Process Optimization

KMM 4902E

Chapter 14

Spring 2024
 Introduction
What is optimization?
• Process of improving an existing situation/device/system
• Such as a chemical process
• Making the best of anything
• Mathematical technique for finding a maximum or
minimum value of a function of several variables
subject to a set of constraints
 Basic Concepts
Decision (design) variables: Independent variables over
which engineers have some control
• Continuous variables: Temperature
• Discrete (integer) variables: Number of stages in column
Objective function: Mathematical function including
decision variables which reaches a minimum or maximum
• Search for maximum: Profit
• Search for minimum: Cost
There may be more than one objective function for a
given optimization problem.
 Basic Concepts
Constraints: Limitations on values of decision variables.
• May be linear or nonlinear
• May involve more than one decision variable
• Equality constraint: Constraint written as equality
involving one or more decision variables
– Reduce dimensionality (number of independent decision
variables) of optimization problem
• Inequality constraint: Constraint written as inequality
involving one or more decision variables
– Reduce (and often bound) search space of the decision
variables
 Basic Concepts
Minimize: f(x)
Subject to: h(x) = 0
g(x) ≥ 0
where x is a vector of n variables (x1, x2, …., xn)
h(x) is a vector of equations of dimension m1
g(x) is a vector of inequalities of dimension m2
 Basic Concepts
Local optimum: Point from which no small, allowable
change in decision variables in any direction will improve
the objective function.
Global optimum: Point at which the objective function is
the best for all allowable values of the decision variables.
There is no better acceptable solution.
 Basic Concepts
Linear programming: If objective function is linear in all
decision variables and all constraints are linear
Nonlinear programming: All other optimization problems
• Quadratic programming: If objective function is
second order in decision variables and constraints are
linear
• Mixed-integer linear/nonlinear programming
(MILP/MINLP): Optimization problems involving both
discrete and continuous decision variables
Many of constraints that we run into in chemical processes
are not linear, and the variables are often a mixture of
continuous and integer.
Example:
Evaluation of the optimal heat exchanger to use in order to heat a
stream from 30°C to 160°C.
• Continuous variables:
– Area of heat exchanger and temperature of process stream
• Integer variables:
– Use of low-, medium-, or high-pressure steam as heating
medium
• Constraints:
– Materials of construction depending nonlinearly on factors:
• Pressure
• Temperature
• Composition of process & utility streams
 Common Misconceptions
• Classical curve of annualized pumping cost versus pipe diameter.
• Annualized pumping cost includes:
• Annualized equipment (pump +
pipe) cost
• Power (operating) cost

Point of intersection of the curves of annual operating cost and

annualized capital cost is not the optimum
• Combined curve represents the total annualized cost
• Optimum: Pipe diameter at which
• Slope of combined cost curve is zero
• Second derivative is positive
 Common Misconceptions
• In reality, cost function would look something like figure below
• Only certain pipe diameters and pump sizes are standard
equipment.
• Other sizes could be produced, but only at much higher costs.
• Thus, only a few cost evaluations are needed — those at the
standard pipe sizes.
 Common Misconceptions
Optimum will usually be found at a point where the first derivatives
of the cost function are zero
• Even when the cost function is continuous and smooth, this is
seldom the case and should never be assumed.

• Nearly all problems of any

reasonable complexity have optima
along at least one constraint.
• Although there is a point of zero
slope (point A), the best design
(minimum annual cost) shown is at
point B.
 Common Misconceptions

One must not assume that the best solution

has been found when it has been bracketed.
• Three points calculated for
optimization of temperature of a flash
unit.
• If objective function is continuous and
smooth, and if there is indeed only one
minimum
• optimum lies between Points A & C
• To be certain that the optimum has been located correctly,
• More points should be evaluated between A and C,
• Using successive quadratic approximations
• Based on the best point achieved plus the closest point on
either side of it.
 Estimating Problem Difficulty
Characteristics of Easy and Difficult Optimization Problems
Easy Problems Difficult Problems
Few decision variables Many decision variables
Independent (uncorrelated) decision variables Correlated decision variables
Discrete decision variables Mixed discrete and continuous decision variables
Topological optimization first Parametric optimization first
Single process units Multiple, interrelated process units
Separate constraints for each decision variable Constraints involving several decision variables
Constraints are obvious Constraints are not obvious or become obvious only
after the optimization has begun
Single objective Multiple objectives
Objective function easy to quantify Objective function difficult to quantify
Linear objective function Highly nonlinear objective function
Smooth objective function Kinked objective function
No local optima Many deep, local optima
 Top-Down and Bottom-Up Strategies

• For any real process of any complexity,

• True “global” optimum can be difficult to find
• Before investing too much time on detailed calculations,
• Check the big picture

