Int. J. Engng Ed. Vol. 20, No. 6, pp. 928±938, 2004 0949-149X/91 $3.00+0.

00
Printed in Great Britain. # 2004 TEMPUS Publications.

Spreadsheets in Chemical Engineering

EducationÐA Tool in Process Design and
Process Integration*
EUGEÂNIO C. FERREIRA
Centro de Engenharia BioloÂgica, Universidade do Minho, Campus de Gualtar, 4710-057 Braga, Portugal.
E-mail: ecferreira@deb.uminho.pt
RICARDO LIMA and ROMUALDO SALCEDO
Departamento de Engenharia QuõÂmica, Universidade do Porto, Rua Dr. Roberto Frias, 4200-465 Porto,
Portugal
Recent developments in embedding numerical optimization procedures with linear and nonlinear
solvers within a spreadsheet environment have greatly enhanced the use of these tools for teaching
chemical process design and process integration. Student skills with respect to these topics are
usually gained by complex and expensive modular simulators, e.g. ASPEN Plus1 or algebraic tools
such as GAMS1 or AMPL1. However, modular simulators have a significant learning curve, and
algebraic modeling languages are usually ignored once students commence careers. This paper
demonstrates how the Solver feature of the Excel1 spreadsheet is used for the optimization of
several chemical engineering systems, including pollution prevention problems and mass-exchange
networks. Three nonlinear problems are examined: the (a) recovery of benzene from a gaseous
emission; (b) design of a chemical reactor network; and (c) solution of material balances in the
production of vinyl chloride from ethylene. Dephenolization of aqueous wastes is presented as a
linear case. The ease with which these and similar process problems can be formulated and solved
within the Excel1 environment constitutes a major step towards teaching practical optimization
and design concepts for university students.

INTRODUCTION to establish to what extent these tools are capable

of solving optimization problems.
UNDERGRADUATE ENGINEERING students The present authors studied recently an interest-
are attracted to the powerful `what-if ' spreadsheets ing problem dealing with the concepts of process
with optimization capabilities, such as the synthesis including heat integration and solvent
EXCEL1 Solver (Microsoft Co.) and What's recovery [1, 2, 8]. The Solver feature of the
Best (Lindo Systems, Inc.). They require a mini- Excel1 spreadsheet is demonstrated for the opti-
mum amount of effort in building a typical simula- mization of several chemical engineering systems,
tion/optimization problem, in comparison with including pollution prevention problems and mass-
standard high level language coding such as exchange networks in the current paper. Three
GAMS1 or AMPL1. Undergraduate instructors nonlinear problems (the recovery of benzene
are adopting Excel Solver for introducing students from a gaseous emission; the design of a chemical
to solving and optimizing process design and reactor network; and the solution of material
integration [1, 2]. In addition, several engineering balances in the production of vinyl chloride from
textbooks now include coverage of the Excel ethylene), and one linear problem (the dephenoli-
Solver [3±6]. The new edition of the classical text- zation of aqueous wastes) are examined. These
book Optimization of Chemical Processes [4] dedi- case studies have been adapted for demonstration
cates several pages to the use of Excel Solver as an purposes in two courses run by the authors.
optimization tool. The book includes a new co-
author, Leon Lasdon, a recognized authority in
operations research optimization software and
implementation of the Excel Solver [7]. THE EXCEL SOLVER
Practicing engineers also use spreadsheets for
many tasks, and process optimization is steadily The Microsoft Excel1 spreadsheet was used as a
becoming a common task in process synthesis, development framework, coupled with the Solver
design and integration. Therefore, it is important add-onÐa companion of Excel since 1991 (version
3.0). The Excel Solver has two nonlinear uncon-
strained optimizers, a quasi-Newton method and a
* Accepted 14 July 2004. reduced gradient method. These are used within a

928
 Spreadsheets in Chemical Engineering Education 929

Fig. 1. Material balances in the production of vinyl chloride from ethylene.

