenuation includes contributions from the scattering or absorption erates are not visible, the measured size always represents the size of the
of the particles in the measuring zone and depends on the size distribu- primary particles and the requirement for dispersion is strongly reduced
tion and the concentration of the dispersed material (ISO 20998-1:2006, [Z . Jinyuan, L . Chulan, and C . Yan, “Stability of the Dividing Distribution
Particle Characterization by Acoustic Methods, Part 1: Ultrasonic Attenuation Function Method for Particle Size Distribution Analysis in Small Angle X-Ray
Spectroscopy) . Scattering,” J. Iron & Steel Res. Inst. 3(1): 1996; ISO 13762:2001, Particle Size
In a typical setup (see Fig . 21-15) an electric high-frequency generator is Analysis—Small Angle X-ray Scattering Method ] .
connected to a piezoelectric ultrasonic transducer . The generated ultra- Focused-Beam Techniques These techniques are based on a focused
sonic waves are coupled into the suspension and interact with the sus- light beam, typically a laser, with the focal point spinning on a circle parallel
pended particles . After passing the measuring zone, the ultrasonic plane to the surface of a glass window . When the focal point passes a particle, the
waves are received by an ultrasonic detector and converted to an electric reflected and/or scattered light of the particle is detected . The focal point
signal, which is amplified and measured . The attenuation of the ultrasonic moves along the particle on circular segments, as displayed in Fig . 21-16 .
waves is calculated from the ratio of the signal amplitudes on the generator Sophisticated threshold algorithms are used to determine the start point
and detector sides . and endpoint of the chord, i .e ., the edges of the particle . The chord length
PSD and concentration can be calculated from the attenuation spectrum is calculated from the time interval and the track speed of the focal point .
by using either complicated theoretical calculations requiring a large number Focused-beam techniques measure a chord length distribution, which cor-
of parameters or an empirical approach employing a reference method for responds to the size and shape information of the particles typically in a
calibration . Following U . Riebel (Die Grundlagen der Partikelgrößenanalyse complicated way (J . Worlische, T . Hocker, and M . Mazzoti, “Restoration of
mittels Ultraschallspektrometrie, PhD thesis, University of Karlsruhe), the PSD from Chord Length Distribution Data Using the Method of Projections
ultrasonic extinction of a suspension of monodisperse particles with onto Convex Sets,” Part. Part. Syst. Char. 22: 81 ff .) . So often the chord length
diameter x can be described by Lambert-Beer’s law . The extinction −ln(I/I0) distribution is directly used as the fingerprint information of the size, shape,
at a given frequency f is linearly dependent on the thickness of the suspension and population status .
layer ∆l, the projection area concentration CPF, and the related extinction cross Instruments of this type are commercially available as robust finger
section K . In a polydisperse system the extinctions of single particles overlay: probes with small probe diameters . They are used in on-line and preferably
in in-line applications, monitoring the chord length distribution of suspen-
 I  sions and emulsions . Special flow conditions are used to reduce the sam-
- ln   ≅ ∆l ⋅ C PF ⋅ ∑ K ( f i , x j ) ⋅ q2 ( x j ) ∆x (21-15) pling errors . Versions with fixed focal distance [Focused Beam Reflectance
 I 0  fi j
Measurement (FBRM)] and variable focal distance (3D ORM technology)
When the extinction is measured at different frequencies fi , this equation are available . The latter improves this technique for high concentrations
becomes a linear equation system, which can be solved for CPF and q2(x) . and widens the dynamic range, as the focal point moves horizontally and
The key for the calculation of the particle size distribution is the knowledge vertically with respect to the surface of the window . For instruments refer,
of the related extinction cross section K as a function of the dimensionless e .g ., to Mettler-Toledo International Inc . (Lasentec FBRM probes) and
size parameter σ = 2πx/λ . For spherical particles K can be evaluated directly Messtechnik Schwartz GmbH (PAT) .
from the acoustic scattering theory . A more general approach is an empiri- Electrical Sensing Zone Methods In the electric sensing zone
cal method using measurements on reference instruments as input . method (Fig . 21-17), a well-diluted and well-dispersed suspension in an
This disadvantage is compensated by the ability to measure a wide size electrolyte is caused to flow through a small aperture [Kubitschek, Research
range from below 10 mm to above 3 mm and the fact that PSDs can be mea- 13: 129 (1960)] . The changes in the resistivity between two electrodes on
sured at very high concentrations (0 .5 to >50 percent of volume) without
dilution . This eliminates the risk of affecting the dispersion state and makes
this method ideal for in-line monitoring of, e .g ., crystallizers (A . Pankewitz
and H . Geers, LABO, “In-line Crystal Size Distribution Analysis in Industrial
Crystallization Processes by Ultrasonic Extinction,” May 2000) .
Current instruments use different techniques for the attenuation mea-
surement: with static or variable width of the measuring zone, measurement
in transmission or reflection, with continuous or swept frequency generation,
with frequency burst or single-pulse excitation .
For process environment, probes are commercially available with a fre-
quency range of 100 kHz to 200 MHz and a dynamic range of >150 dB,
covering 1 to 70 percent of volume concentration, 0 to 120°C, 0 to 40 bar,
pH 1 to 14, and hazardous areas as an option .
Vendors of this technology include Sympatec GmbH (OPUS), Malvern
Instruments Ltd . (Ultrasizer), Dispersion Technology Inc . (DT series), and
Colloidal Dynamics Pty Ltd . (AcoustoSizer) .
Single-Particle Light Interaction Methods Individual particles have
been measured with light for many years . The measurement of the parti- FIG. 21-16 Different chords measured on a constantly moving single spherical
cle size is established by (1) the determination of the scattered light of the particle by focused-beam techniques .
21-12 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

FIG. 21-17 Multisizertm 3 COULTER COUNTER from Beckman Coulter, Inc ., uses
the electrical sensing zone method .

either side of the aperture, as the particles pass through, are related to the
volumes of the particles . The pulses are fed to a pulse-height analyzer where
they are counted and scaled . The method is limited by the resolution of
the pulse-height analyzer of about 16,000:1 (corresponding to a volume
diameter range of about 25:1) and the need to suspend the particles in an
electrolyte (ISO 13319:2007, Determination of Particle Size Distributions—
Electrical Sensing Zone Method ) .
Gravitational Sedimentation Methods In gravitational sedimenta-
tion methods, the particle size is determined from the settling velocity and the FIG. 21-18 Equipment used in the pipette method of size analysis .
undersize fraction by changes of concentration in a settling suspension . The
equation relating particle size to settling velocity is known as Stokes’ law
(ISO 13317-1:2001, Part 1: General Principles and Guidelines): ±2 percent is possible by using this apparatus . The technique is versatile in
that it is possible to analyze most powders dispersible in liquids; its dis-
18 ηu advantages are that it is a labor-intensive procedure, and a high level of
x St = (21-16)
(ρs − ρ f ) g skill is needed (ISO 13317-2:2001, Part 2: Fixed Pipette Method) .
The hydrometer method is simpler in that the density of the suspension,
where xSt is the Stokes diameter, η is viscosity, u is the particle settling veloc- which is related to the concentration, is read directly from the stem of the
ity under gravity, ρs is the particle density, ρf is the liquid density, and g is the hydrometer while the depth is determined by the distance of the hydrometer
gravitational acceleration . bulb from the surface (ASTM Spec . Pub . 234, 1959) . The method has a low
The Stokes diameter is defined as the diameter of a sphere having the resolution but is widely used in soil science studies .
same density and the same velocity as the particle settling in a liquid of the In gravitational photo sedimentation methods, the change of the concen-
same density and viscosity under laminar flow conditions . Corrections for tration with time and depth of sedimentation is monitored by using a light
the deviation from Stokes’ law may be necessary at the coarse end of the size point or line beam . These methods give a continuous record of changing
range . Sedimentation methods are limited to sizes above 1 µm due to the optical density with time and depth and have the added advantage that the
onset of thermal diffusion (brownian motion) at smaller sizes . beam can be scanned to the surface to reduce the measurement time . A cor-
An experimental problem is to obtain adequate dispersion of the particles rection needs to be applied to compensate for a deviation from the laws of
prior to a sedimentation analysis . For powders that are difficult to disperse, geometric optics (owing to diffraction effects the particles cut off more light
the addition of dispersing agents is necessary, along with ultrasonic probing . than geometric optics predicts) . The normalized measurement is a Q2(x)
It is essential to examine a sample of the dispersion under a microscope to distribution (ISO 13317-4:2014, Part 4: Photo Gravitational Method ) .
ensure that the sample is fully dispersed . (See the subsection Wet Dispersion .) In gravitational X-ray sedimentation methods, the change of the concen-
Equations to calculate size distributions from sedimentation data are tration with time and depth of sedimentation is monitored by using an X-ray
based on the assumption that the particles sink freely in the suspension . To beam . These methods give a continuous record of changing X-ray density
ensure that particle-particle interaction can be neglected, a volume concen- with time and depth and have the added advantage that the beam can be
tration below 0 .2 percent is recommended . scanned to the surface to reduce the measurement time . The methods are
There are various procedures available to determine the changing solid limited to materials having a high atomic mass (i .e ., X-ray-opaque material) and
concentration of a sedimenting suspension: give a Q3(x) distribution directly (ISO 13317-3:2001, Part 3: X-ray Gravitational
In the pipette method, concentration changes are monitored by extract- Technique) . See Fig . 21-19 .
ing samples from a sedimenting suspension at known depths and pre- Sedimentation Balance Methods In sedimentation balances the
determined times . The method is best known as Andreasen modification weight of sediment is measured as it accumulates on a balance pan sus-
[Andreasen, Kolloid-Z. 39: 253 (1929)], shown in Fig . 21-18 . Two 10-mL sam- pended in an initial homogeneous suspension . The technique is slow due to
ples are withdrawn from a fully dispersed, agitated suspension at zero time the time required for the smallest particle to settle out over a given height .
to corroborate the 100 percent concentration given by the known weight of The relationship between settled weight P, weight undersize W, and time t
powder and volume of liquid making up the suspension . The suspension is is given by
then allowed to settle in a temperature-controlled environment, and 10-mL
samples are taken at time intervals in geometric 2:1 time progression dP
P =W − (21-17)
starting at 1 min (that is, 1, 2, 4, 8, 16, 32, 64 min) . The amount of powder in d ln t
the extracted samples is determined by drying, cooling in a desiccator, and
weighing . Stokes diameters are determined from the predetermined times Centrifugal Sedimentation Methods These methods extend
and the depth, with corrections for the changes in depth due to the extrac- sedimentation methods well into the submicrometer range . Alterations of
tions . The cumulative mass undersize distribution comprises a plot of the the particle concentration may be determined space- and time-resolved
normalized concentration versus the Stokes diameter . A reproducibility of during centrifugation (T . Detloff and D . Lerche, “Determination of Particle
 PARTICLE CHARACTERIZATION 21-13

The disc centrifuge, developed by Slater and Cohen and modified by Allen
and Svarovsky [Allen and Svarovsky, Dechema Monogram, Nuremberg, nos .
1589–1625, pp . 279–292 (1975)], is essentially a centrifugal pipette device .
Size distributions are measured from the solids concentration of a series
of samples withdrawn through a central drainage pillar at various time
intervals .
In the centrifugal disc photodensitometer, concentration changes are mon-
itored by a light point or line beam . In one high-resolution mode of opera-
tion, the suspension under test is injected into clear liquid in the spinning
disc through an entry port, and a layer of suspension is formed over the free
surface of liquid (the line start technique) . The analysis can be carried out
using a homogeneous suspension . Very low concentrations are used, but the
light-scattering properties of small particles make it difficult to interpret the
measured data .
Several centrifugal cuvette photocentrifuges are commercially available .
These instruments use the same theory as the photocentrifuges but are
limited in operation to the homogeneous mode of operation (ISO 13318-1:2001,
Determination of Particle Size Distribution by Centrifugal Liquid Sedimenta-
tion Methods—Part 1: General Principles and Guidelines; ISO 13318-2:2007
Part 2: Photocentrifuge Method) .
The X-ray disc centrifuge is a centrifuge version of the gravitational instru-
ment and extends the measuring technique well into the submicrometer
range (ISO 13318-3:2004, Part 3: Centrifugal X-ray Method) .
Sieving Methods Sieving is probably the most frequently used and
abused method of analysis because the equipment, analytical procedure,
and basic concepts are deceptively simple . In sieving, the particles are pre-
FIG. 21-19 The Sedigraph III 5120 Particle Size Analysis System determines particle sented to equal-size apertures that constitute a series of go/no go gauges .
size from velocity measurements by applying Stokes’ law under the known conditions Sieve analysis implies three major difficulties: (1) with woven-wire sieves,
of liquid density and viscosity and particle density . Settling velocity is determined at the weaving process produces three-dimensional apertures with consid-
each relative mass measurement from knowledge of the distance the X-ray beam is erable tolerances, particularly for fine-woven mesh; (2) the mesh is easily
from the top of the sample cell and the time at which the mass measurement was damaged in use; (3) the particles must be efficiently presented to the sieve
taken . It uses a narrow, horizontally collimated beam of X-rays to measure directly the apertures to prevent blinding .
relative mass concentration of particles in the liquid medium . Sieves are often referred to their mesh size, which is a number of wires per
linear unit . Electroformed sieves with square or round apertures and toler-
ances of ±2 µm are also available (ISO 3310, Test Sieves—Technical Require-
Size Distributions Based on Space and Time Resolved Extinction Profiles in ments and Testing, 2016: Part 1: Test Sieves of Metal Wire Cloth; 2013; Part 2:
Centrifugal Field,” Proceedings of Fifth World Congress on Particle Technology, Test Sieves of Perforated Metal Plate; 1990; Part 3: Test Sieves of Electroformed
Session Particle Measurement, Orlando, Fla ., April 23–27, 2006) . Sizes are cal- Sheets) .
culated from a modified version of the Stokes equation: For coarse separation, dry sieving is used, but other procedures are nec-
essary for finer and more cohesive powders . The most aggressive agitation
18 ηu is performed with Pascal Inclyno and Tyler Ro-tap sieves, which combine
x St = (21-18)
(ρs − ρ f )ω 2 gyratory and jolting movement, although a simple vibratory agitation may
be suitable in many cases . With Air-Jet sieves, a rotating jet below the siev-
where ω is the radial velocity of the centrifuge . The concentration calcu- ing surface cleans the apertures and helps the passage of fines through the
lations are complicated due to radial dilution effects (i .e ., particles do not apertures . The sonic sifter combines two actions, a vertical oscillating col-
travel in parallel paths as in gravitational sedimentation but move away from umn of air and a repetitive mechanical pulse . Wet sieving is frequently used
each other as they settle radially outward) . Particle velocities are given by with cohesive powders .
Elutriation Methods and Classification In gravity elutriation the
ln (r /s) particles are classified in a column by a rising fluid flow . In centrifugal elu-
u= (21-19)
t triation the fluid moves inward against the centrifugal force . A cyclone is a
centrifugal elutriator, although it is not usually so regarded . The cyclosizer
where both the measurement radius r and the surface radius s can be varying . is a series of inverted cyclones with added apex chambers through which
The former varies if the system is a scanning system, and the latter if the water flows . Suspension is fed into the largest cyclone, and particles are
surface varies due to the extraction of samples . separated into different size ranges .
Concentration undersize Dm is determined by Differential Electrical Mobility Analysis (DMA) Differential electri-
x
cal mobility analysis uses an electric field for the classification and analy-
sis of charged aerosol particles ranging from about 1 nm to about 1 µm in
= ∫ exp(−2 ktz
2
D )q3 ( x ) dz (21-20)
m
x min
a gas phase . It mainly consists of four parts: (1) A preseparator limits the
upper size to a known cutoff size . (2) A particle charge conditioner charges
ρs − ρ f the aerosol particles to a known electric charge (a function of particle size) .
with k= ω2 (21-21) A bipolar diffusion particle charger is commonly used . The gas is ionized
18 η either by radiation from a radioactive source (e .g ., 85Kr) or by ions emitted
from a corona electrode . Gas ions of either polarity diffuse to the aerosol
where q3(x) = dQ3(x)/dx is the volume or mass density distribution and z is particles until charge equilibrium is reached . (3) A differential electrical
the integration variable . mobility spectrometer (DEMS) discriminates particles with different electri-
The solution of the integral for measuring the concentration at constant cal mobility by particle migration perpendicular to a laminar sheath flow .
position over time is only approximately possible . A common way uses The voltage between the inner cylinder and the outer cylinder (GND) is varied
Kamack’s equation [Kamack, Br. J. Appl. Phys. 5: 1962–1968 (1972)] as rec- to adjust the discrimination level . (4) An aerosol particle detector uses,
ommended by ISO 13318-1:2001 (Part 1: Determination of Particle Size by e .g ., a continuous-flow condensation particle counter (CPC) or an aerosol
Centrifugal Liquid Sedimentation Methods). electrometer (AE) .
An analytical solution is provided by measuring the concentration to at A typical setup of the DEMS is shown in Fig . 21-20 . It shows the flow rates
least one time at different sedimentation heights: of the sheath flow F1, the polydisperse aerosol sample F2, the monodisperse
Dm 2 (classified) aerosol exiting the DEMS F3, and the excess air F4 .
 ri  dD
Q3 ( x ) = ∫  s 
1
m (21-22) The electrical mobility Z depends on the particle size x and the number
of elementary charges e:
where ri is the measurement position and s the surface radius; Q3(x) is the
cumulative mass or volume concentration; and (ri/si)2 is the radial dilution p ⋅e
Z (x ) = [1 + Kn( A + BeC/Kn )] (21-23)
correction factor . 3 πηx
21-14 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

0 – 20 kV F1 F2
Sampling TWISTER
finger

a
L Drive
b unit

F3 F4 GND

FIG. 21-20 Schematic of a differential electrical mobility analyzer . Sample

outlet
MYTOS
with the number of elementary charges p, the Knudsen number Kn of 2l/x,
the mean path l of the gas molecule, the dynamic fluid viscosity η, and
numeric constants A, B, C determined empirically .
FIG. 21-21 A typical on-line application with a representative sampler (TWISTER)
Commercial instruments are available for a variety of applications in in a pipe of 150 mm, which scans the cross section on a spiral line, and dry disperser
aerosol instrumentation, production of materials from aerosols, contami- with particle-sizing instrument (MYTOS) based on laser diffraction . (Courtesy of
nation control, etc . (ISO/CD 15900:2009, Determination of Particles Size Sympatec GmbH.)
Distribution—Differential Electrical Mobility Analysis for Aerosol Particles) .
Surface Area Determination The surface-to-volume ratio is an
important powder property since it governs the rate at which a powder
interacts with its surroundings, e .g ., in chemical reactions . The surface area On-Line On-line places the measuring device in the process environ-
may be determined from size distribution data or measured directly by flow ment close to, but not in, the production line . The fully automated system
through a powder bed or the adsorption of gas molecules on the powder includes the sampling, but the sample is transported to the measuring
surface . Other methods such as gas diffusion, dye adsorption from solution, device . Mainly laser diffraction, ultrasonic extinction, and dynamic light
and heats of adsorption have also been used . The most commonly used scattering are used . See Fig . 21-23 .
methods are as follows: In-Line In-line implements sampling, sample preparation, and measure-
In mercury porosimetry, the pores are filled with mercury under pressure ment directly in the process, keeping the sample inside the production line . This
(ISO 15901-1:2005, Pore Size Distribution and Porosity of Solid Materials— is the preferred domain of laser diffraction (mainly dry), image analysis, focused-
Evaluation by Mercury Porosimetry and Gas Adsorption—Part 1: Mercury beam techniques, and ultrasonic extinction devices (wet) . See Fig . 21-24 .
Porosimetry) . This method is suitable for many materials with pores in the
diameter range of about 3 nm to 400 mm (especially within 0 .1 to 100 mm) .
VERIFICATION
In gas adsorption for micro-, meso- and macropores, the pores are character-
ized by adsorbing gas, such as nitrogen at liquid-nitrogen temperature . This The use of reference materials is recommended to verify the correct function
method is used for pores in the ranges of approximately <2 nm (micropores), of the particle sizing equipment . A simple electrical, mechanical, or optical
2 to 50 nm (mesopores), and >50 nm (macropores) (ISO 15901-2:2006, Pore Size test is generally not sufficient, as all functions of the measuring process,
Distribution and Porosity of Solid Materials—Evaluation by Mercury Porosim- such as dosing, transportation, and dispersion, are only tested with sample
etry and Gas Adsorption, Part 2: Analysis of Meso-pores and Macro-pores by Gas material applied to the instrument .
Adsorption; ISO 15901-3:2007, Part 3: Analysis of Micro-pores by Gas Adsorption) . Reference Materials Many vendors supply certified standard reference
An isotherm is generated of the amount of gas adsorbed versus gas pressure, materials which address either a single instrument or a group of instruments .
and the amount of gas required to form a monolayer is determined . As these materials are expensive, it is often advisable to perform only the
Many theories of gas adsorption have been advanced . For mesopores the primary tests with these materials and perform secondary tests with a stable
measurements are usually interpreted by using the BET theory [Brunauer, and well-split material supplied by the user . For best relevance, the size range
Emmet, and Teller, J. Am. Chem. Soc. 60: 309 (1938)] . Here the amount of and distribution type of this material should be similar to those of the desired
absorbed na is plotted versus the relative pressure p/p0. The monolayer application . It is essential that the total operational procedure be adequately
capacity nm is calculated by the BET equation: described in full detail (S . Röthele and W . Witt, “Standards in Laser Diffraction,”
5th European Symposium Particle Char ., Nuremberg, March 24–26, 1992) .
p /p0 1 C −1 p
= + ⋅ (21-24)
na (1 − p /p0 ) nmC nmC p0

The specific surface per unit mass of the sample is then calculated by assess- Sample inlet
ing a value am for the average area occupied by each molecule in the complete
monolayer (say, am = 0 .162 nm2 for N2 at 77 K) and the Loschmidt number L:

a s = nm ⋅ a m ⋅ L (21-25)
Vibratory feeder

PARTICLE SIZE ANALYSIS IN THE PROCESS ENVIRONMENT Dry disperser

The growing trend toward automation in industry has resulted in the devel-
opment of particle sizing equipment suitable for continuous work under
process conditions—even in hazardous areas (Fig . 21-21) . The acquisition of
particle size information in real time is a prerequisite for feedback control
of the process .
Today the field of particle sizing in process environment is subdivided
into three branches of applications . MYTOS
At-Line At-line is the fully automated analysis in a laboratory . The
sample is still taken manually or by stand-alone devices . The sample is (a) (b)
transported to the laboratory, e .g ., by pneumatic delivery . Several hundred
samples can be measured per day, allowing for precise quality control of FIG. 21-22 (a) At-line particle sizing MYTOS module (courtesy of Sympatec GmbH)
slow processes . At-line laser diffraction is widely used for quality control in based on laser diffraction, with integrated dosing and dry dispersion stage . (b) Module
the cement industry . See Fig . 21-22 . integrated into a Polysius Polab AMT for lab automation in the cement industry .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-15

TWISTER
MYTOS

MYTOS

TWISTER
OPUS

(a) (b)

FIG. 21-24 (a) Typical in-line laser diffraction system with a representative sampler
FIG. 21-23 Typical on-line outdoor application with a representative sampler
(TWISTER and MYTOS), all integrated in a pipe of 100 mm . (b) In-line application of
TWISTER 440, which scans the cross section on a spiral line in a pipe of 440 mm, and a an ultrasonic extinction (OPUS) probe monitoring a crystallization process in a large
hookup dry disperser with laser diffraction particle sizer MYTOS . (Courtesy of Sympatec vessel . (Both by courtesy of Sympatec GmbH.)
GmbH.)