Strategy of looking at the big picture first

Followed by the detailed study
Top-down
 Strategies
Base Case:
• Goal of optimization: Improving the process
• Essential: Starting from a defined process
• For any optimization, need a starting point – this is our base case
• The base case may be any technically feasible process
• The further the base case is from optimum,
• The more optimization must be done
Choice of a base case :
• Choose the best available case
• Level of detail of the base case is a crucial decision
• Base case certainly need not be a completed design
• However, it is essential to include enough detail in the base case
 Objective Functions (OF)
• Most common objective functions have units of money
• An OF on operating and capital investment, e.g., EAOC
(equivalent annual operating cost) or NPV (net present value)
should be chosen with the aim of minimizing or maximizing the
function
• Some objective functions are not directly based on economics.
• For example, one may be asked to maximize production of
chemical B from the plant, etc.
• However, the objective function should be quantitative.
 Analysis of the Base Costs
• First analysis of the base costs:
• Determine targets of an idealized process
• Next: A form of Pareto analysis
• This analysis is based on the observation that, for most
problems, a large fraction of the objective function (e.g.,
80%) is due to a small fraction of the contributing factors
(e.g., 20%).
 Analysis of the Base Costs
Category Contribution to • For many chemical processes:
EAOC ($/yr)
• Raw material costs swamp all other
Raw materials 11,215,000
Raw material (at 100%
costs and are treated separately.
conversion) = $ 11,185,000 • Focus should not be on reducing
Target savings = $ 30,300
cost of raw materials, this will not
Medium-pressure steam 695,000
have a significant effect on
Towers and vessels 210,000
reducing the cost of producing.
Heat exchangers 170,000
Pumps (including 160,000 • Single largest operating cost is
electricity) medium-pressure steam.
Reactor 70,000
• Greater improvement in objective
Cooling water 31,000
function can be made by lowering
Waste water treatment 1,000
steam costs than can be obtained by
Total 12,552,000
lowering any other cost category.

Basically, this first analysis of the base case provides a target

for optimization and a road map to proceed to the solution.
 Decision Variables
• Even simple processes have tens if not hundreds of potential DVs
• Important to identify which of these has significant effect on OF.
• A Pareto analysis will identify the major costs,
• How those costs are affected by changes in operating variables
• Needs some preliminary investigation
 Decision Variables
• Usually, the single-pass
conversion in the reactor is an
important decision variable since
recycle rates are influenced
directly. R

VR C1

• If unreacted material leaves the F0A

z0A A B
kf D

kr xD
process, then decision variables
that reduce these streams by F

z
better recovery in the separations
sections (and higher conversion) VS

should be considered.
• High utility costs (steam and
B
electricity) warrant decision xB

variables that focus on recovering

heat and work.
 Topological Optimization
• Topological Optimization involves changes in the arrangement
(topology) of process equipment.

• During the design of a new process unit or the upgrading of an

existing unit, topological optimization should, in general, be
considered first.
• Topological changes usually have a large impact on the overall
profitability of the plant.
• Parametric optimization is easiest to interpret when the topology
of the flowsheet is fixed.

• Combination of both types of optimization strategies may have to

be employed simultaneously,
• But the major topological changes are still best handled early on
in the optimization process.
 Topological Optimization

Questions that should be addressed include:

• Can unwanted by-products be eliminated?
• Can equipment be eliminated or rearranged?
• Can alternative separation methods or reactor configurations be
employed?
• Can heat integration be improved?
 Elimination of Unwanted Nonhazardous By-Products
or Hazardous Waste Streams
• This issue should be addressed early on in the design process.
• Benefit of
• Obtaining 100% conversion of reactants
• With a 100% selectivity to the desired product
• This goal is never reached in practice, but approached through
suitable choices of
• Reaction mechanisms
• Reactor operation
• Catalyst
• A chemical engineer:
• May not be directly involved in the choice of reaction paths,
• May be asked to evaluate and optimize designs for using
alternative reactions in order to evaluate the optimum scheme.
 Elimination of Unwanted Nonhazardous By-Products
or Hazardous Waste Streams
• Unwanted by-products or waste streams are produced
• Due to side reactions suppressed but not eliminated
• Unwanted by-product: Stream that cannot be sold for an overall
profit
• Example: In production of cumene; benzene and propylene react
in a gas-phase catalytic reaction to form cumene:

Cumene reacts further to form p-diisopropyl benzene (DIPB), a

by-product sold for its fuel value:
 Elimination of Unwanted Nonhazardous By-Products
or Hazardous Waste Streams
• Costs of benzene, propylene, and fuel credit:
$1.196/kg, $1.334/kg, and $3.16/GJ, respectively
• Composition of DIPB waste stream:
2.76 kmol/h DIPB and 0.92 kmol/h cumene
• Standard heats of combustion for DIPB and cumene :
DIPB: –6.82 GJ/kmol, Cumene: –5.00 GJ/kmol
• Revenue for sale as fuel:
[(2.76) (6.82) + (0.92) (5.00)](8000)(3.16) = $592,000/yr
• Cost of equivalent raw materials:
benzene = (2.76 + 0.92) (78) (8000) (1.196) = $2,746,000/yr
propylene = (2.76*2 + 0.92) (42) (8000) (1.334) = $2,887,000/yr
• Net cost of producing DIPB stream:
(2,746+2,888-592)×103 = $5,041,000/yr
• What process changes could be made to eliminate the production of
the DIPB waste stream?

