EVALUATION OF WASTE MINIMIZATION ALTERNATIVES UNDER

UNCERTAINTY: A MULTIOBJECTIVE OPTIMIZATION APPROACH

Mauricio M. Dantus and Karen A. High*

School of Chemical Engineering

Oklahoma State University

423 Engineering North

Stillwater, OK 74078

*Author to whom all correspondence should be addressed

1
ABSTRACT

A general methodology is proposed for implementing waste minimization programs.

This paper focuses primarily in the evaluation of source reduction alternatives using a two

criteria approach. The first criteria considers the economic performance of the process that

includes all the waste related costs within an environmental accounting framework. Whereas the

second criteria evaluates the environmental impact of the alternative taking into account a release

factor, as well as toxicological data for each chemical present in the process stream. Realizing

the uncertainty present in the evaluation of potential source reduction projects, the evaluation

phase is accomplished using a multiple objective optimization approach by combining the

compromise programming method and the stochastic annealing algorithm using the process

simulator ASPEN PLUS. As an example, the methyl chloride process is evaluated using the

proposed methodology.

Keywords: waste minimization, stochastic optimization, multiobjective optimization,

uncertainty, environmental costs, and environmental impact

2
INTRODUCTION

Each year, the US industry generates more than 14 billion tons of waste, including

gaseous emissions, solid wastes, sludge, and wastewater (DOE, 1997). These wastes have

traditionally been managed using end-of-the-pipe treatment technologies that seek to eliminate as

much as possible the amount of pollution produced. Even though this approach has been

successful in achieving its goal, it can only be considered a temporary solution, since in the long

run it is not very effective to solve the pollution problem. As a result, an alternative that has been

pursued is the implementation of waste minimization programs as part of an agenda towards a

sustainable development. Attempts have been made to promote the implementation of such

programs by identifying their potential benefits. However, despite these benefits a report by the

Environmental Protection Agency (EPA) (EPA, 1992) suggests that the majority of the US

manufacturers have been slow to move away from the traditional end-of-the-pipe strategies. The

main reason has been the difficulty in establishing the different environmental costs associated

with a particular operation.

The environmental costs should not be the only factor considered in the evaluation of

source reduction alternatives. With the same degree of importance, the overall environmental

impact of the process —generally difficult to quantify in monetary terms— should be consider as

a complementary decision tool. Furthermore, as the investment question is analyzed under a

broader perspective the analyst becomes aware of additional factors of which the analyst or the

decision maker has no control over. In most instances the decision to invest needs to be made

with incomplete or uncertain information. Therefore, one can question the applicability of the

traditional deterministic approach used in the design or retrofit of industrial processes.

3
 Numerous procedures have been proposed for the design and retrofit of chemical

processes (Douglas, 1988; Grossmann and Kravanja, 1995; Pohjola et al., 1994; Wilkendorf et

al., 1998). However, only few are aimed towards minimizing the pollution generated in a

process (Alvaargaez et al., 1998; Douglas, 1992; Fonyo et al., 1994; Hopper et al., 1992; Mallick

et al., 1996; Manousiouthakis and Allen, 1995; Spriggs, 1994). These methods generally look at

options that are evaluated under a single objective approach, either minimize the amount of

waste generated, or maximize profit. Yet, no detail is given on how this profit should be

evaluated in order to incorporate all waste related costs.

Some work has been done in simultaneously evaluating alternatives under the two

previous criteria —maximize profit and minimize the amount of waste generated— (i.e., Ciric

and Huchette, 1993). In their procedure waste reduction options are evaluated based only on a

deterministic approach and do not consider the type of waste that is generated. The type of waste

as a function of its toxicity characteristic is addressed for example by Cabezas et al. (1997),

Mallick et al. (1996), and Stefanis and Pistikopoulos (1997). These methodologies focus also on

a deterministic perspective and the first two do not consider the process’ economic impact. The

deterministic caveat has been addressed for example by Chaudhuri and Diwekar (1996);

Chaudhuri and Diwekar (1997); and Ierapetritou et al. (1996), but only as a single objective

optimization problem.