BULK SOLIDS FLOW AND HOPPER DESIGN

General References: Jenike, Storage and Flow of Solids, Bulletin 123, University of For simplicity, the term bin will be used henceforth as a descriptor for a
Utah Engineering Station, 1964 (revised 1976); Schulze, Powders and Bulk Solids—Behavior, storage vessel of any size .
Characterization, Storage, and Flow, Springer, New York, 2007; Mehos, “Designing and Chute—equipment used to transfer bulk material by gravity between
Operating Gravity Dryers,” Chem. Eng . 116(5): 34 (May 2009); Maynard, “Ten Steps to
an Effective Bin Design,” Chemical Engineering Progress, pp . 25–32, (Nov . 2013); Carson, other pieces of equipment . The cross-section must be only partially full of
Pittenger, and Marinelli, “Characterize Bulk Solids to Ensure Smooth Flow,” Chem. Eng. material; otherwise it acts as a sloping bin or hopper .
pp . 54–59, April 2016 . Cylinder—vertical part of a bin . The cylinder may be round or rectangular
and has a constant cross section .
Expanded flow—flow pattern inside a bin, where all the bulk material is
FUNDAMENTALS in motion in the bottom portion of the vessel when withdrawn, but flow
only occurs in a flow channel in the top portion of the vessel centered over
Many industrial processes involve the transfer and feeding of bulk solids, and the outlet .
the ability of such materials to flow in a controlled manner during these oper- Feeder—device for modulating the withdrawal rate of bulk material,
ations is critical to product quality . Hoppers, bin, and silos are used to store e .g ., rotary valves, screw feeders, and belt feeders . Often, a valve or gate is
granular raw materials, intermediates, and final products . With modifica- used to stop and start flow, but such devices in general should not be used
tions, they can also be used as process vessels, such as purge columns, heaters to control the discharge rate of bulk solids .
and coolers, and moving-bed reactors . Transfer chutes are inclined or vertical Flow channel—the space in a bin in which the bulk solid is actually flowing
assemblies in which bulk solids flow by gravity from one location to another . at any point in time during withdrawal .
Unlike hoppers, bins, or silos, chutes are not filled with the bulk material . Funnel flow—flow pattern inside a bin, where the bulk material only
Bulk solids have unique properties that cause them to flow differently moves in a flow channel above the outlet when withdrawn .
from liquids . Bulk solids are frictional and in general are compressible . Hopper section—the converging part of a storage vessel that has sloped
Liquids are frictionless and are nearly incompressible . Bulk solid flow prop- walls and a variable cross section .
erties are strongly dependent on the consolidation stresses applied and Mass flow—flow pattern inside a bin where all material is in motion when
minimally if at all dependent on strain rate, whereas fluid flow properties any material is withdrawn .
are strongly dependent on strain rate and minimally dependent on abso- Pressure—force per unit area applied to an object in a direction perpen-
lute pressure . In addition, bulk solids are anisotropic whereas liquids are dicular to the surface; same as compressive stress .
isotropic . Process vessel—a bin that has been modified to allow heating, cooling, dry-
Definitions When discussing the storage and handling of granular ing, reacting, or other processes to take place . Frequently they are equipped
materials, the following definitions are commonly accepted: with heat exchangers or distributors for gas injection .
Bulk solid—a material consisting of discrete solid particles, handled in Stress—force per unit area, a tensor quantity .
bulk form . Flow Problems Many storage vessels are fabricated from architectural
Hopper, bin, or silo—storage vessels for bulk solids . The terms are often or fabrication viewpoints (e .g ., hopper walls sloped 30° from vertical for
used interchangeably . Silos usually refer to tall vessels that store several ease of fabrication or 45° to minimize headroom requirements and simplify
tons of material . Hoppers and bins frequently refer to smaller vessels . The design calculations) . However, designing equipment without regard to the
converging section of a storage vessel is often called the hopper section . bulk material being handled often leads to flow problems . Common solids
Examples of hopper, bin, and silo geometries are given in Fig . 21-25 . flow problems include the following:

Conical Pyramidal Wedge-shaped Transition Chisel

FIG. 21-25 Common bin geometries .

21-16 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Caking. Some materials will readily flow from a bin, portable container, or
bag if handled continuously or even after time at rest . Other materials, how-
ever, will exhibit flow problems if allowed to remain at rest for a period of
time . Given enough time at rest, some bulk solids will gain additional cohe-
sive strength, and obstructions to flow (e .g ., arches and ratholes) or hard
lumps may become exceptionally difficult to eliminate .
Segregation . Some materials, when transferred into a bin or pile, will seg-
regate; that is, particles of different size, shape, density, etc . will separate .
Segregation can occur by a number of different mechanisms, depending on
the physical characteristics of the particles and the method of handling .
Flow Patterns Three primary flow patterns can occur in a bin: mass
flow, funnel flow, and expanded flow . Mass flow and funnel flow are illus-
trated in Fig . 21-28 .
In funnel flow, an active flow channel forms above the outlet, with stag-
nant material remaining at the periphery . This occurs when the walls of the
hopper section of the storage vessel are not steep enough or have low enough
friction to allow flow along them . The size of the resultant flow channel is
approximately the largest dimension of the outlet . It is equal to the diameter
of a round outlet or the diagonal of a slotted outlet . In the case of conical
funnel flow bins, the fraction of the vessel volume that is active can be dra-
matically small . If the bulk material is cohesive, a stable rathole will form,
thereby reducing the effective capacity of the bin to a small fraction of its
intended capacity .
A funnel flow bin typically exhibits a first-in, last-out flow sequence . There-
fore, materials that readily cake or degrade over time should not be handled
in funnel flow hoppers . Funnel flow can cause erratic flow and induce high
FIG. 21-26 Cohesive arch . loads (depending on vessel size) on the structure and downstream equip-
ment due to collapsing ratholes and eccentric flow channels . If the bulk solid
is cohesive, ratholes may become stable, and the vessel will not empty .
Funnel flow bins are best suited for bulk solids that are free-flowing and do
No flow. If a stable dome, bridge, or arch forms over the outlet of a bin, not degrade or gain strength over time . They should not be used if segregation
the bulk solid will not flow when the feeder is started or the gate is opened . is a concern . Funnel flow bins require less headroom and in general are less
If a stable rathole forms in a vessel in which flow only occurs in a narrow expensive to build than mass flow bins since they can have shallower walls .
channel above the outlet, material will stop flowing when the flow channel In mass flow, the entire bed of solids is in motion when material is dis-
empties . Obstructions to flow are illustrated in Figs . 21-26 and 21-27 . charged from the outlet, including material along the walls . Mass flow bins
Erratic flow. Erratic flow occurs when both arching and ratholing occur . typically have steep and/or low-friction walls . Provided that the outlet is
If a rathole collapses due to external vibration, the bulk solid may arch as it large enough to prevent arching, all material will be discharged from the
impacts the outlet . After the arch fails due to vibration or operator interven- hopper, since ratholes cannot form .
tion, the flow channel will empty, leaving a rathole momentarily stopping Mass flow bins are characterized by a first-in, first-out flow sequence and
flow until it eventually collapses, reforming a cohesive arch . therefore are suitable for handling materials that degrade with time or are
Flooding. If a stable rathole develops and fresh material is added or if a prone to caking . The steep hopper walls provide a more uniform flow than fun-
rathole collapses and falls into the channel, the material may become aer- nel flow bins, making mass flow hoppers suitable for process vessels . Discharge
ated or fluidized . Since most feeders are designed to handle solids and not rates are predictable and steady, since the bulk density of the material at the
fluids, the fluidized material may flood, that is, discharge uncontrollably in outlet is nearly independent of the head of the material inside the vessel . Seg-
a fluidized state from the bin, and the feeder will not be able to control the regation by the dusting or sifting mechanism is minimized, as fine and course
rate of discharge . particles separated during filling are remixed at the outlet during discharge .
Limited discharge rate. Because a fine powder dilates as it flows toward A disadvantage of a mass flow bin is that it requires more headroom due to
the outlet, vacuum will naturally develop inside the hopper above the out- its steep hopper section . This is especially the case for conical mass flow bins .
let . As a consequence, air will flow counter to the solids, disrupting flow . Expanded flow is characterized by mass flow in the lowermost section of a
Increasing the speed of the feeder will no longer increase the discharge rate bin and funnel flow in the upper section . An expanded flow bin is illustrated
of powder as the discharge rate has become limited . in Fig . 21-29 .
The outlet of the funnel flow section must be large enough to prevent a
stable rathole from developing . Because the bottom section is designed for
mass flow, discharge rates are uniform and predictable . Expanded flow hop-
pers are frequently used when large bin diameters are required .

FIG. 21-27 Stable rathole . FIG. 21-28 Flow patterns—funnel flow (left) and mass flow (right) .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-17

σy
τyx
τxy

σx σx
x

τxy
σyx
σy

FIG. 21-30 Stress on an element of bulk solid .

Funnel
flow
When one is analyzing solids, the maximum and minimum stresses fre-
quently must be identified . These stresses are called principal stresses and
can be readily found using Mohr’s circle, which can be derived from the
Mass stress transformation equations . Mohr’s circle is illustrated in Fig . 21-32 and
flow is described by

FIG. 21-29 Expanded flow pattern .

(σ x − σ avg )2 + τ 2xy = R 2 (21-29)

where
σx + σ y
σ avg = (21-30)
ANALYSIS OF STRESS 2
and
Bulk solids are anisotropic; their stresses vary with direction . Although a
2
bulk solid consists of individual particles, it is convenient to describe a bulk  σx − σ y 
R2 =  + τ 2xy (21-31)
material as a continuum . A bulk solid element is sketched in Fig . 21-30 . The  2 
stresses acting normal (i .e ., perpendicular to) the element in the x and y
directions are denoted σx and σy, respectively . The shear stresses acting in Note that Mohr’s circle is centered at σavg and the two points (σx, τxy) and
the x and y directions are denoted τyx and τxy, respectively . (σ y , − τxy) lie on opposite sides of the circle . To determine the stresses with
Given a state of stress, the magnitude of the normal and shear stresses respect to the rotated or transformed axes, the line connecting the two
acting on the bulk material will depend on the coordinate system used to points (σx, τxy) and (σ y , − τxy) is rotated 2θ .
describe the direction of these stresses . A new set of axes, denoted by x′ and y ′, The maximum and minimum values of the normal stresses, i .e ., the major
rotated an angle θ from the original axes, is shown in Fig . 21-31 . and minor principal stresses, respectively, can be determined from the two
The stresses in terms of the new coordinate system are given by the fol- intersection points of Mohr’s circle and the horizontal axis . The major prin-
lowing stress transformation equations: cipal stress σ1 and minor principal stress σ2 can therefore be calculated from

σx + σ y σx − σ y σ1 = σ avg + R (21-32)
σ x′ = + cos2 θ + τ xy sin2 θ (21-26)
2 2
σ 2 = σ avg − R (21-33)
σx + σ y σx − σ y
σ y′ = − cos2 θ − τ xy sin2 θ (21-27) SOLIDS-INDUCED LOADS
2 2
The geometry of the bin and the solids flow properties, which determine the
solids flow pattern, prescribe the pressure profiles that develop within the
σx − σ y bulk solids during initial fill and discharge . Solids-induced load analy-
τ x ′y ′ = − sin2 θ + τ xy cos2 θ (21-28)
2 ses are used to determine vessel wall thicknesses and reinforcements for

y y

y´

σy τ y´x´ y
σ y´ τx
τyx
τxy σ x´ x´

σx σx θ
x x

τxy σ x´
τyx
τ x´y´ σ y´
σy τ y´x´

FIG. 21-31 Stresses on a rotated element .

21-18 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

where σv is the vertical stress, z is the depth of the solids bed, RH is the hydraulic
σ1 radius, k is the Janssen coefficient, which is equal to the ratio of the stress
acting perpendicular to the walls to the vertical stress, ρb is the bulk density,
g is acceleration due to gravity, and φ′ is the wall friction angle, which is equal
σy to the inverse tangent of the wall friction coefficient . If a constant bulk density
is selected, Eq . (21-34) can be integrated to give the Janssen equation:

ρb gRH   k tan φ′  
τxy
σv = 1 − exp  − z  (21-35)

R
k tan φ′   RH 

The horizontal stress σh is given by

σx
2θ
τxy ρb gRH   k tan φ′  
σ2
σh = 1 − exp  − z  (21-36)
tan φ′   RH 

σavg The value of k is typically in the range of 0 .3 to 0 .6 . Note that unlike for liq-
uids, the maximum solids stress is proportional to the hydraulic radius of
σx the cylinder and is independent of its height .
Hopper Section Walker [Chem. Eng. Sci. 21: 11, 975 (1966)] and Walters
τxy [Chem. Eng. Sci. 28: 1, 13 (1973)] analyzed the stresses in the hopper section
by performing an equilibrium force balance on an elemental volume with
FIG. 21-32 Stress formation using Mohr’s circle analysis . converging sides . A differential equation results, which is generalized by
Jenike [Loeffler, F . J ., and C . R . Proctor (eds .), Unit and Bulk Materials
Handling, Effect of Solids Flow Properties and Hopper Configuration on Silo
structural integrity and the load on feeders, which is needed to determine Loads, ASME, 1980, pp . 97–106] as
power requirements .
A typical bin consists of a vertical (cylinder) section followed by a con- dσ v σ
verging (hopper) section . Solids stresses are illustrated in Fig . 21-33 . - n v = − g ρb (21-37)
dz z
In the cylindrical section, the vertical and wall stresses increase with
depth, tending asymptotically toward a maximum . The wall stresses are   tan φ′  
smaller than the vertical stresses by a factor equal to k. At the centerline of n = (m + 1)  k  1 +  −1 (21-38)
the bin, the major principal stresses act vertically downward, and the minor   tan θ  
principal stresses act horizontally . where θ is the hopper angle ( from vertical) and m is equal to 1 for a conical
When a previously empty bin is initially filled with a bulk solid, the major hopper and equal to 0 for a straight-walled hopper having a slotted outlet .
principal stresses in the converging section act vertically downward at the Integration yields the following [British Standard BS EN 1991-4:2006]:
centerline . This stress state after initial fill is called the active stress state .
A discontinuity exists in the wall stress profile . Both the wall stresses and ρb gh  x  x  
n n

−    + σ vht  
x
vertical stresses decrease toward the hopper outlet . σv =  (21-39)
n −1  h  
h   h 
When material is discharged from the bin, the stress conditions in the
hopper section change if mass flow develops . In order to flow, the bulk solid
where x is the vertical coordinate upward from the hopper apex, h is the
is compressed laterally and expands vertically . As a result, the major principal vertical height between the hopper apex and the cylinder-hopper transition,
stresses act horizontally at the centerline . This state of stress is called the and σvht is the mean vertical stress on the solid at the transition after filling
passive state . A peak stress, called the switch, occurs at the hopper-cylinder
[as determined from Eq . (21-34)] .
interface .
The wall stress σw is calculated from
Cylinder Stresses The stresses on the solids in the straight-walled
section of a bin were originally calculated by Janssen [Z. Ver. Dt. Ing. 39: 1045
σ w = kσ v (21-40)
(1895)] . An equilibrium force balance on a volume element of bulk solids
yields the following differential equation:
The frictional traction τw is determined from
d σ v k tan f′
+ σ v = ρb g (21-34)
dz RH τw = kσ w (21-41)

Mean vertical stress Mean vertical stress

Wall stress Wall stress

Stress Stress
(a) Initial fill (b) Flow

FIG. 21-33 Representative stress profiles in a mass flow bin; wall stress acts normal to the cylinder or hopper walls .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-19

The value of the stress ratio k depends on the flow properties of the bulk In contrast to fluids, materials having the same composition frequently
material handled and the slope of the hopper walls . Methods described by have dramatically different fundamental flow properties . This is so because
Enstad [Chem. Eng. Sci. 40: 10, 1273 (1975)] can be used to calculate the flow properties are often dependent on the material’s particle size, shape,
stress ratio . and particle size distribution . In addition, temperature, moisture content
Funnel Flow Hoppers In funnel flow hoppers, the wall friction is not (or the relative humidity of the material’s interstitial air), purity, surface
fully mobilized . An effective wall friction angle is therefore determined per energy, and morphology all can influence the flow behavior of a bulk solid .
EN 1991-4:2006 . The flow properties of many materials change when they are stored at rest .
Eccentric Loads Although more common with funnel flow, eccentric There is no substitute for measuring the flow properties of the actual
discharge, in which the flow channel formation is not concentric with the materials that are to be handled, and testing should be performed over a
vertical section of the bin, can occur in mass flow as well . Eccentric discharge range of temperatures, moisture contents, relative humidity levels, times at
can occur when a single outlet is not centered or when multiple outlets do rest, and stress levels at which the bulk solid will be stored and handled .
not discharge at the same rate . Eccentric flow can also occur if gates or valves Using flow property data from the literature or assuming that the proper-
are partially opened or if feeder interfaces are not properly designed . ties are the same as those of other materials whose properties are known is
Asymmetric pressures develop on the walls when eccentric discharge extremely risky . For example, “coal” is an extremely nonhomogeneous mate-
takes place . A procedure for calculating discharge loads during eccentric rial; yet some textbooks and design codes provide values (a limited range
flow is described in BS EN 1991-4:2006 . or sometimes a single value) of important design parameters such as bulk
Note that the formulas for calculating normal pressures and shear trac- density and wall friction . Such tabular data are, at a minimum, misleading
tions above do not include inherent factors of safety . A prudent designer and potentially worse than having no data at all .
must recognize that factors of safety are always necessary to account for Cohesive Strength and Internal Friction A bulk solid gains cohesive
unexpected loading conditions or deficiencies in the ability of the structure strength when consolidated in a bin . The size of the outlet of the vessel that
to withstand these loads . The designer is ultimately responsible for choos- will prevent arching or the formation of a stable rathole depends greatly on
ing the appropriate safety factor based on risks associated with structural the bulk material’s cohesive strength .
failure and compliance with applicable design codes . Qualified engineers Figure 21-34 is a schematic of a uniaxial compressive strength tester . In a
should review results of load analyses . uniaxial test, a sample is placed in a cell with nearly frictionless walls and
is then consolidated by applying a stress equal to σ1 . Next, the load and cell
are removed . The compacted specimen is then loaded with increasing com-
BULK SOLIDS FLOW PROPERTIES TESTING pressive stress until it breaks apart, i .e ., fails . The failure stress is called the
Designing systems for bulk solids can be challenging since they have a wide material’s cohesive strength or the unconfined yield strength fC .
range of characteristics, e .g ., cohesive or free-flowing; fine or coarse; fluffy Uniaxial compressive strength test results are often highly variable .
or dense; adhesive to surfaces or surface-repellant; easily aerated or nearly Improvements have been made to uniaxial strength testers to reduce their
impermeable; and highly compressible or nearly incompressible . Defining variability; however, uniaxial compression tests usually do not provide a
a particle size distribution, density, or permeability is relatively straightfor- bulk material’s true unconfined yield strength, and therefore the cohe-
ward . The best metric for cohesion or adhesion is not as obvious . Various sive strength of a bulk solid is best measured by direct shear cell testing .
characteristics or a combination of them are needed to define a bulk mate- Translational (Jenike), annular (ring), and rotational testers are frequently
rial’s ease of flow or “flowability .” However, even this is not enough, since used . They are described in ASTM standards D-1628 (translational), D-6773
flowability also depends on the design of the vessel from which the material (annular), D-6682, and D7891 (rotational) . Schematics of the testers are
is flowing—or is attempting to do so . given in Fig . 21-35 .
Several methods exist for measuring the relative flowability of bulk materials . The direct translational shear tester was originally developed by Andrew
The simplest is to determine the bulk solid’s angle of repose by pouring it onto a Jenike [Storage and Flow of Solids, Bulletin 123, University of Utah, 1964
horizontal surface and measuring the surcharge angle of the pile that is formed . (revised, 1980)] . This tester is particularly robust in that its cell can be
A material that forms a steeper pile is believed to be less flowable than one that placed in extreme environments, allowing a material’s cohesive strength to
is shallow . However, as stated by Jenike [Storage and Flow of Solids, Bulletin 123, be measured over a full spectrum of process conditions . Its disadvantage is
University of Utah, 1964 (revised, 1980)], “The angle of repose is not a measure that significant operator training and experience are usually required to be
of the flowability of solids . In fact, it is useful only in the determination of the able to obtain reproducible results .
contour of a pile, and its popularity among engineers and investigators is Modern annular and rotational shear testers are computer-controlled
due not to its usefulness but to the ease with which it is measured .” and are thus more straightforward to operate and less prone to operator
A compressibility test is another relative measure of flowability . A sample variability than manually controlled shear testers .
of bulk solid is vibrated or compacted (“packed”) inside a rigid container, In a rotational shear cell, shear deformation of the specimen varies with
and its change in bulk density is measured . Two common methods to ana- radius in the cell: at the perimeter it is at its maximum, while at the center it
lyze the results are the Hausner ratio and Carr compressibility . The former is is zero . This can result in data that differ from results obtained using a Jen-
the ratio of the “tapped” density to the aerated or loose bulk density . The lat- ike (translational) or annular shear cell . Rotational and annular shear cells
ter is determined by dividing the difference between the packed and freely permit infinite travel, so they are better suited than a Jenike shear cell for
settled bulk density by the packed bulk density . A low Hausner ratio or Carr testing bulk solids that require large shear strain to reach steady state . The
compressibility supposedly indicates that the material is easy to handle . Jenike shear cell tester can be relatively easily modified to operate at high or
However, these indices are of limited use, since at best they can be only low temperatures, whereas this is more difficult, if not impossible, with the
loosely correlated to the flow behavior of similar bulk solids . In addition, other two types of testers .
these methods are deficient as the stress applied to the sample of bulk solid Cohesive strength is measured by shear cell testing as described in ASTM
is unknown, the tests do not replicate the degree of consolidation that takes D-1628, D-6682, D-6773, or D7891 . A sample of bulk material is placed in a
place when a material is stored in a vessel, and the gain in the material’s cell and then presheared, i .e ., consolidated by exerting a normal stress and
strength during rest cannot be determined . then shearing it until the measured shear stress is steady [as illustrated by
Solids rheometers of various designs are sometimes used to quantify the the point (σss, τss) shown in Fig . 21-36 . Next the shear step is conducted, in
relative flowability of bulk solids . The material is placed in a cell equipped which the vertical compacting load is replaced with a smaller load, and the
with an impeller, and the torque or energy required to rotate the agitator is
measured . In some instruments, the vertical force on the agitator can also
be directly measured . Flowability is deemed to correlate with the torque or
σ1
the power drawn by the agitator . Unfortunately, the stresses acting in the
shear zone during testing are unknown, and therefore the results cannot
be applied to actual process conditions . In addition, both fluidization and fc
agglomeration can occur inside the test cell, confounding the results . High
torque or energy consumption may be the result of high friction between
the bulk material and the walls of the cell, rather than an indication of the
material’s cohesive strength . Test methods based on stirred vessels there-
fore have questionable utility [Schulze, Powders and Bulk Solids: Behavior,
Characterization, Storage and Flow, Springer, 2007] . σ2 = 0
There are five fundamental bulk solid flow properties that provide
useful and reliable information from which one can design a bin for reli-
able flow: cohesive strength, internal friction, bulk density, wall friction,
and permeability . FIG. 21-34 Uniaxial compressive strength test .
 21-20 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Normal stress

Bracket
Cover
Shear stress

Shear stress τ
Ring
Base

Bulk solid
φ
Normal stress
Shear stress
Cover δ
fc σ2 σ1
Annular ring Normal stress σ

FIG. 21-37 Linear yield locus .

