OPTIMIZING PUMPING SYSTEMS TO MINIMIZE FIRST OR LIFE-CYCLE COST

by
Judy Hodgson
Pump Consultant
E. I. du Pont de Nemours
Wilmington, Delaware

and

Trey Walters
President and Director of Software Development
Applied Flow Technology Corporation
Colorado Springs, Colorado

The IntelliFlow® technology featured in this paper has since migrated from AFT Mercury and AFT
Titan to become the Automated Network Sizing (ANS) Module for AFT Fathom™ and AFT Arrow™.
This module utilizes similar methods to provide the cost minimization as presented in this paper.
 Reproduced with permission of the Turbomachinery Laboratory (http://
turbolab.tamu.edu). From Proceedings of the 19th International Pump Users
Symposium, Turbomachinery Laboratory, Texas A&M University,
College Station, Texas, pp. 1-8, Copyright 2002.
 OPTIMIZING PUMPING SYSTEMS TO MINIMIZE FIRST OR LIFE-CYCLE COST
by
Judy Hodgson
Pump Consultant
E. I. du Pont de Nemours
Wilmington, Delaware
and
Trey Walters
President and Director of Software Development
Applied Flow Technology Corporation
Colorado Springs, Colorado

Reproduced with permission of the Turbomachinery Laboratory (http://turbolab.tamu.edu). From Proceedings of the Nineteenth International
Pump Users Symposium, Turbomachinery Laboratory, Texas A&M University, College Station, Texas, pp. 1-8, Copyright 2002.

A new design concept is introduced called the optimal pumping

Judy Hodgson is a Pump Consultant in system operating point (OPSOP). In simple terms, the OPSOP uses
the Rotating Machinery Group in cost data to identify the optimum tradeoff in pipe, pump, and
Engineering at DuPont, in Wilmington, (optionally) energy costs for a system that may have one or more
Delaware. Her specialty is modeling and duty points. Using this information, a new and improved method of
analyzing pumping systems. She has been a pump sizing is described.
pump consultant since 1997. Prior to that, To establish benchmark comparisons for typical petrochemical
she had project, maintenance, and research pumping systems, these optimization methods were applied to four
and development assignments with DuPont. previously designed systems. With a modest amount of effort, first
Ms. Hodgson received her B.S. degree cost reductions were as much as 17 percent, and life-cycle cost
(Mechanical Engineering, 1991) from Penn reductions were as much as 72 percent (based on 10 years), with
State University. savings of over $100,000 in several cases.

INTRODUCTION
Trey Walters founded Applied Flow The potential cost and energy savings from pumping systems is
Technology Corporation (AFT) in 1993, in great. Recent studies have found that pumping systems account for
Woodland Park, Colorado. He is currently about 20 percent of world energy usage (Frenning, et al., 2001).
the President and Director of Software Efforts that minimize wasted energy in these systems would not
Development. He is responsible for com- only have substantial economic savings, but an equally important
mercial software development of new and environmental impact, as well.
existing pipe system modeling products. Mr. Although savings can be made by optimizing existing systems,
Walters’ development work at AFT has been the greatest opportunities are in systems yet to be built. The reason
in the areas of incompressible and com- being that in new designs the piping can be included as one of the
pressible pipe flow, waterhammer, and variables that the engineer can modify to optimize the system. In
pump system optimization. He is also involved in thermal/fluid large existing systems, it would be cost prohibitive to make a
system consulting and customer training. Previously he was a piping change.
Research Engineer for Babcock & Wilcox in Alliance, Ohio, in Unfortunately, pumping system design engineers work in an
steam/water system design, and a Senior Engineer with General environment where budget and schedule constraints limit their
Dynamics in San Diego, California, in cryogenic rocket design. He ability to optimize their designs using traditional methods. The
has 15 years of experience in thermal/fluid system engineering, number of variables in complex pumping systems makes such
and has published eight papers. optimization impractical, even with modern hydraulic analysis
Mr. Walters holds a BSME (1985) and MSME (1986), both from software. Most of the design engineer’s effort is focused on
the University of California, Santa Barbara. He is a registered ensuring the system will merely function properly.
Professional Engineer in the State of California. With the abundant opportunity for cost and energy reduction in
new pumping systems, the need exists for technologies that will
allow engineers to optimize pumping system designs to minimize
ABSTRACT cost and energy usage. The commercial software, AFT Mercury,
Numerical optimization methods offer a powerful new addresses this need.
technology for pump users when combined with pumping system
analysis software. Whether the design goal is to reduce first costs ANALYSIS VERSUS DESIGN
or life-cycle costs, this technology promises to significantly reduce Before discussing the potential of modern optimization
pumping system costs and energy usage. technology for the pumping industry, it is worth pausing to
Optimization methods work by automatically selecting pipe and underscore the difference between engineering analysis and
pump sizes to minimize cost. Design engineers define the engineering design. Engineering analysis involves the application of
constraints for the system, such as flowrate, NPSH margin, or fluid engineering formulas and calculation methods to predict the behavior
velocity. The optimization software then finds the combination of of a given system. The calculation methods might be applied in hand
pipe and pump sizes to minimize the cost while satisfying the calculations, spreadsheets, or modeling software. Such methods are
constraints. satisfactory for evaluating the performance of an existing system or
1
2 PROCEEDINGS OF THE 19TH INTERNATIONAL PUMP USERS SYMPOSIUM

