Reactions and Separations

Can We Believe the

Simulation Results?

Be careful of these key issues that may generate

differences between a distillation-tower
Henry Z. Kister, computer simulation and its actual performance.
Fluor Daniel Simulations do not always square with
nuts-and-bolts reality.

P revious surveys (1, 2) collected case his-

tories of tower malfunctions from the
open literature. Many reports described
simulations that did not reflect what a
tower was actually doing. Often, the problem was
with the simulation. Sometimes, the problem was
that the tower did something unexpected: the simula-
Three major findings
This survey revealed three major issues that require
attention in using simulations:
1. correctly predicting vapor/liquid equilibrium (VLE)
2. having the simulation match plant data
3. applying graphical techniques to troubleshoot
simulations.
tion was actually correct, based on the data fed to it. These three issues are present in about two-thirds of
Finally, there were instances where both mishaps the reported cases. In about 20% of the remaining
took place the simulation had some serious prob- cases, the process chemistry and hardware efficiency
lems, but there were aspects of tower behavior that did not match what was true in an actual tower. Other
were not fully understood initially and the simula- items correctly modeling feeds, obtaining the true
tion helped to explain. vapor and liquid loads, reliably predicting the hydraulic
This article focuses on instances where problems behavior, and finding and squelching bugs in the simu-
were found in the simulation or where the simulation lations were found to be problematic, but to a lesser
was instrumental in identifying a previously misun- degree.
derstood problem (see table).
The cases were extracted from surveys in Refs. 1 Problems with VLE data and predictions
and 2 and have been updated with some recently re- Most case studies falling into this category involve
ported cases. The original numbering for each case close-boiling components. The problems can be with two
has been retained, so that the reader may easily locate chemicals of similar vapor pressure (e.g., hydrocarbons),
the cases in the previously referred to articles. or due to a non-ideality that pushes the volatilities of a
The number of examples presented here is by no pair close to a pinch. Correctly estimating non-idealities is
means a large enough sample for performing a sta- another trouble-spot when it comes to VLE predictions. A
tistical analysis of the main problems in trou- third dilemma is characterizing heavy components in
bleshooting distillation simulations. Nonetheless, crude-oil distillation. This is a key problem in simulating
the cases provide guidance on what to look for refinery vacuum towers. Few reports were made regarding
when troubleshooting a distillation simulation other situations. It seems that VLE prediction for pairs of
and what to watch out for when carrying out the components that have reasonably high volatilities, for ex-
next simulation. ample, methanol/ethanol, is not often troublesome.