Generalized Reduced Gradient algorithm [9] for change rapidly, especially near the boundaries of
solving constrained optimization problems. The the active constraints. It is possible to control;
linear simplex method with bounds on the vari-
a) the solution process by limiting the time taken
ables, and the branch-and-bound method imple-
and the number of interim calculations by the
mented by Fylstra et al. [7], can be used for solving
solution process;
linear and integer problems.
b) the precision within which constraints are con-
The approach used to obtain better initial esti-
sidered binding;
mates of the basic variables in each one-dimen-
c) the convergence criteria for the solutions.
sional search can be specified in Solver options.
Linear extrapolation from a tangent vector or
quadratic extrapolation can be used, which may Example 1: Material balances in the production of
improve the results on highly nonlinear problems. vinyl chloride
It is also possible to specify the differencing This case study illustrates the use of Excel Solver
method to estimate derivatives of the objective in the solution of simultaneous nonlinear equations
and constraint functions: `Forward' when the associated with material balances in the production
constraint values change relatively slowly, or of vinyl chloride from ethylene. DeLancey
`Central', used for problems when the constraints [10] solved this example using Scientific Notebook
930 E. Ferreira et al.

Table 1. Material balances and stoichiometric equations [10]

(MacKichan Software, Inc.) primarily oriented for product stream, F9 , consists only of vinyl chloride
solving systems of nonlinear equations. and water.
The flow diagram in Fig. 1 represents the main Setting F1 ˆ 1 mole/hr results in a problem with
steps in the production of vinyl chloride (C2 H3 Cl) 24 independent (unknown) variables and 26 equa-
from ethylene (C2 H4 ). tions issued from material balances (Table 1), with
The reactions taking place separately in each no degrees of freedom. The EXCEL Solver was
reactor are: used to determine all of the unknown flow rates,
Chlorination: Fj , and mole fractions, xij (mole fraction i in
C2 H4 + Cl2 ! C2 H4 Cl2 stream j). The species are labeled in Table 2.
The Solver is used to compute the root of one
Oxyhydrochlorination: equation subject to several equality constraints
C2 H4 + 2HCl + ‰O2 ! C2 H4 Cl2 + H2 O (Fig. 2). Equation 1 ($C$36) was set as `Target
Cell' with a required zero value and all the 26
Pyrolysis:
equations were set as equality constraints. The
C2 H4 Cl2 ! C2 H3 Cl + HCl
initial values for the decision variables (`By Chan-
The ethylene feed, F1 , is 90% molar ethylene and ging Cells') were 1.0 for all the flow rates (Fj ) and
the remainders are inerts. The chlorine and oxygen 0.50 to all the mole fractions (xi,j). The solution is
feeds, F2 and F3 , respectively, are pure. All of the obtained almost instantaneously.
ethylene, oxygen, and chlorine react and the
conversion of the hydrochloric acid (HCl) fed to Example 2: Dephenolization of aqueous wastes
the oxyhydrochlorination is complete. This example is used to illustrate the synthesis of
Only 50% of the total dichloroethane (C2 H4 Cl2 ) mass-exchange networks based on a mathematical
fed to the pyrolysis reactor is converted, with the programming approach. For an overview of this
remainder being separated and recycled with inerts technique the reader is referred to El-Halwagi [11].
in stream F12 . The inert concentration in the An oil-recycling plant is demonstrated in Fig. 3
recycle stream is 50% molar. Pure hydrochloric as adapted from [11]. Two types of waste oil are
acid (HCl) is recycled in stream F13 . The final handled: gas and lube oils. The two streams are

Table 2. Labeling of components

Index 1 2 3 4 5 6 7 8

Species C2 H4 Cl2 HCl O2 C2 H4 Cl2 C2 H3 Cl H2 O Inerts

 Spreadsheets in Chemical Engineering Education 931

Fig. 2. Solving simultaneous nonlinear equation associated with material balances in the production of vinyl chloride from ethylene.

Fig. 3. Dephenolization of aqueous wastes in an oil recycling plant.