design. However, applying engineering analysis to a new design is

problematic in that it cannot answer the real question asked by the
designer, “What design will best achieve the project goals?” There
are numerous possible design goals, among which are to minimize
cost, energy, or risk, or to maximize performance, safety, or Yes
reliability. With engineering analysis, the engineer must propose a
design, and then use analysis to evaluate the proposed design. Good
designs are typically defined as designs that function properly.
No
In contrast, true engineering design answers the real question
being asked. Rather than evaluating a proposed design, a design-
based method’s output will be the design itself. The input data will
be the design requirement to be satisfied.
OPTIMIZATION TECHNOLOGY OVERVIEW
Optimization technology consists of algorithms of mathematical Figure 1. Logical Structure of a Pumping System Optimizer.
procedures. The searching methods in these algorithms modify a
design in ways likely to offer improvements. Analysis software tools
do not lose their value in this context, but take on even greater value
by acting as a critical subroutine to the optimization algorithms.
The development of modern numerical optimization techniques
(hereafter referred to as “optimization”) was highly influenced by
the introduction of the digital computer. Such techniques started to
appear in the 1960s (Schmit, 1960). For various reasons,
engineering optimization has found its primary application in
structural design. Optimization is considered a mature technology
and is included in many leading structural finite element packages.
In contrast, optimization of fluid and thermal systems is more rare,
but an area ripe for application. A comprehensive survey of modern
optimization technology was recently given by Vanderplaats
(1999a).
The definition of the objective function is critical in an
optimization analysis. The objective function is the function which
one attempts to minimize (such as cost) or maximize (such as
performance). The objective function depends on the value of the
design variables (such as pipe sizes), as well as other parameters
derived from the system response.
Although many different objective functions can be defined, the
authors will focus here on optimization of monetary cost, with the
goal being to minimize cost. All examples will be described in this
context. Figure 2. Historical Pump and Pipe Sizing Methods as Compared
Two popular methods that can be used to find the optimal design to Optimization Pumping System.
are gradient-based methods and genetic algorithm methods.
Additional discussion on how these methods work is given in the Traditional Approach
Appendix. At the top is what can be called the traditional approach. The
OPTIMIZATION OF PUMPING SYSTEMS traditional approach involves specifying the pipe sizes, equipment,
etc., and then selecting the pumps. Pumps and pipes are typically
In order to implement optimization methods in pumping system oversized, and operational problems with the system are dealt with
design, a highly reliable and computationally efficient pumping by installing pressure drop devices (e.g., orifices) after the system
system analysis software tool is required. This will be called the is installed. The ability to consider relative cost between the
hydraulic solver. The hydraulic solver needs to be highly reliable pumping and piping is impractical and therefore not considered. It
because it may be called tens of thousands of times by the is worth noting that even with the availability of cost effective
optimization engine (which will be called the optimizer). If the pumping system analysis software for over a decade, the traditional
hydraulic solver fails to solve any of the systems called for by the approach is still used by many large engineering organizations. It
optimizer, the optimization process itself will fail. Similarly, with is also worth noting that it is not uncommon for pumping systems
tens of thousands of systems to solve, the hydraulic solver must be designed with the traditional approach to have operational
computationally efficient or computer runtimes may be problems.
unacceptably long. In addition to the optimizer and hydraulic
solver, a third element that simplifies the optimization process is a Iterative Approach
user interface. The three elements are related as shown in Figure 1. The second approach in Figure 2 is what can be called the
When structured as in Figure 1, the hydraulic solver functions as iterative approach. Because this method uses system analysis
a subroutine, called repeatedly by the optimizer to evaluate a series software, it allows different pipe and pump configurations to be
of designs. The hydraulic solver consists of traditional pumping assessed quickly on the computer. The iterative approach is a
system analysis software. significant improvement over the traditional method, and allows
PUMP SIZING PROCESS— greater ability to verify that different operating cases can be
satisfied before the system is actually built. Even with modern
HISTORICAL PERSPECTIVE
modeling software, it remains impractical to perform any realistic
Figure 2 relates different approaches to sizing pumps for pump versus pipe cost optimization. Hence this is rarely done,
systems. Each will be discussed in turn. missing opportunity for cost and energy savings.
 OPTIMIZING PUMPING SYSTEMS TO MINIMIZE FIRST OR LIFE-CYCLE COST 3

