Process Optimization in Chemical Engineering
Process Optimization in Chemical Engineering
7.1 Introduction
In chemical processing units, optimization is the method that seeks to solve the
problem of minimizing or maximizing an objective function that relates the vari-
able to optimize with the design and operating variables. The criteria for analysis of
the economic objective function involve fulfilling a process criteria restrictions,
conditions, design equations, and respecting the limits of the variables (Seider and
Warren 2003). Most problems in chemical engineering processes have many
solutions, in some cases becoming endless. The optimization is related to the
selection of an option that is best in a variety of efficient options but being the
only one that comes closest to an economic optimum performance and operation.
Plants operating profits are achieved by optimizing performance of the valuable
products or reduction of pollutants, reducing energy consumption, improving
processing flows, decreasing operation time, and minimizing plant shutdowns. To
this end it is useful to identify the objective, constraints, and degrees of freedom in
the process, reaching benefits such as improving the design quality, ease of trou-
bleshooting, and a quick way to make correct decisions. However, the argument for
the implementation of the optimization process is not well supported if the formu-
lation of the optimization process has an uncertainty in the mathematical model
describing the process. Although the mathematical model is a description of reality,
an optimization on the mathematical model does not guarantee optimization
modeling phenomenon due to the difference between the mathematical model
and the actual phenomenon. In the case of chemical engineering problems, most
processes and operations are well represented by mathematical models with some
complexity, this leads by ensuring mathematical model optimization is going to
optimize the process.
Optimization can take place at any level within an organization, from a complex
combination of plants, distribution facilities to each floor, units combinations,
individual equipment, subsystems of a piece of equipment or smaller units. Process
The formulation of the objective function is based on the mathematical model of the
phenomenon being optimized. In the chemical industry, cost minimization or
profit maximization is held as usual objective function, in which case the design
equations and describing the process should relate to the operation cost or invest-
ment. In other cases it is important that an objective function is raised in order to
maximize the operation performance or choose the most appropriate diameter for
specific equipment, in such a case the equations describing the operation must be
related to the process efficiency and unit sizing. Depending on what you want to
optimize, more rigorous and complex models need to be used. Each optimization
problem has three main categories:
1. A target function (at least) to be optimized.
2. Equality constraints.
3. Inequality constraints.
To achieve a feasible solution for the optimization problem, the problem vari-
ables must satisfy the restrictions to the degree of accuracy required, so a workable
region is obtained in the problem where the optimum point is found. In some cases
the optimal solution is a single point, but in other cases the problem becomes
indeterminate (Edgar et al. 2001).
In design and control calculations, information is important and user must remove
redundant equations before running calculations. In the case where there is no
debug information the solution of the problem cannot be obtained and the infor-
mation should be reviewed again. The degrees of freedom in a model are the
number of variables that can be specified independently and are defined as:
Nf ¼ N v N e ð7:1Þ
where
Nf: degrees of freedom number
Nv: total number of variables
Ne: number of independent equations including specifications
7.2 Formulation of Optimization Problem 345
The formulation of the objective function is one of the crucial steps in the applica-
tion of optimization to a practical problem. In the chemical industry the objective
function is generally expressed in monetary terms, since the goal of companies is to
minimize production costs and maximize profits subject to variations in the con-
straints. In other cases, the problem to be solved is to maximize the performance of
a component in a reactor or minimizing service flow heat exchanger. As formulated,
the mathematical model of each problem the complexity of the equations that
represent, at present, has the advantage of having software capable of solving
highly nonlinear functions must be analyzed.
The ability to understand and apply concepts of cost analysis, profit analysis,
budgets, and balances are key to assess opportunities. Economic decisions are made
at various levels of detail and the more detailed study; it takes longer and requires
more resources. To formulate the objective function and which are related to
economic parameters must take into account two types of costs (Aspen Technology,
Inc. 2005).
346 7 Process Optimization in Chemical Engineering
1. Costs associated with the mass flows, such as purchase costs of raw materials or
revenues for products.
2. Costs associated with operating variables in the model, such as electric power,
steam, fuel gas, etc.