• Reduce the per-pass conversion of propylene in the reactor:

• This has the effect of suppressing DIPB reaction
• By reducing cumene concentration in reactor.
• Increase the ratio of benzene to propylene in the feed to the reactor:
• This reduces propylene concentration in the reactor and hence tends
to suppress DIPB reaction.
• Obtain a new catalyst:
• For which DIPB reaction is not favored as much as with the current
catalyst.
 Elimination of Unwanted Nonhazardous By-Products
or Hazardous Waste Streams
• Alternatively, a hazardous waste stream may be produced.
• In such cases additional costly treatment steps (either on-site or
off-site) would be required in order to render the material
benign to the environment.
• The economic penalties for these treatment steps—incineration,
neutralization, and so on—are often great and may severely
impact the overall economic picture for the process.
• In addition to the economic penalties that the production of
waste streams cause, there are political ramifications that may
overshadow the economic considerations.
• For many companies, the production of hazardous wastes is no
longer an acceptable process choice, and alternative reaction
routes, which eliminate such waste streams, are aggressively
pursued.
 Elimination and Rearrangement of Equipment
• Both the elimination and rearrangement of equipment can lead to
significant improvements in the process economics.
• Algorithms for these topological changes are under development.
However, the approach used here is one based on intuitive
reasoning and illustrated by process examples.

Elimination of Equipment
• The starting point is a PFD in which all process equipment serves a
valid function: that is, the process does not contain any redundant
equipment that can be eliminated immediately.
• The elimination of a piece of equipment is often the result of a
change in operating conditions and can thus be considered the end
product of a series of parametric changes.
Example: Evaluate the topological changes in the cumene PFD,
which can be made by reducing the per-pass conversion of propylene
in the reactor.
As the single-pass conversion of propylene (the limiting reactant) in
the reactor is reduced, the DIPB production decreases due to the
lowering of the cumene concentration in the reactor.
At some point, the second distillation column and the associated
equipment can be removed, because all the DIPB produced can leave
the process in the cumene product, Stream 13.
Rearrangement of Equipment
• There are certain guidelines that should be followed when the
sequence of equipment is considered.
• Some are obvious:
• One should try to pump a liquid rather than compress a gas;
thus, it will always be better to place a pump before a
vaporizer rather than a compressor after it.
• Other topological changes are somewhat more subtle:
• Most common examples of equipment rearrangement are
associated with separation section of a process and the
integration of heat transfer equipment.
Example: Consider dimethyl ether (DME) process
• Process consists of a gas-phase catalytic reaction in which
methanol is dehydrated to give DME with no appreciable side
reactions:
2CH3OH → (CH3)2O + H2O
DME
• Reactor effluent stream is cooled and sent to two distillation
columns.
• First column separates DME product from water and unreacted
methanol.
• Second column separates methanol, which is recycled, from
water, which is then sent to a waste water treatment facility to
remove trace organic compounds.
• Is there any economic advantage gained by changing the
order of distillation so that water is removed first and DME
and methanol are separated in the second column?
Example:
• No simple way to determine whether the separation sequence
should be changed.
• Rigorous parametric optimization for both topologies should be
made to choose the configuration with the best economics
• However, there are some guidelines that may help determine which
sequences are worthy of further consideration.
1. Perform easiest separation first - that is, the one least demanding
of trays and reflux - and leave most difficult to the last.
2. When neither relative volatility nor feed composition varies
widely, remove components one by one as overhead products.
3. When the adjacent ordered components in the feed vary widely
in relative volatility, sequence splits in order of decreasing
volatility.
4. When concentrations in the feed vary widely but the relative
volatilities do not, remove the components in order of
decreasing concentration.
Example: Apply the guidelines for column sequencing to the
DME process using the information given in Table and Figure
• Stream leaving reactor and entering separation section:

• Easier separation:According to Rule 1 DME should be removed first

• Other guidelines: Do not add any additional guidance
• However, need of special consideration for water (no mention in guidelines)
• Water has a very high latent heat of vaporization:
• Duties of condensers and reboilers in columns will be higher than for
similar flows of organic materials.
• By removing the DME first, water must be reboiled in both columns.
• Suggestion: There may be some advantage to removing water first and then
separating DME and methanol.
 Alternative Separation Schemes and Reactor Configurations
• Early in the design process,
• It is important to consider the use of alternative technologies
• To separate products from unused reactants and waste streams
• To evaluate alternative reactor configurations.
• Despite the wide range of separation technologies available to the
process engineer,
• When considering liquid-gas processes, vast majority of
separations are composed of
• Distillation
• Gas absorption and liquid stripping
• Liquid-liquid extraction
 Alternative Separation Schemes and Reactor Configurations
• Distillation (including extractive and azeotropic distillation):
• 90% to 95% of all separations, product recoveries, and
purifications in the chemical process industry
• Default option for process separations involving liquids and
vapors
• Relative maturity of distillation technology
• Relatively inexpensive energy requirements
• Alternatives to distillation technology may have to be used:
• If relative volatilities of two components are close to 1 (for
example, less than 1.05)
• If excessively high pressures or low temperatures are required
to obtain a liquid-vapor mixture
Example: In the HAD, the fuel gas leaving the unit, Stream 16,
contains a significant amount of hydrogen, a raw material for the
process.
• Currently, there is no separation of Stream 8.
• Mixture of methane and hydrogen leaving V-102 is split in two,
with one portion recycled and other portion purged as fuel gas.