Automated Approach optimizing for first cost or LCC and, if LCC, the design lifetime of
The third approach in Figure 2 is what can be called the the system. For this example the authors will optimize the system
automated approach. This method couples system analysis over a relatively long life cycle of 20 years. Details are given
software with modern optimization technology to actually search elsewhere (Applied Flow Technology, 2001, Chapter 12), and will
for a system that meets the design requirements while minimizing be summarized here.
the system cost. The automated approach accounts for the relative Standard cost data for the steel pipe (per length) and fittings of
costs between pumping and piping because cost data are the basis various sizes can be found in standard sources (e.g.,
for the optimization. Note that the automated approach allows the MEANSDATA, 2001). However, it is not possible to specify the
user to specify the criteria for optimization, meaning that the cost of the pumps because the pumps are not yet selected. To get
system can be optimized to minimize either first cost or life-cycle around this dilemma, one must create a “generic pump,” with cost
cost (LCC). This will be discussed subsequently. Also note that the data that vary with power. In the Figure 3 example, a cost data
automated approach allows the optimization to be performed over curve was constructed using MEANSDATA (2001) for pumps at
multiple operating cases, thus ensuring all operating cases are the design flowrate (300 gpm in this example) for power levels
adequately accounted for in the final optimum design. (varying with head) from one to 100 hp. This established the cost
for purchasing and installing the pumps as a function of power. The
DEFINING THE OBJECTIVE basis for approaching pump sizing in this way is discussed in the
literature (e.g., Darby, 2001). The step of creating a generic pump
Before performing an optimization, the objective must be can be referred to as Phase 1 of the pump sizing process. Figure 4
defined. In the context of this study, the objective is to minimize shows the generic pump cost data.
monetary cost. But what costs should be included? At first blush
one might respond that all costs should be included. If one includes
all costs, then one is designing the system based on LCCs.
However, very few companies actually do this today. Most
pumping systems are designed to minimize the first cost of the
system. When doing so, a number of important (even dominant)
cost items are neglected (Chemical Engineering Magazine, 2000;
Frenning, et al., 2001; Hovstadius, et al., 2000). When one designs
for first cost, one neglects operating (i.e., energy) costs and
maintenance costs and is in effect deciding that these costs will be
excluded from the objective. The objective function to be
minimized is thus a different function than when optimizing for the
life cycle.
One powerful aspect of using optimization is that either first cost Figure 3. Cooling Water System Schematic Optimization Example.
or LCC can be defined as the objective. Indeed, once the model is
set up, one can perform separate optimizations for each of these
objectives and assess the design differences. Should business
reasons lead one to opt for a design optimized for first cost,
optimization allows this to be an informed decision rather than one
made in ignorance.

DESIGN CONSTRAINTS
Another important aspect of optimization is design constraints.
Design constraints are derived from the design requirements. For
example, typical pumping system design constraints are minimum
flowrate, maximum pipe pressure, and adequate pump NPSH.
These are defined in the input area of the “User Interface” shown
in Figure 1.
All pumping systems will have design constraints, and the
optimization will thus be a constrained optimization. More
information on constrained optimization numerical techniques is
given by Vanderplaats (1999b).

FINDING THE PUMP “SWEET SPOT”

Figure 4. Assumed Cost for a 300 GPM Generic Pump Based on
Once the pipe system is laid out and the objective is defined, Power.
optimization can identify the pump “sweet spot.” This is the pump
operating point that, when combined with the optimized piping Maintenance and operation costs can be significant (Frenning, et
system, yields the absolute lowest cost. The technical term for this al., 2001), and can be included as recurring cost items in the
point is the optimal pumping system operating point (OPSOP). optimization. Maintenance costs were neglected for this example
for brevity’s sake. Operation costs were included by specifying the
Modeling Considerations cost for power (assumed to be $0.06/kW-hr) and the overall
Consider the cooling water system shown in Figure 3. This efficiency of the generic pumps (assumed to be 70 percent). In
system has two parallel (centrifugal) pumps; each controlled by a addition, the time value of money can be an important aspect of
flow control valve, which supply cooling water to three heat designing for LCC. This too can be included in the optimization,
exchangers. As part of the layout, the pipe lengths are already but was neglected here for brevity’s sake.
defined, as are the heat exchangers and their associated hydraulic
characteristics. The system will use steel pipe of standard sizes. Optimization Results
The question here is what size pumps and pipe sizes should be The system in Figure 3 was optimized for a 20-year life cycle
used. The answer, of course, depends on whether we are and it was found that the optimal pumps will each generate about
4 PROCEEDINGS OF THE 19TH INTERNATIONAL PUMP USERS SYMPOSIUM