Min f ðxÞ
ð7:2Þ
s:t: gðxÞ ¼ b
Another important classification refers to the nature of the variables. There are two
types of variables: Continuous and discrete variables. Discrete variables are those
that take only integer values; in the case of process engineering would be appro-
priate to: number of process equipment, number of tube passes in a heat exchanger,
number of stages on a gas–liquid separation, etc. These problems are known as
mixed integer programming problems.
In design and operation of plants, some problems involve nonlinear relation-
ships, binary or integer variables, and continuous variables. Continuous variables
generally represent process variables such as flow, and integer variable represents a
decision (Gomez and Esteban 2006). Some problems in plant operation and activ-
ities programming involve variables that are not continuous and often are integer
values. Integer variables can take binary values to represent decisions define
equipment configuration, or to define whether or not a specific stage feed of a
distillation column. Other integer variables can take values from the set of natural
numbers (positive integers) and can be useful if you want to define the number of
stages of evaporation in a train of evaporators or the number of steps in a heat
exchanger. The solution of such problems involves mathematical variables and the
use of more resources for which methods are designed to make the simplest
problem complexity. And the problems of continuous variable integer variable
issues can also be classified as linear (MILP) and nonlinear (MINLP) depending
348 7 Process Optimization in Chemical Engineering
min Z ¼ f ðx; yÞ
s:t: g j ðx; yÞ 0 j2J ð7:3Þ
x 2 X, y 2Y y2ℤ
This formulation is general and each problem can develop your own equations
describing the problem.
One method for solving linear and nonlinear problems with integer variables is
the “Branch and Bounds” (B&B) method using linear relaxations. This method can
be used in both linear and nonlinear problems with integer variables. You can
consider using the B&B method when all integer variables are binary but with other
by this method the solution is more complex. Relaxation method of discrete vari-
ables is based on the fact that the discrete variables can be viewed as continuous if a
regression of the discrete variable data is performed. By solving the problem with
integer variables can be obtained relaxed variables. For relaxed variables you need
to create a new LP (branching problem) in which the variables are evaluated at each
of the nearest integer values and select which will provide better value to the
objective function. For each subproblem limits, the relaxed variables (bounding)
are set and the behavior of this variable is checked to see if you need further
consideration (Gomez and Esteban 2006).
The second method of solving the MILP and MINLP problems is “outer
approximation” (OA). Each iteration of the OA method involves two subproblems:
A nonlinear programming problem with continuous variable and a linear problem
with mixed variables. In general, the method is based on solving the problem by the
lower limit of NLP MINLP problem. The second subproblem, the MILP, takes into
account the whole and continuous variables, all nonlinear functions in the linear-
ized rated range. MILP problem optimization generates the upper limit of the
optimal solution MINLP. At each iteration, the step size analysis of the problem
is reduced to that of a finite number of iterations and errors depending on the
accepted solution to optimal MINLP (Edgar et al. 2001) is located.
One of the great advantages of process simulators lies in the ability to integrate with
design and optimization process. Due to the implementation of powerful numerical
methods, simulators allow solving larger problems and a lot of variables; however,
some of these variables are not behind the calculations of each operation so that
problems rarely have more than 100 degrees of freedom. Among the many
7.3 Optimization in Sequential Simulators 349
Here are some tips on using optimization simulators which are presented below.
Most simulations have the option of using design specifications. Iterative calcu-
lation related to the specifications commonly found in recycle loops and conver-
gence shall be achieved in the specifications for each recycle iteration. In other
instances the specifications are implemented as external loops, where the conver-
gence should be achieved in recycles for each iteration of the outer loop. It is also
possible to solve both loops simultaneously, usually using only one pass through the
units in the recycle loops and specification.
When optimizing a process is often preferable to include design specifications at
first, then when the optimization algorithm must be removed include those speci-
fications. Thus the optimization algorithm replaces those specifications.