Table 3. Data on waste streams for the dephenolization example [11]

Flow rate Supply Target

Stream Description Gi , kg/s composition composition

R1 Condensate from 2 0.050 0.010

first stripper
R2 Condensate from 1 0.030 0.006
second stripper

first de-ashed and de-metallized. Atmospheric The principal pollutant in both wastewater
distillation is used to obtain light gases, gas oil, streams is phenol that can be separated using
and a heavy product. The heavy product is distilled several techniques. Solvent extraction using gas
under vacuum to yield lube oil. The gas and the lube oil (S1) or lube oil (S2) as process Mass Separation
oils can be further processed to attain other proper- Agents (MSA) is an option. The data for the waste
ties. The gas oil is steam stripped to remove light and streams and the process MSA are given in Tables 3
sulphur impurities, then hydrotreated. The lube oil and 4 respectively.
is dewaxed/deasphalted using solvent extraction Three external technologies are also considered
followed by steam stripping. The process has two for the removal of phenol. These processes include
main sources of waste water. These are the conden- adsorption using activated carbon, S3, ion exchange
sate streams from the steam strippers. using a polymeric resin, S4, and stripping using air,
932 E. Ferreira et al.

Table 4. Data process mass separation agents for the dephenolization example [11]

Equilibrium
Upper bound Supply Target distribution Cost Cj ($/kg of
on flow rate composition composition coefficient recirculation
Stream Description Lcj , kg/s xsj xtj mj ˆ y/xj MSA)*

S1 Gas oil 5 0.005 0.015 2.00 0.000

S2 Lube oil 3 0.010 0.030 1.53 0.000
S3 Activated carbon 1 0.000 0.110 0.02 0.081
S4 Ion-exchange 1 0.000 0.186 0.09 0.214
S5 Air 1 0.000 0.029 0.04 0.060

* Including regeneration and make-up costs.

S5. The equilibrium data for the transfer of phenol to using the LINGO package (Lindo Systems, Inc.).
the jth lean stream is given by y ˆ mj xj where the The objective function is:
values of mj are given in Table 4. Also, listed are the
supply and targetcompositions and unit cost data for minf0:081L3 ‡ 0:214L4 ‡ 0:060L5 g
each MSA. Throughout this example, a minimum
allowable composition difference, "j , of 0.001 (kg subject to:
phenol)/(kg MSA) is used. 1 ˆ 0.0052
An analysis based on `pinch diagrams' (see [11] 2 ± 1 ‡ 0.0101L2 ˆ 0.0308
for details) indicates that 0.0184 kg phenol/s is the 3 ± 2 ‡ 0.0010L1 ‡ 0.0013L2 ˆ 0.0040
excess capacity for the process MSA and that 4 ± 3 ‡ 0.0066L1 ‡ 0.0086L2 ˆ 0.0396
0.0124 kg phenol/s are to be removed using an 5 ± 4 ‡ 0.0024L1 ‡ 0.0537L4 ˆ 0.0144
external MSA. 6 ± 5 ‡ 0.0222L4 ˆ 0.0060
The problem of minimizing the operating cost of 7 ± 6 ‡ 0.0444L4 ˆ 0.0040
mass separation agents was formulated in [11] by 8 ± 7 ‡ 0.0420L4 ˆ 0.0000
adopting the linear-programming approach solved 9 ± 8 ‡ 0.0510L3 ‡ 0.0114L4 ˆ 0.0000

Fig. 4. Minimizing operating cost in the dephenolization of aqueous wastes.

Spreadsheets in Chemical Engineering Education 933

Fig. 5. Recovery of benzene from a gaseous emission.