40 ft of head, resulting in about 4.3 hp each. The optimization took Table 1. First Cost and Life-Cycle Cost Comparison for Optimized
about 15 seconds on a 933 MHz computer, and required 521 calls Cooling System.
to the hydraulic solver. The optimizer selected this design from
among 4.8 million possible designs. For reference, the LCC for the Optimized for: Material Installation Total Operation Total
optimal system was $116,500. (pipes, fittings
& pumps)
System ($.06/kW-hr) (system +
operation)
The 40 ft/4.3 hp pumps represent the optimal size that
First cost 20 yr 22,000 11,800 33,700 117,000 150,700
minimizes the cost over 20 years given the previously discussed
assumptions. This is the sweet spot (OPSOP) for the Figure 3 Life Cycle cost 20 yr 33,800 16,400 50,100 66,300 116,500

system. Once the optimum point is identified using a generic

pump, one can select actual pumps with actual cost data. With Table 1 shows that in order to achieve the reduction from
actual data, the system can be reoptimized for each candidate $150,700 to $116,500 over 20 years, one must spend about 50
pump. percent more initially (increased from $33,700 to $50,100).
The step of using actual pump performance and cost data can be By performing optimizations over different life-cycle periods,
referred to as Phase 2 of the pump sizing process. Assuming the one can see that the sweet spot is a function of the design lifetime
performance, efficiency, and cost do not significantly differ from (Figure 6). As the lifetime increases, lower power pumps (with cor-
the generic pump operating at the OPSOP, the reoptimized (i.e., respondingly larger pipe diameters) become a better choice.
Phase 2) systems should not differ significantly from the optimum
system found in Phase 1.
While it is unnecessary, it is nevertheless informative to look at
optimal pumping systems away from the OPSOP. This is shown in
Figure 5. A number of different optimization models were created
with the pump head fixed at levels from 35 to 70 ft. Each system
was optimized, and thus represented the best possible system for
pumps of that head level. Figure 5 confirms that the best possible
system should generate about 40 ft of head and cost about
$115,000. As mentioned previously, the optimal cost for Figure 3
was $116,500, consistent with the Figure 5 sweet spot.

Figure 6. Pump Sweet Spot for Different Life-Cycle Periods for the
Cooling System Example.

MULTIPLE OPERATING POINTS

It is not uncommon for pumping systems to have more than one
operating point. This can be handled by a pump system optimizer
as outlined in Figure 1. A pump optimal pumping system operating
point still exists. However, because there are multiple operating
points, the OPSOP becomes a composite operating point. The
composite point curve would still look like Figure 5, where the
pump power would be derived from the maximum required power
among all operating points.
WEIGHT OPTIMIZATION
To simplify the data gathering required to perform an
Figure 5. Optimized Systems for Pumps of Different Head Levels optimization analysis on pumping systems, rather than minimize
and Power Show Sweet Spot Locations. cost, one can minimize weight. Because weight increases with pipe
diameter as does cost, pipe weight tracks fairly closely with first
There is something else vitally important that can be learned cost. And pipe weight is much easier to estimate than cost.
from Figure 5. The best possible systems that can be designed for However, when optimizing for weight, monetary cost issues are
pumps sized higher (or lower) than the sweet spot (OPSOP) will ignored. Optimizing pumping systems for weight has important
yield significantly higher costs. For example, a pump sized at 70 ft advantages and disadvantages. These are discussed in Table 2.
will result in LCCs almost 30 percent more for the optimal system!
If the system using a 70-ft pump is not optimized, as is the case Table 2. Issues Encountered with Weight Optimization.
with a majority of today’s systems, the cost will be even higher.
This demonstrates the importance of concurrently sizing the pump Advantages No detailed cost data required
and system upfront when the design engineer has the most latitude. Fast and easy to implement
Comparison with First Cost Optimization Useful to minimize initial cost