It is recommended that prior to any optimization algorithm to conduct prelim-
inary searches varying the most important variables. The more used to perform this
350 7 Process Optimization in Chemical Engineering
Table 7.1 Important methods present in process simulators (Aspen Technology, Inc. 2005)
Method Description Restrictions
Fletcher- Corresponds to the Polak Ribiere modi- This scheme is efficient for minimiza-
Reeves fication of the Fletcher-Reeves conjugate tion without restrictions. This method
gradient scheme does not handle constraints
Quasi It refers to the method of Broyden- In limited applicability is similar to the
Newton Fletcher-Goldfarb-Shanno Flechter-Reeves method. This scheme is
efficient for minimization without
restrictions. This method does not han-
dle constraints
In both limited applicability and is sim-
ilar to the method Flechter-Reeves
BOX This method corresponds to a sequential Handles inequalities constraints, but not
search problem solver with nonlinear equalities. Usually requires a large
objective functions number of iterations, however it is a
fairly robust method
SQP It is the Sequential Quadratic Program- Considered one of the most efficient
ming (SQP) method methods to minimize linear and
nonlinear problems. Allows equality
and inequality constraints
Mixed This strategy seeks to take advantage of Only allows inequality constraints
the strength of the BOX method and SQP
efficiency. Use an initial assessment with
BOX method using a low tolerance, then
find the solution with tolerance using
SQP
Complex Traditional black boxes method. Only It does not require derivative calcula-
available in Aspen Plus tions. It can handle as inequality con-
straints and bounds on the optimization
variables. Blocks should be used for
converging external convergence recy-
cles or design specifications. Requires
many iterations to converge
procedure is by way sensitivity analysis so that the results will be closer to the
optimum and the optimization algorithm is more likely to converge with fewer
iterations.
In the example of extractive distillation (Gil 2003; Uyazan 2006), the recycle of
glycerol specification from column to the heat exchanger E-101 and subsequent
admission to the dehydration column should be reviewed. Now it is required to
optimize this process to maximize earnings (Fig. 7.1).
The first step in any optimization process is to properly define the objective
function and constraints to narrow the search for the optimal operating conditions.
In this case it will raise the profit goal or earnings to be having the process for
10 years taking into account the investment costs of the equipment, the cost of raw
materials, and the utilities cost consumed in that period of time (Langston
et al. 2005). The value of anhydrous ethanol sold product is also taken into account,
so that you can calculate the annualized earnings or profit of the process of
extractive distillation and solvent recovery.
In this case the objective function is described below:
Now you must disaggregate each of the terms to include operating conditions which
will be varying during optimization modifications. The only income in this process
is due to the sale of product, in this case, anhydrous ethanol.
US$
Income ¼ Anhydrous alcohol sales ð7:6Þ
year
Regarding expenses, there are two types: fixed and operating costs. Fixed costs are
the expenses incurred that are not dependent on the amount of product
352 7 Process Optimization in Chemical Engineering
manufactured. Operating expenses are those expenses which are incurred for the
production, which are a function of operating time.
US$
Expenses ¼ Cfixed þ Coper ð7:7Þ
year
In the overhead costs of equipment, buildings, roads, etc. are taken into account, for
this case the cost of the equipment.
The operational costs include raw materials, defined as substances that enter the
process, utilities costs, mainly steam, electricity, instrument air for elements, etc.;
however, equipment costs have an important weight in this term of the equation.
When replacing (7.7) such terms are converted into:
US$
Expenses ¼ Craw materials þ Cutilities þ Cequipment ð7:8Þ
year
Now each of the terms is decomposed depending on their nature. The cost of raw
materials for this process corresponds to:
US$
Raw materials cost ¼ CEthanol þ CGlycerin ð7:9Þ
year
The utilities costs correspond to the costs of electricity for pumping steam to raise
the temperature in reboilers, cooling water for condensers, and finally refrigerant
for heat exchanger E-104. The use of the same is minimized by installing double
exchanger to cool the anhydrous ethanol; thus it can reduce the temperature by
using water up to 30 C (while maintaining a temperature difference of 10 C) and
then use to carry coolant to the storage temperature.
US$
Utilities cost ¼ CElectricity þ CSteam þ CCooling Water þ CRefrig: ð7:10Þ
year
Finally, the costs of equipment for this problem are only three types: pumps, heat
exchangers, and distillation columns.
US$
Equipment cost ¼ CPumps þ CDist: Towers þ CHeat Exchangers ð7:11Þ
year
The described terms of the objective function must be calculated by the simulation
program and updated with actual process data for each simulation run. For this
Aspen Plus has a very useful tool for this purpose. This is the Calculator tool which
is accessed through the route Flowsheeting Options|Calculator Data Browser
button. This tool allows the calculation of equations in this case correspond to the
cost models of equipment, products, and utilities used in the objective function.