Bulk solid

Normal stress on the yield locus should lie to the right of the point of tangency to the
smaller Mohr’s circle .
Shear stress Also determined are the effective angle of friction and kinematic angle
Cover of internal friction (δ and φ, respectively) . The effective angle of friction is
found by constructing a line through the origin and tangent to the larger
Ring Mohr’s semicircle . The kinematic angle of internal friction is the angle
formed between a line that is horizontal and one drawn tangent to the
smaller Mohr’s circle at its intersection with the yield locus (see Fig . 21-37) .
The yield locus generally is slightly concave downward, but with many
particulate solids a straight line is a sufficient approximation . If the yield
Bulk solid locus is approximated as a straight line for all particulate solids, then subse-
quent calculations are much simpler, but, in some cases, somewhat conser-
FIG. 21-35 Shear cell testers—Jenike direct (top), annular (middle), and rotational vative results may be obtained, that is, a higher fc value will be determined
(bottom) . than when using a fitted curve . The shear data that make up the yield locus
(i .e ., all data points without the steady-state or preshear data) are regressed
sample is again sheared until it fails . These preshear and shear steps are to give the following linear relation:
repeated at the same consolidation level for a number of reduced normal
stresses, and the yield locus is then determined by plotting the failure shear τ = c + σ tan φ (21-42)
stress versus normal stress (see Fig . 21-36) .
Ideally, all measurements of the preshear shear stress τss should be iden- where τ is the shearing stress and σ is the normal stress . Equation (21-42)
tical . However, because of unavoidable variability during testing, there is is the Coulomb equation . The slope of the line is equal to the tangent of the
inevitably scatter in the τss values . Prorating is used to account for the vari- kinematic angle of internal friction φ, and the intercept is equal to c, which
ability of the data [ Jenike, Storage and Flow of Solids, Bulletin 123, University is called the material’s cohesion .
of Utah, 1964 (revised, 1980)] . The unconfined yield strength and major consolidation stress are then
To determine the major consolidation stress σ1 and the unconfined calculated from
yield shear strength fC from the yield locus, a line is drawn through the 2c (1 + sin φ)
shear step data . Mohr’s semicircle is then drawn through the steady-state fC = (21-43)
result (σss, τss) tangent to the yield locus line (see Fig . 21-37) . cos φ
The larger point of intersection of the semicircle with the horizontal and
axis gives the value of the major consolidating stress σ1 . The unconfined
 A − A 2 sin 2 φ − τ 2SS cos 2 φ  c
yield strength fC is determined by drawing Mohr’s semicircle tangent to the
yield locus and passing through the origin . The point of intersection of this
σ1 =   (1 + sin φ) − (21-44)
 cos 2 φ  tan φ
circle and the horizontal axis is the unconfined yield strength, which can
be considered the cohesive strength of the bulk solid . Note that all points respectively, where
c
A = σ SS + (21−45)
tan φ
The minor consolidation stress σ2 can be calculated from
τ 2SS
σ 2 = σ SS − (21-46)
σ1 − σ SS
Shear stress τ

Data from pre-shear steps

Finally, the effective angle of friction δ is calculated from
τss
Data from shear steps  σ − σ2 
δ = sin −1  1 (21-47)
 σ1 + σ 2 
A straight line through the origin of the σ-τ diagram, tangent to the larger
Mohr’s circle, is the effective yield locus (EYL) . The larger Mohr’s circle can be
constructed by drawing a circle having a radius R that passes through σ1 on
the horizontal axis . The radius R is given by
σss σ1 − σ 2
Normal stress σ R= (21-48)
2
FIG. 21-36 Yield locus . Construction of the effective yield locus is illustrated in Fig . 21-38 .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-21

L
EY

Shear stress τ
TYL

φ´t

YL
δ
σ

FIG. 21-38 Construction of the effective yield locus .

fC σ2 fCt σ1
Normal stress σ
Plotting values of fC versus the major consolidation stress σ1 gives the FIG. 21-40 Construction of the time yield locus .
flow function of the bulk solid . The flow function describes the relationship
between a bulk material’s unconfined yield strength and its state of consoli-
dation . Construction of the flow function from a number of yield locus mea-
surements is illustrated in Fig . 21-39 . The preshear, time consolidation, and shear steps are repeated at the same
Some bulk materials gain cohesive strength if stored at rest . Unless a bin normal stress σss for a number of normal stresses, and the time yield locus
is expected to be operated continuously, the time unconfined yield strength (TYL) is then determined by plotting the failure shear stress versus normal
of the bulk material should be measured . To conduct a time test, a sample of stress . An example of a time yield locus is given in Fig . 21-40 .
bulk material is placed inside a cell and presheared using a normal stress σss To calculate the time unconfined yield strength, Mohr’s circle is drawn
used during instantaneous testing . After preshear, the sample is then kept through the origin and tangent to the time yield locus . The point of intersec-
consolidated at that state of stress, typically by applying a vertically acting tion with the horizontal axis is the material’s time unconfined yield strength
load equal to the major consolidation stress σ1 associated with the corre- fCt . This value, along with the value of the major consolidation stress for
sponding instantaneous test . After the time of interest has passed (e .g ., 2 to instantaneous flow σ1, becomes one point on the time flow function .
3 days if the bulk material is to be stored at rest over a weekend), the vertical The time angle of internal friction φt is the angle formed between a hori-
compacting load is replaced with a smaller load, and the shear step is con- zontal line and a line drawn tangent to the smaller Mohr’s circle at its inter-
ducted, in which the shearing force again is applied until the sample fails . section with the time yield locus (see Fig . 21-40) .
By measuring time yield loci using other normal stress levels and cor-
responding major consolidation stresses, the time flow function can be
determined by plotting the time unconfined yield strength fCt versus major
consolidation stress σ1 . If a bulk material gains strength when stored at rest
in a bin over time, its time flow function will lie above its instantaneous flow
function, as illustrated in Fig . 21-41 .
Because of the obvious time-consuming nature of a time consolidation
test, extra test cells should be used so that the sample after preshear can be
kept consolidated outside the shear tester .
Shear stress τ

Bulk Density A method to measure the bulk density of a material as

a function of compressive stress (i .e ., pressure) is given in ASTM D6683 . A
sample is placed in a cylinder of known volume, and its mass is recorded .
A lid with a known weight is placed on the specimen, and the displacement
is noted, allowing an updated volume to be calculated . The consolidation
pressure is equal to the weight placed on the sample, divided by the cross-
sectional area of the cylinder . The bulk density is equal to the mass of sam-
ple, divided by the volume . Increasing loads are placed on the lid, and the
displacement is recorded for each load . From the data, the bulk density as
a function of consolidation pressure is determined . A typical bulk density
fC1 fC2 fC3 σ11 σ12 σ13 curve is shown in Fig . 21-42 .
Normal stress σ
Unconfined yield strength fC

Unconfined yield strength fC or fCt

fC
3
fC2
fC1

σ11 σ12 σ13

Major consolidation stress σ1 Major consolidation stress σ

FIG. 21-39 Construction of flow function from yield loci . FIG. 21-41 Instantaneous and time flow functions .
21-22 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Normal
force

Cover
Shearing
Bulk density ρb

force
Bulk Solid

Coupon of
wall material

FIG. 21-43 Wall friction test equipment .

where the wall friction coupon is placed above the test cell and those, such
Normal stress σ as the Jenike shear cell, where the wall friction coupon is located below . The
direction of grain orientation of a wall surface can have a significant effect
FIG. 21-42 Typical bulk density—consolidation stress relationship . on the wall friction angle . This effect can be captured in the Jenike shear cell,
but not in rotational or annular shear cells .
After a number of steady shear load values have been recorded, the
instantaneous wall yield locus (WYL) is constructed by plotting shear stress
The relationship between bulk density and consolidation stress is non- versus normal stress . The angle of wall friction φ′ is the angle that is formed
linear . The bulk density increases with increasing consolidation pressure, when a line is drawn from the origin to a point on the wall yield locus . A typi-
varying rapidly at low stress and less so at high stress . Data can be fit to a cal wall yield locus is shown in Fig . 21-44 .
number of equations that describe the relationship between bulk density The wall yield locus is typically concave downward . In addition, the wall
and consolidation pressure . Jenike and Johanson [Powder Technol. 5: 133 yield locus does not necessarily intersect the origin, as many bulk materials
(1971/72)] assumed a power law relationship: adhere to a wall surface in the absence of a normal stress . As a consequence,
φ′ is usually higher at lower applied stresses . This is important in the design
β
σ of hoppers, since for mass flow, the stresses at the hopper outlet are low and
ρb = ρb0   (21-49) the angle of wall friction is therefore usually higher near the outlet . Only
 σ0 
when the yield locus is a straight line that passes through the origin is φ′
where σ is the consolidation pressure, σ0 is an arbitrarily chosen reference constant .
consolidation level, ρb0 is the bulk density at that consolidation, and β is To measure the static friction between a wall surface and a bulk solid after
called the compressibility . A limitation of the model is that at zero stress, storage at rest, wall friction time tests are performed . A sample is sheared
Eq . (21-49) gives a bulk density equal to zero . Hence, at a low consolidation under a normal load until a steady shear load is observed . The normal load
is then reduced by 10 to 20 percent, and shearing is continued until steady
pressure, the bulk density is limited to no less than ρbmin, the loose-fill bulk
state is again reached . This step is analogous to the procedure of obtaining
density .
Ideally, a model that describes the bulk density–consolidation pressure rela- a point on the instantaneous wall yield locus .
tionship should give a bulk density of ρmin at zero consolidation . Some relation- The shear is then reduced to zero and the sample is stored in the cell
for the required period of time . After the storage time, the sample is again
ships that meet this criterion are [Gu et al ., Powder Technol. 72: 39 (1992)]
sheared, and the maximum shear load is reported .
The pair of normal stress and maximum shear stress values provide one
ρb = ρb min(1 + ασ)β (21-50) point on the time wall yield locus (TWYL). Repeating the test at different
normal loads completes the time wall yield locus . The time angle of wall
ρb = ρb min + ασβ (21-51) friction is the angle obtained by drawing a line from the time wall yield locus
ρb = ρb min σ=0 to the origin (see Fig . 21-45) .
Changes in both the bulk solid’s properties ( for example, due to changes
(ρb 0 − ρb min )σ in moisture content, temperature, etc .) and the wall surface ( for example,
ρb = ρb min + 0 < σ < σ0 (21-52)
ρb 0 due to surface finish, corrosion, or abrasion) affect the wall friction angle .
It is important that one not assume that a “smooth” surface, i .e ., one with
ρb = ασ β σ ≥ σ0 a low Ra (average surface roughness) value, will necessarily be low in wall
friction angle . Sometimes a surface with small asperities (which result in a
ρb = ρb max − (ρb max − ρb min ) e−ασ (21-53) larger Ra value) will be less frictional than a smoother surface .
1 + ασ
ρb = ρb minρb max (21-54)
ρb max + ρb min ασ

where α and β are empirical constants and ρbmax is the material’s maximum
bulk density .
Wall Friction The flow pattern inside a bin depends on the friction
between the bulk solid and the wall material . Therefore, measuring wall
Shear stress τ

friction is one of the necessary steps to design reliable bulk solids storage
vessels .
To measure the friction between a bulk solid and a wall material, the
method described in ASTM D-6128 is usually followed . The test is conducted
using a direct translation shear tester . A sample of bulk solid is placed inside
a retaining ring on a flat coupon of wall material (see Fig . 21-43), and a load
is then applied normal (i .e ., perpendicular) to the bulk solid . The ring and
bulk solid in the ring are forced to slide along the stationary wall material,
and the resulting steady shear force is measured as a function of the applied
normal load . The normal load is then reduced, and the test is continued φ´
until a new steady shear load is measured . The test is repeated for various
normal loads .
In most industrial applications, the wall surface upon which a bulk solid Normal stress σ
is sliding is below the bulk solid . Due to the segregating nature of some bulk
solids, differences in wall friction angles may be observed between testers FIG. 21-44 Wall yield locus .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-23

control and measure the relative velocity between the bulk solid and wear
surface as well as the solids stress exerted on the wall material . Any changes
in the wall friction angle can be determined by running a wall friction test
before and after a certain number of cycles of the wear test .
A description of an abrasive wear tester is given in U .S . Patent No .
4,446,717 . A sample of bulk solid is conveyed to a circular disc fabricated
Shear stress τ

from the wear material, and a supporting frame applies and measures
torque, thrust, and the number of revolutions . Tests are conducted over a
range of expected solids stresses . Wear is measured as the weight loss of the
sample disc after a certain number of revolutions of the disc .
It is convenient to define a wear ratio WR as the dimensionless ratio of
unit thickness of material lost per unit distance travel of the bulk solid slid-
ing relative to the wear surface .
The solids velocity at the hopper wall can be calculated using a method
φ t´ described by Johanson and Royal [Bulk Solids Handling 2: 3, 517 (Sept . 1992)]
φ´ and is inversely proportional to the cross-sectional area of the hopper . The
wear ratio is a function of the applied solids stress . Methods to calculate
Normal stress σ solids-induced loads were described previously . The wear rate is equal to
the product of the wear ratio WR and the solids velocity at the wall . Because
FIG. 21-45 Time wall yield locus . the solids velocity and stress vary inside the hopper, the degree of abrasive
wear will vary with location .

Permeability The maximum achievable steady-state discharge rate of BIN DESIGN

a bulk material from a bin depends on its permeability . The pressure drop
that arises when a gas passes through a bed of bulk solid is described by When developing the functional ∗ design of a new bin, there are a number
Darcy’s law if the gas flow is laminar: of factors that determine what type of bin is required . These factors include
the cohesiveness of the bulk solid, headroom or footprint constraints, segre-
K dP gation concerns, the likelihood of degradation over time (e .g., caking, spoil-
us = − (21-55) age, spontaneous combustion), the location of the outlet relative to bin’s
ρb g dz
centerline, and discharge rate requirements .
where us is the superficial gas velocity, K is the bulk material’s permeabil- In general, for a given volume, mass flow hoppers and silos are taller than
ity, and dP/dz is the gas pressure gradient . Permeability is determined by those designed for funnel flow . If there are headroom restrictions, designing
measuring the pressure drop and gas flow rate, using the apparatus shown a mass flow bin with the desired capacity may be challenging . Whenever
in Fig . 21-46 . this is the case, an engineer should confirm that the constraints are neces-
The permeability of a bulk solid is highly dependent on its bulk density . sary or consider whether a funnel flow or expanded flow bin will suffice .
The relationship between permeability and bulk density can be determined Mass Flow Hopper Angle Flow problems can often be prevented by
by filling the tester with compacted material . Often the relationship between ensuring that a mass flow pattern will develop in the vessel . The first step in
permeability and bulk density is described by a power-law relationship, e .g ., achieving mass flow is to ensure that the converging walls are steep enough
and have friction low enough to allow the bulk material to slide along them .
ρ
−α This is accomplished by first testing the material to measure wall friction
K = K 0  b  (21-56) and then calculating the minimum hopper angle that will allow mass flow .
 ρb 0  By assuming a radial stress field, Jenike (Gravity Flow of Bulk Solids, Bulle-
tin 108, University of Utah, 1961) was able to calculate stresses in the region
where α is a constant, ρb0 is a reference bulk density, and K0 is the perme- of the hopper outlet as a function of the effective angle of friction δ, hopper
ability of the bulk solid measured at the reference bulk density . angle ( from vertical) θ′, and wall friction angle φ′ . Jenike determined that
Abrasive Wear Abrasive wear of equipment is a problem that should when the boundary conditions are not compatible with the radial stress
not be overlooked . It is important that solids contact surfaces be durable, equations, mass flow cannot occur in a hopper, and a funnel flow pattern
thick enough to provide reasonable wear life, and, if necessary, replaceable results .
to minimize maintenance . Design charts originally developed by Jenike in 1961 provide allowable
Prediction of the wear rate can be accomplished in a two-step process . hopper angles for mass flow, given values of wall friction angle and the effec-
The first step is to determine the bulk solids velocity and compressive stress tive angle of friction . These charts are summarized in Figs . 21-47 and 21-48
acting on the wear surface . The second step is to determine the rate of wear for conical (or pyramidal) and planar hoppers (e .g ., wedge-shaped hoppers
of the wear surface as a function of bulk solids stress, using a suitable wear and transition hoppers), respectively . The outlet of a wedge-shaped or tran-
testing apparatus . sition hopper must be at least 2 times as long as it is wide for Fig . 21-48
A suitable wear tester is one that continuously introduces a fresh sup- to apply if it has vertical end walls and 3 times as long if its end walls are
ply of a bulk solid sample to a wear surface . The tester should continuously converging .
Values of the allowable hopper angle for mass flow θ′ (measured from
vertical) are on the abscissa, and values of the wall friction angle φ′ are on
the ordinate . Any combination of φ′ and θ′ that falls within the mass flow
region of the chart will provide mass flow .
An analytical description of the theoretical boundary between mass flow
and funnel flow regions for conical hoppers is given in Enstadt [Chem. Eng.
Sci. 30: 1273 (1975)]; an explicit equation that gives recommended mass
flow angles for hoppers with slotted outlets is provided by Arnold et al .
(Bulk Solids: Storage, Flow, and Handling, TUNRA Publications, 1980) .
Differential Hoppers with round or square outlets should not be designed at the theo-
pressure retical mass flow hopper angle value . Otherwise, a small change in the bulk
transducer material’s flow properties may cause the flow pattern inside the hopper to
Bulk powder change from mass flow to funnel flow, with its associated risk of flow prob-
lems . A 3° to 5° margin of safety with respect to the mass flow hopper angle
given in Fig . 21-47 is therefore recommended .
Sloping walls required for mass flow in wedge-shaped hoppers can be
10° to 12° less steep than those required to ensure mass flow in conical or
Flow meter
∗A functional design includes hopper shape, hopper angles, outlet details, overall
Gas configuration, insert size and location (if one is needed), interior surface specifications
for flow, liner material and installation considerations, details to avoid flow impediments,
FIG. 21-46 Permeability tester . and considerations for proper gas flow where important .
21-24 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

50

δ = 50° δ = 60°
40
δ = 40°
Wall friction angle (°)

30
δ = 30°
End wall
20

10
δ = 30° Side wall
δ = 60° FIG. 21-49 Side and end walls of transition hopper .
0
0 10 20 30 40 50 60

δ = 50° δ = 40° Mass Flow Outlet Dimensions to Prevent Arching The outlet of
Hopper angle from vertical (°)
the hopper must be large enough to prevent stable obstructions to flow
FIG. 21-47 Theoretical mass flow hopper angles for hoppers with round or square from developing . The required outlet size depends on the unconfined yield
outlets . Note: a minimum safety factor of 3 to 5∞ should be used. strength, the effective angle of friction, the bulk density of the bulk solid, and
the flow pattern .
For an obstruction to flow to develop, the cohesive strength that the bulk
pyramidal hoppers . In fact, hoppers with angles less steep than those given in solid gains as a result of its consolidation in the hopper must be able to sup-
Fig . 21-48 may still allow flow along the walls . Planar flow hoppers are there- port the weight of the obstruction . Jenike’s flow–no flow postulate is as fol-
fore highly suitable for materials that have high values of wall friction angle . lows [ Jenike, Storage and Flow of Solids, Bulletin 123, University of Utah, 1964
As illustrated in Fig . 21-49, transition hoppers have both straight sides (revised, 1980)]: “Gravity flow of a solid in a channel will take place provided
(side walls) and round sides (end walls) . The appropriate chart or equation the yield strength which the solid develops as a result of the action of the
must be used in specifying the angles of the end walls (Fig . 21-47) and side consolidating pressure is insufficient to support an obstruction to flow .”
walls (Fig . 21-48) when designing a transition hopper for mass flow . In a mass flow bin, as an element of bulk material flows downward, it
Additional care must be taken when designing a pyramidal hopper for mass becomes consolidated under a major consolidation stress σ1 and develops
flow . The angles that are formed at the intersections of the sloping walls of an unconfined yield strength fc . The consolidating stress follows the Janssen
pyramidal hoppers are significantly less steep than those of the hopper walls equation in the vertical section of the bin, sharply changes at the cylinder-
themselves . The valley angle from vertical θv can be calculated from hopper junction, and then decreases toward the outlet .
Jenike (Gravity Flow of Bulk Solids, Bulletin 108, University of Utah, 1961)
calculated the stress on the abutment of a cohesive arch over the outlet σ as
θv = tan −1 tan 2 θside + tan 2 θend (21-57)
ρb gB
σ= (21-58)
where θside and θend are the side and end wall angles from vertical, respectively . H (θ′)
Side, end, and valley angles are defined in Fig . 21-50 .
where B is the diameter of the outlet of a conical hopper or the width of
the slotted outlet of a planar hopper and the function H(θ′) given by Jenike
is shown in Fig . 21-51; H(θ′) can be calculated from [Arnold and McLean,
50 Powder Technol. 13: 255 (1976)]
δ = 60° 130° + θ′
δ = 50° H (θ′) = (21-59)
65°
40
for round outlets and
δ = 40° 200° + θ′
H (θ′) = (21-60)
Wall friction angle (°)

200°
30
for slotted outlets .
δ = 30° The stress and strength profiles inside a bin are shown in Fig . 21-52 . Note
that there is a critical outlet size where the stress on the abutments of a
20

10
δ = 30°
θside
δ = 60° θv
0 θend
0 10 20 30 40 50 60 70
Hopper angle from vertical (°) δ = 40°
δ = 50°

FIG. 21-48 Recommended mass flow hopper angles for wedge-shaped hoppers . FIG. 21-50 Side, end, and valley angles of pyramidal hoppers .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-25

3.0 At the hopper outlet,

Bρb gs(θ′)(1 + sin δ)

σ1 = (21-63)
2sin θ′
2.5
Jenike (1961) defined the ratio of the major consolidation stress to the arch
support stress as the flow factor ff, that is,
Conical
σ1
2.0 ff = (21-64)
σ
H(θ′)

Hence, the flow factor is given by

1.5 H (θ′) s (θ′)(1 + sin d)

ff = (21-65)
2sin θ′

The flow factor is a function of the hopper angle θ′, angle of wall friction φ′,
Planar
1.0 and the effective angle of friction δ . The latter depends on the major consoli-
dation stress σ1 at the hopper outlet . The angle of wall friction depends on
the stress normal to the hopper wall σ′, which is not equal to σ1 .
Charts that provide flow factors for conical and planar flow hoppers
based on Jenike’s solutions to the stress function [ Jenike, Storage and Flow
0.5 of Solids, Bulletin 123, University of Utah, 1964 (revised, 1980)] are given in
0 10 20 30 40 50 Figs . 21-53 through 21-60 . Explicit expressions for the flow factor from an
Hopper angle θ′ (deg) analytical form of the stress function were derived by Arnold and McLean
[Powder Technol. 13: 255 (1976); Powder Technol . 72: 121 (1992)] .
FIG. 21-51 Function H(θ′) . Superimposing the material’s flow function and flow factor on the same
graph allows the unconfined yield strength and arch stress to be compared .
The flow factor is constructed by drawing a line having a slope equal to 1/ff
through the origin .
cohesive arch is equal to the unconfined yield strength of the bulk solid .
The relationship between the effective angle of friction δ and the major
This outlet dimension represents the minimum outlet size that will prevent
consolidation stress σ1 is provided by the effective yield locus . In a converg-
a stable cohesive arch from developing .
ing hopper, the stresses in the bulk solid are represented by Mohr’s circle
Jenike postulated that near the hopper outlet the stress distribution of the
that is tangent to the material’s effective yield locus . The intersections of
bulk solid could be described by a radial stress field, i .e ., the stress distribu-
Mohr’s circle and the horizontal axis gives the principal stresses . In mass
tion could be approximated by a straight line through the hopper vertex .
flow, the material is also slipping along the hopper wall, and therefore the
The average stress was modeled as
wall stress σ′ is represented by the wall yield locus . The shear and normal
stresses at the wall are therefore located at the upper intersection of the
σ avg = rρb gs(θ′) (21-61) wall yield locus and Mohr’s circle . The relationship between σ1, δ, σ′, and φ′
is illustrated in Fig . 21-61 .
where r is the radial coordinate with the origin located at the vertex of the To determine the size of the outlet required to prevent arching, the flow
hopper, σavg is the average stress, and s(θ′) is called the stress function . function and flow factor are compared . The flow factor is dependent on
Jenike (1961) developed solutions to the stress function and presented the material’s effective angle of friction δ and its angle of wall friction φ′ as
them in chart form . well as the hopper angle and geometry . The angle of wall friction is a func-
The major consolidation stress is related to the average stress by tion of the stress normal to the hopper wall σ′ . Hence, unless the angle of
wall friction and effective angle of friction are constant, calculation of the
σ1 = σ avg (1 + sin δ) (21-62) critical outlet diameter or width is iterative . The procedure is as follows
Distance from hopper outlet

fc
σ1
–
σ1

Stress < Strength

Stress > Strength

Stress or Strength

FIG. 21-52 Stress and strength profiles of mass flow hopper .