Interestingly, when the Figure 3 system was optimized for first Disadvantages Only pipes are optimized
cost, the optimal pumps generated about 70 ft of head and required Cannot be used for optimal pump sizing
about 7.6 hp. This is the sweet spot for a pump optimized for first Cannot perform Life Cycle optimization
cost. The cost was slightly over $150,000, consistent with Figure
5. A detailed comparison of the results of the Figure 3 system for
REAL WORLD APPLICATIONS
initial cost and 20-year LCC is given by Applied Flow Technology
(2001, Chapter 12). The major cost categories are summarized in To evaluate the cost-saving ability of the software, a benchmark
Table 1. must be established with which to compare costs. A worthy
 OPTIMIZING PUMPING SYSTEMS TO MINIMIZE FIRST OR LIFE-CYCLE COST 5

benchmark must be rooted in the real world. The benchmarks for For this system, depicted in Figure 7, the tempered water pump
this study were actual pumping systems from various plants of a (“Recirc Pump”) receives its river water supply from the river water
large international chemical company. All these systems were booster pump (“P-54”), which in turn takes its water from the raw
originally designed and built using traditional design methods. water supply header. Only the tempered water system, which is
They are typical systems, not especially poorly designed or well circled in Figure 7, was evaluated for this study. The rest of the
designed. Because these are real systems, they have real system (i.e., the header, booster pump, and other users) was already
requirements and constraints. Therefore, each one has different in service; the project only included the tempered water system.
opportunities for improvement. These systems were specifically
chosen based on their type of opportunities in order to evaluate the
software’s ability to optimize them.
Four systems were evaluated. One system had a control valve
doing the turndown of the flow. This system was optimized with a
variable frequency drive. Another system had a gravity-fed flow.
The third system had material of construction tradeoffs to consider.
The fourth one had high maintenance costs due to an improperly
sized pump.
All four of these cases represent typical design issues found in
pumping systems in the petrochemical industry. The first case, the
one with the control valve, is extremely prevalent with the majority
of the industry’s systems controlled by control valves. The second
case with the gravity-fed flow is relatively unusual. Because of the
lack of experience with sizing such lines, this line was sized as a
pump-fed pipe—which was conservative. Being too conservative
is indeed a typical design issue the industry faces. The third case
has a choice of material for the pipe. The one choice is inexpensive Figure 7. Tempered Water System Schematic.
metal, but must use large diameter pipe to limit the corrosion
aggravated by velocity; the other choice is more expensive The original design had the temperature control valve
controlling the flow to the heat exchanger. Not only was the 4:1
material, but can use standard size pipe. Material tradeoff decisions
turndown expensive in terms of energy costs, but the 3 inch Monel®
are common in the industry. The last case study is all too
control valve was expensive in terms of first cost. Also, the
common—an improperly sized pump. In this case, the pump is
turndown forced the pump back on its pump curve, which
actually undersized rather than the more typical oversized pump
adversely affected the reliability of the pump and increased
that results from all the fudge factors added in by everyone maintenance costs. (Details regarding how the pump reliability is
associated with the design. Undersized or oversized, the affected by where the pump is operated on the pump curve will be
optimization process is the same. discussed in the IMPROPERLY SIZED PUMP section.)
To assess the software, each system as it was originally designed The optimization software did not find more cost-efficient piping;
(without the optimization software), was first evaluated by the it did, however, find that by replacing the control valve with a variable
software with the optimization feature turned off so as to establish frequency drive, the energy savings would be 71 percent (or $3875)
the benchmark cost of the system. Then the constraints were set over five years and again 71 percent (or $14,582) over 10 years. This
and the system was optimized. The two costs, “benchmark” and energy was saved because, instead of having the excess pressure that
“optimized,” were then compared. In every case, the software was the pump is generating dissipate across the control valve, the VFD
able to reduce the cost of the system, whether it was first cost, controlled the speed so the pump never generated the pressure in the
LCC, or both. first place. The savings in maintenance cost from running the pump
The following are the details of the evaluations. slower resulted in savings of 84 percent (or $24,890) over five years
and the same 84 percent (or $41,830) over 10 years. Even the first
Control Valve Versus VFD
cost would have been reduced, by $4700. This savings stems from the
The first system is a tempered water supply for a heat exchanger difference in price between the VFD and the control valve. The
in a chemical intermediates plant in the Northeastern United States. percent savings in the first cost of the material for the entire project,
The flowrate required to cool the heat exchanger not only depends on which includes the Monel® pipe but not the heat exchanger (which
the production rate, but which of two products is being formulated, was already in service before this project), would have been 17
and the temperature of the river water. As a result of these variables, percent ($4700). The percent savings in LCC for the project would
the flow demand had more than a 4:1 turndown, ranging from 80 to have been 54 percent ($38,267) for a five-year life cycle, and 62
400 gpm. The original system had a control valve making the percent ($61,113) for a 10-year life cycle. Refer to Table 4 for the
turndown; the optimized system has a variable frequency drive summary of the possible savings in first cost and a five-year LCC.
(VFD). Refer to Table 3 for the flow demand of the system, and Refer to Table 5 for the summary of savings using a 10-year life cycle.
corresponding control valve or VFD settings, in addition to the
distance the pump must operate from the best efficiency point (BEP). Table 4. First Cost and Five-year Life-Cycle Cost Comparison of
Tempered Water System’s Control Valve Option (Actual) Versus
Table 3. Flow Demand and Corresponding Control Valve or VFD (Optimized). (“Weighted” Cost Refers to Using the Duty
Variable Frequency Drive Settings Required to Attain Those Flows, Cycle Listed in Table 3 to Calculate the Costs.)
Plus the Distance the Pump must Operate from BEP.
First Weighted Weighted Cost of Piping Life
Cost Energy Maintenance Control Cost Cycle
Flow Rate Duty Cycle Control Valve dP Setting (psid) VFD RPM Setting Cost Cost Valve/VFD Cost
(gpm) (% of time) & Pump % of BEP & Pump % of BEP
Control valve $28,550 $12,260 $29,482 $11,000 $17,550 $70,292
400 10 1 * 86% 1750 * 87% VFD $23,850 $3,584 $4,592 $6,300 $17,550 $32,025
280 30 17 * 31% 1225 * 86% VFD system $4,700 $8,676 $24,890 $4,700 0 $38,267
savings
120 50 30 * 26% 508 * 90%
VFD % 17% 71% 84% 43% 0 54%
80 10 31 * 17% 315 * 95% savings
6 PROCEEDINGS OF THE 19TH INTERNATIONAL PUMP USERS SYMPOSIUM