52 www.cepmagazine.org October 2002 CEP

 Case Type of
No.-Ref. Column Brief Description
Section 1. How Good Are Your VLE Predictions?
102-3 Acetylene Solvent losses were far greater than design. Unsuccessful most challenging job. On the other
solvent/water extrapolation of VLE data was one of the causes. Increasing hand, we have seen a comparatively
stripper number of trays and raising reflux helped reduce losses. small number of problems in VLE
129-4 Chemical Column pressure was lowered from 100 to 30 mm Hg to predictions of medium- and high-
AMS (alpha improve separation, and valve trays were replaced by screen trays to volatility ideal systems.
methyl- match the capacity. Separation did not improve. Extrapolating
styrene)- VLE from 100 to 30 mm Hg gave optimistic expectations that
phenol did not materialize.
Matching plant data
A major problem appears to be
118-5 Chemical A laboratory error gave incorrect VLE data based on which a tower
super with 200 theoretical stages was built where over 300 stages obtaining a reliable, consistent set of
fractionator were required. With the 200 stages, product purity could not be plant data. It can be difficult to get
achieved. The plant was forced to rerun the purified material a correct numbers from flowmeters and
second time through the tower, effectively halving plant capacity. proper data from laboratory analyses.
140-6 Butadiene 1,2-butadiene is less volatile than 1,3-butadiene and leaves The data may need to be checked and
mostly through the bottom, but a commercial simulator rechecked. The troubleshooters
predicted it would leave through the top. Problem was due to
incorrect critical constants used in the equation of state.
prime tool is compiling mass, compo-
nent and energy balances, and check-
109-7 Trichloroethylene The concentration of TCE in CTC was higher than expected.
(TCE)/carbon A total reflux test showed that separation near the column
ing the results from laboratory analy-
tetrachloride bottom was worse than expected. Either VLE nonideality or ses to catch the lying flowmeter.
(CTC) decomposition of chlorinated ethanes at the reboiler A prime area where mismatches
temperature was the culprit. between simulated values and plant
123-8 Water/dichloro- DCM concentration in the wastewater was data have been reported is in chemi-
methane (DCM) very low. Based on ideal behavior, the vapor vent to atmosphere cal operations with two liquid phas-
from the wastewater storage tank would have contained little es. A close comparison of the tem-
DCM. Measured DCM at the vent was 27 mole %.
perature profile derived from a sim-
137-9 N-heptane/ Use of fractional-composition data from batch distillation ulation with that measured in the
toluene showed that the popular VLE choice for this system gave poorer
test system simulation of plant data than alternative VLE procedures. tower is an excellent tool for estab-
lishing the presence or absence of a
314, VLE error leads to mismatch between plant data and second liquid phase in a tower. For
315 simulation, see Section 2. instance, a simulation may not pre-
122-2 A 2% difference in relative volatility in a low-relative-volatility
dict that two phases will be present,
(1.1) system accounted for a difference of 50% in the tray while the measured temperatures
efficiency. The designers efficiency worked only with the may show that there are, in fact,
designers volatility; the operators efficiency worked only with two phases.
the operators volatility.
Another hurdle is in simulating
124-10 Ethylbenzene/ Identical column simulations using SIMSCI, HYSIM and ASPEN, refinery vacuum towers. The most
styrene all employing Soave-Redlich-Kwong (SRK) VLE, calculated
entirely different product purities. Reason was small differences
troublesome prediction to make is
in the critical temperature and pressure and in the acentric how much entrainment of liquid from
factor for styrene. the flash zone reaches the slop-wax
117-11 Refinery In five deep-cut towers, the design wash-oil flowrate was too small, draw and becomes part of the mea-
vacuum leading to drying, coking, high-pressure drop, loss in gas quality sured draw (overflash). This mea-
and short runlength. The drying resulted from simulations that surement is difficult, and is often es-
underestimatedthe fraction of wash oil vaporized. In all cases, tablished by a component balance.
inaccurate boiling point characterization of the heavy fractions
of the crude led to these underestimates. When this value is not measured, or
measured incorrectly, a simulation
130-12 Refinery Inaccurate TBP characterization of the heavy fractions of the
vacuum crude led to a wash-oil flowrate that was too small to prevent can yield an incorrect reflux rate,
coking in a deep-cut wash bed. Coke plugged the level bridle and which, when used in debottleneck-
draw nozzle on the slop-wax collector tray. Unable to drain, slop ing, can cause the wash bed to dry
wax was re-entrained into the wash bed. and coke up.

These findings pretty much match the authors experi- Graphical techniques
ence. The largest VLE problems seen are with the distilla- Three key graphical techniques that can shed light on
tion of close-boilers, whether due to their vapor pressures or what the columns and simulations are doing are the Mc-
their non-idealities. Many problems have surfaced with Cabe-Thiele and Hengstebeck diagrams, multicomponent-
VLE predictions for non-ideal pairs. In petroleum refining, distillation composition profiles, and, in azeotropic sys-
characterization of the vacuum tower feed and bottom is the tems, residue-curve maps. These methods permit visualiza-