Figure E14.6. Toluene Hydrodealkylation

Reactor and Hydrogen Recycle Loop
Question: What benefit would there be in separating hydrogen from
methane in Stream 8 and recycling a hydrogen-rich stream to the
reactor?
• Significant benefits may be derived by sending a hydrogen-rich
recycle stream back to the front end of the process.
• The methane that is currently recycled acts as a diluent in the
reactor.
• By purifying the recycle stream all the equipment in the reactor
loop, E-101, H-101, R-101, E-102, and C-101 could be made
smaller (because the amount of methane in the feed to each
equipment would be reduced), and the utility consumption for the
heat-exchange equipment would also be reduced.
• In addition, the amount of hydrogen feed, Stream 3, would be
reduced because far less hydrogen would leave in the fuel gas,
Stream 16.
Question: What technology could be used to achieve this
separation?
• Several technologies exist to purify hydrogen from a stream of light
hydrocarbons.
• Distillation: Not a viable option for this separation
• Temperature at which methane begins to condense from Stream
8 is less than –130°C.
• Two most likely candidates for this separation:
• Membrane separation, in which hydrogen would preferentially
diffuse through a polymer membrane,
• Pressure swing adsorption, where methane would preferentially
adsorb onto a bed of molecular sieve particles.
 Parametric Optimization
• In optimizing a chemical process:
• Key decision variables must be identified early in the
optimization procedure.
• Necessary in order to reduce the computational effort and
time and make the problem tractable.
• Choice of key decision variables is crucial to the efficiency of
the optimization process.
 Parametric Optimization
Some important variables that should be considered for most processes
1. Operating conditions for the reactor—for example, temperature, pressure,
concentration of reactants. Temperature range may be restricted by catalyst
properties; that is, catalyst may sinter at high temperatures or be inactive at low
temperatures.
2. Single-pass conversion in the reactor. The selectivity will be determined by the
conditions mentioned in (1) and the single-pass conversion.
3. Recovery of unused reactants.
4. Purge ratios for recycle streams containing inerts.
5. Purity of products (this is often set by external market forces).
6. Reflux ratio and component recovery in columns, and flow of mass separating
agents to absorbers, strippers, extractors, and so on.
7. Operating pressure of separators.
 Parametric Optimization
Most processes utilize recycles to recover unused reactants:
• Any change in operating conditions occuring within a recycle loop
• Impact all the equipment in the loop.
• Consequently:
• Whole flowsheet may have to be resimulated and
economics reworked every time a new value of a variable is
considered.
• For variables that do not lie within a recycle loop:
• Optimization may be simplified.
• Example: Distillation column in a separation sequence in which
two products are purified and sent to storage.
• Operation of such a column does not impact any part of the
process upstream
• Can be considered independently after the upstream process
has been optimized.
 Single-Variable Optimization:
Case Study on T-201, DME Separation Column
• For optimization of distillation columns, variables to be considered:
• Reflux ratio
• Operating pressure
• Percent recovery of key components
• Purity of the products Feed, Stream 9
Temperature 89°C
• For initial case study, fix:
Pressure 10.3 bar
• Column pressure
Vapor fraction 0.148
• Feed to the column
Molar flows (kmol/h)
Dimethyl ether 130.5 kmol/h
Methanol 64.9 kmol/h
Water 132.9 kmol/h
Total Flow 328.3 kmol/h
 Single-Variable Optimization:
Case Study on T-201, DME Separation Column
• Product specification for DME: 99.5 wt% pure
• Assumption: 98.9% of DME in the feed must be recovered
• Possible optimization variables for this column:
• Pressure,
• Reflux ratio,
• % recovery of DME
• First:
• Focus on reflux ratio
• Later:
• Relax constraint on operating pressure
• Carry out a two-variable optimization
 Single-Variable Optimization:
Case Study on T-201, DME Separation Column
• Choose an objective function
• Considering reflux ratio as the only decision variable:
• Costs that will be affected by changes in reflux:
• Capital cost associated with column
• Operating cost associated with column
• To account correctly for both one-time costs and operating costs:
• Objective function taking the time value of money into
account should be used.
• Use the before-tax NPV as objective function:
 Single-Variable Optimization:
Case Study on T-201, DME Separation Column
• COMd(value of cost of manufacturing term without depreciation):
• Fixed capital investment based on total module cost (FCITM)
and cost of utilities.
• Other terms are not relevant to the optimization:
• They do not change with reflux ratio.
• Using:
• Plant life of ten years after start-up
• 10% internal rate of return
• Assumption of a one year construction period
 Single-Variable Optimization:
Case Study on T-201, DME Separation Column
• Fixed capital investment term includes total module
costs for
• T-201
• E-204
• E-205
• V-201
• P-202 A/B
• Utility costs include:
• Electricity for P-202
• Heating utility for E-204
• Cooling utility for E-205
• Run a series of case studies (ChemCad):
• For different reflux ratios
• Costs are calculated
• For equipment (CAPCOST using mid-1996 prices) and utility
 Data for DME Column Optimization, R/Rmin, versus OBJ
Steam Cost Cooling Water Electrical Cost Total Utility OBJ from Eq.
R/Rmin 3
FCI ($10 ) ($103/yr) Cost ($103/yr) ($103/yr) Cost ($103/yr) (19.2) ($103)
1.01 684 72.50 4.04 0.48 77.02 –1911
1.02 509 72.72 4.05 0.48 77.24 –1551
1.03 441 72.96 4.06 0.48 77.50 –1417
1.04 411 73.15 4.07 0.48 77.70 –1358
1.11 354 74.68 4.13 0.49 79.31 –1255
1.27 342 78.16 4.28 0.51 82.95 –1256
1.60 322 85.04 4.54 0.55 90.17 –1265