Table 5. First Cost and 10-Year Life-Cycle Cost Comparison of velocities for ambient temperature sulfuric acid in carbon steel
Tempered Water System’s Control Valve Option (Actual) Versus pipe. With a velocity of 6 ft/s in the original design, the system
VFD (Optimized). (“Weighted” Cost Refers to Using the Duty incurs a high cost of maintenance for replacing the corroded pipe.
Cycle Listed in Table 3 to Calculate the Costs.) Besides replacing the existing carbon steel pipe with larger
diameter carbon steel pipe, another viable option is to replace it
First Weighted Weighted Cost of Piping Life
with stainless steel pipe. Stainless steel is not sensitive to the
Cost Energy Maintenance Control Cost Cycle velocity of sulfuric acid in the line. Refer to Figure 11 for the
Cost Cost Valve/VFD Cost sensitivity of various metals to the velocity of sulfuric acid.
Control valve $28,550 $20,604 $49,547 $11,000 $17,550 $98,701

VFD $23,850 $6,022 $7,717 $6,300 $17,550 $37,588

VFD system $4,700 $14,582 $41,830 $4,700 0 $61,113

savings

VFD % 17% 71% 84% 43% 0 62%

savings

Figure 9. Sulfuric Acid Unloading System Schematic.

Gravity-Fed Flow
The second system was a simple municipal wastewater transfer
system at a chemical manufacturing plant in Spain. Refer to Figure
8 for the schematic. The water was pumped from a tank up a hill to
a collection box that then gravity-fed to the municipal sewer on the
bottom of the other side of the hill. Upon optimization, the gravity-
feed line was found to be oversized. The design engineer used the
same size pipe to go up the hill as he did for flowing down it. But
the economics are different. For a pumped fluid, the size of the
pipe must be balanced between minimizing the cost of the pipe,
where smaller is better, versus the cost of the energy it takes to
pump through the line, where bigger is better. In a gravity-feed
line, there is no energy tradeoff. As long as the gravity head can
push the required flow through the pipe, the pipe can be quite
small. In this case, the 4000 ft of 18 inch PVC pipe could have
been 12 inch, with no increase in energy expenditure, while still
transporting the required flow. The optimum sized pipe would have
saved the project seven percent in first cost, or $124,955. Refer to
Table 6 for a breakdown of the pipe cost savings.
Figure 10. Graph of Corrosion Rate of Carbon Steel with Respect
to Velocity of Ambient Temperature 93 Percent Sulfuric Acid.
(Courtesy of Indiana Ordinance Study PE-13)

Figure 8. Municipal Wastewater System Schematic.