7.4 Introductory Example 353
The utilities cost is associated with the electricity, steam, and cooling water
consumption on each of the process equipment. Prices for each service are taken
from (Seider and Warren 2003).
7.4.1.3 Pumps
The cost of a pumping system should include both the cost of the pump and engine
that can be estimated with the following correlations:
S ¼ QH 0:5 ð7:12Þ
where
S: size factor
Q: volumetric flow in gpm
H: pump head in ft
n o
CB ¼ exp 9:2951 0:6019 ½lnðSÞ2 ð7:13Þ
where
CB: base cost
S: size factor
CP ¼ FT FM CB ð7:14Þ
PT PB QHρ
PC ¼ ¼ ¼ ð7:15Þ
ηP ηP ηM 33, 000 ηP ηM
where
PC: power consumed
PT: theoretical pump power
ηP: pump efficiency
ηM: electric motor efficiency
PB: pump brake power
FT: motor type factor (Table 16.22 from Seider and Warren (2003))
The sum of purchase cost of the pump and motor, which depends on the flow and
hydraulic head, is the total cost of the pumping system, thus the cost of the pump is
related to the independent variable in this equipment, which is the flow.
According to Seider and Warren (2003) the base cost of a heat exchanger is given
by the type of heat exchanger that will be used. In this case, Kettle type heat
exchangers will be used as reboilers in distillation columns, and floating heat
exchangers will be used for condensers and other exchangers in the process.
According to this:
7.4 Introductory Example 355
C P ¼ FP FM FL C B ð7:22Þ
This makes it possible to obtain the cost of the exchangers depending on the heat
transfer area mainly. You need to evaluate the different configurations of heat
exchangers for the required processes: reboilers, condensers, and heat exchangers.
Since the distillation columns operate under pressure, they must be designed as
pressure vessels, thus starts by determining the design pressure, which depends on
the operating pressure as follows (Seider and Warren 2003):
n o
Pd ¼ exp 0:60608 þ 0:91615 ðlnPo Þ þ 0:0015655 ðlnPo Þ2 ð7:25Þ
Pd D i
tP ¼ ð7:26Þ
2 S E 1:2 Pd
356 7 Process Optimization in Chemical Engineering
where
Di: inner diameter in inches
S: maximum material stress allowable
E: welding efficiency
Then, the tower weight is calculated.
CP ¼ FM CV þ CPL ð7:30Þ
With the data entered for this facility, the medium pressure steam is used which
condenses isothermally at a temperature of 358.42 F (181.34 C) thus make use
only, its latent heat of vaporization.
The next service to enter the cooling water is called CW, with the information
shown in Table 7.4.
A utility that does not correspond to any particular application, called U-1,
which will be useful later, is added. The information on this utility is shown in
Table 7.5.
To enter this information keeps the option Specify heating/cooling selected in
the Calculation section.
Once finished entering utilities, should be included in each unit as required. To
this end, each equipment block in Setup Utility is a tab called where the relevant
service is entered. In Table 7.6 the relevant utilities for each unit is shown.
Note that the mixer and the heat exchanger E-101 utilities required by the
calculation (Figs. 7.5 and 7.6).
7.4 Introductory Example 359
To enter the proper utility in the distillation columns should be directed to the
Condenser and Reboiler of each module tabs and enter the information in the
Utility tab.
The cost model is associated to each item with a set of empirical equations
describing the cost based on the same design variables. The variables must be
known and must be updated within each calculation simulator. To this end, Aspen
Plus provides the Calculator tool which has the ability to program in Fortran or
Excel equations describing such cost models. For this case Excel is used to calculate
the costs and the objective function.
To access the Calculator tool, follow the Flowsheeting Options > Calculator
route in the Data Browser menu. A new calculator block is activated with the New
button and the default name, C-1 is left. It speaks directly to the Calculate tab in
Excel Calculation method is selected (Fig. 7.7).
360 7 Process Optimization in Chemical Engineering
Clicking the Open Excel Spreadsheet button, the window version of Microsoft
Excel® installed on the computer is displayed. This is completely defined as the
Calculator tool because it takes the values of the spreadsheet Microsoft Excel. Then
the pop-up window is displayed (Fig. 7.8).