CEP October 2002 www.cepmagazine.org 53

 Reactions and Separations

tion of the simulation and allow the

Case Type of
No.-Ref. Column Brief Description
engineer to use his or her judgment
in interpreting the results.
Section 2. Does Your Distillation Simulation Match Plant Data?
These diagrams are not drawn
General from scratch when troubleshooting
315-13 Aromatics Diagnosis based on the initial simulation was control instability. simulations. They are simple plots of
A simulation reality check against plant data exposed needs for the composition profiles generated
better energy-balance data, a low reflux test, VLE review for one by the simulation. The methods for
pair, and a surface-temperature survey. Once modified to
adequately reflect plant data, the simulation pointed to a problem plotting these diagrams for distilla-
of unexpectedly low tray efficiency. The corrective action for a tion troubleshooting are discussed
forthcoming revamp became improving trays, not controls. elsewhere (16, 26).
311-14, 16 Olefins Two different simulation models matched plant data well.
demethamizer Both suggested efficient packing in the lower sections. One model Incorrect or puzzling
suggested efficient, the other inefficient packings in the upper chemistry
sections. Plant logs of the temperature-reflux dependence proved
that the model predicting poor upper efficiency was correct. A In chemical towers, reactions
revamp based on this model succeeded; had the high efficiency such as decomposition, polymeriza-
model been used, the revamp would not have met is goals. tion or hydrolysis are often unac-
312-15 Olefins A simulation based on a set of tower readings led to theories for counted for in a simulation. Or,
water explaining liquid carryover from the top. A detailed test invalidated sometimes, a component that is be-
quench the simulation and theories. A discrepancy between data and lieved to be present in one chemical
simulation, initially attributed to an incorrect temperature
measurement, was proven in the tests to be due to error in flow
form turns out to be in another. In ei-
measurement. This completely changed the explanation for the ther case, the resulting separation
carryover. will differ from what the computer
314-16 Stabilizer To develop a simulation for revamp, column was tested at high simulates. When dealing with unsta-
and low reflux. Low reflux data matched the simulation well, high ble chemicals, such as some nitro
reflux data matched poorly. A Hengstebeck diagram compounds, this can lead to exother-
explained the mismatch in terms of a VLE inaccuracy.
mic decompositions and explosions.
329-17 Refinery, Basing a simulation on ASTM D86 gave optimistic tray efficiency There are also cases in which the
depentanizer and a misleading simulation. Matching simulated-to-measured chemistry of a process is not well
bottom component analysis gave correct tray efficiency and good
simulation. understood. One of the best ways to
get a good simulation in these situa-
With Second Liquid Phase
tions is to first run the chemicals
1426-18 Chemical Replacement of tower by a larger one caused instability, reduced through a mini-plant, as recom-
solvent capacity and yielded high solvent losses. The reason was refluxing of mended by Ruffert (27). In quite a
dehydration by water in the cyclohexane entrainer. An undersized condensate line few cases of reactive systems in
azeotropic caused condensate buildup in the condenser all the way to its
distillation mid-point vent, from where it drained directly into the cyclohexane which pilot work was carried out by
side of the decanter. Matching simulated temperature profile to a client, undertaking a through engi-
plant data revealed excess water in the hydrocarbon reflux, pointing neering-based understanding of the
to a decanter malfunction. chemistry led to redesign of a pro-
1279-18 Chemical The column operated normally at close to maximum capacity until it cess that would otherwise would not
monomer suddenly became unstable at high rates. Cause was a crack in the have worked well into one that was
and decanter baffle plate that allowed water into the refluxed organic
water phase. The reflux water generated second liquid phase and,
trouble-free.
separation therefore, temperature instability on the trays. Matching simulated
from acid temperature profile to plant data revealed excess water in the Inefficient efficiency
15 ft I.D. organic reflux, pointing to a decanter malfunction. estimates
877 -18 Chemical Acid recovery was poor due to a malfunctioning collector that The table reveals no clear trend
acid recovery collected liquid from an internal condenser, splitting it into a heavy- regarding estimating efficiency. The
from organics, phase reflux and a light-phase acid product. The phase-separation
packed tower overflow weir in the collector was too tall, and was installed in an incorrect
author has found that in established
location, and the product draw-nozzle was undersized. All led to processes, such as the separation of
poor decanting. Once these were corrected, the tower operated normally. benzene from toluene, ethanol from
Matching simulated temperature profile to plant data showed the water or ethane from propane, esti-
presence of two liquid phases where only one was expected,
pointing to a malfunction in split between the heavy-phase reflux
mating the efficiency is quite trou-
and light-phase acid product. ble-free for conventional trays and
packings. Problems arise when devi-
ating from these. For instance, in a