In this context, a comprehensive methodology is presented that takes into account the

uncertainties present when evaluating process alternatives to reduce the waste generated in a

chemical process. The methodology incorporates stochastic and multiple objective optimization

techniques, to select the alternative that maximizes the process’ profit and minimizes its

environmental impact. This is accomplished using the process simulator ASPEN PLUS.

4
 The second part of this article presents an example of the application of the proposed

methodology as applied to the manufacture of methyl chloride. This process was selected due to

its potential for improvement and for its environmental restrictions and limitations.

IMPLEMENTING WASTE MINIMIZATION PROGRAMS

The implementation of waste minimization programs can encompass a series of activities

that can vary in complexity, and can be successfully implemented during various phases of the

process, from the design stage to the retrofit of existing operations. Whatever the project’s scope

is, starting from the basic five step framework outlined by CMA (1993), the proposed

methodology for implementing waste minimization programs consists of six major steps (see

Figure 1): characterization of process streams, evaluation of environmental impacts, development

of process model, and identification, evaluation, and implementation of pollution prevention

alternatives. As needed, this methodology can be followed in a different order and repeated until

the specific goals and specifications have been met.

CHARACTERIZATION OF PROCESS STREAMS

Process streams are characterized by source, destination, flowrate, composition, and

properties. The characterization study should include waste and non-waste streams to help

identify possible areas where it could be worthwhile to conduct source reduction studies. And,

since the characterization study might involve some capital investment, a decision needs to be

made on the number of streams to include and on the study’s level of detail. A possibility is to

include those streams that represent a potential environmental impact, a possible target of

environmental regulations, and/or an important cost effect.

5
EVALUATION OF ENVIRONMENTAL IMPACTS

This phase identifies the potential environmental impact of the different streams incurred

from the possible release to air, water, and land. At this point starting from the idea taken from

Alliet Gaubert and Joulia (1997) it is useful to generate Specific Chemical Flowsheets (SCF) to

identify the path taken by a critical component. For example, Figure 2 shows the SCF for carbon

tetrachloride. In this manner, pollution prevention alternatives that seek to reduce the amount of

carbon tetrachloride generated in the methyl chloride process might be easier to identify. Under

the SCF approach, a stream is considered to be part of a path if the concentration of the chemical

is greater than a reference concentration, that is a function of the chemical in question, its

environmental impact, its economic importance, and the degree of recovery desired.

DEVELOPMENT OF PROCESS MODEL

The information generated in the previous steps is used to develop a process model that

serves as an analysis tool to evaluate the current performance of the process and the behavior of

possible source reduction alternatives. The units and streams to include in the model are an

important factor, since the more units or streams considered will make the model more accurate

but will also increase its computational requirements.

IDENTIFICATION OF POLLUTION PREVENTION ALTERNATIVES

The identification of pollution prevention is generally accomplished with hierarchical

review approaches and other methods of structured thinking (Spriggs, 1994). These techniques

6
are useful in generating and identifying possible alternatives. However, they are considered only

as a screening tool and they do not attempt to find the best option.

In the hierarchical review approach, one of the most important contributions is that of

Douglas (1992) that employs at each level a series of heuristic rules that can be used to identify

pollution prevention alternatives. Several other techniques that have emerged for process design

and retrofit can be used together with this procedure as a mean to generate additional alternatives.

Two important strategies are those of mass exchanger network (MEN) (El-Halwagi and

Manousiouthakis, 1989) and heat exchanger network (HEN) (Linnhoff, 1994) analysis. In

addition, several lists have been published that can serve as complementary sources for source

reduction ideas (Chadha, 1994; CMA, 1993; Doerr, 1993; Dyer and Mulholland, 1998; Nelson,

1990; Siegell, 1996).