Table 6. Piping Cost Comparison of Municipal Wastewater System

(Includes Materials and Labor).

Original System Optimized System Savings ($) Savings (%)

$1,711,604 $1,586,649 $124,955 7%

Material of Construction
The third system was a 93 percent sulfuric acid unloading Figure 11. Graph of Corrosion Rate of Various Metals with Respect
station at a pigment plant in the Gulf Coast. Refer to Figure 9 for to Velocity of 95 Percent Sulfuric Acid at 120°F. (Courtesy of The
the schematic. The original design exceeded the recommended International Nickel Company, Inc.)
maximum velocity limit of 3 ft/s for carbon steel pipe. The
corrosion of carbon steel increases dramatically with the velocity The optimization software evaluated the existing system and
of sulfuric acid flowing in the pipe. Figure 10 depicts the determined whether it would have been more cost-effective to use
relationship between the velocity and rate of corrosion for various stainless steel or use large diameter carbon steel. For first cost, the
 OPTIMIZING PUMPING SYSTEMS TO MINIMIZE FIRST OR LIFE-CYCLE COST 7

least expensive option is, of course, the “disposable” pipe of the

existing system. This option is 15 and 75 percent cheaper than the
other options. As for LCC, for a five-year life cycle, the most cost-
effective design was the stainless steel system. This design would
have saved 35 percent, or $60,000, over the existing design, which
does not include the cost of lost production due to the downtime.
Refer to Tables 7 and 8 for the cost summary of the options for
five- and 10-year life cycles.

Table 7. First Cost and Five-Year Life-Cycle Cost Comparison of

93 Percent Sulfuric Acid Unloading System.

Pipe Option First Cost Maintenance Operational LCC Savings LCC % Savings
Cost Cost Over Existing Over Existing

Existing 6 ft/s c.s. $47,470 $96,788 $25,808 0 0

Larger diameter c.s. $54,438 $74,840 $5,546 $35,242 21%

Figure 13. The Effect of Distance from the BEP on Reliability.
(Courtesy of Barringer & Associates)
Stainless steel $84,185 0 $25,834 $60,047 35%

Table 8. First Cost and 10-Year Life-Cycle Cost Comparison of 93

Percent Sulfuric Acid Unloading System.

Pipe Option First Cost Maintenance Operational LCC Savings LCC % Savings
Cost Cost Over Existing Over Existing

Existing 6 ft/s c.s. $47,470 $162,661 $43,373 0 0

Larger diameter c.s. $54,438 $125,775 $9,320 $63,971 25%

Stainless steel $84,185 0 $43,416 $125,902 50%

It would be up to the business team to weigh first cost against

LCC, not to mention the importance of safety and uptime. How
ever the importance is placed, the software generates the data to
make the right decision.

Improperly Sized Pump

The final system was a molten sulfur transfer system at an acid
plant on the Gulf Coast. Figure 12 is the schematic of the system.
The actual pump installed was significantly undersized and running Figure 14. The Effect of Distance from the BEP on Pump
to the far right of the BEP of the pump curve. Using Barringer and Maintenance Costs. (Reference Pump: ANSI 3
4-10 Stainless
Associate’s chart (Barringer, 1997) relating the effect of distance Steel Single-Stage, Horizontal Pump, with Dual Pressurized Seal,
from the BEP to the reliability (Figure 13) and coupling that with Running at 1750 RPM in a Mildly Corrosive Service.)
the cost of labor and material for repair, Figure 14 was used to relate
the distance from BEP to the cost of maintenance. Table 9. First Cost, Five-Year, and 10-Year Life-Cycle Cost
Comparison of Molten Sulfur Transfer System.