 21-26 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

35 50

30
40
ff = 1.5 ff
25 =
1.
Angle from wall friction φ´

Angle of wall friction φ´

ff = 1.6
30
20
ff =
1.3
ff =
ff 1.4
15 =
1.
8 20 ff =
1.
ff
=
2
ff = 6
10 1.8
ff =
ff
= 2
ff 2. 10
=
3
5 ff =
5 ff = 2.5
ff 3
=
ff

4
=
4

0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50
Hopper angle from vertical θ´ Hopper angle from vertical θ´
FIG. 21-53 Flow factors for conical hoppers, δ = 30° . FIG. 21-55 Flow factors for conical hoppers, δ = 50° .

[ Jenike, Storage and Flow of Solids, Bulletin 123, University of Utah, 1964 4 . The flow factor and flow function are plotted together . As shown in
(revised, 1980)]: Fig . 21-62, there are three possibilities:
1 . The effective angle of friction δ, wall friction angle φ′, and bulk density a. There is no intersection, and the flow function lies below the flow
ρb are estimated . factor . A dimension B that is the minimum that prevents cohesive arch-
2 . The hopper angle is selected, one that ensures mass flow, by using the ing cannot be determined . Instead, B is selected based on other con-
appropriate charts (Fig . 21-47 or Fig . 21-48) . Note that if a conical hopper is siderations such as discharge rate requirements, choice of feeder, or
to be specified, a safety factor of at least 3° should be used with respect to prevention of particle interlocking . The major consolidation stress σ1 is
the theoretical mass flow boundary . determined from Eq . (21-66):
3 . The flow factor ff is determined from the appropriate chart given by
Figs . 21-53 through 21-60 . ρb gB
σ1 = ff (21-66)
H (θ′)

40 50

35

40
30
Angle from wall friction φ´

ff = 1.1
Angle of wall friction φ´

ff =
25 1.4
ff = 30 ff =
1.5 1.2
20 ff ff =
= 1.
1.
ff = 6 ff = 3
1.8 20 1.4
15
ff =
ff 1.5
=
2
10 ff
=
ff 1
= 10 ff = .6
2. ff = 1
5 2 .8
5 ff =
ff
=

ff = 3
3
ff

4
=
4

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50
Hopper angle from vertical θ´ Hopper angle from vertical θ´

FIG. 21-54 Flow factors for conical hoppers, δ = 40° . FIG. 21-56 Flow factors for conical hoppers, δ = 60° .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-27

35
50 ff
=
4
30 ff
=
ff = ff = ff 3
2.5 =
2.
3 ff
5
40 ff = 2
25 =
ff 1.8

Angle of wall friction φ´

=
Angle of wall friction φ´

1.
ff = 1.6 ff 6
=
1.
ff 4
=
20 30 1.
3
ff = 1.8
ff =
2 ff
15 =
1.
2
ff = 20
2.5 ff
=
10 ff 1.
= 3
1.
ff = ff 4
=
3 ff 1.
10 = 6
ff 1.8
5 ff = ff = 2
=
4 ff 2.5
ff = 3
=
4
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Hopper angle from vertical θ´ Hopper angle from vertical θ´

FIG. 21-57 Flow factors for planar flow hoppers with slotted outlets, δ = 30° . FIG. 21-59 Flow factors for planar flow hoppers with slotted outlets, δ = 50° .

b. The flow factor and flow function intersect . The minimum outlet 5 . The value of φ′ at the outlet is checked . The effective angle of friction is
dimension Bmin is calculated using Eq . (21-67): determined from a plot of δ versus σ1, and the effective yield locus is drawn
by drawing a straight line through the origin at an angle equal to δ . Mohr’s
H (θ′)σ crit
Bmin = (21-67) circle is drawn through σ1 that is tangent to the effective yield locus . The
ρb g value of φ′ is found from the intersection of Mohr’s circle and the wall yield
locus, as shown in Fig . 21-61 .
Larger outlet diameters or widths of course can be used, and they are
6 . The recommended hopper angle θ′ is updated based on the new value
generally selected by considering standard feeder sizes or discharge rate
of φ′ . The steps are repeated until convergence is reached .
requirements .
To prevent mechanical interlocking, the following rules of thumbs are
c. There is no intersection and the flow function lies above the flow
used: for a conical hopper, the outlet diameter should be at least 6 to 8 times
factor . Gravity flow will no longer be possible in a hopper with converging
the size of the largest particle that will be handled; for hoppers with slotted
walls . Consideration should be given to changing the flow properties of
outlets, the outlet width should be at least 3 to 4 times the largest par-
the material, such as increasing its particle size, reducing its moisture
ticle size .
content, or using a flow aid .

60 ff =
40 ff = 5
ff = ff = ff = 4
3
2.5 4
ff = 50 ff
2 ff = 2 ff =
= 2.
ff = 1.3 ff 1.8 5
=
Angle of wall friction φ´

Angle of wall friction φ´

30 ff ff .6
1
= =
1. 40 1.
4
8 ff
=
1
ff .3
ff

ff = =
=

1.4 1.
2
1.

ff
30 =
5

20 1.
1

ff =
ff = 1.6 20
1.8 ff
=
1.
10 ff ff
= 2
= 1
ff 2 ff
=
.3
= 10 ff
1.
4
2. =
5 ff 1.6
ff f =
ff = = 3 f f 1
ff ff = f = = 2 .8
= 3 2.
4 4 5
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Hopper angle from vertical θ´ Hopper angle from vertical θ´

FIG. 21-58 Flow factors for planar flow hoppers with slotted outlets, δ = 40° . FIG. 21-60 Flow factors for planar flow hoppers with slotted outlets, δ = 60° .
21-28 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

High permeability
Low permeability

EYL
Shear stress τ

WYL

δ
φ´
σ’ σ1
Normal stress σ
Consolidating Bulk Gas
FIG. 21-61 Construction of effective yield locus and wall yield locus . pressure density pressure

FIG. 21-63 Consolidating pressure, bulk density, and gas pressure profiles for coarse
(high permeability) and fine (low permeability) powders .
If a bulk solid is to be stored at rest in a bin, the flow function and wall
yield locus must be based on time tests . The intersection of the time flow
function and flow factor is used to determine the critical stress and hence For fine powders, the discharge rate from a mass flow hopper is given by
the minimum outlet size .
Mass Flow Discharge Rates While an outlet diameter greater than Bg  1 dP 
the minimum will prevent cohesive arching, it may not necessarily be large m S = ρbO AO 1+ (21-69)
enough to allow the desired discharge rate . From a force balance together 2(m + 1)tan θ′  ρb g dz O 
with continuity of the solids, the following equation is derived for the dis-
charge rate of coarse powders from mass flow hoppers: Because the gas pressure gradient at the outlet (dP/dz)|O is often less than
zero for fine powders, Eq . (21-69) shows they can have discharge rates dra-
matically lower than those of coarse powders .
Bg
m S = ρbO AO (21-68) The gas pressure gradient is related to the material’s permeability and the
2(m + 1) tan θ′ rate of air counterflow by Darcy’s law [Eq . (21-55)] . Gu et al . [Powder Tech-
nol. 72: 39 (1992)] derived a relationship between the air and solids flow
where m s is the solids discharge rate, AO is the cross-sectional area of the rates that when combined with Darcy’s law gives
outlet, ρbO is the bulk density of the material at the hopper outlet, m is equal
to 0 for conical hoppers and equal to 1 for hoppers with straight walls and m ρ g 1
= S bO 
dP 1 
− (21-70)
slotted outlets, and B is the diameter of the outlet of a conical hopper or the dz O K O AO  ρba ρbO 
width of a slotted outlet beneath a planar flow hopper .
The maximum flow rate of a fine powder can be several orders of magni- where KO is the permeability of the powder at the hopper outlet, ρbO is its bulk
tude lower than that of coarser materials . Two-phase flow effects are signifi- density at the outlet, and ρba is the bulk density at a location inside the hopper
cant due to the movement of interstitial gas as the powder compresses and where the gas pressure gradient is equal to zero . Calculating this value is chal-
expands during flow . Figure 21-63 illustrates solids and gas pressure profiles lenging, and therefore an estimate that is proportional to the stress given by
in bins for coarse (high-permeability) and fine (low-permeability) powders . the Janssen equation is frequently used . Note that the calculated limiting mass
In the cylindrical portion of a bin, the stress level increases with depth, flow discharge rate can be very sensitive to the choice of ρba .
causing the bulk density of the material to increase and the void fraction The stress balance used in the analysis does not account for the cohesive-
to decrease, squeezing out a portion of the interstitial gas . This gas leaves ness of the powder, and hence the expressions are only valid for outlet sizes
the top free surface of the bulk material . In the converging section of the that are significantly larger than the critical arching dimensions . Jenike &
vessel, the consolidated material expands as it moves toward the outlet, Johanson, Inc . (Tyngsboro, Mass ., United States) uses a proprietary computer
reducing the material’s bulk density and increasing its void fraction . This model based on unconfined yield strength, permeability, and compressibility
expansion results in a reduction in interstitial gas pressure to below atmo- data to calculate discharge rates from mass flow bins .
spheric (i .e ., vacuum), causing gas counterflow through the outlet if the gas To increase the flow rate of fine powders, injection of a small amount of
pressure below the outlet is atmospheric . At a critical solids discharge rate, air above the hopper outlet is often effective, as it will eliminate the oppos-
the solids stress drops to zero, and efforts to exceed this limiting discharge ing air pressure gradient if injected at the correct rate and at the proper
rate will result in erratic flow . location .
Funnel Flow Outlet Size to Prevent Arching and Ratholing For
funnel flow hoppers, the outlet must be large enough to prevent both a cohe-
sive arch and stable rathole from developing . The critical rathole diameter is
Flow function lies above flow factor calculated by first determining the major consolidating pressure σ1 on the
bulk solid . The consolidating load can be estimated by the Janssen equation
Unconfined yield strength fc

Flow factor ρb gRH   − k (tan φ′) h  

σ1 = 1 − exp  (21-71)
slope = 1/ff k tan φ′   RH  


where RH is the hydraulic radius of the vertical section of the hopper, h is its
height, and k is the Janssen coefficient .
Flow function intersects flow factor Jenike (Gravity Flow of Bulk Solids, Bulletin 108, University of Utah, 1961)
calculated the stress on a rathole as
Flow function lies below flow factor ρb gD
σ1 = (21-72)
G(φt )
Major consolidation stress σ1
where D is the diameter of a round outlet or the diagonal of a slotted out-
FIG. 21-62 Plot showing both flow factor and flow function . let and G(φt) is a function given by Jenike, which is plotted in Fig . 21-64 .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-29

12 not occur . When funnel flow hoppers with elongated outlets are designed,
prevention of arching must also be considered, i .e ., the width of the slotted
outlet must be large enough to prevent a cohesive arch from developing .
The same procedure that is used to determine the minimum outlet width to
prevent arching in a planar flow mass flow hopper is followed, except that a
10 flow factor of 1 .7 is used .
Expanded Flow Hopper Dimensions An expanded flow hopper is essen-
tially a funnel flow hopper above a mass flow hopper . The upper diameter of the
mass flow section must be larger than the critical rathole diameter DF , while its
8 outlet size must be larger than the critical arching dimension . An example of an
expanded flow hopper is shown in Fig . 21-66 .
G (φt)

The main advantage of expanded flow compared to complete mass flow

is that the overall bin height is less while at the same time a stable rathole
cannot develop .
6 Capacity The mass of bulk material that must be stored is generally
based on production rate, frequency of operation, required inventory, and,
in some cases, residence time . Once the required mass has been determined,
the necessary volumetric capacity can be determined . Because bulk solids
are generally compressible, a suitable average bulk density should be used
4 for converting mass to volume . To be conservative, a low bulk density value
should be chosen to ensure that the volume of the bin is sufficient . Also note
that the working capacity of a bin will be less than the actual volume since a
pile will form when the bin is filled .
2 A reasonable height-to-diameter ratio (H/D) of the cylinder section
30 40 50 60 70 should be used, with ratios between about 1 .5 to 4 usually being the most
Static angle of internal friction φt (°) economical . Height may be limited because of building constraints, zoning
considerations, or constraints imposed by other structures or equipment .
The volume V and height H of some common hopper designs are given
FIG. 21-64 Function G(φt) .
in Fig . 21-67 .
The location and the number of inlets to a bin are not as important
as the number and location of its outlets . In general, the use of a single
Using Jenike’s flow–no flow postulate, the rathole will collapse provided that outlet is preferable . If multiple outlets are required, the best option is to
the flow channel stress is greater than the cohesive strength of the bulk solid use a single, centered outlet beneath the hopper section and then split the
that makes up the rathole . The critical rathole diameter DF can therefore be stream beneath it .
calculated as If multiple outlets are to be located below a circular cylinder, each outlet
should be located equidistant from the bin centerline . Examples of proper
G ( φt ) f C and improper multiple outlet designs are shown in Fig . 21-68 . If a bin is center-
DF = (21-73) filled, the outlet arrangement on the right is preferable to that on the left
ρb g because similar material discharges from each outlet provided that all outlets
are fully open and material is discharging at approximately the same rate
where fC is the unconfined yield strength of the bulk solid at the consolida- from each one .
tion pressure given by the Janssen equation . Useful Guidelines The quality of construction of a bin—particularly
A conical funnel flow hopper with an outlet diameter smaller than DF or a its interior surface—is critical to its ability to function as desired . If hori-
planar funnel flow hopper with an outlet whose diagonal is less than DF will zontal welds, incorrectly lapped liner plates, or poorly constructed mating
not empty completely . This is illustrated in Fig . 21-65 . Because the major flanges exist in a bin designed for mass flow, it may discharge in a funnel
consolidation stress is higher in the lower part of the bin, the unconfined flow pattern . The lower of two mating flanges should be oversized to pre-
yield strength of the bulk solid will be correspondingly higher . As material vent any protrusions into the flowing solid . All flanges should be attached
discharges in a funnel flow pattern, ratholes that form in the upper part of to the outside of the hopper, with the hopper wall material being the surface
the vessel may continually collapse, provided that the stress on the stagnant in contact with the flowing solids .
material is greater than its cohesive strength . However, if the size of the out- A feeder, slide gate, or both may be used below the hopper outlet . If a gate
let is smaller than the critical rathole diameter, a level will be reached where is used below a mass flow hopper, the gate must be either fully open or fully
the ratholes will no longer fail . closed . A partially opened gate creates a flow obstruction and will convert
If a hopper with a square or round outlet is designed with an opening large what would otherwise be a mass flow design into funnel flow . A gate should
enough to prevent development of a stable rathole, cohesive arching will never be used to modulate flow in a mass flow bin .

Upper part empties Stress < strength

Stable ratholes form Stress > strength

σ1

–
σ1 fc

Stress or strength
FIG. 21-65 Formation of a stable rathole in a funnel flow hopper .
21-30 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

When using a feeder under a slotted outlet, its capacity must increase
along the outlet length in the discharge direction . This is generally accom-
plished using a specially designed screw or belt . When a screw feeder is
used, the screw should be comprised of a section having a tapered shaft
and a constant shaft diameter, increasing pitch section . A properly designed
interface between a slotted outlet of a hopper and a belt will progressively
discharge more material onto the belt along its length .
A rotary valve is often used as a feeder under a conical hopper outlet to
Funnel flow hopper control flow . Common problems experienced with such use of a rotary valve
Diameter > DF
are arching of material over the valve opening, erratic discharge, and prefer-
Mass flow hopper ential flow from one side of the hopper into one side of the valve . Most rotary
valves have a transitioning throat area, which is significantly smaller than
the nominal size of the valve and may be too small to prevent arching . In
FIG. 21-66 Expanded flow hopper . addition, this transition section may have a rougher than expected surface,

Conical

D–B
H=
2tan α
D
π (D3 – B3)
V=
24 tan α
α

!""
B

Pyramidal

Valley

A–a B–b
H= =
2 tanαside 2 tan αend
B( A
H[2(AB + ab) + Ab + aB]
V=
6

αside
αend

b a!

Transition

D–B D–L
H= =
2 tan αside 2 tanαend
D
"
πD2 BL D(B + 2L)
V≈ + + H
12 3 12

αside αend

B
! L(

FIG. 21-67 Hopper capacities .

 BULK SOLIDS FLOW AND HOPPER DESIGN 21-31

Critical angle
(a) (b) for mass flow
FIG. 21-68 Multiple outlets below a circular cylinder; (a) poor design; (b) recom- 2 × Critical angle
mended design . Insert for mass flow

preventing flow along it . Preferential flow can be overcome by having a ver-

tical section with a length greater than its diameter . FIG. 21-69 Inserts: cone-in-cone (left) and bullet (right) .
Gas leakage through the rotary valve often restricts the material dis-
charge rate, thereby lowering the efficiency of the valve, and it can even
cause arching and erratic discharge . Proper venting of rotary valves is usually
Air (or nitrogen) cannons operate differently in that they rely on a pres-
important, but it is critical if a fine powder is being handled .
sure wave to provide the stress required to break an arch . Cannons work
Handling an abrasive bulk solid in a mass flow bin may result in sig- by releasing a volume of high-pressure gas into the bin . The required size,
nificant abrasive wear of the wall material . Generally, the hopper surface
number, and location of the cannons depend on the cohesive strength of the
becomes smoother with wear; however, occasionally the wall becomes
bulk material and the dimensions of the bin .
rougher, which may upset mass flow . Wear is a function of the bulk solid Air cannons are best used for reinitiating flow after a cohesive arch devel-
flow properties, wall surface, solids pressure, and velocity . The life of a given
ops when the material is stored at rest . If flow problems occur without storage
wall material can be estimated by conducting wear tests in which a screw
at rest, air cannons are unlikely to be a viable solution, as they must be fired
continuously feeds a fresh sample of bulk solid to a rotating coupon of wall
repeatedly to maintain flow . Air cannons are usually not effective in preventing
material .
flow problems in funnel flow bins since ratholes are inherently stable .
The material of construction of most bins is metal, with reinforced con-
Chemical flow aids are often used to prevent arching or the formation of
crete the most common alternative . When because of its size the bin must a stable rathole . Parting agents such as silicates, stearates, and phosphates
be erected in the field, reinforced concrete should be considered . A double
are effective as they increase the distance between adjacent particles,
layer of reinforcing steel (rebar) is likely required, especially if eccentric
thereby reducing the magnitude of their cohesive forces . Note that while a
loads are possible . Reinforced concrete is also preferred over metal if abra- flow aid may be effective in reducing a bulk solid’s cohesive strength, the
sion or corrosion is a concern . additive may increase wall friction, potentially resulting in flow problems
associated with funnel flow .
Inserts A disadvantage of mass flow bins is that relatively steep hopper
FLOW AIDS sections are generally required, so in some cases bins may be too tall for
Flow aids are mechanical or pneumatic devices or chemical additives used the available space . When properly designed, an insert can be used to allow
mass flow in a bin with shallow hopper walls that, without modifications,
to induce bulk solids to flow more easily . Vibrators and air cannons are two
would discharge in a funnel flow pattern .
examples of common mechanical and pneumatic flow aids . Common chem-
ical additives include silicates, stearates, and phosphates . Cone-in-cone and bullet designs are shown in Fig . 21-69 . The cone-in-cone
insert, or Binsert, is designed to allow mass flow through the inner cone and
Vibrators impart forces to the bulk solid through the walls of the stor-
also through the annular space between the inner and outer cones . The angle
age vessel, most frequently on the hopper walls . Some vibrators produce
low-frequency, high-amplitude forces, while others deliver high-frequency, of the inner cone is equal to or steeper than the hopper angle recommended
for mass flow in a conical hopper, and the angle of the outer cone is equal to
low-amplitude forces . Their effect on flow obstructions can vary . In some
double that of the inner cone . The outlet diameter of the inner cone must be
cases, they may be an effective means of restoring flow when a bin becomes
plugged . In many cases, however, their effect is minimal or can even exac- greater than the critical arching diameter . For cohesive materials that would
otherwise arch over the outlet of the inner cone of an insert, an inverted cone
erbate flow problems . Applying sufficient but not excessive force where it is
or “bullet” can be placed above the inner cone .
required to collapse an arch or rathole is difficult, particularly in the case of
a rathole where the force must usually be transmitted through a significant Note that the location of the inner cone and, when applicable, the size of
the bullet greatly influence the effectiveness of an insert . An insert that is
mass of material to reach it .
too high or too low relative to its optimum size and position will expand the
The force required to overcome a cohesive arch depends on the bulk
solid’s cohesive strength and the size of the outlet . If the hopper outlet is flow channel very little, if at all . Also note that the loads acting on inserts are
usually very high, and therefore the supports required to resist these loads
slightly undersized, that is, its diameter or width is only marginally smaller
than its critical arching dimension after storage at rest, a vibrator may be may be large and can impede flow .
able to provide enough energy to restart flow . Because ratholes are inher- Air Assist and Fluidization Air pads and air nozzles are sometimes
ently stable, the outlet diameter required to prevent a rathole from forming used to inject low-pressure air into a bin . While they are generally ineffective
can be several times the outlet size of a bin; thus vibrators in general cannot in correcting problems caused by arching or ratholing, they may be useful
in increasing the discharge rate of fine powders by reducing the adverse
be used to overcome ratholing .
pressure gradient that develops above a hopper outlet and the resulting
A steep flow function is evidence of a bulk solid that is pressure-sensitive;
counterflow of air .
i .e ., its strength increases substantially when additional stresses are applied .
Vibrating such a bulk solid often makes flow problems more severe . A better technique to increase the discharge rate of fine powders is to use
an air permeation system, which consists of a sloping shelf or insert through
A vibrating discharger or bin activator employs an inverted cone or dish
which a relatively low flow rate of air is introduced . The air reduces or elimi-
that moves in a gyratory, horizontal, or vertical motion . The bulk solid then
nates the vacuum that naturally develops when a bulk solid dilates in the
flows around the cone or dish into a conical section below it, which essen-
hopper section, increasing its void fraction . To be effective, the air should
tially operates as a chute . Vibratory dischargers can be effective in overcom-
be distributed uniformly, its flow rate should be low enough to prevent
ing flow problems in some bins, provided that they are used appropriately .
fluidization, and the permeation system should not impede solids flow in
When used at the outlet of a funnel flow bin, the flow channel above it will
the hopper .
be approximately the size of the top diameter of the discharger . If this diam-
eter is smaller than the material’s critical rathole diameter, a stable rathole
will form, and the discharger will be ineffective in overcoming it . Binsert is a registered trademark of Jenike & Johanson, Inc .
21-32 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Air-assist dischargers are designed to reduce the wall friction angle to