Whichever methodology is employed to identify pollution prevention alternatives there

are four possible types of changes that can lead to pollution reduction in a manufacturing process

(EPA, 1988): (1) product changes, (2) input material changes, (3) technology changes, and (4)

good manufacturing practices. In addition, when source reduction alternatives are considered

they should be analyzed from a macro perspective by looking at the complete operation.

EVALUATION OF POLLUTION PREVENTION ALTERNATIVES

The evaluation of pollution prevention alternatives is probably the most important step

since its main purpose is to select which if any of the alternatives should be implemented. As

shown in Figure 1, this step starts with the definition of the decision problem. Following a

decision theory framework, the methodology considers that the decision to implement a project is

made by a single person (individual decision making process). Furthermore, to handle the

7
uncertainties present in the decision making process (see Table 1), it considers that some

information is available or can be assumed (decision making under risk). Thus selecting the

alternative that maximizes or minimizes the expected value of a given criterion. Lastly, the

decision problem incorporates two objectives that represent the desired state of the system. That

is, alternatives should seek to “maximize the profit” and minimize the environmental impact.”

Measuring the process’ profit

The first objective seeks to maximize the amount of profit that can be obtained from a

particular investment. Among the different profitability tools available in the literature the

annual equivalent profit (AEP) for a specific project lifetime Ny given by:

? Ny Fn ?? ir (1 + ir )Ny ?
AEP = ?? n ?? ?
?n = 0 (1 + ir ) ??(1 + ir ) − 1?
Ny
(1)

was selected as the first objective’s attribute. Where the cash flow F, a function of the cash

inflow Fi and outflow Fo , can be calculated in a tabular form (e.g., Peters and Timmerhaus,

1991) or with

Fn = (Fi , n − Fo, n )(1 + i f ) (1 − Tx ) + I D f DTx

n
(2)

Where, once the tax rate Tx, the depreciation factor fD, the depreciable investment ID, and the

annual inflation rate if, together with the interest rate ir are specified, the annual equivalent profit

for each alternative can be calculated.

8
 The identification of the various cash flows (e.g., environmental costs) associated with a

pollution prevention project is not so easily accomplished. Hence to aid in their identification,

the economic performance of the process should be measured under an environmental accounting

framework, where the environmental costs and benefits in increasing complexity can be divided

in five groups: (1) usual costs, (2) direct costs, (3) hidden costs, (4) liability costs, and (5) less

tangible benefits. A detailed description of these costs is given by Dantus (1999)

USUAL COSTS: These include the total fixed capital investment and the production

costs generally associated with the process or product. The total fixed capital investment is the

amount of money required to supply the necessary equipment and manufacturing facilities, plus

the amount of money required as working capital for operation of these facilities. Whereas the

production costs include all the expenses related to the manufacturing operation (i.e., raw

materials, power and utilities, and operating supplies).

DIRECT COSTS: The direct costs W of a process include the capital, operating, material,

and maintenance costs involved in the treatment, recycling, handling, transportation, and disposal

of waste:

n m n
W =? ? wi m j , i c j + ? wi (Tri + M i di + Di ) (3)
i =1 j =1 i =1

The first term in Equation 3 represents the loss of material being wasted within a

particular process stream. This includes for example, the cost of raw materials or product that is

being released to the environment as a component of a waste stream, instead of being converted

9
into finished product and subsequently becoming a source of income. In this manner, the true

cost of the waste stream can be accounted for.

The treatment costs Tr include the required operating costs and fixed capital investment

for the treatment of waste. This treatment can eliminate, partially or completely all

environmental damage costs. However, a decision needs to be made regarding the optimum

degree of treatment required; since an excess may not be economically feasible, while on the

other hand too little treatment can result in excessive damage costs. Furthermore, the degree of

treatment applied to a particular waste stream is also influenced by the current regulatory scheme.