934 E. Ferreira et al.

10 9 ‡ 0.0555L3 ‡ 0.0123L4 ‡ 0.0277L5 ˆ 0.0144 into oil. The oil is then fed to a regeneration system
11 10 ‡ 0.0025L3 ‡ 0.0013L5 ˆ 0.0000 in which it is heated and passed on to a flash
±11 ‡ 0.0010L3 ˆ 0.0000 column that recovers benzene as a top product.
k  0, k ˆ 1, 2, . . . , 11 The bottom product is the regenerated oil, which
Lj  0, j ˆ 1, 2, . . . , 5 contains 0.08 mole/mole% benzene. Finally, the
L1  5, regenerated oil is cooled and pumped back to the
L2  3. absorber.
The EXCEL Solver is used to assess the optimal
where Lj is the flow rate of the jth MSA, k±1 and
flow rate of recirculating oil that minimizes the
k are the residual masses of the key pollutant
total annualized cost (TAC) of the system.
entering and leaving the kth interval. The first set
The design equations for this process follow:
of 11 equality constraints represents successive
material balances around each composition inter- . maximum practically feasible outlet composition
val. Setting k  0 enables the waste streams to of the MSA which satisfies the assigned driving
pass the mass of the pollutant downwards if it does force, "j :
not fully exchange it with the MSAs in a given yin bj
interval. Setting Lj  0 guarantees that the optimal xout;max
j ˆ i
"j
mj
flow rate of each MSA is non-negative. The last
two constraints are upper bounds on flow rates less . Flow rate of oil:
than the total available quantity of the corres-
ponding lean stream. …yin
i yout
i †
The EXCEL Solver is used to optimize the Lj ˆ Gi out
…xj xin
j †
objective function, which was set as the `Target
Cell' for minimization (Fig. 4). The initial values .
for the variables were set to 1.0 for all the flow …xj xj †log mean
rates (Lj ) and to 0.001 for all the mole fractions   in    out 
(k ). The solution is quickly obtained by selecting yi bj yi bj
xout
j x in
j
the `Linear Model' option in the Solver Parameters mj mj
since all relationships are linear as is the objective ˆ 8h  in i9
yi bj
< xj out =
function. The binding of the equality constraints mj
ln h  out i
determines the residual masses of the key pollutant : xin yi bj ;
entering and leaving the kth interval (k ). The j mj
solution for the MSA optimized flow rates
. Number of Transfer Units,
is {L1 , L2 , L3 , L4 , L5 } ˆ {5.0000, 2.0800, 0.1127,
0.0000, 0.0000}. Therefore, activated carbon (L3 ) xin xout
j j
is the optimum external MSA. The same minimum NTUy ˆ
operating cost can also be achieved by other …xj xj †log mean
combinations of the process MSA, since both L1
and L2 are inexpensive. . Absorver column height, H ˆ NTUy HTUy
. Column diameter,
Example 3: Recovery of Benzene from a Gaseous s
Emission 4…volumetric flowrate of gas†
Dˆ
This example is used to illustrate the application …gas superficial velocity†
of the EXCEL Solver to solution of a nonlinear
optimization problem [11] which is to remove Cost equations:
benzene from a gaseous emission by contact with
an absorbent (wash oil, molecular weight 300). The . Annual operating cost, AOC ˆ Co Lj (8000 yrs/
gas flow rate is 0.2 kmole/s and it contains 0.1% annum)
molar (1000 ppm) of benzene. The molar mass of . Fixed cost of installed shell and auxiliaries,
the gas is 29 g/mole, its temperature is 300 K, at a FC1 ˆ 2300H 0:85 D0:95
pressure of 141 kPa. It is desired to reduce the . Cost of packing, FC2 ˆ 800(/4)HD2
benzene to 0.01 mole/mole% using the system . Annualized Fixed Cost, AFC ˆ (FC1 ‡ FC2)/
shown in Fig. 5, where benzene is first absorbed Depreciation
. Total Annualized Cost, TAC ˆ AOC ‡ AFC

Constraints:
. Mass velocity of oil,

Lj Moil kg
 2  2:7 m2 s
D
4
Fig. 6. Using `Goal Seek' to find the value 2.70 for the mass . "j  0.005
velocity of oil ($I$47) by adjusting "j . . "j  0.00072
 Spreadsheets in Chemical Engineering Education 935

Fig. 7. Reactor network optimization.