54 www.cepmagazine.org October 2002 CEP

first-of-a-kind process, efficien-
cy prediction can be an issue. Case Type of
No.-Ref. Column Brief Description
This is tackled by running pilot
tests, if schedules permit and Section 2. Does Your Distillation Simulation Match Plant Data? (Continued)
the economics favor, or by Refinery Vacuum Tower Wash Sections
gross oversizing, if the column
218-19 Refinery The design wash-oil flow was too small, leading to coking of the
costs are low or if the schedule vacuum wash bed. This resulted from a single-tower simulation model
is tight. Likewise, efficiency predicting low-wash dryout ratios. Segmenting the simulation
predictions run into trouble model into a number of flash units with recycles gave the correct
when engineers are on the steep dryout ratio, requiring triple the previous wash rate. The revised
simulation model correctly predicted plant data.
portion of the learning curve for
a new device. 318-20 Refinery Following replacement of trays by grid in the wash zone, the column
vacuum experienced chronic coking leading to high-pressure drop, reduced
gas oil yield, and high metals content of gas oil. The design allowed
Feed entry for little vaporization in the wash bed. In reality, all the wash oil
A correct representation of supplied vaporized and the bed dried up. Problem solved by
the feed inlet is crucial if the redesigning the spray header for 34 times the original wash rate.
number of stages between the
320-21 Refinery Wash-bed coked within 6 mo following replacement of low-efficiency
feed and the first drawoff is vacuum by high-efficiency packings. Asphaltene balance showed that
small, especially if it is only practically all the overflash was entrainment. The efficient packing
one or two. One of the two vaporized more wash liquid, causing unwetting and coking in the
cases reported here happened in lower packings.
a refinery vacuum tower in Have You Used Graphical Techniques to Troubleshoot Simulations?
which the first major product 313-16, 22 Olefins Addition of an interreboiler caused column design to approach a
exited the tower between 0.5 C2 splitter pinch. Pinch was undetected by a simulation; the simulation
and 2 stages above the feed. converged and worked well. Pinch detected by a McCabe-Thiele
diagram.
The other happened with
sponge oil returning to the 204-23 Refinery The column feed was rich (72%) in butane. A few degrees of extra
debutanizer preheat caused a large increase in feed vaporization, accompanied
main fractionator in a refinery by a large drop in stripping vapor rate. This increased butane in
catalytic-cracking unit immedi- bottom product. The problem was solved by controlling the flow of
ately above a product draw. steam to the preheater.
The feed entry issue is not 317-24 Refinery Isobutane in bottom was 34 times the design, and column capacity
unique to refining. It can be alky De- was restricted. Excessive subcooling of feed overloaded the bottom
more severe in chemical towers, isobutanizer, section. Also, a few trays above the feed had low liquid rates and
especially if some of the chemi- 8-ft I.D ., could have been blowing. Several other problems were identified.
valve trays Problem solved by adding a feed preheater, installing anti-jump
cals react in the vapor phase baffles in the lower trays, and adding picket-fence weirs to the
and not in the liquid state. Entry low-liquid-rate trays.
of the feed into the vapor space 1559-5 Chemical Feed to a two-column train fluctuated. Top purity of the first tower
may give completely different was held constant, which concentrated the disturbances in the feed
results than entry onto the tray to the second tower, destabilizing this tower. To overcome, a surge
or downcomer liquid. drum was added with 5-7 days residence time. Heat losses from the
drum caused a 50F drop in feed temperature to the second column,
aggravating its reboiler limitation, which, in turn, necessitated a
Vapor and liquid loadings reduction in reflux ratio and, therefore, a lower purity of the top
Calculated vapor and liquid product.
loadings are the basis for all of 210-16, 25 Chemical A vapor-side product impurity content was 10% (vs. design value of 1%),
the hydraulic evaluations of due to a non-forgiving concentration profile. Over the eight design stages
trays, packings and tower inter- in the bottom bed, the concentration rose from 30% at the bottom to 50%
four stages below the side draw, then dipped to 1% at the side draw.
nals. Incorrect loadings mean A miss by 12 stages would bring the concentration to 10%.
that tower-capacity estimates
will be incorrect. 212-26 Azeotropic Components A, B and C: Column separated a minimum-boiling AC
column azeotrope from a heavy C. Small amounts of light-boiler B (lighter
Usually, the hydraulic evalu- than the azeotrope) escaped in the top product. Changes in the
ation of a section of tower is reactor produced much more B in the feed. The light B was expected
based on the highest vapor and to go up, but much of it ended up in the bottom. Reason was the
liquid loadings in that section. formation of a much lighter AB azeotrope that distilled up, leaving a
BC mix in the bottom.
These are derived from the sim-
ulation. The cases in this sec-