Microsoft Excel automatically in a new tab bar tool called Aspen (in newer
versions) appears. If the menu not appears, check in the complements menu that the
Aspen Plus tool (Fig. 7.9) is activated.
7.4 Introductory Example 361
Fig. 7.11 Variable definition tab from Aspen Plus® in Microsoft Excel®
In this tab you can integrate Aspen Plus calculation engine with Excel spread-
sheet and allows calculations and check properties databases. For this exercise,
Aspen tool can be found in the Complement tab is used (Fig. 7.10).
This tab is useful to define variables in the simulation developed in Aspen Plus
and use them in calculations as required. To define a variable of Aspen Plus in
Microsoft Excel you must first select the Define button, which appears in the
definition variables window.
In this case, for example, the molar flow of defined Azeo + stream is selected. As
an entering data sheet, Microsoft Excel is defined as an Import variable, if a result
obtained in the spreadsheet should be exported to Aspen Plus, Export Variable
option must be selected as is the case of the objective function value (Fig. 7.11).
Closing the Variable Definition tab and if the variable does not already appear in
the cell in which it is defined, it is necessary to refresh the Microsoft Excel sheet
using the Refresh button on the Complements tabs and then run the simulation from
Aspen Plus. This procedure must be done to define each variable that you want to
import from or export to Aspen Plus® using Microsoft Excel®.
Thus all imports of information are performed from Aspen Plus to Microsoft
Excel and equipment costs, raw materials, utilities, product, and objective function.
The value of the objective function is exported to Aspen Plus using the cost of
the U-1 utility, Microsoft Excel sheet cools, and Aspen Plus simulation runs. The
U-1 utility is defined in the Utilities section.
7.4 Introductory Example 363
The cost of the U-1 utility takes the value of the objective function as Aspen Plus
has no direct way to relate the calculated heat of the objective function in Microsoft
Excel with a variable whose name is in Aspen Plus. As the cost of U-1 utility is
used, any variable can be used in a stream, block, or usefulness that is completely
independent and does not affect the development of the simulation or process. The
variable takes the value of the objective function is then used to perform sensitivity
analysis and optimization.
The results of sensitivity analysis are considered for determination of the operating
conditions of the process and subsequently the target responses (called earnings
function) for the optimization problem performed for this example.
The sensitivity analysis shows that a low recovery of solvent (95 %) and low
solvent feed ratio, none of the independent variables analyzed has great influence.
Differences in energy requirements at each analysis reboiler to 95 % recovery are
shown, because the amount of water in the reboiler to increase the purity require-
ment on top ethanol is increased (Fig. 7.12).
The sensitivity analysis in which varies the glycerol flow, temperature, and the
reflux ratio with a solvent feed ratio of 2.2, introduces substantial changes to what
had been observed with a solvent realcion 0.2 feed. In the solvent temperature
analysis, Fig. 7.13 shows that there is stability in the ethanol composition at 150 C,
where the temperature of the glycerol can evaporate water extracted and increase in
distillate product fraction. In Fig. 7.12 the lower solvent to feed ratio must be to
carry out water extraction shown in ethanol, the solvent feed ratio is about 0.4. In
Fig. 7.14, a large area in which the reflux ratio composition maintains high ethanol
5 1
4 0.99
Ethnaol mole fracition
3 0.98
Qcond
2 Qreb 0.97
Gcal/h
1.5 0.976
0.974
0 Qcond 0.968
Qreb 0.966
-0.5 Etanol mole frac
0.964
-1
0.962
-1.5 0.96
0 50 100 150 200 250
Glycerol Temperature (°C)
Reflux Ratio: 0.4
Solvent to FeedRatio: 0.2
6 0.98
4 0.975
Fig. 7.14 Sensitivity analysis on reflux ratio from first tower (95 % rec)
in the distillate product of high solvent ratio under which feed the process simula-
tion is observed.
A similar analysis for a recovery of 99 % is realized. These sensitivity analyses
help to establish the conditions and limits of process variables when performing
sensitivity analyses with objective functions.