These last constraints on the minimum allowable using the `Goal Seek' tool available in EXCEL.
composition difference at the lean end of the mass This feature allows one to find a specific result for
exchanger are upper and lower bounds to the a cell by adjusting the value of any other cell by
search space. The lower bound, "j  0.00072, is solving iteratively the sequence of nonlinear equa-
equivalent to the constraint on mass velocity of oil tions. The goal cell (mass velocity of oil) was set to
being less than 2.7 kg m±2 s±1 and was calculated 2.7 by changing the "j cell value (see Fig. 6).
936 E. Ferreira et al.

Table 5. Local optima for the reactor network optimization [12]

Variables Local 1 Local 2 Local 3 Local 4 Global

F6 0 0 0 1 1
F7 0 0 0 0 1
CA4 0 undefined 1 1 0.771509
CA6 0 undefined 0.658770 0.392874 0.516993
V2 0 0 5.362991 16 5.097112
Fobj 0 0.374617 0.386745 0.388102 0.388811

The EXCEL Solver was then used to find the best subject to;
value "j to minimize the total annualized cost. The
initial value of "j was set to 5  10±4 and the optimum F1 ‡ F2 ± F0 ˆ 0
one was 1.25  10±3 which corresponds to a mini- F1 ‡ F8 ± F3 ˆ 0
mum TAC of $41,386/yr. In [11] an optimum of F2 ‡ F7 ± F4 ˆ 0
$41,560/yr was found by trying several initial F5 ‡ F7 ± F3 ˆ 0
values of "j until a minimum TAC was identified. F6 ‡ F8 ± F4 ˆ 0
By limiting the search space for "j with upper and F1 CA0 ‡ F8 CA6 ± F3 CA3 ˆ 0
lower bounds, the EXCEL Solver converges to the F1 CB0 ‡ F8 CB6 ± F3 CB3 ˆ 0
optimum irrespectively of the initial value for this F2 CA0 ‡ F7 CA5 ± F4 CA4 ˆ 0
parameter. A previous attempt without these F2 CB0 ‡ F7 CB5 ± F4 CB4 ˆ 0
bounds, and using only the constraint on mass F3 …CA3 ± CA5 † ± k1 V1 CA5 ˆ 0
velocity of oil, was too sensitive to the initial value F3 …CB3 ± CB5 † ‡ (k1 CA5 ± k3 CB5 †V1 ˆ 0
of "j . F4 …CA4 ± CA6 † k2 V2 CA6 ˆ 0
F4 …CB4 ± CB6 † ‡ (k2 CA6 ± k4 CB6 †V2 ˆ 0
k1 ˆ 0.09755988 h 1
Example 4: Design of a chemical reaction network k2 ˆ 0.99k1 h 1
The optimal way to solve network problems is by k3 ˆ 0.0391908 h 1
specifying binary variables that multiply in an k4 ˆ 0.90k3 h 1
appropriate (logical) way the continuous variables, CA0 ˆ 1 kg.m 3
such that nonexistent units (corresponding to a CB0 ˆ 0 kg.m 3
zero binary variable associated with that unit) are F0 ˆ 1 m3.h 1
treated with mathematical consistency. Thus, a
and to the inequality constraints:
network problem becomes a mixed integer nonlinear
programming (MINLP) problem. However, there 0  Fi  1 (volumetric flow rates, m3 h 1 );
are many examples of such problems that are i ˆ 1; . . . ; 9
specified as pure nonlinear programming (NLP) 0  CAi ; CBi  1 (concentrations, kg.m 3 );
problems, e.g. Fig. 7 [12]. i ˆ 1; . . . ; 9
The reactions are first order, the reactors are 0  Vi  16 (reactor volumes, m3 );
perfectly mixed (steady-state) and there is no vari- i ˆ 1; 2
ation in density of the reacting mixture. The V10:5 ‡ V20:5  4
problem may be formulated as follows:
The ki are 1st order rate constants.
  This problem has 5 degrees of freedom, and a
F5 CB5 ‡ F6 CB6 decomposition algorithm can be employed to
Fobj ˆ maxfCB9 g ˆ max
F9 optimize the simulation step [13]. A sequential