CEP October 2002 www.cepmagazine.org 55

 Reactions and Separations

tion of the table show that a major

Case Type of weakness is that subcooling of reflux
No.-Ref. Column Brief Description and feeds is not always properly al-
Section 2. Does Your Distillation Simulation Match Plant Data? (Continued) lowed for in estimating the tower
Refinery Vacuum Tower Wash Sections
loadings. Further, the presentation of
liquid and vapor in the simulation
213-26 Chemicals Components A, B and C: Light product B was separated from heavy
azeotropic reactants A and C. With a feed rich in A and lean in B, pure B could
output is not generally user-friendly,
column not be produced due to minimum-boiling A/C and A/B azeotropes, and may often conceal higher vapor
the latter boiling 1C less than B. Problem solved by a column and liquid loads under superheated or
concentrating B (with some A and C) at the top, bottom being an A/C subcooled conditions.
mix. C was added to the B concentrate en route to the next column,
where it drew the A to the bottom (as an A/C mix), leaving B at the top.
Trusting simulator hydraulic
214-26 Freon -22 Column separated a light HF/HCI/R22 mix from heavy HF recycle to predictions
(R22) the reactor. Two simulations, identical except for initial value
Reflux column differences of 0.15% HF, gave completely different results. The only amazing result in the
This difference shifted the column across a distillation boundary, reported cases is that only two cases
giving completely different end points. involved problems with a simula-
tions hydraulic predictions. Mis-
220-27 Acetic acid/acetic C9 bottoms separated from ternary azeotrope that formed two liquid
anhydride/C9 phases. Decanter acid/anhydride was distilled in second tower to leading hydraulic predictions from
alkane remove C9. Residue curve map showed that when the feed is low on simulators is one of the major trou-
anhydride, there is no phase split in the decanter and the process ble spots, together with VLE data
fails. Cure was diverting low-anhydride feeds away from the towers. matching plant data and the need for
Section 4. Are Your Chemistry and/or Process Sequence Correct? scouting out problems via graphical
110-7 Absorption of HF HF absorption was poor. HF escaping in the column overhead techniques. This was stated in our
from HCl gas by destroyed the downstream glass plant. The cause was that most of previous work (16). Case 514, for
wash with the "HF" in the feed was in the form of carbonyl fluoride. This example, is one of about a few dozen
aqueous HCl component was sparingly soluble in water, but hydrolyzed slowly
to HF.
that we have seen in which simula-
tions gave optimistic capacity pre-
141-28 Mini-plant Distillate from first two towers was all the phenol. Distillate from
phenol and thirdshould have been phenol-free reactant, but contained 1.5%
dictions with packings.
reactant recovery, phenol,formed by a previously unknown cracking reaction of the It is difficult to understand why
3-tower train high boilers at the bottom. Solved by switching process sequence, only two cases were reported here
so that high boilers are removed in the second tower and reactant (both by the author!). Many engi-
separated from phenol in the third. Easy to switch in a mini-plant,
almost impossible once a full-scale plant is built.
neers regard it as a given that a sig-
nificant number of the hydraulic
112-17 O-Nitro-toluene The residues were held at 150F and air admitted. A previously predictions from simulators are un-
recovery unknown exotherm set in, causing an explosion.
trustworthy. Others, especially those
134-30 Solvent/ Column separating the reaction solvent and separation solvent from in high-tech companies, often prefer
residue batch residue experienced excessive solvent losses to residue. Change in
still, vacuum the upstream reactor, and using the same solvent for both reaction using proprietary prediction meth-
and separation, reduced feed inconsistencies, permitted semi-batch ods to those cached in the simula-
operation, and reduced solvent losses. tions, and have little incentive to
135-31 Pharm- Changing plant operation from a single to multiple distillation take the simulators hydraulic
aceutical, process rendered the residue more thermally unstable. This led to predictions seriously.
batch distillation an unstable mixture in a reactor, which exothermically decomposed, The most troublesome hydraulic
caught fire and exploded, injuring one person.
and efficiency predictions are for
Section 5. How Good is Your Efficiency Estimate? packed towers. There have been prob-
307-7 Desorption of Column failed to achieve design separation. Design efficiency was lems with optimistic predictions made
methanol, predicted from air humidification and oxygen stripping studies in a by simulation software and by using a
acetone and single-plate laboratory column. Wall and downpipe mass-transfer
ammonia from enhanced efficiency in the lab column. This led to optimistic
vendors program. When it comes to
water, using air efficiency predictions. trays, jet and downcomer backup-
flooding predictions, at least some
308-32 Isopropyl The 2-ft azeotropic distillation column using benzene and IPE
alcohol/water entrainer with metal-mesh packing achieved 12-in. HETP (height methods in simulators are reasonable,
azeotropic equivalent of a theoretical plate). This was considerably higher than the even good. Simulator predictions are
distillation using design HETP. Column HETP was scaled up from small-diameter columns far less satisfying for downcomer
benzene and IPE that had good and frequent liquid distribution and redistribution. choke flooding, which normally limits
entrainer
high-pressure and foaming systems,
as well as predicting tray efficiency.