nearly zero, thereby allowing powders to flow along hopper walls . The hopper
section either is lined with air panels or is fabricated using a permeable mg cos α tan φ′
membrane through which a small amount of air is injected . Jenike [Storage α
and Flow of Solids, Bulletin 123, University of Utah, 1964 (revised, 1980)]
recommends conical hopper angles between 40° and 50° from vertical,
as steeper hoppers may require large outlet diameters to prevent arching . mg
Shallow hopper angles can be used provided that a fully open, unrestricted
on/off valve is used; or if enough gas is added to completely fluidize the bulk
FIG. 21-71 Element of bulk solid sliding on a straight chute .
material, a fluidized discharger is used .
A fluidized discharger can be used when the bulk material is fluidizable
and a low bulk density of the discharged material is acceptable . Fluidized
dischargers can generally be used for Geldart group A, B, and C materials The coupon is inclined about a pivot point until it just starts to slide . Usually
[Powder Technol. 7(5): 285 (1973)], although group C materials may require a safety factor of 5° is applied to this minimum value to ensure clean off .
mechanical agitation . Discharge from a bin equipped with a fluidized dis- While sliding on a straight surface, the particles will accelerate or decel-
charger is typically controlled through use of a rotary valve . Note that if erate, depending on the relative values of the chute angle α measured from
the entire contents of the bin are fluidized, then potential structural issues horizontal and the wall friction angle φ′ (see Fig . 21-71):
might develop due to the hydrostatic stress field .
a = g (sin α − cos α tan φ′) (21-77)
STANDPIPES
A standpipe is frequently used to seal a bin outlet against gas pressures . It where a is the acceleration .
is composed of a vertical cylinder mounted between the bin and feeder . Its Assuming that the chute cross section does not decrease along a distance
major advantages over other methods such as sealing screws is that it has S on the chute surface, the stream velocity V is given by
no moving parts and it is inexpensive to fabricate and operate .
A standpipe provides a seal against gas pressure by providing a sufficient V = V02 + 2 aS (21-78)
length to lower the pressure gradient such that the flow of gas through the
bed of solids at the hopper outlet is reduced to an acceptable rate . Stand- where V0 is its velocity at the starting point .
pipe design is discussed by Carson and Marinelli [Power Engr. 85: 11 (1981)] . When the velocity of the stream changes as it passes through a chute,
its cross-sectional area will change . To prevent flow stoppages, the chute
CHUTES should be sized such that it is no more than about one-third full at the point
of minimum velocity .
Chutes are used to direct the flow of bulk solids . They need to be properly While chutes can be fabricated and installed in rectangular sections,
designed to avoid problems such as plugging, excessive wear, dust genera- having curved surfaces upon which the material slides is advantageous .
tion, and particle attrition . Chutes composed of cylindrical pipes or rounded surfaces control the
A chute must be sufficiently steep and low enough in friction to permit stream well, as they can be used to center the load, allowing the momen-
sliding and clean off . Referring to Fig . 21-70, the velocity of a stream of parti- tum of the material to keep the chute clean . The path that the bulk material
cles (assuming no bouncing) after impacting a chute V2 relative to its veloc- will flow depends on its frictional properties and flow rate . Discrete element
ity before impact V1 is method (DEM) models should be used to design chutes with complicated
geometries .
V2 = V1 (cos θ − sin θ tan φ′) (21-74) Free-fall height and sudden changes in the direction of material flow
should be minimized to reduce solids impact pressures, which can result
where θ is the impact angle and φ′ is the wall friction angle . in high abrasive wear, attrition, generation of dust, and potential adhesion
If the particles fall freely when they are dropped onto the chute, their of solids to the wall . Since impact pressure is proportional to sin θ and V12,
velocity before impact V1 is given by reducing the impact angle θ and drop height H will reduce wear, and the
momentum of the flowing material will keep the chute surface cleaned off .
Dust is created when air is entrained into the flowing material . To avoid
V1 = 2 gH (21-75)
dusting, the chute should be designed such that the material remains in
contact with the chute surface, the material stream is concentrated, and
where H is the drop height . the velocity through the chute is kept nearly constant . If the material is to
If the sum of φ′ and θ equals 90°, V2 will be reduced to zero, and the bulk land on a belt conveyor at the exit of the chute, the velocity of the stream
material will not slide on the chute surface unless its angle of inclination is should be in the direction of and equal to or greater than the velocity of
greater than a minimum value . To determine this minimum value required the belt .
to overcome adhesion at impact, chute tests [Bulk Solids Handling 12(3): 447 Attrition of friable particles is most likely to occur at impact points where
(1992)] can be performed . A sample of the bulk material is loaded onto a the impact pressures are high . Therefore, attrition can be minimized by
wall coupon, and a load representing the impact pressure is briefly applied . minimizing the impact angle θ, ensuring that the flowing stream is concen-
The impact pressure σ is given by trated and remains in contact with the chute surface, and maintaining a
constant stream velocity .
σ ≈ ρb gV12 sin 2 θ (21-76)
SEGREGATION
Some materials, when transferred into a bin, will segregate; that is, particles
of different size, shape, density, etc . will separate . Segregation can occur by a
V1 number of different mechanisms, depending on the physical characteristics
of the particles and the method of handling . The three most common mech-
anisms are fluidization (air entrainment), dusting (particle entrainment),
and sifting. These segregation methods are illustrated in Fig . 21-72 . They are
θ discussed in detail in the Solids Mixing subsection .
Fluidization, or air entrainment, can cause vertical segregation, i .e ., hor-
izontal layers of fines and coarse material . Fine powders generally have a
lower permeability than coarse materials and therefore retain air longer .
Thus, when a bin is being filled, the coarse particles are driven into the bed
while the fine particles remain fluidized near the surface . Air entrainment
often develops in materials that contain a significant percentage of
V2 particles below 100 mm . Fluidization segregation is also likely to occur
when a bin is filled or discharged at high rates or if gas counterflow is
present . Segregation by the fluidization segregation mechanism is illus-
FIG. 21-70 Velocity of a particle after impact on a chute . trated in Fig . 21-73 .
 BULK SOLIDS FLOW AND HOPPER DESIGN 21-33

Coarse
Coarse Dust
Fines Fines

FIG. 21-74 Particle entrainment during filling of a bin .

Fines percolate through the bed as they fall from the center and accumulate
in the middle . The result is side-to-side separation of particles by size .

Fluidization Dusting Sifting CAKING

FIG. 21-72 Segregation by fluidization, dusting, and sifting .

Caking is frequently moisture-induced . When the moisture content of a bulk
material reaches a critical value, moisture will condense primarily at the
contact points between adjacent particles, causing liquid bridges . If local
drying occurs due to temperature swings during storage or transit, solid
bridges may form when soluble components in the liquid precipitate . Water
Dusting, or particle entrainment, involves airborne particles, differences is also a plasticizer for many materials, and its presence can cause particles
in settling velocities between particles, and air currents to cause movement to deform and increase interparticle contact area . Elevated temperature
of suspended particles . Dusting can occur when powder is dropped and and impurities also frequently increase the likelihood of a material to cake .
impacts onto a pile surface, causing the release of finer particles into the Caking occurs when the magnitude of interparticle forces increases
air . Particles can also be reentrained in air if large pockets of air bubble up significantly over time . These cohesive forces are primarily van der Waals
through a stationary bed of material from below . These particles will tend forces, polar interactions, and forces associated with plastic creep or liquid
to remain suspended in the air and will be carried by air currents to the least bridges (when moisture is present) . Van der Waals forces include all inter-
active portion of the receiving vessel’s area, generally the lowest part of the molecular forces that act between electrically neutral molecules . Polar inter-
pile surface that is farthest away from the impact point . Generally, powders actions occur when adjacent particles contain regions that are permanently
that are susceptible to this mechanism contain a portion of finer particles electron-rich or electron-poor . Van der Waals forces and polar interactions
below 50 mm that do not readily adhere to larger particles . Dusting is illus- increase as the distance between particles decreases . Although these forces are
trated in Fig . 21-74 . proportional to particle size, the likelihood of caking generally decreases with
Sifting occurs when smaller particles move through a matrix of larger increasing particle size since the number of inter-particle contacts is inversely
ones . Four conditions must exist for sifting to occur: proportional to the square of the particle diameter .
• A difference in particle size between the individual components, typically With some bulk materials, plastic creep, which is the tendency of a mate-
a minimum ratio of 2:1 or greater, but has been observed at size ratios as rial to deform when under consolidation, may occur . Plastic creep can be
small as 1 .2 to 1 .3 severe if impurities that behave as plasticizers are present or if the bulk solid
• A sufficiently large mean particle size, typically one greater than approxi- is subjected to high temperatures for long periods, especially when above
mately 500 mm its glass transition temperature Tg . Differential scanning calorimetry (DSC),
• Free-flowing material thermal mechanical analysis (TMA), and inverse gas chromatography (IGC)
• Interparticle motion are frequently used to measure Tg. IGC is preferable over the other methods
All four of these conditions must exist for sifting segregation to occur . if moisture is known to act as a plasticizer since it can be conducted at a
If any one of these conditions does not exist, the mix will not segregate by constant relative humidity .
this mechanism . Liquid bridging occurs when moisture accumulates at the contact points
Sifting segregation is illustrated in Fig . 21-75, which is a photograph of a between adjacent particles . The likelihood of liquid bridging can often be
typical pile that forms when a vessel is filled . Because coarser particles tend inferred from a powder’s moisture sorption isotherm, which relates relative
to be more mobile, they roll downward toward the periphery of the pile . humidity and equilibrium moisture content . An example of an isotherm

Fines

FIG. 21-73 Segregation due to fluidization of fine, light particles . FIG. 21-75 Sifting segregation after formation of a pile .
21-34 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

An analysis can be performed to determine the moisture distribution in a

bulk solid that will result if a temperature gradient (e .g ., if product pack-out
temperatures exceed storage temperatures or if storage temperatures vary)
is imposed . One assumes that the temperature gradient remains constant .
Since this is not true, the analysis results in a conservative view of possible
conditions that can exist if temperature differences were to remain for an
extended time .
The analysis is as follows . If a bulk solid is exposed to a warm surface
Moisture content

(temperature = TH) on one side and a cool surface (temperature = TC) on the
other, the temperature profile at steady state would be given by

T = TC + (TH − TC ) z (21-79)

where z is the ratio of the distance from the cold surface to the width
between the hot and cold surfaces .
At steady state, the concentration of water in the interstitial air Cw is con-
stant . The vapor phase moisture concentration is the product of the abso-
lute humidity H and the dry air density ρa:

Cw = ρaH (21-80)

0 20 40 60 80 100 The relative humidity RH is related to absolute humidity by

Relative humidity
RH 29 HPt
FIG. 21-76 Example isotherm .
= (21-81)
100 (18 + 29 H ) Pwsat

where Pt and Pwsat are the total pressure and saturation pressure of pure
that has the sigmoidal shape characteristic of many bulk materials prone to water, respectively . Due to the temperature gradient, the relative humidity
caking is shown in Fig . 21-76 . of the interstitial air will vary . As a result, the amount of condensed moisture
The moisture isotherm is initially linear as water molecules are adsorbed that is in equilibrium with the interstitial air will also vary . The relationship
until a monolayer is formed . The effect of moisture on caking is generally between the solid’s equilibrium moisture content X and the relative humidity
negligible in this region . As relative humidity increases, multilayer adsorp- of the interstitial air is given by the material’s isotherm . Since the amount of
tion takes place as a consequence of hydrogen bonding . In this region, moisture in the gas phase is negligible compared to that in the solid phase,
the slope of the isotherm is initially shallow but steepens with increasing the total amount of moisture in the solid after migration can be assumed to
relative humidity . As moisture uptake increases, the particles become sur- be equal to the initial solid moisture content X0
rounded by moisture . If the solids are water-soluble, the layer of moisture
can be viscous, and the bulk material may become cohesive . X = ∫ X ( z )dz = X 0 (21-82)
The third region occurs at high relative humidity, where the equilibrium
z
moisture content increases dramatically . In this region, most of the incre-
mental condensation takes place at the contact points between particles . A specification for a bulk material’s moisture content that, if exceeded, causes
This phenomenon, which is known as capillary condensation, is accompa- caking (as determined from unconfined yield strength measurements) can
nied by liquid bridging, and it results in strong forces between particles . This be determined by finding the value of Cw that satisfies Eqs . (21-80) to (21-82)
leads to caking over time . If soluble matter exists in the liquid bridges, and if and the material’s moisture isotherm .
the liquid evaporates, then strong, solid bridges may form .
To quantify caking, the temperature, relative humidity, and consolidation
pressures used during the test must simulate those expected when the bulk PROCESS VESSELS
material is stored . A material’s unconfined yield strength is best measured
using a shear cell tester or uniaxial compression tester . A significant change Moving-bed process vessels are hoppers, bins, or silos modified to allow pro-
between the unconfined yield strength measured for continuous handling cessing of a bulk solid . Examples of processes include heating, cooling, con-
and after storage at rest indicates that caking has occurred . ditioning, drying, and conducting a chemical reaction . In some moving-bed
Any moisture limits to avoid caking that are based on equilibrium mois- processors, a gas is injected and passes countercurrent or perpendicular
ture content should account for the possibility of moisture migration . This to the flow of solids . Others are equipped with plate-and-frame or tubular
occurs when a temperature gradient exists during packaging, storage, or heat exchangers . Fluidization of the solids is not required, nor is it desired .
transit of powders . The mechanism of caking due to moisture migration is Rather, the bulk material flows downward as a moving bed toward the
as follows: processor’s outlet .
• The relative humidity of the interstitial air at the warm boundary To ensure trouble-free operation of moving-bed processors, the following
decreases . design criteria must be met:
• As a consequence, moisture desorbs from the warmer solids, as the solids 1 . The outlet of the processor must be large enough to prevent solids flow
and interstitial air are no longer in equilibrium . stoppages and to allow the desired production rate .
• The absolute humidity of the interstitial air increases . 2 . The solids velocity should be as close to uniform across the processor’s
• The driving force in the gas phase leads to moisture migration toward the cross section as possible and practical . At a minimum, there must not be any
interior, which has a lower absolute humidity . regions of stagnant material inside the processor .
• The relative humidity of the cooler interstitial air increases . 3 . If required, the gas must be introduced in such a manner that it does
• Moisture adsorbs onto solids in the interior in an effort to reestablish not disrupt the flow of solids .
equilibrium . 4 . The processor must provide the required residence time .
Moisture migration is illustrated in Fig . 21-77 . 5 . If applicable, a driving force for heat or mass transfer must exist
throughout the cylindrical section of the processor .
Counterflow, cross-flow, and radial-flow designs exist (see Fig . 21-78) .
In a counterflow system, the gas introduction system must be designed
Gas phase Diffusion Adsorption such that there are no regions with high gas velocities, as localized fluidi-
zation will cause channeling, bypassing of the solids, and flow instabili-
Hot Cold ties . Designs that involve nozzles or perforated plates therefore should be
boundary boundary avoided . Gas is most uniform when it is injected into the moving solids
bed via an annulus and a set of crossbeams located near the intersection
Desorption Conduction Solid phase
of the cylinder and hopper sections of the processor . Two examples of
well-designed gas distributors are shown in Fig . 21-79 [Mehos, Chem. Engr.
FIG. 21-77 Schematic describing moisture migration . 116(5): 34 (2009)] .
 SOLIDS MIXING 21-35

Solids in
Solids in
Gas out

Gas in Perforated walls Gas out

Cylinder section

Gas in
Gas distributor

Hopper section

Solids out
Solids out

FIG. 21-78 Process vessels with gas injection—countercurrent (left) and cross-flow (right) .

In counterflow systems, the gas velocity must be low enough to prevent

fluidization in the cylindrical section of the processor . A rule of thumb is
that the vessel’s cross-sectional area at the level of gas introduction should
be such that the superficial gas velocity is no more than approximately one-
third the bulk material’s minimum fluidization velocity .
The gas flow rate must be low enough to prevent excessive entrainment of
fines at the top of the column but high enough to ensure that a driving force
for mass or heat transfer exists throughout the column if required .
Cross-flow and radial designs are preferred if the required gas flow rate
is high because a low pressure drop will be realized . Cross-flow and radial-
flow processors are fabricated with permeable walls through which the gas
enters and exits the moving bed of solids .
For cross-flow and radial-flow designs, the gas velocity must be low
enough to prevent pinning or cavity formation . Pinning occurs when the
frictional force between the bulk solid and the permeable wall through
which the gas exits is great enough to prevent the particles from flowing
downward and along the wall . A cavity can develop if the pressure gradient
that develops when gas is injected into the bed causes a gap to form between
the bulk solid and the wall from which the gas is introduced . If a cavity
FIG. 21-79 Gas distributors with crossbeams . forms, gas will flow preferably upward rather than across the bed .

SOLIDS MIXING

General References: L . T . Fan, Y .-M . Chen, and F . S . Lai, “Recent Developments PRINCIPLES OF SOLIDS MIXING
in Solids Mixing,” Powder Technology 61: 255–287 (1990); N . Harnby, M . F . Edwards, and
A . W . Nienow (eds .), Mixing in the Process Industries, 2d ed ., Butterworth-Heinemann, Industrial Relevance of Solids Mixing The mixing of powders,
1992; B . Kaye, Powder Mixing, Springer, Netherlands, 1997; R . Weinekötter and H . Gericke, particles, flakes, and granules has substantial economic importance in a
in B . Scarlett (ed .), Mixing of Solids, Particle Technology Series, Kluwer Academic Publishers,
Dordrecht, 2000; J . M . Ottino and D . V . Khakhar, “Mixing and Segregation of Granular
broad range of industries, including, e .g ., the mixing of human and animal
Materials,” Annul Review of Fluid Mechanics 22(1): 207–254 (2000); E . L . Paul, V . A . foodstuff, pharmaceutical products, detergents, chemicals, and plastics .
Atiemo-Obeng, S . M . Kresta (eds .), Handbook of Industrial Mixing: Science and Prac- As in most cases the mixing process adds significant value to the product,
tice, Wiley, Hoboken, N .J ., United States, 2003; A . Kudrolli, “Size Separation in Vibrated the process can be regarded as a key unit operation to the overall process
Granular Matter,” Reports on Progress in Physics 67(3): 209–247 (2004); J . Bridgwater, stream .
“Mixing of Powders and Granular Materials by Mechanical Means—A Perspective,” The most important use of mixing is the production of a homogeneous
Particuology 10(4): 397–427 (2012) . blend of several ingredients in order to eliminate variations in concentration .
21-36 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

If a material consists of an ingredient or compound exhibiting fluctuations Segregation in Solids and Demixing If a solids mixture consists of
caused by an upstream process, or raw material, the term homogenization particles of different size, shape, density, surface roughness, cohesion, or
is used for the elimination of these fluctuations . By mixing, a new product other material properties, then the mixture has propensity to segregate
or intermediate is created for which the quality and price are very often under external excitations . Segregation competes with both dispersive and
dependent on the efficiency of the mixing process . The efficiency is deter- convective mixing, resulting in a decrease of the mixing quality . Particle
mined both by the materials to be mixed (e .g ., particle size and particle size property differences along with process flow characteristics determine the
distribution, density, surface roughness) and by the process and equipment degree and trend of segregation . Among all driving factors for segregation,
used for performing the mixing . The design and operation of the mixing unit the particle size difference is the most dominant factor for segregation (J . C .
itself have a strong influence on the quality produced, but upstream mate- Williams, Mixing, Theory and Practice, vol . 3, V . W . Uhl and J . B . Gray (eds .),
rial handling process steps such as feeding, sifting, weighing, and transport Academic Press, Orlando, Fla ., 1986) . In practice, there is always polydisper-
determine also both the quality and the capacity of the mixing process . For sity of particle size, even in a single ingredient . Hence, nearly all industrial
example, the filling and discharge flow patterns in storage bins can aid mix- powders can be considered as solid mixtures of different size particles, and
ing or cause demixing depending on their design . Downstream processing size segregation is one of the most ubiquitous problems in industrial solids
may also destroy the product quality due to segregation (demixing) . There- handling . Figure 21-81 illustrates the four most common mechanisms for
fore, the mixture quality of the end product depends on the whole flowsheet, size segregation, and each is discussed below .
not just the mixer . A mixing process can fail, broadly, in two ways: Percolation Segregation Percolation segregation or sifting segregation
1 . The quality of the mix is poor. In cases where the mixing produces the is by far the most important segregation effect, which occurs when finer
end product, this will be noticed immediately at the product’s quality inspec- particles percolate through the gaps between the larger ones under gravity
tion . Frequently, however, mixing is only one in a series of further processing (Fig . 21-81a) . These gaps act as a sieve, so this effect is also called kinetic
stages . In this case, the effects of unsatisfactory blending are less apparent, sieving. As a rule of thumb, a small sphere with diameter ∼1/6 .5 of larger
and might possibly be overlooked to the detriment of final product quality . spheres will just fit through the cracks, even in the larger sphere’s densest,
2 . Overmixing. A batch may be overmixed if mixing proceeds longer than hexagonally close-packed, state . Beyond this limit, particles may “freely sift .”
is necessary for content uniformity . Likewise, a continuous mixer may have For less ideal mixtures, sifting is still possible with motion promoting micro-
residence time in excess of what is necessary for content uniformity . Excess structural change . If a solids mixture is moved, gaps open up between the
residence time means either the throughput is lower than required or the grains, allowing finer particles to selectively pass through the particle bed
vessel is larger than required . Both represent wasted resources . Note, for (see Fig . 21-81), resulting in segregation . Furthermore, percolation occurs
materials that degrade, attrite, or agglomerate, overmixing may also result even where there is a small difference in the size of the particles (250- and
in demixing due to segregation . 300-µm particles) [ J . C . Williams, Fuel Soc. J., University of Sheffield, 14: 29
Mixing Mechanisms: Dispersive and Convective Mixing It is useful (1963)] . The most significant economical example is segregation during
to consider the mixing process as two separate mechanisms: dispersion and
convection (see Fig . 21-80) . Dispersive mixing is a local effect, and it occurs
when neighboring particles randomly change places with each other
(e .g ., micromixing) . Therefore, dispersion is kinematically similar to thermal
diffusion in liquids and gases . However, unlike with diffusion, in particulate
systems there is no relative random motion unless energy is added by some Percolation-driven segregation
external means (e .g ., vibration or shearing) . Dispersion then occurs when
both a concentration gradient and agitation are supplied . Convection corre-
sponds to relative movement of large groups of material (i .e ., macromixing) . Motion
For good convective mixing, the material should be continuously divided
and then combined again after portions have changed places (Fig . 21-80) .
This forced convection can be achieved by bulk movement of material in a
tumbling, fluidized, or static mixer; or by rotation or vibration of internals (a)
with agitated mixers . These two mechanisms work in conjunction as follows .
Convection takes lumps of material of like components, separates them, and Agglomeration-driven segregation
combines them with lumps of material of other components . Along with con-
tinuous agitation, dispersion then smears the composition gradients at the
boundaries of these lumps, allowing for finer scales of mixing .
Motion