HIDDEN COSTS: These include the expenses associated with permitting, monitoring,

testing, training, inspection, and other regulatory requirements related to waste management

practices. These costs are generally not allocated to the unit responsible for incurring them, and

are usually charged to an overhead account (EPA, 1989).

The hidden costs are generally not a direct function of the amount of waste being

generated and depend mainly on the process regulatory status. Once a regulatory status analysis

has been done estimates for hidden costs can be obtained from EPA (1989).

LIABILITY COSTS: These costs include the fines and penalties to be incurred when a

facility is in non-compliance. They also incorporate the future liabilities for remedial action,

personal injury, and property damage associated with routine and accidental release of hazardous

substances. Consequently, the liability costs can be divided in two groups: (1) penalties and

fines, and (2) future liabilities. Both of these are judgmental in nature and will require a

probabilistic evaluation of future events associated with the process in operation.

10
 The penalties and fines due to non-compliance can be estimated with

( ) ?k ( )
n
E kf = f ,i ?? k f ,i (4)
i =1

where the information required to calculate the expected value of the fines and penalties E(kf) can

be estimated using the knowledge of existing plant operations and/or previous penalties imposed

to the facility. Several estimates can also be found in EPA (1989).

The future liabilities incorporates the liabilities associated with remediation,

compensation, and natural resource damages. There are not many methods available to estimate

the future liabilities. However, a review of several of them is given by EPA (1996).

Due to the uncertainty associated with the determination of future liabilities costs as well

as the lack of a general methodology to estimate them, the approach followed is similar to the

one used for calculating fines and penalties

E ( kr ) = ( )
n

?
i =1
kr ,i ?? kr ,i (5)

E ( kc ) = ? ? ( k )[ L ]
n

c ,i c ,i + cL ,i k c ,i (6)
i =1

E (k N ) = ( )
n

?i =1
k N ,i ?? k N ,i (7)

The expected values for the future liabilities can be estimated using knowledge of

previous cases handled within the company, or in some instances information regarding similar

cases can be obtained from EPA’s Civil Docket Database.

11
 LESS TANGIBLE BENEFITS: These include the benefits obtained as a result of the

increase in revenues or decrease in expenses due to an improvement in consumer acceptance,

employee relations, and corporate image. Although it is quite difficult to estimate these benefits

it is reasonable to assume that they may be significant (EPA, 1989). However, despite their

significance no information is available in the literature to calculate these benefits and their

particularity makes it difficult to develop a mathematical approach to estimate them.

Measuring the process’ environmental impact

Starting from the work by Mallick et al. (1996), the approach taken to measure the

process’ environmental impact θ is the use of a non-monetary valuation technique that calculates

the environmental impact of each chemical present in the waste stream in terms of

Environmental Impact Units (EIU) per kilogram of product produced

n m

?? wi mj ,i Φ j
i =1 j =1
θ=
P (8)

Based on the work by Davis et al. (1994), the environmental impact index Φ is given by:

Φ = ( Human Health Effect + Environmental Effect ) ?

( Exposure Potential)
(9)

Human Health Effects = HVoralLD 50 + HVinhalationLC 50 + HVcarcinogenity + HVother (10)

Environmental Effects = HVoralLD 50 + HV fishLC 50 + HV fishNOEL (11)

12
 Exposure Potential = HVBOD + HVhydrolisis + HVBCF (12)

Where the hazard values HVi for each endpoint i are calculated using toxicological

information specific to each chemical as described by Davis et al. (1994). The last two terms in

Equation 10 —the carcinogenity and other specific effects— account for the chronic human

health effects. Davis et al. (1994) calculates these using a semiquantitave approach, where based

on the carcinogenity classification or the presence of specific effects, a numerical value is

assigned. However, the use of such semiquantitative evaluation might not always lead to a valid

toxicity comparison between chemicals. For this reason, and by keeping the same scale assigned

by Davis et al. (1994) to both chronic effects, the hazard value for each toxicological endpoint is

calculated based on the classification presented in the Hazard Ranking System Final Rule