56 www.cepmagazine.org October 2002 CEP

Bugs in the simulator
Fortunately, this is not a major issue, Case Type of
No.-Ref. Column Brief Description
but it is encountered from time to time
especially when simulation vendors Section 5. How Good is Your Efficiency Estimate?(Continued)
come out with an upgraded version of 302-3 Acetylene Some causes of excessive solvent losses were an incorrect efficiency
their programs. Always be on the look- solvent/water estimate and excessive entrainment. Performance was improved by
out for bugs. A bug in a VLE method in stripper adding trays and mist-elimination pads under top section trays, and
raising reflux.
one simulator disappeared in the next
upgrade, but returned in the following 304-4 Refinery Tray efficiency was lower than expected. Column contained valve
vacuum trays and operated at low liquid loads and with wide variations in
one. Corrective action calls for using an vapor loads.
independent means of verifying equilib-
Section 6. Incorrect Simulation of Feed Entry May Give Misleading Predictions
rium values and verifying the simulated
mass, component and heat balances. 321-33 Refinery, Following replacement of trays by structured packings, LCO/sponge-
FCC main oil draw temperature dropped 60F, LCO product contained 5%
fractionator, more gasoline, and the LCO stripper stopped stripping. Reason was
Lessons learned several towers that the two trays between the sponge-oil return and the LCO draw
Examining these cases reveals were eliminated. Gasoline-rich returned sponge-oil mixed with tower
seven key items that require vigil and liquid to form the LCO product. Problem minimized by minimizing
sponge oil. In one case, solution was returning sponge oil below the
attentiveness: LCO draw, generating a pumpdown.
1. The major issues affecting the
Section 7. Have You Specified the Correct Vapor and Liquid Loadings?
validity of distillation simulations are
getting good VLE data, having the 316-34 Olefins After replacing trays by packing, column efficiency fell, incurring high
simulation match plant data, and demethanizer ethylene losses. Gamma scans showed poor liquid distribution in the
upper two beds, flooding in the third, and poor vapor distribution
using graphical techniques to trou- in the bottom bed. The design made no allowance for vapor
bleshoot simulations. Another key condensation by the highly subcooled (70F) feeds. This overloaded
issue is obtaining good hydraulic pre- distributor capacities. Some improvement achieved by rerouting
dictions from the simulation. some of the cold main feed to an upper bed.
2. Providing the correct chemistry 330-12 Petrochemical Tower flooded 510% below design because additional vapor and
and the correct tray or packing effi- liquid traffic induced by 100F reflux subcooling was not accounted
for in the internals design. Solved by using bubble-point reflux.
ciency is needed to ensure the validity
of the tower simulation, but is less 306-23 Refinery The design vapor rate in the slurry section of a FCC fractionator did
fluid catalytic not allow for vaporization that occurs when a bottom feed with
troublesome than the factors men- cracker 300F superheat contacts column liquid. Column therefore
tioned above. Modeling feeds, pre- fractionator prematurely flooded. Problem solved by injecting subcooled quench
dicting vapor and liquid loads, and liquid to desuperheat the feed. At a later stage, subcooled quench
detecting and correcting simulation was replaced by a lighter liquid that vaporized, and premature
flooding reoccurred.
bugs cause trouble in tower simula-
tions, but to a lesser extent. Section 8. Simulator Hydraulic Predictions: To Trust or Not to Trust?
3. VLE predictions from commer- 514-16 Refinery 2-in. Pall rings were replaced by 3-in. modern random packings.
cial simulations are most trouble- vacuum Expected capacity increase was 30% but only 17% materialized. Both
some with close-boiling compo- the default and suppliers options in a commercial computer
simulation were optimistic, leading to the high expectation.
nents, non-ideal systems, or heavy-
component characterization in 515-16 High pressure A commercial simulator gave optimistic prediction of packing
capacity because it allowed extrapolation of a good correlation well
crude-oil distillation. Predictions for beyond its applicability limits.
components that have medium or
Section 9. Bug in Simulation
high relative volatilities and no
major non-idealities usually do not 326-35 Specialty Well-known commercial simulations with successful convergence
chemical and no error messages had erroneous energy balances on all three
cause a problem. towers. Cause was a bug in the default convergence software. Repeat
4. The major hurdle in matching with an alternative convergence procedure gave valid mass and energy
a simulation with plant data is ob- balances. Using the original simulation, all three reboilers would have
taining a reliable, consistent set of been grossly undersized and tower feed grossly mislocated.
the plant data. Getting correct num-
bers from flowmeters and laborato-
ry analyses is the major headache. Specific issues re- 5. The key graphical techniques invaluable for trou-
ported here are situations when a second liquid phase is bleshooting simulations are the McCabe-Thiele and Heng-
present, and when simulating the wash sections of refin- stebeck diagrams, multicomponent-distillation composition-
ery vacuum towers. profiles, and, in azeotropic systems, residue-curve maps.