• Optimum value of R/Rmin:

• Close to 1.12
• For values greater than 1.1:
• OF changes slowly with R/Rmin
 Two-Variable Optimization:
Effect of Pressure and Reflux Ratio on T-201
• Now: Effect of pressure on operation of T-201 is considered
• Tempting to use result from single variable optimization
• R/Rmin = 1.12
• Carry out a univariate search on pressure
• However, this technique not yield optimum P or R/Rmin values
• As pressure changes, optimum reflux ratio also changes
• In order to optimize correctly this situation,
• Both P and R/Rmin must be varied
• Best combination should be determined
 Two-Variable Optimization:
Effect of Pressure and Reflux Ratio on T-201
• Different approaches for solution:
• One method:
• Pick different pressures
• Repeat procedure used in 14.4.1 for each pressure
• Plot results
• Not a particularly efficient procedure
• But it yields plots that are easy to interpret
• Example of a bivariate search technique
 Two-Variable Optimization:
Effect of Pressure and Reflux Ratio on T-201
Table 14.5. Data for Two-Variable Optimization of DME Column, T-201
Pressure 13.5 11.5 10.3 9.0 7.5
(Bar)
R/Rmin OBJ OBJ OBJ OBJ OBJ
1.01 –2052 –1975 –1911 –1926 –5203
1.02 –1699 –1613 –1551 –1547 –4847
1.03 –1560 –1474 –1417 –1409 –4719
1.04 –1499 –1411 –1358 –1347 –4665
1.11 –1394 –1312 –1255 –1250 –4616
1.27 –1365 –1288 –1256 –1243 –4715
1.60 –1385 –1309 –1265 –1259 –4955
All costs in $1000 using data from 1996, first edition of text.
 Two-Variable Optimization:
Effect of Pressure and Reflux Ratio on T-201
• Relationship between P and NPV at optimum R/Rmin
• Highly nonlinear
• Optimum value of R/Rmin
• Not remain constant with pressure (shown by dotted line)
• For this example, (R/Rmin)opt does not change very much with P
 Two-Variable Optimization:
Effect of Pressure and Reflux Ratio on T-201
• While carrying out parametric optimization:
• Topology of distillation column and associated equipment
should remain fixed
• This may
• Unduly constrain the overall optimization of process
• Limits the range of pressures over which optimization may
be considered
• Moreover, when carrying out this optimization:
• Necessary to change some of the utilities used in the process
• Significantly impacts the results.
Table 14.6. Breakdown of Costs and Process Information for R/Rmin = 1.27
Pressure 13.5 11.5 10.3 9.0 9.0 7.5
(Bar)
mp steam 89.79 82.44 78.16 — 72.26 —
lp steam — — — 62.80 — 56.67
Cooling water 4.23[a] 4.26[a] 4.28[a] 8.61[b] 8.61[b] —
Refrig. water — — — — — 539.35[c]
Electricity 0.53 0.52 0.51 0.50 0.50 0.49
FCI 357 344 342 354 340 307
NPV –1339 –1263 –1230 –1178 –1217 –4692
Rmin 0.3528 0.3214 0.3024 0.2810 0.2810 0.2562
Tcondenser (°C) 57 50 46 40 40 34
Treboiler (°C) 164 157 153 147 147 140
All costs in $1,000 or $1,000/yr based on 1996 data. [a] ΔT = 10°C [b] ΔT = 5°C [c] ΔT = 10°C
• Consider switch from medium-P steam to
low-P steam for process pressure of 9.0
bar.
• Small but significant increase in NPV
• When low-P steam substituted for
medium-P steam
• Reason for this increase:
• Reduction in utility costs (due to less-
expensive low-P steam)
• Outweighs the increase in equipment
costs (due to smaller temperature
driving force and higher operating P in
E-204, column reboiler).
• Switch to low-P steam is actually possible
for all of operating pressures shown
• With exception of 13.5 bar,
• Reboiler temperature for these
cases is less than 160°C
• Temperature at which low-P steam
condenses
• Consider a change in the cooling water utility.
• As operating pressure of column decreases,
• Bottom and top temperatures also decrease.
• One consequence:
• Low-P steam may be used in the reboiler
• In addition, as top temperature in column decreases,
• Point is reached where a temperature increase of 10°C for the cooling water
can no longer be used
• Occur at an operating pressure of 9.0 bar. Cooling water can be used to condense
the overhead vapor from the column, but the temperature increase of the utility
must be reduced.
Pressure (Bar) 13.5 11.5 10.3 9.0 9.0 7.5
mp steam 89.79 82.44 78.16 — 72.26 —
lp steam — — — 62.80 — 56.67
Cooling water 4.23[a] 4.26[a] 4.28[a] 8.61[b] 8.61[b] —
Refrig. water — — — — — 539.35[c]
Tcondenser (°C) 57 50 46 40 40 34
Treboiler (°C) 164 157 153 147 147 140
All costs in $1,000 or $1,000/yr based on 1996 data.
• Results shown are for a change in cooling water T of 30°C (in) to 35°C (out).
• As operating pressure is reduced still further to 7.5 bar,
• Column top temperature decreases to 34°C.
• At this point,
• Use of cooling water becomes almost impossible
• Flow of cooling water and area of exchanger would become extremely large,
• Any slight changes in cooling water temperature would have a large impact
on the process.
• For this reason,
• Refrigerated water is substituted as the cooling utility for this case.
• Result:
• Very large increase in the cost of utilities
• Resulting decrease in the objective function (NPV) as shown
Pressure (Bar) 13.5 11.5 10.3 9.0 9.0 7.5
mp steam 89.79 82.44 78.16 — 72.26 —
lp steam — — — 62.80 — 56.67
Cooling water 4.23[a] 4.26[a] 4.28[a] 8.61[b] 8.61[b] —
Refrig. water — — — — — 539.35[c]
Tcondenser (°C) 57 50 46 40 40 34
Treboiler (°C) 164 157 153 147 147 140
All costs in $1,000 or $1,000/yr based on 1996 data.
 Figure 14.8. Effect of Operating Pressure
of T-201 on NPV (Data for R/Rmin = 1.27)