(Federal Register, 1990) and in the Bouwes and Hassur’s (1997) methodology:

(
HVother = max HVRfC , HVRfD ) (13)

HVRfC = 1569
. − 125
. log RfC (14)

If RfC > 18 ? HV=0; If RfC < 0.0018 ? HV=5

HVRfD = 1165
. − 1167
. log RfD (15)

If RfD > 5 ? HV=0; If RfD < 0.005 ? HV=5

HVcarcinogenity = max( HVSF , HVUR ) (16)

HVSF − A / B = 4.301 + log SF (17)

If SF < 0.0005 ? HV=1; If SF > 5 ? HV=5

HVSF − C = 3.301 + log SF (18)

13
 If SF < 0.005 ? HV=1; If SF > 50 ? HV=5

HVUR − A / B = 4.854 + log UR (19)

If UR < 0.00014 ? HV=1; If UR > 1.4 ? HV=5

HVUR − C = 3854
. + log UR (20)

If UR < 0.0014 ? HV=1; If UR > 14 ? HV=5

Since the “waste streams” might not be the only emission source in the process, a

question arises weather Equation 8 should include only the environmental efficiency of these

“streams”. As Siegell (1996) suggests, based on studies in the United States, the largest source

of volatile organic compounds (VOCs) released —accounting for 40 to 60%—, is that of fugitive

emissions from piping and other fluid handling operations. In addition, VOCs fugitive emissions

can also come from storage tanks, loading operations, and wastewater treatment units.

Subsequently, if these emissions, as well as other possible accidental releases are not taken into

account, the environmental impact of the process might not be correctly evaluated. In this

context, a release factor r that accounts for the release potential of a particular stream —

including waste and non-waste streams— is incorporated into Equation 8.

n m

?? ri f i mj ,i Φ j
i =1 j =1
θ= (21)
P

The release factor can take values from 0 to 1. For waste streams, r=1, whereas for non-waste

streams, 0 < r < 1. Estimating r is equivalent to calculating the probability of obtaining a release

from a specific stream. This usually can be done considering past data and experiences related to

14
the process under study. Based on the categories presented by Kolluru (1995), a guideline for

estimating r based on the expected frequency of the release is given in Table 2.

IMPLEMENTATION OF POLLUTION PREVENTION ALTERNATIVES

Once the feasible pollution prevention alternative has been evaluated and identified, the

next step is to implement such alternative. During its implementation, it is important to

document the real benefits or savings that have been obtained from the waste minimization

project. This might be helpful in the future in order to evaluate other pollution prevention

projects, and serve as a proof of the potential advantages of applying such programs.

MULTIPLE OBJECTIVE OPTIMIZATION

The typical multiple objective optimization problem is given by

min z1 = f (x, y ), z2 = f (x, y ), Κ , zn = f (x, y ) (22)

subject to

g(x, y) = 0

h(x, y) ? 0

where the optimum answer corresponds to the values of the continuous and discrete variables x*

and y* respectively that minimize a set of objectives zi subject to a set of equality g() and

inequality h() constraints.

There are several publications that present an overview of multiple criteria optimization

theory, including Chankong and Haimes (1983), Goicoechea et al. (1982), Sawaragi et al. (1985),

15
Stancu-Minasian (1990), Vanderpooten (1990), and Yoon and Hwang (1995). Furthermore,

several reviews have been published that deal with specific applications. For example,

engineering applications are reviewed by Goicoechea et al. (1982); chemical engineering

applications by Clark and Westerberg (1983); management science applications by Anderson et

al. (1991) and Kirkwood (1997); and environmental management applications by Munda et al.

(1994) and Janssen (1992).