Similarly to the variables shown, it is recommended to perform sensitivity
analyses on all optimization variables (or influences on the objective function), in
order to determine the ranges of these so as to properly identify the area of response
objective function and gain access to a global (or local) maximum easier to
calculate.
7.4 Introductory Example 365
7.4.3 Results
When comparing the results obtained from the sensitivity analysis developed by Gil
and Uyazán (2003) with the values found by the optimization process you may
notice that it is not far from each other. This indicates conducting sensitivity
analysis on the process parameters on the same optimal solution. It is expected
that the number of iterations will be reduced from the point obtained by this
method.
However, the greatest changes occurred in the reflux ratio of the extractive
column and the operating pressure of the regeneration column. For the first, you
may notice that by optimizing the system as a whole, the number of stages is
increased by one, so it is expected that equipment with as many stages can make the
same separation with less reflux. Also the column initial cost increased slightly
Fig. 7.15 Sensitivity analysis results varying solvent flow on tower TD-10
considering operating costs due to solvent flow. Operating cost reduce by half
considering the varying in reflux flow.
For the case of regeneration column, the number of stages reduced. This shows
that the operation of the extraction column has improved such that it allows
performing the recovery with fewer steps, due to increased solvent flow. Also
operating costs are reduced because the pressure that must operate this column is
a little higher as in the base case.
7.5 Summary
Aspen Plus and Aspen HYSYS® have powerful optimization tools into their
interfaces that allow optimization of both simple and complex systems. The exam-
ple developed allowed us to observe the recommendations, specifically the devel-
opment of case studies on process units, are bringing the system to the optimal
solution.
In the example, it was possible to see how to optimize the system as a whole, the
total system performance increases. So the optimization should be considered an
important role of process design aspect, since that improves the operation of each
process unit considering its interaction with other operations. This is one reason
why the computational tools can help you perform complex calculations related to
process engineering.
7.6 Problems
References
Aspen Technology, Inc. (2005) Aspen HYSYS simulator help. Aspen Technology, Cambridge
Edgar TF, Himmelblau DM, Lasdon LS (2001) Optimization of chemical processes, 2nd edn.
McGraw Hill, New York
Gil I, Chasoy W (2010) Determinacion experimental de datos de Equilibrio Lı́quido-Vapor.
Sistema Agua-Glicerina, VI Congreso Argentino de Ingenierı́a Quı́mica
Gil ID, Uyazán AM (2003) Simulacion de la deshidratacion de etanol azeotr opico por destilaci
on
extractiva. Thesis in Chemical Engineering, Department of Chemical and Environmental
Engineering, Universidad Nacional de Colombia
Gomez P, Esteban H (2006) El método SQP de optimizaci on con restricciones. XXI Simposium
Nacional de la Union Cientı́fica Internacional de Radio
Grossmann IE (2002) Review of nonlinear mixed-integer and disjunctive programming tech-
niques. Optim Eng 3(3):227–252
Langston P, Hilal N, Shingfield S, Webb S (2005) Simulation and optimisation of extractive
distillation with water as solvent. Chem Eng Process 44(3):345–351
Lee FM, Pahl RH (1985) Solvent screening study and conceptual extractive distillation process to
produce anhydrous ethanol from fermentation broth. Ind Eng Chem Process Des Dev, Phillips
Petroleum Company 24(1):168–172
Miller-Klein Associates (2006) Impact of biodiesel production on the glycerol market
Pagliaro M, Rossi M (2010) The future of glycerol, 2nd edn, RSC green chemistry series. Royal
Society of Chemistry, Cambridge
Ravagnani MS (2010) Anhydrous ethanol production by extractive distillation: a solvent case
study. Process Saf Environ Prot 88(1):67–73
References 369
Seider JD, Warren D (2003) Product & process design principles: synthesis, analysis and evalu-
ation, 2nd edn. Wiley, Somerset
Taha HA (1998) Investigacion de operaciones, 6th edn. Pearson Education, Harlow
Tarquin AJ, Blank LT (2001) Ingenierı́a economica. McGraw Hill, New York
Uyazán AM (2006) Produccion de alcohol carburante por destilaci on extractiva: simulaci
on del
proceso con glicerol. Revista Ingenierı́a E Investigacion 26(1):45–50