Table 6. Comparison between GAMS and the EXCEL Solver for the reactor network optimization

Starting point
Lower bounds
Algorithm (10 5) Upper bounds Midpoints

GAMS Solvers:
MINOS Local2 Local3 Global
MINOS5 Local4 Local3 Local3
SNOPT Local2 Local2 Local4
CONOPT Global Local4 Global
CONOPT2 Global Local4 Local4
EXCEL Solver:
Quasi-Newton Non-feasible Global Global
Conjugate gradient Non-feasible Local3 Global
 Spreadsheets in Chemical Engineering Education 937

solution is not possible, no matter what the choice initial vector [10±2, . . . .. 10±2]T in both worksheets
of decision variables, and several solutions with a available.
subsystem of a minimum of 4 equations were
encountered [14]. This subsystem is linear and
CONCLUSIONS
readily solved. However, since in this example we
compare the results obtained by the EXCEL
The problems analyzed in this paper are of
Solver to the optimizers available within GAMS
sufficient complexity to allow the observation of
[15], the full nonlinear set of equations was solved as
some convergence problems within the EXCEL
an alternative. The problem was solved starting Solver. The solution may be dependent on the
from 3 different solution vectors, corresponding
initialization and is in general a local optimum.
respectively to the lower bounds (assumed 10±5 in
Since the optimizers within the EXCEL environ-
order to avoid numerical difficulties with the opti-
ment are local search algorithms [7], convergence
mizers), to the upper bounds and to the midpoints
to the global optimum is only guaranteed with
of the search intervals. Table 5 demonstrates the
convex problems. This is also applicable to the
four local optima obtained by [12] using a systema-
nonlinear optimizers available within GAMS [15].
tic search coupled with deterministic algorithms for
However, initializations that correspond to reason-
some selected variables corresponding to the 5
able and practical designs generally progress
degrees of freedom. Also shown is another local
towards feasible and locally optimal solutions.
optima obtained with GAMS in a previous study The EXCEL Solver is not as good as robust
[14], which simply corresponds to no reaction and to
global optimizers [14]. However, it does provide an
the closure of the mass balances (Table 6). Thus, in
integrated framework for problem setting, visual-
only 27% of the runs was the global optimum ization, inspection and solving of particular
obtained with the algorithms available within
utility for practicing engineers. The ease with
GAMS. The EXCEL Solver found it in 50% of the
which these and similar process problems can
runs. However, GAMS could always find some
be formulated and solved within the Excel
local optimum, while the EXCEL Solver failed to
environment constitutes a major step towards
find feasible solutions in 33% of the runs, irrespec-
teaching practical optimization and design
tive of the options available (derivatives and step
concepts, ultimately benefiting students with
sizes). This behavior cannot be extrapolated to
knowledge acquisition procedures and later in an
other problems, and many more examples are
effortless continuing practice throughout careers.
required to compare these solvers.
Despite these benefits, the use of modular simu-
This example also demonstrates how the Excel lators such as ASPEN Plus is probably warranted
Solver may handle floating point exceptions (viola-
for complex processes that need detailed stage
tions on the equality or inequality mathematical
calculations and extensive use of physical and
domain, viz. divisions by zero and forbidden thermodynamic data libraries.
arguments of transcendental functions). This may
All these Excel workbook files and other exam-
be simply circumvented by checking for forbidden
ples are available for download on the Internet at
operations or values before the expression is eval-
www.deb.uminho.pt/ecferreira/download.
uated. If a violation occurs, a flag is enforced and
the objective function is penalized directly, and AcknowledgementsÐThe authors wish to thank Dr Russell
avoids premature stoppage of the optimization Paterson for his helpful comments and corrections. R. Lima
was supported by a PRAXIS-XXI BD/21481/99 grant from
procedure. This approach may be easily verified FCTÐFundacËaÄo para a CieÃncia e a Tecnologia, Portugal.
by downloading Example 4 and specifying as