The multiobjective decision problem in Equation 22 could in theory be solved using

similar methods as those employed in single objective optimization problems. However, when

dealing with several objectives there is usually no alternative that maximizes each criterion

simultaneously. That is, there is generally no investment option that maximizes the process’

profit and minimizes its environmental impact. Hence, a sacrifice of the first objective is

required to obtain a better performance of the second objective. As a consequence, the optimum

solution obtained will be considered as the best compromise solution according to the decision

maker’s preference structure (Vanderpooten, 1990).

This compromise solution corresponds to a set of feasible answers, generally referred to

as noninferior or Pareto optimal solutions (see Definition 1)(Vanderpooten, 1990)

Definition 1: z’ ? Z is nondominated iff there is no z ? Z such that z > z’.

That is, a nondominated point is such that any other point in the set of possible outcomes

Z, which increases the value of one criterion also decreases the value of at least one other

criterion.

Within the realm of decision making, the analytical methods for obtaining the best

compromise solution can be classified in to three general groups: (1) generating techniques, (2)

16
techniques with prior articulation of preferences, and (3) methods of progressive articulation of

preferences. Due to the intensive computational requirements of the generating techniques, the

techniques of prior or progressive articulation of preferences might be more appealing especially

when analyzing alternatives under uncertainty.

Among the different methods of prior articulation of preferences, the one selected is the

compromise programming tool (Zeleny, 1973) approach that identifies solutions that are closest

to the ideal solution z* —defined as the vector z * = (z1* , z *2 ,K , zn* ), where zi* = max zi (x,y)— as

determined by some measure of distance (Goicoechea et al., 1982). This ideal solution is

generally not feasible. However, it can be used to evaluate the set of attainable nondominated

solutions. In this case, the closeness or distance Lj of the nondominated solution from ideal

solution is given by (Goicoechea et al., 1982).

n
Lj = ? γ ij (zi* − z i (x ))
j
(23)
i=1

Consequently, a compromise solution with respect to j is defined as xj* such that

min L j (x ) = Lj (x j )
*
(24)

The preference weight γ is used to represent the relative importance of each objective.

The Decision Maker’s (DM) preferences are also expressed in the compromise index j

(1 ? j ? ? ), which represents the DM’s concern with respect to the maximal deviation

(Goicoechea et al., 1982). As a result, the noninferior solutions defined within the range

1 ? j ? ? correspond to the “compromise set” (in practice only three points are calculated: j=1,

17
j=2, and j=? ) (Goicoechea et al., 1982), from which the decision maker still has to make the final

choice to identify the best compromise solution.

Applying the compromise programming approach, the two-objective optimization

problem used to evaluate pollution prevention alternatives is given by

? AEP * − AEP (x, y ) ? j ? θ (x , y ) − θ

j j
**
?
min L j = γ j
?? ?? + γ θ ?? ??
? AEP − AEP ? θ −θ
AEP * ** * **
? ? (25)

subject to:

g(x, y) = 0

h(x, y) ? 0

Where, AEP** = min AEP(x,y), AEP* = max AEP(x,y), θ** = min θ(x,y), and

θ* = max θ(x,y).

STOCHASTIC OPTIMIZATION

Stochastic optimization methods try to solve the problem

min z = f (Ω, x, y ) (26)

subject to

g(Ω, x, y) = 0

h(Ω ,x,y) ? 0

That is, the solution to Equation 26 is given by the optimum values of the discrete and

continuous variables, y* and x* respectively, that minimize the objective function z over all

possible values taken by the uncertain parameters Ω.