• During parametric optimization:

• Changes in operating conditions
• May require corresponding changes in utilities
 Flowsheet Optimization Using Key Decision Variables
• When decision variables for an optimization is in a recycle loop
• Evaluation of objective function becomes more complicated
• Example: Consider three variables for DME flowsheet:
• Single-pass conversion in reactor (x)
• Fractional recovery of DME product in tower, T-201 ( f )
• Reactor pressure ( p)
• Even with only three variables, it is not clear:
• How many runs should be performed
• Over what range of decision variables runs should be performed
 Flowsheet Optimization Using Key Decision Variables
• First step:
• Choose the range over which decision variables should be varied
• The ranges chosen for this example are:
• 0.7 < x < 0.9, 0.983 < f < 0.995, 12.70 bar < p < 16.70 bar
• Choice of these ranges is somewhat arbitrary,
• It has been found from research in the laboratory:
• For the current catalyst
• Maximum single-pass conversion of 90%
• Maximum operating pressure of 16.70 bar
• Because of KEQ and to avoid undesirable side reactions
 Flowsheet Optimization Using Key Decision Variables
• Lower limits for p and x are arbitrary & could be changed

• If the optimization results warrant looking at lower values.

• Choice of the range of f values is again arbitrary,

• Experience tells us

• Fractional recovery of product (f) should be close to 1

 Flowsheet Optimization Using Key Decision Variables
• Next question to be answered:

• How many points should be chosen for each variable?

• If three values for each variable are used,

• There would be a total of (3)(3)(3) = 27 simulations to run.

• A conservative estimate of the length of time to run a simulation,

collect the results, evaluate the equipment parameters, price the
equipment, estimate the utility costs, and finally evaluate the
objective function is, for example, 2 hours per run.

• Total time investment of (2)(27) = 54 hours of work!

Much larger if, for example, five variables were considered
 Flowsheet Optimization Using Key Decision Variables

• Only end points of the range for each variable are used,

• (2)(2)(2) = 8 simulations will be run

• Still a significant investment of time

• More reasonable than the previous estimate.

• Essentially: Carry out a 2k factorial experiment.