CEP October 2002 www.cepmagazine.org 57

Reactions and Separations

HENRY Z. KISTER is director of fractionation technology at Fluor Daniel (One

6. In chemical towers, problems can arise when reac- Fluor Daniel Dr., Aliso Viejo, CA 92698; Phone: (949) 349-4679; Fax: (949)
tions are not properly accounted for in the simulation 349-2898; E-mail: [email protected]). He has over 25 years of
and/or when a component believed to be in one physical experience in design, troubleshooting, revamping, field consulting, control
and startup of fractionation processes and equipment. Previously, he was
form turns out to be in another. Brown & Roots staff consultant on fractionation, and prior to that he
7. Estimating tray and packing efficiencies is not a worked for ICI Australia and Fractionation Research Inc. (FRI). He is the
major issue for established processes operating in conven- author of textbooks Distillation Design and Distillation Operation, as
tional hardware. Most of the prediction issues arise when well as over 60 published technical articles, and has taught the IChemE-
sponsored Practical Distillation Technology course 240 times. He
simulating new systems or hardware. CEP
obtained his BE and ME degrees from the Univ. of New South Wales in
Australia. He is a Fellow of IChemE, a member of AIChE, and serves on the
FRI Technical Advisory and Design Practices committees.

17. Kister, H. Z., et al., Sensitivity Analysis is Key to Successful DC5

Literature Cited
Simulation, Hydrocarb. Proc., p.124 (Oct. 1998).
1. Kister, H. Z., Are Column Malfunctions Becoming Extinct or 18. Opong, S., and D. R. Short, Troubleshooting Columns Using
Will they Persist in the 21st Century?, Trans. IChemE, 75, Part A, Steady State Models, in Distillation: Horizons for the New Millen-
p. 563 (Sept. 1997). nium, in Topical Conference Preprints, AIChE Spring National
2. Kister, H. Z., Distillation Operation, McGraw-Hill, New York Meeting, Houston, TX, p. 129 (Mar. 1418,1999).
(1990). 19. Golden, S. W., et al., Improved Flow Topology for Petroleum Re-
3. Martin, H. W., Scale-up Problems in a Solvent-Water Fractiona- finery Crude Vacuum Distillation Simulation, 44th Canadian Chem.
tor, Chem. Eng. Progress, 60 (10), p. 50 (Oct. 1964). Eng. Conference, Calgary, Alberta, Canada (Oct. 25, 1995).
4. Guy, J. L., and J. A. Bonilla, Case History of a Retrayed Column: 20. Golden, S. W., et al., Refinery Vacuum Column Troubleshooting,
Troubleshooting Techniques and Methods, paper presented at the paper presented at AIChE Spring National Meeting, New Orleans,
AIChE Spring National Meeting, New Orleans, LA (Mar. 29Apr. 2, LA (Mar. 31, 1993).
1992). 21. Golden, S. W., et al., Refinery Analytical Techniques Optimize
5. Sloley, A. W., et al., Why Towers Do Not Work, paper presented at Unit Performance, Hydrocarb. Proc., p. 85 (Nov. 1995).
AIChE Spring National Meeting, Houston, TX (Mar. 2024, 1995). 22. Kister, H. Z., et al., Problems and Solutions in Demethanizers with
6. Moura, C. A. D., and H. P. Carneivo, Common Difficulties in the Interreboilers, in Proceedings of the 8th Ethylene Producers Con-
Use of Process Simulators, B. Tech. Petrobras, 34 (3/4), (Jul./Dec. ference, New Orleans, LA (1996).
1991), quoted in R. Agrawal et al., Uncovering the Realities of 23. Lieberman, N. P., Troubleshooting Process Operations, 3rd ed.,