REFERENCES

1. E. C. Ferreira and R. L. Salcedo, A tool to optimize VOC removal during absorption and
stripping, Chemical Engineering, 108(1), January 2001, pp. 94±98.
2. E. C. Ferreira and R. L. Salcedo, Can spreadsheets solve demanding optimization problems?
Comp. Appl. in Eng. Education, 9(1), 2001, pp. 49±56.
3. W. G. Filby (ed.), Spreadsheets in Science and Engineering, Springer, Berlin, (1998).
4. T. F. Edgar, D. M. Himmelbau and L. S. Lasdon, Optimization of Chemical Processes, 2nd ed.,
McGraw-Hill Education, New York (2001).
5. B. V. Liengme, Excel 2002 for Scientists and Engineers, 3rd ed., Butterworth-Heinemann, London
(2002).
6. P. Cornell, Accessing and Analyzing Data with Microsoft1 Excel, Microsoft Press, Redmond
(2003).
7. D. Fylstra, L. Lasdon, J. Watson and A. Waren, Design and use of the Microsoft Excel Solver,
Interfaces, 28(5), 1998, pp. 29±55.
8. R. Lima and R. L. Salcedo, An optimized strategy for equation-oriented global optimization,
European Symp. on Comp. Aided Chem. Eng, Vol. 10, J. Grievink and J. V. Schijndel (eds), Elsevier
Science BV, pp. 913±918 (2002).
9. L. S. Lasdon, A. D. Waren, A. Jain and M. Ratner, Design and testing of a generalized reduced
gradient code for nonlinear programming, ACM Trans. Math Software, 1(4), 1978, pp. 34±50.
10. G. B. DeLancey, Process analysis an electronic version, Chemical Engineering Education, Winter,
pp. 40±45, (1999).
938 E. Ferreira et al.

11. M. M. El-Halwagi, Pollution Prevention through Process Integration: Systematic Design Tools,
Academic Press, San Diego (1997).
12. S. H. Choi, J. W. Ko and V. Manousiouthakis, A stochastic approach to global optimization of
chemical processes, Comp. Chem. Eng., 1999, pp. 1351±1356.
13. R. L. Salcedo and R. Lima, On the optimum choice of decision variables for equation-oriented
global optimization, Ind. Eng. Chem. Res., 38, 1999, p. 4742.
14. R. L. Salcedo, R. Lima and M. F. Cardoso, Global optimization of chemical processes using
simulated annealing, Proc. Indian Nat. Sci. Ac., 69A(3±4), 2003, pp. 359±401.
15. A. Brooke, D. Kendrick and A. Meeraus, GAMS: A user's guide, release 2.25, The Scientific Press
Series, Boyd Fraser Pub. Co., San Francisco (1992).

EugeÂnio C. Ferreira is Associate Professor of biological engineering at the University of

Minho in Braga, Portugal. He received a B.Che.E. degree and Ph.D. degree from the
University of Porto in 1986 and 1995, respectively. His current research interests include
modeling and control in bio(chemical) and wastewater treatment processes. E. C. Ferreira is
currently teaching Strategy of Process Engineering, Process Control, Wastewater Treatment,
and Air Pollution. More details at author's homepage: www.deb.uminho.pt/ecferreira.

Romualdo Salcedo is Professor of chemical engineering at the University of Porto, Portugal.

He received a B.Che.E. degree from the University of Porto in 1975, a M.Eng. and a Ph.D.
degree from McGill University, Montreal, Canada respectively in 1977 and 1981. His
current research interests include simulation and optimization of nonlinear processes and
air pollution control. R. Salcedo is currently teaching Process Design and Optimization and
Air Pollution Control. He holds patents on numerically optimized gas cyclones and
recirculation systems, which resulted in industrial applications.

Ricardo Lima is a Ph.D. graduate student in chemical engineering at the University of

Porto. His work relates to the development of an integrated strategy for simulation and
optimization of large-scale processes, using stochastic optimizers.