First 5-Yr 5-Yr Savings 10-Yr 10-Yr Savings

Cost LCC $&% LCC $&%

Original pump $8,757 $105,116 --- $170,696 ---

Optimized pump $8,104 $31,424 $73,692 * 70% $47,295 $123,401 * 72%

• Based on real world systems as the benchmark, using

optimization software in place of traditional design techniques
results in significant cost savings for both first cost and LCC.
Figure 12. Molten Sulfur Transfer System Schematic. • Besides being easy to use, the software proved to be versatile
enough to manage different design issues commonly found in the
Installing a properly sized pump for this application would have petrochemical industry.
saved the project $73,692, or 70 percent in LCC with a five-year • This software has the potential to be a powerful tool in the effort
life cycle. For a 10-year life cycle, the savings would have been to move from first-cost-focused design to LCC-focused because no
$123,401, or 72 percent. Although the original pump was additional work is required to compare the two different designs
undersized, the optimum pump actually had a smaller casing than and their respective monetary tradeoffs.
the original and therefore was less expensive than the original. This
resulted in a small savings in first cost. The savings was $653 or APPENDIX
seven percent. Refer to Table 9 for the cost summaries. Following is a discussion of optimization methods that can be
used to optimize pumping systems.
CONCLUSIONS
Gradient-Based Methods
• Numerical optimization technology is a proven technology that
has been successfully applied in other design arenas that, until now, For typical engineering systems, gradient-based optimization
have not been utilized in pumping system design. methods are the most efficient by a significant margin. Gradient-
8 PROCEEDINGS OF THE 19TH INTERNATIONAL PUMP USERS SYMPOSIUM

based methods search for the optimum using the partial derivative Barringer, H. P., 1997, “Reliability Engineering Principles”
of the objective function (i.e., cost) with respect to each design Training Course, Slide 45, Barringer & Associates, Humble,
variable (i.e., pipe diameter). From the starting point (the initial Texas.
pipe sizes specified by the user), the gradient-based methods use Chemical Engineering Magazine, November 2000, “Analyzing
the partial derivatives to determine new pipe sizes that will cause Pump Life-Cycle Costs,” Chemical Week Associates.
the overall objective to decrease.
Gradient-based methods search the design space in such a way Darby, R., 2001, Chemical Engineering Fluid Mechanics, Second
as to meet all user design requirements (i.e., constraints). Designs Edition, New York, New York: Marcel Dekker, pp. 200-203.
that satisfy all constraints are called feasible designs. Designs that Frenning, L., et al., 2001, Pump Life Cycle Costs: A Guide to LCC
fail to satisfy one or more constraints are infeasible. Analysis for Pumping Systems, Hydraulic Institute and
In complex systems, the partial derivatives of the objective Europump, Parsippany, New Jersey.
function with respect to the design variables are frequently not
available from analytical relationships. Therefore, the derivatives Hovstadius, G., Erickson, R. B., and Tutterow, V., 2000, “Pumping
must be approximated by forward or central difference methods. System Life Cycle Costs, An Overlooked Opportunity?”
For those interested, the mathematics behind these methods is PumpLines, pp. 10-12.
discussed by Vanderplaats (1999b). Indiana Ordinance Study PE-13, 1958.
Genetic Algorithm Methods International Nickel Company, Inc., 1983, The Corrosion
Resistance of Nickel-Containing Alloys in Sulfuric Acid and
Genetic algorithm methods work by simulating so-called Related Compounds, Suffern, New York.
biological evolution. An initial population of designs is randomly
generated, and each design is evaluated by running the hydraulic MEANSDATA, 2001, R. S. Means Company, Inc., Kingston,
solver for that design. The resulting objective function value (i.e., Massachusetts.
cost) and constraints are used to determine which designs are Schmit, L. A., 1960, “Structural Design by Systematic Synthesis,”
“fittest” (i.e., more optimal—least costly), and the better designs Proc. Second Conference on Electronic Computation, ASCE,
are crossbred while the worse designs are eliminated. The New York, pp. 105-122.
population breeds for some specified number of generations or
until no further improvements are obtained. Vanderplaats, G. N., 1999a, “Structural Design Optimization—
Genetic algorithm methods are inherently discrete, and thus Status and Direction,” AIAA J. Aircraft, 13, (1), pp. 11-20.
have the advantage (for pipe system optimization) of only Vanderplaats, G. N., 1999b, Numerical Optimization Techniques
considering discrete system designs. They also have the advantage for Engineering Design, Third Edition, Vanderplaats Research
of being better suited to finding the global optimum, rather than a & Development, Inc., Colorado Springs, Colorado.
local minimum. A major disadvantage is that they are slower than
gradient-based methods. As the number of pipes increases, the ACKNOWLEDGEMENTS
performance significantly degrades as compared to gradient-based The authors would like to thank Jeffrey Olsen of Applied Flow
methods because they require many more calls to the hydraulic Technology and the staff at Vanderplaats Research and
solver. Development for their critical input to the technology underlying
The performance of genetic algorithm methods can be this paper. Tom Chen of DuPont was also a very valuable member
significantly improved when used in conjunction with gradient- of the team in helping analyze and optimize the systems.
based methods (Applied Flow Technology, 2001, Chapter 14).

REFERENCES
Applied Flow Technology, 2001, AFT Mercury 5.0 User’s Guide,
Woodland Park, Colorado.