18
 An approach usually taken for solving Equation 26, is to replace the stochastic problem

by a suitable deterministic problem (Stancu-Minasian, 1990). In this case, the problem is solved

by finding the solution vectors x*, y* that minimize the expected value of the objective function,

subject to some a priori distribution of Ω:

min E (z ) = f (Ω, x, y ) (27)

subject to

g(Ω, x, y) = 0

h(Ω ,x,y) ? 0

The expected value of the objective function E(z) is obtained by finding its average value

over all its possible values. However, obtaining all these values is time consuming,

computationally intensive, and meaningless. Therefore, the approach usually taken is to take a

sample of the objective function’s distribution and calculate its average value z using

ns
z = ? zi n s
i =1 (28)

As the sample size ns increases, the average value of z in Equation 28 becomes more

accurate. However, as ns becomes larger, the number of times the objective function needs to be

evaluated is incremented. This results in an augmentation in the problem’s computational

requirements. All in all, this computationally intensive sampling and evaluation process has

been one of the problems associated with the application of stochastic optimization techniques.

Once the stochastic problem has been reformulated in a deterministic form, the problem

in Equation 27 can be solved using for example mixed integer non linear programming (MINLP)

19
techniques. However, in spite of their success, the MINLP approach to process synthesis may

pose certain problems especially with sequential process simulators (Chaudhuri and Diwekar,

1997). In addition, the presence of integer variables increases the computational complexity of

the problem and can provoke further discontinues in the process model. Furthermore, MINLP

methods can get trapped into some neighborhood within the search region, leading to a local

solution and failing to find the global optimum.

An alternative approach that circumvents the problems associated with MINLP

algorithms is the use of random search methods. These methods have the feature of exploring

more globally the feasibility region of a given problem, thus having a good possibility of finding

the global optimum (Maffioli, 1987). Among the different random approaches used in process

optimization, the method that has probably received the most attention is the simulated annealing

algorithm (Bohachevsky et al., 1995; van Laarhoven and Aarts, 1987) that is based on the

analogy between the simulation of the annealing of solids and the solving of large combinatorial

optimization problems.

The simulated annealing algorithm —as the physical annealing process— initiates at a

high “temperature” (where the temperature is used simply as a control parameter), and it is

cooled until it reaches a point where no further changes occur. At each temperature level TL,

sufficient configuration changes are made until the system reaches equilibrium. It is important to

mention, that these neighborhood moves need to be Markov chains, since only in such cases, the

algorithm is guaranteed to attain a global optimum. Yet, in the execution of the algorithm

asymptotic convergence can only be approximated. Thus, any implementation results in an

approximation algorithm (van Laarhoven and Aarts, 1987).

20
 From an optimization point of view, the simulated annealing algorithm is a random

search method in which the configuration of the various variables are accepted if they result in a

reduction in the objective’s function value, and in the case of resulting in an increase they are

accepted with a certain probability. Initially, when one is far away from the optimum point, the

algorithm accepts more uphill moves, but —according to the Metropolis algorithm— as it

approaches the optimum, less uphill moves are accepted.

The possibility of accepting uphill moves is one of the method’s main advantage, since it

prevents the algorithm from being trapped in a specific neighborhood or local optima. However,

the algorithm’s main criticism is its high computational requirements due to the large amount of

trials that need to be evaluated. To reduce the number of configurations to be analyzed and to

increase the algorithm’s efficiency research directions have looked at either making a careful

selection of the main parameters, also known as the cooling schedule (van Laarhoven and Aarts,

1987; Collins et al., 1988); or modify the method’s structure and design features (Andricioaei

and Straub, 1996; Collins et al., 1988; Painton and Diwekar, 1995; Rakic et al., 1995; Tovey,

1988; Yamane et al., 1998).

An interesting approach that will be incorporated in the proposed methodology is the one

given by Painton and Diwekar (1995), who developed a modified algorithm referred to as the

“stochastic annealing algorithm.” The main purpose of this new algorithm is to include a

function that penalizes the objective function,

ns
2σ
z = ? zi ns + b(TL )
i =1 ns (29)

in such a manner that the error incurred in sampling the objective function’s distribution —for

example due to a small sample size— is accounted for. In addition, through a weighting

21
function b(TL ) , the algorithm considers that at high temperatures it is not necessary to take large

samples since it is exploring the solution space. Yet, as the system gets cooler more information

is needed so as to obtain a more accurate value of the objective function. Hence, this weighting

function increases the size of the penalty as the temperature level rises.