 Flowsheet Optimization Using Key Decision Variables

• Each evaluation of objective function (dependent variable) can be

considered the result of an experimental run

• In which the independent variables (x, f, and p) are varied

• Low range for each variable is designated 0

• High range for each variable is designated 1

• Run named DME011 represents the process simulation:

• Conversion is 0.7 (x is at its low value),

• Fractional recovery is 0.995 (f is at its high value)

• Pressure is at 16.7 bar (p is at its high value)

Test Matrix for Three Variable Optimizations of DME Process
Variable 1 = Variable 2 = Variable 3 =
x f p
0 = 0.7 0 = 0.983 0 = 12.7 bar FCI Utilities OBJ (NPV)
Number 1 = 0.9 1 = 0.995 1 = 16.7 bar ($1000) ($1000/yr) ($1000)
Base Case 0.8 0.989 14.7 1297 732 –7630
DME000 0.7 0.983 12.7 1378 782 –8136
DME001 0.7 0.983 16.7 1381 872 –8760
DME010 0.7 0.995 12.7 1393 780 –8152
DME011 0.7 0.995 16.7 1440 868 –8851
DME100 0.9 0.983 12.7 1210 539 –6130
DME101 0.9 0.983 16.7 1261 634 –6885
DME110 0.9 0.995 12.7 1232 537 –6160
DME111 0.9 0.995 16.7 1299 634 –6961
• Once results have been obtained,
• Interpretation and determination of optimum operating conditions
must be carried out.
• Values of objective function are shown at the appropriate corners of
the box.
• Also base-case simulation lies at the middle of the box
• This is due to choosing the ranges for each variable symmetrically
about the base-case value
1. Analyze these results in terms of general trends using a method
known as
ANALYSIS OF MEANS
2. Fit a model for the objective function in terms of the decision
variables,
3. Use it to estimate the optimum conditions or, more correctly, to
give direction as to where the next test run should be performed.
• This approach is termed
RESPONSE SURFACE ANALYSIS
• A powerful method to interpret and correlate simulation results
 Sensitivity of Objective Function to Changes in Decision
Variables: Analysis of Means
• Using the data from the table and figure,
• Estimate how a change in each variable affects the NPV.
• This is simply done by
• Averaging all results at a given value of one variable
• Comparing them with the average value at the other level of that
variable.
• This yields the following results:
conversion, x = 0.7, NPVavg,x = 0 = (–8136 – 8760 – 8152 – 8851)/4 = –8475
conversion, x = 0.9, NPVavg,x = 1 = (–6130 – 6885 – 6160 – 6961)/4 = –6534

recovery, f = 0.983, NPVavg,f=0 = (–8136 – 8760 – 6130 – 6885)/4 = –7478

recovery, f = 0.995, NPVavg,f=1 = (–8152 – 8851 – 6160 – 6961)/4 = –7531

pressure, p = 12.7 bar, NPVavg,p= 0 = (–8136 – 8152 – 6130 – 6160)/4 = –7145

pressure, p = 16.7 bar, NPVavg,p= 1 = (–8760 – 8851 – 6885 – 6961)/4 = –7864
 Sensitivity of Objective Function to Changes in Decision
Variables: Analysis of Means
conversion, x = 0.7, NPVavg,x = 0 = (–8136 – 8760 – 8152 – 8851)/4 = –8475
conversion, x = 0.9, NPVavg,x = 1 = (–6130 – 6885 – 6160 – 6961)/4 = –6534

recovery, f = 0.983, NPVavg,f=0 = (–8136 – 8760 – 6130 – 6885)/4 = –7478

recovery, f = 0.995, NPVavg,f=1 = (–8152 – 8851 – 6160 – 6961)/4 = –7531

pressure, p = 12.7 bar, NPVavg,p= 0 = (–8136 – 8152 – 6130 – 6160)/4 = –7145

pressure, p = 16.7 bar, NPVavg,p= 1 = (–8760 – 8851 – 6885 – 6961)/4 = –7864

• It can be concluded:
• Single-pass conversion has the greatest influence on the NPV
• Followed by reactor pressure
• With recovery of DME having only a slight effect
• Moreover, results suggest:
• NPV will be maximized (least negative value) by using:
• High conversion, low pressure, and low DME recovery
 Sensitivity of Objective Function to Changes in Decision
Variables: Analysis of Means
• Although the maximum NPV that was obtained from these
simulations was –$6,130,000 (DME100),
• It should not be assumed that
• This is the true maximum,
• The maximum lies within the range of decision variables
considered thus far.
• In fact, the above analysis tells us that
• We should move to a new range for pressure & DME recovery
• In order to estimate the maximum for the objective function,
• Further simulations at different conditions will need to be done.
• Recovery of DME is not very important comparing with
other effects,
• Single-pass conversion is constrained at its maximum value
• Problem has been reduced to a one-dimensional search (p).
• Question: What should be the next value of p?
• Although p can be decreased by some arbitrary amount,
• There may be another constraint as p is lowered.
• Detrimental change in NPV was identified
• When pressure in T-201 became so low
• Refrigerated water required as utility for overhead condenser
• Lowest pressure at which reactor can be run = 10.7 bar
• Still operate T-201 @ 9.0 bar
• Which is close to the lower limit for using cooling water
• Next simulation should be carried out at (x=0.9, f=0.983, p=10.7)
 Modeling Objective Function in Terms of Decision Variables

• Useful to be able to estimate NPV values for new conditions

• Before actually running simulations

• Can be done by using the results from table

• To model the NPV as a function of x, f, and p.