Mixing by disperson (b)

Vibration-driven segregation

Vibration Vibration Vibration

(c)
Mixing by dividing and blending
Interstitial fluid-driven segregation
Convection Dust segregation
Trajectory segregation Fluidization segregation

VO

FIG. 21-80 The mixing process can be observed in diagrammatic form as an overlap (d)
of dispersion and convection . Mixture consists of two components A and B; A is sym-
bolized by the white block and B by the hatched block . Dispersion results in a random
arrangement of the particles; convection results in a regular pattern . FIG. 21-81 Four major segregation mechanisms for different sized particles .
 SOLIDS MIXING 21-37

heap formation when filling solids into bunkers or silos . A mobile layer with To mitigate the segregation effect on mixing, a few methods can be used .
rolling particles forms on the surface of such a pile, which restricts larger Williams [ J . C . Williams, Fuel Soc. J., University of Sheffield, 14: 29 (1963)]
particles from passing into the core of the pile, so they remain near the free suggested adding a small quantity of water to form capillary bridges between
surface, slide or roll downward the heap, and eventually deposit near the the particles to reduce their mobility and thus stabilize the condition of
sidewalls of the container . As a result, if filling a container at its center, seg- the mixture . The bridges increase cohesion of particles, so the tendency to
regation can result in a pattern in which fine particles concentrate in the segregate decreases . However, increasing cohesion too much may reduce
middle region below the feed point and coarse particles concentrate near powder flowability and also reduce the dispersive mixing . Another method
the sidewalls of the container . Thus filling a silo and then emptying it from a to limit segregation lies in changing the segregation direction or time scale .
central discharge point are particularly critical . Remixing of such segregated For example, using baffles in rotating drum mixers or using zigzag chutes
heaps to certain extent can be achieved through mass flow discharge; i .e ., can alter the flow direction to change the segregation direction [Shi et al .,
the silo’s contents move downward in blocks, slipping at the walls, rather Physical Review Letters 99: 148001 (2007)] . When filling a vessel, increasing
than emptying from the central core ( funnel flow) . the feed rate can reduce the segregation time scale relative to the dispersive
Agglomeration Segregation This can occur if interparticle cohesion and convective time scales so that a better mixing state can be achieved .
within a given species of particles is stronger than interparticle adhesion However, in general, having ingredients of a uniform grain size (e .g ., through
between species (Fig . 21-81b) . In this case alike particles tend to agglom- granulation) and minimizing the motion of solids mixtures after blending
erate rather than disperse, forcing segregation and limiting mixing . Con- are essential to avoid the segregation effect .
versely, an ordered mixture may occur if adhesion is stronger than cohesion .
In either case, more information on mechanisms of agglomeration can be
found in the Agglomeration subsection . MIXTURE QUALITY: THE STATISTICAL DEFINITION
Vibration-Induced Segregation This can occur if a solids mixture is OF HOMOGENEITY
vibrated, where the coarser particles often float up against the gravity force
and accumulate near the top surface . This mechanism is often called the To judge the quality of a solids mixture, the status of mixing has to be quan-
Brazil nut effect and is illustrated in Fig . 21-81c for the case of a large par- tified . Thus a degree of mixing has to be defined . Here one has to specify
ticle in a mix of finer material . During vibration, smaller particles flow into what property characterizes the mixture (e .g ., composition, particle size,
the vacant space created underneath large particles, preventing the large temperature) . Note there are circumstances in which a good mixture
particles from reclaiming their original position . If the large particles have requires uniformity of several properties simultaneously, but for this discus-
a higher density than the fines, they will compact the fines, further reducing sion we assume only a single property of interest x . The mixture quality is
their mobility and the ability of the large particles to sink . However, when traditionally checked by taking a number N of samples and analyzing the
the large particle is less dense than the fines, it may sink, called the reverse property of interest in each sample xi . Note that one must adhere to proper
Brazil nut effect, but its underlying mechanism is still not well understood sampling practices as described in the Particle Characterization subsection
[Shinbrot and Muzzio, Physical Review Letters 81: 4365–4368 (1998)] . When when extracting the samples . The arithmetic average value of the property
excessive vibration is applied, the powder bed can be fluidized and buoy- of interest p = Σi xi/N will be near to the dosed value of that property to the
ancy will dominate segregation . batch . In fact, if the entire batch is analyzed, the average value of the prop-
Interstitial Fluid-Driven Segregation This encompasses several erty will be identical to the dosed value (barring any analytical uncertainty) .
effects which share the common factor of a fluid contributing to the segre- However, regardless of whether the whole batch or a subset is analyzed, the
gation processes, including trajectory segregation, fluidization segregation, variance of the property of interest over the samples, σ2 = Σi (xi − p)2/N, is
and dust segregation (Fig . 21-81d ) . Trajectory segregation often occurs in dependent on the quality of the mixture .
cyclones or conveying termination into a silo where particles are following In general, higher-quality mixtures have lower variance . In some cases
the different trajectories due to different particle sizes . That is, fine particles the sample relative standard deviation (RSD) = σ/p is reported as a measure
follow fluid streamlines stronger than larger particles . Fluidization segrega- of mixture quality . Regardless, the expected magnitude of the variance is
tion occurs due to different size particles experiencing different drag and strongly dependent on the size of the sample . In the illustration shown
gravity forces, which may lead to segregation via elutriation, entrainment, in Fig . 21-82, the numeric values represent xi from a batch split into N = 36
or buoyancy . Dust segregation occurs when extremely fine particles move samples . The variance for this small sample size is σ2 = 0 .25 . If larger samples
with air currents and separate from coarse particles . had been used instead, and only 9 samples were taken, the variance would
Other Segregation Mechanisms These mechanisms include bounc- have been σ2 ≈ 0 .04, for the same batch . If even larger samples were collected
ing induced segregation, angle-of-repose induced segregation, and den- so that only 4 samples were taken, the variance would have been only
sity segregation . They can also impact industrial processes, but occur σ2 ≈ 0 .02 . Clearly the expected variance, therefore “mixed-ness,” is dependent
less frequently and have fewer effects when compared to size segregation on the size of the sample collected . The size of the sample should then be
described above . tied directly to the scale at which the product must be mixed . Danckwerts
Predicting segregation quantitatively is difficult, because there are often termed this the scale of scrutiny [P . V . Danckwerts, Appl. Sci. Res. 279(3):
multiple mechanisms affecting the final particle spatial distribution . For 279–296 (1952)] . Depending on the application, the scale of scrutiny can
example, one or multiple types of the above-mentioned segregation can range from a single pharmaceutical tablet to an entire railcar or beyond .
interplay with dispersive and convective mixing simultaneously in one It is up to the engineer’s understanding of the process, product, and appli-
process . Segregation, dispersion, and convection all depend on both par- cation to determine the scale of scrutiny .
ticle properties and specific flow conditions in mixing, packaging, or trans- When computing the variance of a set of samples, it is useful to com-
port processes, but there are no first principles theories to calculate them . pare to some idealized scenarios . Three idealized mixtures of useful
Recently, significant progress has been made in quantifying how the per- consideration are perfectly ordered, perfectly segregated, and ideally
colation segregation interacts with dispersive and convective mixing by random, as shown in Fig . 21-83 . In a perfectly ordered scenario (Fig . 21-83a),
using a phenomenological continuum model [Fan et al ., J. Fluid Mech . 741: each component of the mixture is proportionally represented down to the
252–279 (2014)] . individual particle scale . Theoretically such a mixture could spontaneously

FIG. 21-82 Mixtures sampled based on increasing scale of scrutiny . From left to right the scale of scrutiny is 1x, 4x, 9x based on
36, 9, or 4 samples of the same mixture .
21-38 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

(a) (b) (c)

FIG. 21-83 Three idealized mixtures from left to right are perfectly ordered, perfectly segregated, and ideally random .

arise if intercomponent adhesion significantly outweighs cohesion between computed with the relation σ2upper = s2(N − 1)/F −1N−1(α), where F −1k(α) cor-
similar components . In such a case, the mixture variance will be equal to responds to the inverse chi-squared distribution with k degrees of freedom
zero (σ2ordered = 0) for all scales of scrutiny above the smallest ordered pair of and a left-tail probability α . Figure 21-84 shows the upper confidence bound
individual particles . for a 95 percent confidence interval (α = 0 .05) . For example, from the figure
A perfectly segregated mixture (Fig . 21-83b) is quite the opposite . In this a sample size of N = 10 could have actual variance up to 1 .7 times the mea-
scenario all particles of one component are entirely unmixed from those of sured sample variance . Likewise, for N = 600 one could have actual variance
another . This may typically be the starting state for a batch mixture when up to 1 .1 times the measured sample variance .
multiple components are added sequentially . In such a case, the mixture When one is interpreting the sample variance, it is important to note that
variance can be calculated as σ2segregated = P(1 − P) for all scales of scrutiny the measured variance is the summation of contributions from the mixture
smaller than the entire vessel . variance, variance due to the sampling method itself (see Particle Charac-
An ideally random mixture (Fig . 21-83c) occurs when a sample contains terization subsection), and variance due to the analytical method used to
relative quantities of components in a completely random sense . For well- determine the component concentration,
agitated, free-flowing yet nonsegregating systems, this configuration is
optimal (as the name suggests) . For the idealized case of a two-component σ 2measured = σ 2mixture + σ sampling
2 2
+ σ analytical (21-84)
mixture where each sample is composed of the same number of particles n,
the variance of an ideally random mixture is σ2random = P(1 − P)/n . In practice
most samples are not taken as fixed in number, but rather by bulk volume The uncertainty from sampling and the analytical method must be deter-
or mass . Furthermore, particles within a mixture typically vary in size such mined separately based on the respective methods being used .
that a fixed mass may not contain a fixed number of particles . This conun-
drum may be solved by use of Stange’s equation [K . Stange, Chemie-Ing-Tech . EQUIPMENT FOR MIXING OF SOLIDS
26: 331–337 (1954); R . Saunders, The Chemical Engineering Journal 39(2):
129–132 (1988)] A wide variety of equipment is commercially available to suit a multiplicity
of mixing tasks . In this overview, mixers and methods for mixing solids are
P (1 − P ) divided into four groups: (1) bunker and silo mixers, (2) rotating tumbling
σ 2random =  Pm1 (1 + c12 ) + (1 − P ) m2 (1 + c22 )  (21-83) mixers, (3) agitated mixers, and (4) other mixers or mixing methods .
M  Bunker and Silo Mixers These are sealed vessels (Fig . 21-85), which
may serve to homogenize large quantities of solids . They are operated
where M is the mass of the sample, m1 is the number average mass of a single batchwise, continuously, or with partial recirculation of the mixture . Their
particle of component 1, c1 is the relative standard deviation of particle size sealed construction also enables material to be conditioned (e .g ., humidi-
(by number) of all particles of component 1, while m2 and c2 are the respec- fied, granulated, dried, fluidized, or rendered inert) as well as mixed . Silo
tive quantities for all particles that are not component 1 . mixers can be divided into two groups: gravity silo mixers and pneumatic
It is useful to compare the measured value σ to the idealized values mixers .
σordered, σrandom, and σsegregated . Mixtures where σordered < σ < σrandom can be called In gravity silo mixers, an individual layer in the silo mixes with other lay-
pseudo-random . In this case one can infer that some amount of ordering ers for better homogenization by utilizing desired flow patterns from the
is occurring . For σrandom < σ < σsegregated the mixture is partially segregated; specific design of the mixers . A variety of technologies have been developed
perhaps mixing is incomplete or spontaneous segregation does not allow to achieve homogenization in gravity silo mixers by using inserts, modi-
complete mixing . Finally if σ > σsegregated, a measurement or arithmetic error fied silo structure, multiple blending pipes, or mechanical activation to
has occurred, as this is physically impossible . These bounds on σ suggest change flow patterns in the silo . Figure 21-85a shows the Binsert hopper-
the Lacey mixing index M = (σ2segregated − σ2)/(σ2segregated − σ2random) . In general an in-hopper axisymmetric silo mixer . Placing an inner hopper modifies the
increase of M corresponds to improved mixing . A segregated mixture results
in a value of M = 0, while an ideally random mixture corresponds to M = 1 .
A batch mixing process with components loaded in series begins with
M = 0 . If no segregation occurs, after mixing for some time M will increase
toward 1 . The engineer must decide a value of M for which the mixture 3.0
quality is sufficient . Equivalently, the engineer must decide a value of σ for 2.0
which the mixture quality is within acceptable limits for the given scale of 1.8
scrutiny . The time at which the mixture reaches this quantity is deemed the
mixing time . It must be determined experimentally for a given mixing sys- 2.5 1.6
tem and mixture . The mixing time for a continuous system can be defined
similarly by replacing the batch time with the continuous mixer residence 1.4
time . Here it is worthwhile to note that overmixing may occur . In some sys-
(σ/s)2

1.2
tems, agglomeration or particle degradation due to mixing can force segre- 2.0
gation . In these circumstances the mixing quality may actually pass through 1.0
a maximum . That is, more mixing time does not always mean better mixing . 10 100 1000
To assess overmixing, examine the product for particle agglomeration or
degradation in light of what potential segregation mechanisms could exist . 1.5
In almost all cases, the samples used to measure and then compute the
mixture variance will not encompass the whole batch . In this case the
sample variance s2 = Σi(xi − p)2/(N − 1) corresponds to a statistical estima-
tor of the true batch variance σ2 . Under the assumption that the property 1.0
of interest is normally distributed among the collected samples, the confi- 0 10 20 30 40 50 60 70 80 90 100
dence limits of σ2 can be computed using the normal methods of statistics . Sample size, N
Because we are typically questioning if mixing is good enough, it is sufficient
to only compute the upper bound confidence limit of σ2 . This value can be FIG. 21-84 Confidence interval for sample variance based on N samples .
 SOLIDS MIXING 21-39

Distributor

Cone-in-cone

Vertical
section

Recirculation
(a) (b) (c) (d)

Filter Inlet
Deflector

Compressed
air
Spout

Fluidizing Throtting zone

Mixing quadrants in
hopper Air fluidizing the
bottom
Outlet
(e) (f) (g) (h)


FIG. 21-85 Bunker and silo mixers . First row shows gravity silo mixers: (a) Binsert blender; (b) mixing silo blender; (c) Phillips blender; and (d ) mechani-
cal blender . Second row shows pneumatic mixers: (e) air jet mixer; ( f ) air merge mixer; (g) fluidized bed mixers; and (h) pneumatic blender . (E . L . Paul, V . A .
Atiemo-Obeng, S . M . Kresta (eds .), Handbook of Industrial Mixing: Science and Practice, Wiley, Hoboken, N .J ., USA, 2003, with permission .)

radial velocity profiles in the silo to obtain a higher radial velocity gradient and allow full use of the bin contents during circulation . Gravity mixers
to promote mixing between layers . The aspect ratio of height to diameter are designed for free-flowing powders and are offered in capacities ranging
of the cylindrical section is critical to the mixing efficiency in such mixers . between 5 and 200 m3 .
Johanson [ J . R . Johanson, Chem. Eng. Prog. 66(6): 50–55 (1970)] suggested Pneumatic Mixers These mixers can be used for fluidizable bulk
a ratio of 1 .5 for hopper half angle below 35° and 0 .5 for hopper half angle solids . Powders rise due to gas drag forces . As the gas velocity increases, it
above 35° . Figure 21-85b shows a gravity silo mixer with modified internal causes formation of bubbles in the powder bed . The minimum fluidization
structure [W . Muller, Germ. Chem. Eng. 5: 263–277 (1982)] . Bulk solids velocity to fluidize a powder bed is a function of both particle and gas prop-
flow through a system of tubes at various heights and radial locations and erties . Detailed information can be found in Sec . 17 . Powder mixing occurs
mix together afterward . Another type of silo mixer using the same mecha- during the formation, rise, and burst of the bubbles . The bubbles carry pow-
nism (e .g ., Zeppelin Centro mixer) uses a central takeoff tube into which ders from the bottom to the surface of the powder bed and cause powder
the solids travel through openings arranged at various heights up this recirculation . Therefore, the mixing quality is closely related to the size,
pipe . Figure 21-85c shows a Phillips mixer using multiple vertical pipes to shape, and dynamics of bubbles, which depend on the powder properties
homogenize bulk solids . Streams from the central feed point and the indi- and gas superficial velocity (see Sec . 17) . Pneumatic mixers can be divided
vidual pipes mix at the outlets in the conical section of the silo . The mixing into different categories by how gas is injected . In air jet mixers (Fig . 21-85e),
efficiency at one pass generally depends on the number of blending pipes, air is blown in through jets arranged around the circumference of a mixing
but installation of a large number of pipes requires additional consider- head placed in the bottom of the vessel, which generates a swirling turbu-
ation of structural support . Figure 21-85d shows a mechanical silo mixer, lent motion . The largest mixers have a capacity of 100 m3 . Alternatively, in
which has a screw shaft in the center . The rotating screw shaft generates air merge mixers (Fig . 21-85f ), the vessel is divided into several segments,
a velocity profile across the silo diameter to promote mixing . It also lifts and each segment is fluidized independently . If gas flowing through a pow-
solids at the bottom of the silo to the top surface to generate an internal der bed reaches a critical speed (minimum fluidization velocity), powders
circulation for enhancing convective mixing . become fully fluidized in fluidized-bed mixers (Fig . 21-85g) . Because of
If the mixing quality in gravity silo mixers within one pass does not increased powder mobility, fluidized-bed mixers possess excellent mixing
meet requirements, the withdrawn material can be fed back into the bun- properties for solids in both a vertical and radial axis . Circulating fluid-
ker until the required homogeneity is achieved . The material drawn off in ized beds are often used in reaction processes, which are combined with
most cases is carried to the top of the bunker often by pneumatic convey- elevated heat transfer and material circulation to enhance the reaction
ing using an external circulation system . In addition, mass flow is neces- processes . Figure 21-85h shows a pneumatic mixer with a central conveying
sary to allow reliable powder flow, avoid segregation during silo discharge, tube to inject gas after powders are filled [W . Krambrock, Powder Technology
21-40 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

15: 199–206 (1976)] . A cone deflector on top of the tube is used to keep powders These mixers can also be used for agglomeration by injecting a binder . Agi-
from falling into the tube during filling and also spread powders when they are tated mixers include quite a few different styles as discussed below .
carried through the tube during the mixing process . This mechanism generates Paddle and Plow Mixers These mixers typically have a trough with
both axial and vertical convective motion to achieve good mixing . an impeller, which is a single or twin shaft mounted with plows or paddles
After blending, when one is discharging the mixtures from the mixer, a (Fig . 21-87a, b) . The rotating plows or paddles lift bulk solids to generate
velocity smaller than the minimum fluidization velocity, called the deaeration convective motion, so these mixers are suitable for mixing free-flowing to
velocity, is needed to better empty the mixer . It is also important to note that cohesive powders, but not very cohesive powders . Plows of the plow mixers
special care must be taken for the mechanical aspects of the pneumatic (Fig . 21-87a) have an inclined angle to the rotating shaft, which can move
mixers . Under fluidized conditions, the solids exert hydrostatic stress on the bulk solids in the axial direction to enhance axial mixing . The rotation
silo, whereas under static or bulk flow conditions the Janssen effect vastly limits speed of the impellers strongly impacts the mixing intensity of plow mixers .
the radial stress . The largest fluidized-bed mixers as used in cement making At lower rotation speed, similar to the tumbler mixers, bulk solids flow in
reach a capacity of 104 m3 . The specific energy lies between 1 and 2 kWh/ton the cascading regime, where particles roll down along the sloped surface .
and air consumption rises sharply in the case of particle sizes above 500 mm . When the rotation speed increases, particles near the plow tips may fly away
Tumbling Mixers Tumbling mixers achieve solids mixing through the from the powder bed . Depending on the powder characteristics, aeration
motion when particles roll down a sloping surface (called a flowing layer) as can take place at plow tips, and bulk solids may be locally fluidized, which
the mixer shell rotates on its own axis or eccentrically . Tumbling mixers are enables particles to move freely and improve mixing . However, at very high
mostly suitable for free-flowing, nonsegregating bulk solids . For the simplest rotation speed, most bulk solids are lifted toward the shell and slide off the
tumbling mixer, a drum mixer rotating along its horizontal axis (Fig . 21-86a), wall without much rolling motion, so the mixing degree will not be pro-
most mixing occurs in the radial direction driven by both convective mixing moted . Therefore, a proper rotation speed for operating plow mixers is criti-
and dispersive mixing through random particle collisions in the flowing cal for the mixing efficiency . Paddle mixers (Fig . 21-87b) are similar to the
layer . Mixing in the axial direction is fairly slow, because no convective plow mixers except that paddles can rotate individually along their shaft to
motion occurs and the mixing is only dispersive in this direction . For produce additional lateral and axial mixing . Paddle mixers generally operate
solids mixtures prone to segregation, drum mixers are incapable of at higher rotation speeds than plow mixers, which can cause segregation and
achieving good mixing, because percolation-driven segregation often some degree of attrition . It is recommended to determine the optimal rota-
dominates convective and dispersive mixing in the flowing layer . To tion speed and blending time through test trials for plow and paddle mix-
enhance mixing in the axial direction and minimize potential segrega- ers . Both plow and paddle types of mixers can have double counter-rotating
tion in the radial direction, internal baffles with proper size and orienta- shafts with two sets of horizontal impellers, where paddles or plows of one
tion are mounted on the shell to promote axial convective mixing and shaft overlap those on the other shaft . Both mixers can be operated in batch
modify the segregation time scale . Axial mixing can also be improved or continuous processes, and typical mixing time is up to 5 min .
in asymmetrically rotating drum mixers in which the cylinder is tilted Ribbon Mixers These mixers (Fig . 21-87c) use two sets of counter-
obliquely to the main axis . Other tumbling mixers include the V-blender, rotating ribbon blades to transport bulk solids along the horizontal shaft
double-cone blender, and bin blender (Fig . 21-86b, c, d ) . Compared to the in both directions to achieve axial mixing and displace bulk solids in the
rotating drum mixer, these mixers add convective mixing in the rotating lateral direction mostly by centrifugal force . The outer ribbons transport
axis direction due to the nonconstant cross-sectional area in this direction . material toward the center, and the inner ribbons push materials toward
Within each rotation, particle streamlines converge and diverge alternat- the end of the mixer . Compared to the plow and paddle mixers, the shearing
ingly, resulting in better mixing in the axial direction . Segregation can generated from the ribbons and amount of bulk solids transported by the
also occur in these mixers, so it is generally not recommended to blend ribbons is smaller, so a longer mixing time is needed for ribbon mixers in the
segregating solid mixtures in tumbling mixers . range from 15 to 30 min . Ribbon mixers can be used for agglomeration by
A proper flow pattern in tumbling mixers is critical for mixing efficiency spraying liquid and for solid-liquid mixing . A high-speed chopper can also
and quality . Cascading flow is the ideal flow for tumbling mixers . In this be fitted to the mixers to reduce agglomerates . The clearances between the
flow regime, particles flow through a thin flowing layer near the free surface outer edge of ribbons and the trough are rarely less than 3 to 6 mm, but par-
after being lifted by the mixer shell, and the rest of the particles have solid ticle attrition may occur depending on the size and fragility of the particles .
body rotation . The flow regimes in rotating tumbler mixers can be charac- There are also vertical ribbon mixers . The shaft of such mixers is vertical,
terized by the Froude number Fr = rω2/g, where r denotes the mixer radius, g and outer ribbons move bulk solids upward near the shell; inner ribbons
denotes the gravitational acceleration, and ω denotes the angular velocity transport bulk solids downward in the center region of the mixers . The ribbon
of the mixer . The Froude number represents the ratio of the centrifugal mixers can be used in both batch and continuous processes .
forces to the gravitational forces and depends on the rotation speed of the Fluidizing Paddle Mixers The fluidizing paddle mixers (e .g ., Forberg
mixer . The cascading flow generally occurs when Fr ranges from 10−4 to 10−1 mixers), as shown in Fig . 21-87d, are batch mixers consisting of paddles
[ J . Mellmann, Powder Technology 118(3): 251–270 (2011)] . When Fr < 10−4, mounted on twin shafts in a twin trough . The counter-rotating paddles
the entire bulk solids slip on the wall and no mixing can occur . When Fr >10−1, par- agitate bulk solids fed from the top of the mixers to fluidize them rapidly .
ticles start to leave the powder bed after being lifted . The centrifugal effect The paddles also induce convective motion of particles to enhance mixing .
occurs at Fr >1, where all particles are thrown to the shell, forming a ring of After mixing, the mixture is discharged through a set of twin doors at the
particles rotating with the mixer . In this flow condition (Fr > 1), good mixing bottom of the mixer to minimize segregation . The tip speed of the paddles
quality cannot be achieved . The capacity of tumbling mixers is up to ∼10 m3 is about 1 m/s, and the mixing time is as quick as 1 minute [H . Forberg,
and they are typically used in batch processes . Powder Handling Process 4(3): 318–320 (1992)] . However, the mixing time
Agitated Mixers Agitated mixers employ mechanical means to create depends on the cohesion of the powders . Higher cohesion needs longer
mixing actions by rotating impellers (e .g ., plows, paddles, or ribbons) along mixing time, but can limit segregation during discharge . Test trials are
its shaft while the mixer shell remains stationary . The shear from the rotat- recommended to determine the mixing time, tip speed, filling point, and
ing parts can provide much stronger agitation compared to the shear due to discharge method . The Forberg mixers are much more efficient than the
gravity in tumbling mixers . This shear generates strong convective mixing plow mixers in both mixing quality and time, because fluidization enables
and dispersive mixing in both axial and radial directions, so agitated mixers much stronger convective mixing . The capacity of Forberg mixers is up to
can handle a variety of bulk solids from free-flowing to cohesive or pastes . 50 m3 . Forberg mixers have been adapted to continuous processes as well .