Simulation, Chem. Eng. Progress, 97 (5), p. 42, (May 2001). PennWell Books, Tulsa, OK (1991).
7. Rose, L. M., Distillation Design in Practice, Elsevier, Amsterdam, 24. Sloley, A. W., and S. W. Golden, Analysis Key to Correcting
The Netherlands (1985). Debutanizer Design Flows, Oil & Gas J., p. 50 (Feb. 8, 1993).
8. Staggs, D. W., The Impact of Non-Ideal Vapor/Liquid Behavior on 25. Kister, H. Z., et al., Improve Vacuum-Tower Performance, Chem.
Solvent Emissions, paper presented at AIChE Spring National Eng. Progress, 92 (9), p. 36, (Sept. 1996).
Meeting, Houston,TX (Mar. 2024, 1995). 26. Short, D. G. R., Using Residue Maps for Solving Separation Prob-
9. Kalthod, V. G., et al., Distillation Column Performance Testing: lems, paper presented at AIChE Spring National Meeting, Houston,
Continuous and Batch Approaches, in Preprints of the Topical TX (Mar. 913, 1997).
Conference on Separation Science and Technologies, Part I p. 225, 27. Partin, L. R., Use Graphical Techniques to Improve Process Anal-
AIChE Annual Meeting, Los Angeles, CA (Nov. 1719, 1997). ysis, Chem. Eng. Progress, 89 (1), p. 43 (Jan. 1993).
10. Sadeq, J., et al., Anomalous Results from Process Simulations, 28. Ruffert, D. I., The Significance of Experiments for the Design of
paper presented at AIChE Annual Meeting, Miami Beach, FL (Nov. New Distillation Column Sequences, in Distillation 2001: Fron-
1995). tiers in a New Millennium, in Proceedings of the Topical Confer-
11. Golden, S. W., et al., Feed Characterization and Deepcut Vacuum ences, AIChE Spring National Meeting, Houston, TX, p. 133 (Apr.
Columns: Simulation and Design, paper presented at AIChE Spring 2226, 2001).
National Meeting, Houston, TX (Mar. 2024, 1995). 29. Miscellaneous Case Histories, in Fire Protection Manual, Vol. 2,
12. Sloley, A.W., et al., Troubleshooting Practice in the Refinery, C. H. Vervalin, ed., Gulf Publishing, Houston, TX, p. 29 (1981).
paper presented at AIChE Spring National Meeting, Houston, TX 30. Williams, J. A., Optimize Distillation System Revamps, Chem.
(Apr. 2001). Eng. Progress, 94 (3), p. 23 (Mar. 1998).
13. Kister, H. Z., et al., Does your Distillation Simulation Reflect the 31. HSA Criticizes Hickson, The Chem. Engineer, p. 5 (Nov. 10,
Real World?, Hydrocarb. Proc., p. 103 (Aug. 1997). 1994).
14. Kister, H. Z., et al., Debottleneck and Performance of a Packed 32. APV DH-682, Distillation Handbook, 2d ed., APV, Chicago, IL
Demethanizer, in Proceedings of the 4th Ethylene Producers Con- (no date of publication given).
ference, New Orleans, LA, p. 283 (1992). 33. Golden, S. W., Case Studies Reveal Common Design, Equipment Er-
15. Kister, H. Z., et al., Troubleshooting a Water Quench Tower, in rors in Revamps, Oil & Gas J., p. 62, (Apr. 7, 1997 and Apr. 14, 1997).
Proceedings of the 7th Ethylene Producers Conference, Houston, 34. Freeman, L., and J. D. Bowman, Use of Column Scanning to
TX (1995). Troubleshoot Demethanizer Operation, paper presented at AIChE
16. Kister, H. Z., Troubleshooting Distillation Simulations, Chem Spring National Meeting, Houston, TX (Mar. 31, 1993).
Eng Progress, 91 (6), p. 63 (June 1995). 35. Le, N. D., et al., Doublecheck Your Process Simulations, Chem.
Eng. Progress, 96 (5), p. 51, (May 2000).

58 www.cepmagazine.org October 2002 CEP