• Several types of functional forms can be chosen for this model.

 Modeling Objective Function in Terms of Decision Variables

• Following form is chosen here:

• a0, a1,... a7 are constants that are fit using the data from table
• This model uses eight arbitrary constants
• Allow the function to predict the NPV exactly at each corner
point of our experimental design.
• This form of model is a simple multivariable polynomial,
• Regression techniques to find a0, a1,... a7 are well established.
 Modeling the Objective Function in Terms of the Decision
Variables

Test Run Result from Table 14.7 Prediction from Equation 14.4
Base Case –7630 –7502
DME000 –8136 –8134
DME001 –8760 –8758
DME010 –8152 –8150
DME011 –8851 –8849
DME100 –6130 –6127
DME101 –6885 –6881
DME110 –6160 –6157
DME111 –6961 –6957
All figures are in $1,000.
• From results in Table,
• Seen that the model predicts eight test runs exactly
• Prediction for the base case is also good,
• Model tends to underpredict the NPV a little
• Expected value of NPV can also be predicted for the next
simulation (x = 0.9, f = 0.983, p = 10.7),
• From Equation, NPV = –5750.
• Actual value for NPV for this new simulation = –5947,
• Higher than the predicted value by about 3%.
At this point, clearly we are close to the optimum. Whether further
simulations are warranted depends on the accuracy of the estimate being
used to obtain the costs and the extra effort that must be expended to analyze
further simulations.
 Optimization in Batch Systems
• Unlike continuous systems, batch operations do not
run under steady-state conditions
• Their performance varies with time.
• Important issues with batch systems:
• Optimal scheduling of different equipment to
produce a variety of products
• Determination of optimal cycle times for batch
processes
 Optimization in Batch Systems
• Optimization of batch operations often involves
determining:
• “best” processing time for a certain operation,
• “best” time at which a certain action should take
place,
• “best” distribution of actions over a period of time.
 Problem of Scheduling Equipment
• Consider a simple case of scheduling a set of equipment
used to produce multiple products.
• For example, there are
• Three products of interest (A, B, C) manufactured
• From same chemical feedstock F.
• Production of each of the three chemicals
• Follows a different sequence
• Requires use of different equipment for different
periods of time:
• Reactor R
• Separator S
• Precipitator P
Table 14.9. Equipment Time Requirements for Products A, B, and C
Time in Time in Time in Value of Product Cost of Feed
Product Reactor (hr) Separator (hr) Precipitator (hr) ($/kg Product) ($/kg Product)
A 7 4 0 0.75 0.25
B 15 3 3 1.12 0.27
C 25 4 2 1.41 0.23

• Costs of the feeds are different for each product

• Because of differences in process efficiencies.
• In general, the longer it takes to make a product,
• The more valuable is the product
• Another complication in this type of scheduling problem:
• Production steps are sequential: R→S→P.
• When a reactor has finished producing a product,
• Separator should be available to carry out the next step in
the process etc.
• Assumption here:
• Separator and precipitator are always available to handle the
products from the reactor
• For this case:
• Solution to the problem is simple
• Because the only piece of equipment must be considered:
• Reactor
• Reaction step is the limiting step in each process.
Problem Formulation
• Number of batches produced per year for products A, B, and C:
• x1, x2, and x3
• Assumption:
• Size of each batch is the same and equal to W kg,
• Objective function: Total profit
Maximize f = W x1 (0.75-0.25) + W x2(1.12-0.27) + W x3 (1.41-0.23)
Maximize f = 0.5 W x1 + 0.85 W x2 + 1.18 W x3
Subject to
Addition of Market Constraints: Now consider a slightly more
realistic version of the problem.
This time, consider the situation where market forces tell us that a
maximum of 800 batches of A in a year can be sold and that to
meet contractual obligations, at least 10 batches of B and 7
batches of C must be produced in a year. These additional
constraints are given as:
(14.8)

Resulting solution is x1=800, x2=199, x3=7, and OF = $577.4W.

Extra constraints have caused the OF to be reduced. This solution
tells us that after our contractual obligations have been fulfilled,
and as much of Product A has been made as is allowed, as much
of Product B as is allowed should be made. In other words, the
order of profitable products is A > B > C.
Considering Other Equipment
Time taken to process the products through the other equipment
becomes a limiting factor.
To illustrate this type of problem, the basic problem is modified, as
shown in Table 14.10.

Table 14.10. Equipment Time Requirements for Products A, B, and C for the
Case When the Time for Separation and/or Precipitation Constrains the
Solution

Time in Time in Time in Value of Product Cost of Feed

Product Reactor (hr) Separator (hr) Precipitator (hr) ($/kg Product) ($/kg Product)
A 7 10 8 0.75 0.25
B 15 7 6 1.12 0.27
C 25 5 2 1.41 0.23