Ω Ω
Ω Ω

(a) (b) (c) (d)

FIG. 21-86 Rotating tumbling mixers: (a) drum blender; (b) V blender; (c) double-cone blender; and (d ) bin blender . [E . L .
Paul, V . A . Atiemo-Obeng, S . M . Kresta (eds .), Handbook of Industrial Mixing: Science and Practice, Wiley, Hoboken, N .J ., USA, 2003,
with permission .]
 SOLIDS MIXING 21-41

(a) (b)

Inlet Zero-gravity
(top cover not shown) (fluidized mixing) zone

Bomb-bay
(c) (d) door outlet

Sigma blade

(e) (f)

(g) (h)

FIG. 21-87 Agitated mixers . (a) plow mixer; (b) paddle mixer; (c) ribbon mixer; (d ) fluidizing paddle mixers; (e) screw
mixer; ( f ) sigma-blade mixer; (g) Henschel mixer; and (h) Muller mixer . [E . L . Paul, V . A . Atiemo-Obeng, S . M . Kresta (eds .),
Handbook of Industrial Mixing: Science and Practice, Wiley, Hoboken, N .J ., USA, 2003, with permission .]

Screw Mixers The screw mixers (Fig . 21-87e) have a hopper-shaped be used to add liquid for granulation . The screw mixers have capacity ranging
vessel with a screw placed along the wall . During operation, the screw orbits from 25 L to 60 m3 .
around the hopper and also rotates along its own axis . This motion lifts bulk Sigma-Blade Mixers or Z-Blade Mixers As shown in Fig . 21-87f, these
solids along the wall upward and spreads them on the surface, after which mixers consist of twin troughs with a Z-shaped blade agitator in each trough .
bulk solids flow downward in the center of the vessel . This circulating flow The blades can overlap and counter-rotate at the same speed or different
generates convective mixing for the bulk solids . The rotation of the screw speeds to fold and shear materials for mixing . The mixing time of Sigma-blade
also generates shearing to cause local dispersive mixing . A second short mixers ranges from 10 to 30 min . A spray bar can be placed above the blades
screw, called the satellite screw, can be used to further enhance mixing . The for liquid addition . These mixers are normally used for producing dough or
screw mixers can also have two screws in two hopper vessels joined along thick viscous pastes .
the wall . The screw mixers are suitable for mixing free-flowing, cohesive Impaction Mixers The impaction mixers (e .g ., the Henschel mixer
powders or pastes . The mixing time is a minimum of 10 min . Because of the shown in Fig . 21-87g) resemble a typical kitchen food processor, where
small clearance between the screw and wall (3 to 6 mm), particle attrition blades at the bottom of the mixer rotate at a high speed (2000 to 3000 rpm)
may occur . The solids at the container wall are continuously replaced by the to generate circulating flow patterns in the vertical plane to mix bulk solids
action of the screw so that bulk solids can be indirectly heated or cooled through convective mixing . This mixer needs significant energy compared
through the container’s outer wall for drying applications . Spray nozzles can with other similar types of mixers due to its high rotation speed .
21-42 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

TABLE 21-4 Comparison of Agitated Mixers* The cycle time of a batch mixer must be computed by dividing the batch
size by the sum of the fill time, mixing time, discharge time, and any idle
Fluidizing
Factor Ribbon/paddle Plow paddle Sigma-blade
time . The fill and discharge times are dictated by the mixer but also by the
device and whole system layout . The mixing time is dictated by the solids to
Material Powders/ Powders/ Powders/ Pasty, sticky, be mixed and their relative concentrations . This time must be determined
consistency granules granules granules gritty slurries by testing or experience . In a batch mixing operation, the component con-
up to centrations are specified simply by their mass fill ratios .
2 × 106 cP
In a continuous mixing process, the ingredients are continuously fed
Allowable fill 40–85 30–70 40–140† 40–65 into the mixer, then mixed and prepared for the next processing stage .
level or batch The operations of feeding, mixing, and discharging occur simultaneously
size (% of within the mixer . In continuous mixing, the weighing and filling of the
total mixer batch mixer are replaced by the component’s controlled continuous feeding .
capacity) The blending time is then equal to the material’s residence time in the
Liquid addition Spray bar Spray nozzles Spray bar Spray bar vessel . The residence time is equal to the mass of material in the mixer
configuration above at mixer above above blade divided by the mass throughput of the mixer . Mixture quality is uncertain
ribbons top paddles during start-up and shutdown of continuous mixers, making continuous
Delumping High-speed High-speed Pin mills above None
mixers less flexible and less preferable in low production rate (<100 kg/h)
agitator chopper chopper paddles situations .
configuration blades at blades at The flowchart given in Fig . 21-88 provides a decision process for selecting
sides sides mixer technology . The selection process starts at the upper left of the figure .
For systems with low-proportion (<0 .5 percent) minor components, it is
Mixing cycle 15–20 <5 <1 10–30 advised to preblend the minor component . Note in some cases it is advised to
length
(min) blend materials by agglomeration or milling . These are not explicitly mixing
unit operations, but must be used to mix when the segregation potential is
Final moisture 90–95 or better 95–98 or 98–99 or better 99 or better high . In cases where these operations would be desirable but must be avoided,
homogeneity better maintaining a uniform blend will be difficult .
(% of complete
homogeneity)
Rotating or Stationary Stationary Stationary Stationary
stationary TABLE 21-5 Comparison Between Batch and Continuous Mixing
vessel Processes
Degree of Some High Slight Very high Batch Continuous
particle shear
Number of ingredients Unlimited 2 to 10; more should be
∗Reproduced from E . L . Paul, V . A . Atiemo-Obeng, S . M . Kresta (eds .), Handbook of premixed
Industrial Mixing: Science and Practice, John Wiley & Sons, Inc ., Hoboken, N .J ., USA,
2003 . Frequency of recipe Several per hour Must remain constant for
†
Percent fill more than 100% of the total capacity for another agitating batch mixer change several hours
of equal volume .
Frequency of cleaning Several times a day Once a day or less
or idle time

High-Shear Mixers The high-shear mixers, such as the Muller mixer Production rate or Any rate, limited by More than 100 kg/h
throughput footprint
shown Fig . 21-87h, include a set of rollers and pans . While rotating, the rollers
grind materials to very fine powders, which are mixed by the shearing from Footprint Large, also requires Generally lower foot-
the pans . Usually, tumbling mixers are used to provide a reasonable mix before intermediate storage print, even at higher
it is fed into the high-shear mixers . throughput
Table 21-4 shows comparisons between different parameters for several Requirements of Simple feeding but high Accurate feeding required,
common agitated mixers used in industry . equipment demands on mixer but low mixer demand
Other Mixing Methods The mixed stockpiles method achieves mixing
by following a defined scheme for building up and emptying large stock- Safety Quantity of material Smaller quantities of
piles . A long stockpile is built up by a movable conveyor belt or other corre- contained in the mixer material are actually
relatively large contained in mixer
sponding device traveling lengthwise . During loading the belt continuously
travels up and down the whole length of the pile . As the pile forms, strata Automation Variable degree of Contained in processing
develop in order of the material’s delivery . If the material is systematically automation
removed crosswise to these layers, each portion removed from the stockpile Material type Cohesive or free-flowing Difficult with poorly feeding
will contain material from all the strata and therefore from the times it was cohesive solids
supplied . This mixing method is mostly useful when the scale of scrutiny
is larger than the thickness of the strata . The mixing by feeding method is Feeding Straightforward; simply Requires precision metering
metering different ingredients and bringing these streams of solids together discharge contents into and monitoring
locally . The quality of the metering determines the mix’s homogeneity . the feeder
Therefore, metered feeder units should ideally be used, preferably operated Maintenance and Straightforward More difficult
gravimetrically with appropriate feedback control of weight loss . There is cleaning
limited axial mixing (transverse or back mixing) in this method . According
Material identity Straightforward if batch Difficult, if not impossible
to the requirements of the case in question, mixing is also required oblique integrity is maintained
to the direction of travel . If this oblique mixing is not sufficient, static mix-
ers can be used for free-flowing powders or granules . The energy input into Application to minor Not good: minor fines can Mixing intensity
the mixer is very low, but such systems need sufficient height to achieve components coat the interior
mix quality . Segregation risk High if material has Low if located immediately
segregation propensity; adjacent to the next unit
BLENDING TECHNOLOGY SELECTION intermediate storage operation
and transportation are
Here we offer advice for selecting the mixers described in the previous required
subsection . Before selection of a specific device can be made, it is necessary to Flexibility Generally versatile Low flexibility
broadly select whether a batch or continuous technology will be used . Table 21-5
offers a pointwise comparison between batch and continuous mixing . Robustness Relatively high Low due to necessity of
In addition to the comparisons offered in the table, it is important to several other pieces of
consider the batch versus continuous nature of the mixer’s preceding and equipment (e .g ., feeders
and monitoring
following unit operations . If both are continuous (batch), a continuous equipment)
(batch) device may be favorable .
 PNEUMATIC CONVEYING OF SOLIDS 21-43

Is proportion Change the

Yes Yes Do the Yes Can the Yes Yes
of minor One-stage specification
materials particle size be
component mixing of the feed
segregate? changed?
>0.5% particle size

No No No
Can
Are the Yes Yes
Two-stage components • Hammer
materials free
mixing be crushed • Ball mill
flowing?
together?

No No

• End or
edge, Yes Yes Do the Yes Are the Yes Yes
Is degradation Can moisture • Wet
runner mill B materials materials free More than 2%?
acceptable? be added? agglomeration
• Muller mill, aerate easily? flowing?
B
No No No No No

• Orbiting • Tumblers B,C

• Ribbon C
screw B • Airmix B Is a mixer Yes
• Ploughshare C
• Pan mixer • Ribbon, B,C really needed? • (Storage
B, C • Orbiting screw B
should be
• Ploughshare B
mass flow)
• Silo Blender B No

• High speed
impeller, B Yes Is degradation Portion into
• End or edge acceptable packets
mill, ribbon B,C

No

FIG. 21-88 Flowchart for mixer selection .

PNEUMATIC CONVEYING OF SOLIDS

General References: Klinzing, Rizk, Marcus, and Leung, Pneumatic Conveying of at the destination point is 0 .5 bar, then the gas velocity will be doubled at the
Solids, 3d ed ., Springer, 2010; Dhodapkar and Jacob, “Fluid-Solids Flow in Ducts,” in destination point . For many materials this may not be a problem; however,
Crowe (ed .), Multiphase Flow Handbook, 2d ed ., Taylor and Francis, CRC Press, 2016; attrition in conveying systems has been estimated to be proportional to
Mills, Pneumatic Conveying Design Guide, Elsevier, 2016 .
velocity to the third to fifth power . Consequently, the doubling of velocity in
a conveying system can result in significant degradation to either the bulk
INTRODUCTION solids or the pipe in which they are conveyed .

When solids are introduced into a flowing gas stream in a pipe or tube at THE GAS/SOLID SYSTEM
the proper conditions, the solids can be successfully moved from one point
to another in a process plant, loading a ship, or to a mine, for example . On If solids are introduced into a pipe with sufficient gas velocity such that
account of the advantages of pneumatic conveying over other bulk solids the solids are homogeneously suspended across the pipe cross section, the
conveying technologies, it finds wide use in nearly every facet of modern resulting condition is observed during dilute-phase conveying flows . This is
production processes . The systems are usually highly reliable such that they depicted as condition 1 in Fig . 21-89a .
can be operated round the clock for many months without equipment fail- This situation is somewhat ideal as it is common to have inhomogene-
ures . The fact that the solids and gas are contained within the conveying ities (such as in condition 2) during actual conveying flows possibly due to
pipe minimizes the contact of the conveyed material with the environment, deviation of the velocity required for perfect homogeneous suspension or
and vice versa . Hence, while it is most common to convey solids with air, “roping” of the solids around an elbow, for example . As the gas velocity is
systems using other gases such as nitrogen or hydrogen or conditioned air further decreased, solids will begin to drop out of suspension and lay down
are also routinely used . Typical conveying distances range from 5 to 10 m to on the bottom of horizontal sections of the conveying system . This condition
500 m . For short distances less than ∼15 m, pneumatic conveying is consid- is called saltation. These “salted out” solids may simply slide along the bot-
ered alongside mechanical conveyors such as vibratory or screw conveyors . tom of the pipe or may congregate as small dunes, as depicted in condition 3 .
For longer distances greater than ∼300 m, pneumatic conveying competes Continued reduction of the gas velocity results in additional dune formation,
with belt conveyors and slurry systems depending on a variety of consider- as illustrated in conditions 4 and 5 . Eventually, solids will begin to occupy the
ations such as attrition, particle size distribution, etc . Pneumatic convey- entire pipe cross section, as depicted in conditions 6 and 7, forming slugs of
ing is used extensively across the world to move bulk solid materials from bulk solid . At this point, the resulting condition in the pipe depends on the
producers to customers . Trucks and railcars interface with dedicated con- particle size, permeability, and air retention characteristics of the conveyed
veying systems that move bulk solids with little effort into production facili- solid . Fine solids, typically less than 150 mm, will have sufficient air reten-
ties or intermediate storage areas . Materials such as cement, plastic pellets, tion such that they will continue to move through the system as a fluidized
grains, and foodstuffs are commonly transported in this manner . slug . In contrast, coarse bulk solids such as plastic pellets have sufficient
Despite the numerous advantages of pneumatic conveying systems, they air permeability that they will be conveyed easily in “dense phase” through
do have some distinct disadvantages . As discussed in detail later in this the pipeline . For particles which deaerate easily and do not have sufficient
section, the gas expansion (and concurrent gas velocity increase) can limit air permeability (typical examples would be sand and granulated sugar), it
system length due to pressure drop, velocity, or attrition considerations . For is highly likely that the conveying system will plug or block upon reaching
vacuum conveying systems, they are limited by the amount of vacuum that the conditions shown in condition 6 or 7 . For these bulk solids, conveying
can be achieved by conventional air movers . For example, if a system oper- systems rely on the use of air injection along the length of the pipe to control
ates at 1 bar when the solids are introduced into the system and the pressure plug length . Flow conditions shown in conditions 1 and 2 are typical of dilute
21-44 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

M3
M2
M1

B A

Pressure gradient
Air only

Conveying gas velocity

FIG. 21-90 Example of conveying system optimization .

The Zenz diagram can be useful in troubleshooting and optimization of

conveying systems . One of the most common applications of the state dia-
gram is the uprating of the mass flow of a dilute-phase conveying system .
Gas flow direction Consider a system operating at point A in Fig . 21-90 at mass flow rate M1 .
As is the case for many systems, the maximum permissible pressure drop is
FIG. 21-89a Pattern of solids flow in pneumatic conveying . [From Wen, U.S. Dept. of constrained by the air mover . However, if the system gas velocity is decreased
Interior, Bureau of Mines, PA, IC 8314 (1959) with permission.] from point A to point B, the mass flow rate increases from M1 to M3 . This
counterintuitive result is explained by recognizing that the total system pres-
sure drop is given by
phase while the conditions shown in conditions 3 through 9 are representa-
tive of dense-phase flow . Most bulk solids will not exhibit all the dense-phase
∆Ptotal = ∆Pair + ∆Psolids (21-85)
flow regimes . For vertical flows, coarse solids such as seeds, grains, or plastic
pellets will form uniform axisymmetric slugs . For fine, fluidizable solids, the
entire pipe will be filled with bulk solid in a fluidized state . By reducing the gas velocity, additional solids can be conveyed without
Zenz or State Diagram Zenz (Fluidization and Fluid Particle Systems, increasing overall pressure drop . The velocity can be decreased until the
Reinhold, 1960) proposed plotting pressure gradient versus velocity for a pressure minimum curve is reached . An experimental Zenz diagram for
conveying system . This is shown in Fig . 21-89b . round, polystyrene Styropor resin is shown in Fig . 21-91 (dp = 2 .385 mm, pipe
This plot is frequently referred to as the Zenz or state diagram . Starting diameter = 52 .6 mm, particle density = 1050 kg/m3) .
at the right-hand side (high velocity) of the plot, the system is operating in Additionally, it is common for a well-operating dilute-phase system to pro-
dilute phase . As the gas velocity decreases, so does the pressure gradient gressively (or suddenly) begin to either experience high pressure drop or, in
until the pressure minimum point is reached . At this point, particles begin to worst cases, block completely . A careful examination of Fig . 21-91 shows that
salt out on the bottom of the pipeline . The locus of pressure minimum points as gas velocity is decreased below the pressure minimum curve, there is a sig-
is referred to as the pressure minimum curve and serves as the basis for salta- nificant increase in pressure gradient such that the system is at severe risk of
tion velocity correlations . Additional reduction of gas velocity now results in blockage . The decrease in gas velocity can commonly be the result of gas leak-
dune formation along with a significant increase in pressure gradient . Visual age in a dilute-phase conveying system through rotary valves or diverter valves .
observation of the conveying line often reveals considerable line vibration
and shaking during the onset of dense-phase conveying . Figure 21-90 shows SALTATION VELOCITY
a Zenz diagram for three different mass flow rates of solids . In this case, the
dashed line represents the “air only” pressure gradient . Three lines of con- A key parameter for the design or analysis of a conveying system is the sal-
stant mass flow rate are plotted where M3 >M2 >M1 . tation velocity . As indicated previously, this velocity is the locus of points
connecting the minima of the mass flow rate curves on the state diagram .

Dense phase Dilute phase

Pressure drop per metre ∆p/∆L [mbar/m]

5.6
G = 1244 kg/h
G = 995 kg/h
4.8
e

G = 743 kg/h
t
ra
w

4.0
flo

G = 497 kg/h
s

Continuous dense
as
Pressure gradient

phase flow
M

3.2
Plug
flow G = 251 kg/h
2.4 G = 0 kg/h
Discrete plug
flow 1.6
Dune flow Dilute phase
flow
0.8
Discontinuous dense
phase flow 0
0 4 8 12 16 20 24 28 32 36
Saltating flow (a) Average air velocity v [m/s]
Gas velocity

FIG. 21-91 Conveying diagram for Styropor polystyrene in 52 .6-mm pipe . (Rizk, F .,
FIG. 21-89b Zenz diagram of pressure gradient versus velocity . Ph .D . thesis, Karlsruhe, 1973 .)
 PNEUMATIC CONVEYING OF SOLIDS 21-45

There is an important physical distinction between saltation velocity and 40

pickup velocity, even though the two terms tend to be used interchangeably
in practice . Saltation represents the velocity at which particles just begin Rizk
to lay down on a horizontal section of pipe . Pickup velocity is the point at 35
which particles will be picked up from a stationary layer . Due to variations in Matsumoto
particle size distribution and the visual determination of the exact saltation/
pickup points, the velocities are rarely uniquely valued . In practice, the con- 30
veying system designer must consider the saltation of the largest particles

Saltation velocity (m/s)

of the size distribution to ensure successful conveyance of the bulk solids .
Numerous correlations have been proposed for the saltation velocity and 25
have been reviewed by Leung and Jones (Proceedings of Pneumotransport 4,
BHRA Fluid Engineering, 1978) and Gomes and Mesquita [Braz. J. Chem. Eng.
31(1): 35–46 (2014)], Cabrejos and Klinzing [Powder Technology 79: 173–186 20
(1994)] and Kalman and Rabinovich [Powder Technology 160: 1003–1113
(2005)] . For large particle systems greater than 200 mm, the equation of Rizk
(Proceedings of Pneumotransport 3, BHRA Fluid Engineering, 1973; and Ph .D . 15
Thesis, University of Karlsruhe, 1973) finds broad applicability . Rizk’s equa-
tion for saltation velocity is given by
10
κ

µ s = δ  s 
1 v
(21-86)
10  gD 
5
where δ = 1 .44dp + 1 .96 and κ = 1 .1dp + 2 .5 for dp in millimeters, ms is the load-
ing at saltation, D is the pipe diameter, and vs is the saltation velocity . A very 0
important distinction is that the equation as written is not explicit in vs . The 0.0 0.5 1.0 1.5 2.0 2.5 3.0
loading (mass flow rate of solids/mass flow rate of gas) at saltation is given by
Particle diameter (mm)
Ws
µs = (21-87)
ρ g Av s FIG. 21-92 Comparison of Rizk and Matsumoto correlations as a function of par-
ticle size .
where Ws is the mass flow rate of solids, A is the pipe cross-sectional area, and
ρg is the density of the gas . Substituting the above expression into Eq . (21-86)
and rearranging terms yields an explicit expression for the saltation velocity Rizk Eq . (21-88):
For 10-mm particles, δ = 1 .44dp + 1 .96 = 1 .44(0 .01 mm) + 1 .96 = 1 .9744; κ = 1 .1dp +
1
2 .5 = 2 .51
 W gD κ 10 d  κ+1
For 2500-mm particles, δ = 1 .44dp + 1 .96 = 1 .44(2 .5 mm) + 1 .96 = 5 .56; κ = 1 .1dp + 2 .5 = 5 .25
vs =  s  (21-88)
 ρg A  1

1  kg   9.81 m 
2.511
 2.511+1
From Eq . (21-88), it can be deduced that saltation velocity increases both as  W gD κ 10 δ    2.08
   (0.1 m)  101.9744 
sec   sec
κ+1

2
 
the mass flow rate of solids increases and as the pipe diameter increases . vs =  s  = = 14.7 m/s
For finer particle sizes, Matsumoto et al . [ J. of Chem. Eng. of Japan 10(4):  ρg A    1.92 kg  (0.0079 m 2 ) 
  3 
273–279 (1977)] showed that the saltation velocity increases with decreas-  m 
ing particle size below a critical particle diameter dp∗ . This reflects the
increased interparticle forces such as van der Waals forces, electrostatics, Similarly for the 2500-mm particles, vs = 16 .9 m/s .
Matsumoto analysis, Eq . (21-89), needs to be calculated first .
and capillary bonding forces at fine sizes . Matsumoto’s saltation formula-
tion first finds the critical particle diameter dp∗: −0.74
d ∗p ρ
−0.74
 945 kg/m 3 
d ∗p −0.74 = 1.39  s  = 1.39  = 0.0142
ρ  ρ   1.92 kg/m 3 
= 1.39  s  (21-89) D g

D  ρ 
g d ∗p = 0.0142(0.1 m) = 0.00142 m = 1.42 mm
For dp ≥ dp∗,
1.06 −3.7 3.61
Since 2 .5 mm >1 .42 mm, Eq . (21-90) is used .
ρ  Frp   Frs 
µ s = 0.373  s   10    (21-90)
 ρ  10  ρ
1.06
 Frp 
−3.7 3.61
g µ s = 0.373  s   10 
 Frs 
 
 ρ  10 
where g

ut vs
ut vs Frp = and Frs =
Frp = and Frs = gd p gD
gd p gD
For dp < dp∗, The terminal velocity for 2500-mm particles is 5 .90 m/s .
1.43 4
 dp   Frs  Frp =
ut
=
5.90 m/s
= 37.67
µ s = 5560     (21-91)
 D 10  gd p  9.81 m  (0.0025 m)
 s 2

Figure 21-92 shows a comparison of the saltation velocity for the Matsumoto vs vs
and Rizk equations as a function of particle size . Frs = = = 1.01v s
gD  9.81 m  (0.1 m)
 s 
2

Example 21-3 Calculate the saltation velocity for 10-mm and 2500-mm particles,
using both correlations for a pressure conveying system at the pickup point . The abso-
Substituting into Eq . (21-90) and solving for vs give
lute pressure at the pickup point is 150 kPa . The conveying gas is air at 22°C . The solids
are high-density polyethylene with a density of 945 kg/m3 . The solids conveying rate is 1.06 −3.7 3.61
7500 kg/h through a 100-mm inner-diameter (ID) tube .
Ws
=
2.08 kg/s
= 0.373 
945   37.67   1.01v s 
ρ g Av s (1.92 kg/m3 )(0.0079 m 2 )v s  1.92  
10 
 
10 


Ws = 7500 kg/h = 2 .08 kg/s A = (π/4)D2 = (π/4)(0 .1 m)2 = 0 .0079 m2

Solving for vs yields vs = 15 .1 m/s which is slightly lower than the 16 .9 m/s obtained
using the Rizk equation .
(150 kPa)  29
kg 
For 10 mm < 1420 mm, Eq . (21-91) is used .
P ( Mw )  kg ⋅ mol 
ρg = = = 1.92 kg/m 3 1.43
RT  kPa ⋅ m 3   dp   Frs 
4

 8.314 kg ⋅ mol ⋅ K  ( 22 + 273.15 ) K µ s = 5560  

 D
 
10 
21-46 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

Substituting for ms and Frs yields

4
1.43
 vs 
Ws  dp   gD 
= 5560    
ρ g Av s  D 10 

Substituting values and solving for vs lead to

4
 vs 
  9.81 m  (0.1 m) 
  2 
s 
1.43

= 5560 
2.08 kg/s 0.00001 m 
 
(1.92 kg/m3 )(0.0079 m 2 )v s  0.1 m   10 

so vs = 42 .18 m/s . This is considerably larger than the value obtained by using the Rizk
equation .

CHOKING VELOCITY
FIG. 21-93 Push or pressure conveying system .
For vertical lines, the limiting flow rate of solids occurs when solids can no
longer be carried upward; this phenomenon is called choking . Chong and
Leung [Powder Technol . 47(1): 43–50 (1986)] reviewed choking correlations
and recommended the equation of Yousfi and Gau [Chem. Eng. Sci . 29(9): The two equations can be solved simultaneously for εc and vc , giving values of 0 .84 and
1939–1953 (1974)] for fine materials (Geldart groups A and B, see Sec . 17 for 7 .65 m/s, respectively . This number is about 50 percent of the value obtained for the
saltation velocity . Hence, usually saltation is the governing concept when designing or
a description) and Yang [Yang, Powder Technol. 35: 143–150 (1983)] for larger optimizing dilute-phase conveying systems .
group D materials . More recently, Klinzing et al . (Pneumatic Conveying of
Solids, 3d ed ., Springer, 2010) have commented that most of the choking work
has been done in small bore (<75-mm) pipe and that most correlations give TAXONOMY OF PNEUMATIC CONVEYING SYSTEMS
prediction in comparison with experimental data to +/−50 percent . They rec-
ommend the work of Leung et al . [Ind. Eng. Chem. Proc. D . 10: 183–189 (1971)], Pneumatic conveying systems can be classified generally into two broad
Punwani et al . (Proceedings of International Powder and Bulk Solids Handling divisions based on the flow type and system pressure . From the Zenz
and Processing Conference, Chicago, 1979) and Yang . Here the work of Yousfi diagram, systems are either dense-phase or dilute-phase . Those systems
and Gau and the work of Yang are presented . Using experimental results for operating above atmospheric pressure are known as pressure systems and
glass, polystyrene, and catalyst ranging from 20 to 290 mm, Yousfi and Gau are often called push systems as material is pushed from the pickup point
developed an empirical relationship for the choking velocity for fine materials to the system destination(s) . Conversely, systems operating below atmo-
spheric pressure are called vacuum or pull systems . Figures 21-93 and 21-94
vc = 32 gd p Re−p0.06 µ 0.28 (21-92) show push and pull systems . The characteristics of each of the four combi-
nations are examined here .
where Rep = ρf Utdp/mf and m = Ws/Wg or the system phase ratio . Note that Wg Dilute phase/pressure system This is one of the most common conveying
= ρgAvc so Eq . (21-92) is not explicit in vc . Using the data of Yousfi and others, systems used by industry . Conveying pressures are typically less than 1 bar
Yang proposed the following system of equations for choking which Cheong gauge due to the widespread use of rotary lobe “Roots” style blowers . With the
and Leung have found to work well for larger particles . application of stepped lines, conveying distances of 300 m can be achieved .
Loadings or phase ratios usually range from 1 to 10 kg/kg . Solids are in sus-
2.2 pension when conveyed in this regime . As shown in Fig . 21-93, this system is
2 gD (εc-4.7 − 1) ρ  preferred when there is a single source and multiple destinations because only
= 6.81 × 10 5  f  (21-93)
(v c − U t ) 2  ρs  one air mover is needed at the beginning of the system . It is possible to use a
pressure system to pick up solids from multiple silos in a row, but one must
Ws ensure that air leakage is minimized from the conveying line into the silos .
Gc = = (vc − U t ) ρs (1 − εc ) (21-94) Dilute phase/vacuum system This conveying is also very commonly used
A
especially for short (typically 30 m or less) systems . These systems will typi-
This set of equations calculates both the voidage at choking εc and the choking cally operate at end of line pressures of no smaller than 0 .5 bar absolute due
velocity vc . If the mass flow rate of system Ws is known, then the solution of to the considerable gas expansion along the length of the conveying line .
the equations is fairly straightforward . For example, for a system starting at ambient pressure with 0 .66-bar pres-
sure drop, the gas velocity along the length of the line will triple . An added
constraint is that rotary lobe blowers operating in vacuum configuration
Example 21-4 Compare the choking velocity with the saltation velocities will only develop 0 .5-bar pressure drop . Nevertheless, vacuum conveying
obtained in the previous example . For the 10-mm particles, it is prudent to use Eq . (21-92) systems find widespread use for batch loading operations, for example .
by Yousfi and Gau . The terminal velocity can be calculated by using methods in Sec . 17 .
The terminal velocity for the 10-mm particles is 0 .0028 m/s . A typical system is shown in Fig . 21-94 . These systems operate by vacuum
conveying a small batch of material into a destination hopper usually to a
 1.92 kg  (0.0028 m/s)(0.00001 m) preset time or weight; the materials are then discharged from the destina-
ρ f Ut d p  m3  tion hopper, and the cycle is repeated .
Re p = = = 0.0029
µf 1.84 × 10 −5
kg
m ⋅s

vc = 32 gd p Re−p0.06 µ 0.28
0.28
 2.08 kg / s 
= 32 9.81 m/s 2 (0.00001 m) × (0.0029)−0.06  3 2 
 (1.92 kg/m ) (0.0079 m ) v c 

Solving the above for vc gives a choking velocity of 1 .57 m/s which is significantly
smaller than the saltation velocity obtained from either correlation .
For the 2500-mm particles, the terminal velocity is 5 .90 m/s . For Eq . (21-94), we have

= (vc − 5.90 m/s)  945 3  (1 − ε c )

Ws 2.08 kg/s kg
Gc = =
A 0.0079 m 2  m 

For Eq . (21-93), we have

2  9.81 2  (0.10 m)(εc−4.7 − 1)

m
2.2
 s   1.92 kg/m 3 
= 6.81 × 105  3 
(vc − 5.90) 2
 945 kg/m  FIG. 21-94 Pull or vacuum-type conveying system .
 PNEUMATIC CONVEYING OF SOLIDS 21-47

Dense-phase/pressure system Numerous researchers and practitioners

have shown that attrition in pneumatic conveying systems has been found
to be proportional to the conveying velocity to the third to fifth power;
hence velocity control in pneumatic conveying systems for attrition-prone
materials is of utmost importance . Consequently, dense-phase or low-veloc-
ity conveying plays an important role in pneumatic conveying of bulk solids
for the minimization of product degradation . Pressures in dense-phase sys-
tems range from 1 to 10 bar . For bulk solids with high air retention prop-
erties, phase ratios as high as 600 have been observed for bulk solids such
as cement . As discussed in the earlier subsection on gas solid systems, the
actual flow mode observed during dense-phase conveying is highly depen-
dent on the air retention and permeability properties of the bulk solid . In
general, these systems can be broken down into three broad classifications:
(1) systems with fine materials which have sufficient air retention and per-
meability as to permit the solid-gas suspension to completely fill the line; (2)
systems handling granular materials where active air management along
the pipeline length is required to prevent pipeline blockage; and (3) systems
with very permeable solids such as plastic pellets which naturally form slugs
and require only proper pressure and velocity for successful dense-phase
conveyance .
Dense-phase/Vacuum System While such systems are possible, they (a)
are not frequently used in practice due to the customary 0 .5-bar pressure
drop limitation . These systems are used for the transfer of fragile solids ( for
example, beaded carbon blacks) over short distances .

PNEUMATIC CONVEYING SYSTEM COMPONENTS

Air Movers Similar to a pump in the movement of liquids, air movers
are a primary component of any pneumatic conveying system . Three dis-
tinct types of devices find use in conveying systems with each having dis-
tinct advantages and disadvantages with respect to cost and performance .
Fans Fans are the simplest devices and typically the least expensive . A
typical centrifugal fan is shown in Fig . 21-95a . While units capable of deliver-
ing 0 .3 bar are available, most units in practice range from 25 to 75 mbar, so
this limits their utility in conveying systems . One unique advantage of fans
is the ability of the bulk solid to pass through the fan itself (however, exer-
cise caution when conveying combustible dusts) . For light materials such as
paper trim scrap, ground celluloses, films, etc ., it is not uncommon for a fan
to pull solids from a source, have them pass through the fan, and then be
pressure-conveyed to a destination silo . The primary disadvantage for use
of fans in pneumatic conveying systems is their comparatively poor pres-
sure versus volume curve, as shown schematically in Fig . 21-96 (dotted line) .
From the dotted curve, it can be observed that for a small change in system
pressure drop, the volume delivered can markedly increase or decrease .
Blowers Blowers are the most common air or prime mover for convey-
ing systems . Invented in the mid-19th century by the brothers Philander and (b)
Francis Marion Roots, these positive displacement devices consist of two
intermeshing lobes, as shown in Fig . 21-95b, and are often known as Roots FIG. 21-95 (a) Typical centrifugal fan (https://www .indiamart .com/proddetail/
blowers . While the two-lobe configuration is relatively common, three-lobe industrial-centrifugal-fan-14497466212 .html) . (b) Roots blower (https://www .indiamart
blowers are used to reduce pulsations and noise . In the pressure configura- .com/proddetail/twin-lobe-industrial-root-blower-10456324097 .html) .
tion, typical maximum pressures range from 0 .5 to 1 .2 bar, and in vacuum,
typically 0 .5-bar suction can be achieved . Since the process is essentially
adiabatic compression, it is important to recognize that the temperature
of the compressed gas can increase significantly (up to 80°C depending on
discharge pressure), which may impact product degradation . A significant
advantage of lobe-style blowers is their relatively flat performance curve,
meaning that the volume delivered typically only decreases by ∼15 percent Compressor
over the expected range of operating pressures . Equipment suppliers pro-
vide volume delivered, adiabatic temperature rise, and brake horsepower as
a function of operating pressure and blower speed . These blowers are often
installed on trucks handling bulk solids and are connected to the engine Blower
power takeoff .
Compressors Compressors are routinely used for dense-phase systems
where pressure drops in excess of 1 bar are not uncommon . They are the
Volume

most expensive of the air movers, but do deliver nearly constant volume over
a wide range of pressures .
Conveying Pipe and Connections Conventional pipe in a variety of
different metals and plastics is used in conveying systems . While Schedule
40 pipe is sometimes used, lighter-gauge pipe (Schedule 10, for example) is
more commonly used, given the amount of weight of material in the pipe
at any given moment is relatively small . In addition to common ANSI pipe,
tube is commonly used in smaller pipe sizes (100 mm) . In the United States, Fan
for example, it is specified by its outside diameter (2, 2 .5, 3, 3 .5, and 4 in)
with various wall thicknesses . While pipe and tubing suffice in most ser-
vices, some materials demand special pipe alterations and treatment . For
abrasive materials like sand, glass batch, and cement, hardened pipe and Pressure
ceramic linings find common use . In other cases such as the conveying of
plastics, pipe can be treated by grooving or shot-peening to prevent the FIG. 21-96 Pressure versus volume characteristics for various air movers .
21-48 SOLIDS PROCESSING AND PARTICLE TECHNOLOGY

FIG 21-97 Typical compression coupling (http://www .morriscoupling .com/prod1 .html) .

formation of deposits such as streamers, angel hair, or snake skins . Flanges

and compression couplings are the most common types of connectors used
to join conveying pipe . For flanged pipe systems, it is not unusual to have
self-aligning flanges to avoid impact of the solids with the lips/edges created
with misaligned piping . The repeated impact of the bulk solids can result in
attrition of the solids or deposition of the solids at the point of impact . FIG. 21-98 Rotary valve . (Courtesy of Coperion.)
A typical compression coupling is shown in Fig . 21-97 .
Feeders for Pneumatic Conveying Systems Numerous devices exist
to feed both dense- and dilute-phase conveying systems . Dhodapkar and from boxes using vacuum conveying . While not sophisticated in design, suc-
Jacob [“Fluid-Solids Flow in Ducts,” in Crowe (ed .), Multiphase Flow Handbook, tion wands are usually designed with a “pipe in pipe” arrangement where
2d ed ., Taylor and Francis, CRC Press, 2016] have summarized these in the outer pipe permits the flow of air to the suction pipe and the inner pipe
Table 21-6 . carries the bulk solid/air mixture . This is particularly critical for bulk solids
Despite the significant number of options available to the conveying sys- with low permeability .
tem designer, three devices find widespread use in conveying system . The The rotary valve is the workhorse for both dilute- and, more recently,
simple suction wand is commonly used for manual unloading of bulk solids dense-phase conveying . A typical valve is shown in Fig . 21-98 .

TABLE 21-6 Pneumatic Conveying System Feeders

Suitable for
Feeder system types Operating pressure Suitable materials Unsuitable materials Conveying mode Comments
Suction nozzle Vacuum Up to 0 .5 bar All free flowing Caking, large chunks, Continuous dilute Feed rate control is difficult
vacuum sticky, cohesive phase
Rotary feeder Vacuum and 0 .5 bar vacuum Wide range of materials Abrasive, caking, Continuous dilute and Most common feeder . Wide
pressure −3 .5 bar pressure sticky, very cohesive dense phases variety of design features
(6 bar maximum)
Eductor Pressure 0 .25 bar Free flowing Abrasive, friable, Continuous dilute Special design for abrasive
highly cohesive phase materials, high-pressure
motive gas may be required
Blow tank-bottom Pressure 0–10 bar Wide range of materials, Very cohesive sticky Batch dilute and Poorly flowing materials will
discharge abrasive materials or compressible dense phases require air injection in the
materials cone, difficult to control
conveying rate
Blow tank- Pressure 0–10 bar Fine fluidizable materials Geldart class B, B/D, Batch dilute and
fluidization type, (Geldart class A/C, A) D or material with dense phases
top discharge large chunks
Double flapper valve Pressure 0–0 .5 bar Abrasive and free-flowing Cohesive Continuous dilute Pulsing feed
material phase
Screw feeder Pressure and 0 .5 bar vacuum Wide range of materials Sticky, compressible Continuous dilute Improper design will result in
vacuum to 1 bar pressure phase feed rate control (what do we
mean by this?) . Rate sensitive
to material properties .
Fuller–Kinyon pump Pressure 2 bar Fine powders Coarse materials Continuous dilute and Typically used in cement
dense phases industry
Slide gate Vacuum Up to full vacuum Coarse materials, preferably Cohesive, sticky Continuous dilute Simple, discharge rate depends
free-flowing materials phase on material properties
Other specialized Pressure 0–0 .5 bar Materials with high air reten- Sticky, cohesive, low Continuous dilute and
feeders: diaphragm tion or low deaeration, air retention dense phases
pump preferably low bulk density
 PNEUMATIC CONVEYING OF SOLIDS 21-49

The rotary valve consists of a 6-, 8-, or 10-pocket rotor which fits closely DESIGN AND ANALYSIS OF DILUTE-PHASE
inside the valve body . Typical rotor/body clearances are 75 to 150 mm . Rotors CONVEYING SYSTEMS
are available in a variety of manifestations: open-ended for free-flowing
products; closed-ended rotors to prevent powder leakage out of the valve In many ways, analysis of dilute-phase systems is a relatively simple exten-
seal; replaceable flexible tip rotors for wear applications; and shallow pocket sion of incompressible flow through piping systems . However, there are four
rotors for materials which are difficult to release from the pocket . Typical important differences between incompressible flow and dilute-phase con-
rotor speeds range from 10 to 50 rpm . Slower than 10 rpm results in a pulsatile veying flow which make the solids flow situation more complicated .
conveying flow, while if the speed is greater than 50 rpm, it can be difficult for 1 . The conveying gas expands continuously along the pipeline length .
the solids to flow into the rotor pockets . The overall valve capacity is given by Hence, analysis of the system must be done in a piecewise fashion along the
length of the pipe .
2 . Pressure drop due to acceleration of the solids must be considered .
V = v(rpm)(
 ρb )(fill efficiency) (21-95)
3 . As opposed to the equivalent length concept for incompressible flow,
each elbow is calculated separately, which serves to account for the decel-
where v is the volumetric displacement per revolution . Fill efficiency can eration and reacceleration of the solids as they transit the elbow .
be typically 70 to 90 percent, but in cases where the valve is not properly 4 . In addition to the pressure drop associated with the fluid, the pressure
vented and is metering fine solids, the fill efficiency can drop to well below drop due to the solids friction needs to be calculated .
50 percent . Typical performance curves for different sizes of a commercial The overall pressure drop equation for a dilute-phase conveying system
valve are shown in Fig . 21-99 . Additional detail on rotary valves can be found is given by Eq . (21-96) .
in the Feeding, Metering, and Dosing subsection .
Another key design factor is rotary valve leakage . In pressure conveying
∆Ptotal = ∆Pgas + ∆Pacc + ∆Plift + ∆Pbends + ∆Psolids + ∆Psolid-gas separation (21-96)
systems, gas will leak past the valve in two distinct ways . Even without rota-
tion, gas will leak past the slits formed by the rotor tip and the valve body .
In addition, as the valve rotates, gas will leak as a result of the gas-filled Each term will now be discussed in turn .
pockets rotating back to the low-pressure side . This gas volume loss must be Gas Pressure Drop Gas-only pressure drop can be calculated by using
accounted for during the design process or during process troubleshooting . the customary methods found in chemical engineering fluid mechanics
Typical gas leakage curves are shown in Fig . 21-100 . textbooks or the corresponding section here . It is important to recognize
Many suppliers will supply leakage curves for new valves, and it is impor- that the gas pressure drop includes both the pressure drop across the con-
tant to recognize that these valves will increase during the life of the valve veying line and any additional pressure drop associated with the piping to
due to wear . Rotor to valve body clearance must be periodically checked and/or from the blower . As opposed to the methods for incompressible flow
with feeler gauges . Wear can be caused by abrasive solids, temperature using a single calculation to calculate the pressure drop across the system,
changes, or deposition of solids on either the rotor or body . In the last it is extremely important to segment the entire conveying line and calcu-
several years, high-pressure rotary valves have been introduced and are late the pressure drop individually across each segment . This must be done
now routinely used in dense-phase conveying applications . because of the change in both the gas velocity and the gas density along the
Blow pots are likely the most common feed device used for dense-phase length of the piping system .
conveying . The most common arrangement consists of a conical hopper with Acceleration Pressure Drop Both the gas and solids must be acceler-
block valves on the inlet and outlet . It is common to have some aeration in the ated, and there is a pressure drop associated with this process . Klinzing et al .
cone in order to fluidize the solids and facilitate their exit out of the blow pot . (Pneumatic Conveying of Solids, 3d ed ., Springer, 2010) show by integrating
Since the blow pot usually works in batch fashion ( fill/pressurize/blow solids/ a modified momentum balance that the acceleration of the solids can be
de-pressurize), the actual “continuous” flow rate is lower than the instanta- represented by
neous flow rate during the blow cycle . On account of their simple design, blow
pots are generally very reliable with the exception of the inlet and outlet valves, U pWs
∆Pacc-solids = (21-97)
which can see wear as a result of the number of cycles they see . A

Performance diagram
PE/PP pellets with bulk density 520 kg/m3 and particle size 3 mm, ∆p = 0.8 barg

ZVD 800

ZVD 630

ZVD 550

ZVD 480

ZVD 400

ZVD 320

ZVD 250

ZVD 200

0 50 100 150
Capacity [t/h]

FIG. 21-99 Rotary valve performance chart . (Courtesy of Coperion.)