MULTIPLE OBJECTIVE STOCHASTIC OPTIMIZATION

The solution to the multiobjective stochastic optimization problem will be obtained

combining the compromise programming approach and the stochastic annealing algorithm, using

the following objective function

? E (AEP * ) − E ( AEP ) ? j ? E (θ ) − E (θ ) ?
j j

min E (L j ) = γ
**
j
?? ? + γ ? ?
? E (AEP ) − E (AEP ) ? ? E (θ ) − E (θ ) ?
AEP * ** ? θ ? * ** ? (30)

MULTIOBJECTIVE STOCHASTIC OPTIMIZATION USING ASPEN PLUS

The evaluation of pollution prevention alternatives under uncertainty is applied

combining the compromise programming approach and the stochastic annealing algorithm (see

Figure 1). This analysis is carried out through multiple objective stochastic optimization

techniques using the ASPEN PLUS process simulator.

DEFINITION OF OPTIMIZATION VARIABLES AND UNCERTAIN PARAMETERS

The number of variables included in the optimization study, as well as the number of

uncertain parameters can become a key factor in the complexity of the evaluation of source

22
reduction alternatives. Therefore, a selection should to be made as to which of these variables or

parameters have an important effect in the process’ performance. This is accomplished through a

three step procedure: (1) identification phase, that consists in the listing of all the possible

discrete and continuous variables, as well as the uncertain parameters; (2) the screening phase,

that with the aid of experimental design techniques, evaluates which of the variables identified

have an important effect on the process performance; and (3) the definition of the ranges phase

that involves the specification of the optimization ranges for the variables identified in the

screening phase and the probabilistic distributions used to describe the uncertain parameters.

DEFINITION OF STOCHASTIC ANNEALING PARAMETERS

Three sets of parameters are required for the correct definition of the stochastic annealing

algorithm: (1) the cooling schedule, that considers the choice of the simulated annealing

parameters, (2) the stochastic annealing set, that includes the weighting function’s parameters

and sample size information, and (3) the specification of the neighborhood moves and sampling

procedures. The appropriate choice of these parameters will not only reduce the amount of

iterations required, but will also guarantee that a global optimum has been obtained

The cooling schedule involves the selection of simulated annealing algorithm main

parameters: (1) T0, the initial value of the temperature —the control parameter—; (2) the final

value of the temperature or stopping criterion; (3) a rule for changing the current value of the

control parameter; and (4) the equilibrium criteria, that is the number of trials at a given

temperature or what is referred to as the length of the Markov chain (ML). The search for

adequate cooling schedules has been addressed in many publications. Some of these include the

works by Bohachevsky et al. (1995); Collins et al. (1988); and van Laarhoven and Aarts (1987).

23
 The neighborhood moves should correspond to a Markov chain described as a series of

random events, where the occurrence of each event depends only on the preceding outcome. In

this manner, neighborhood moves for binary variables are described by

If rand[0,1] < 0.5 then yi=0

If rand[0,1] > 0.5 then yi=1

For continuous variables, a move is defined as a random change for one variable. To

accomplish this several move sequences are reviewed by Bohachevsky et al. (1986);

Bohachevsky et al. (1995); Edgar and Himmelblau (1988); and Vanderbilt and Louie (1984).

Chaudhuri and Diwekar (1996) propose that a random change should be made according to:

xi1 = xi0 + [2 ? rand (0,1) − 1]S c ,i (31)

STOCHASTIC ANNEALING ALGORITHM WITH ASPEN PLUS

The idea behind the implementation of the stochastic annealing algorithm using ASPEN

PLUS can be seen in Figure 3. The initialization block sets the initial values for the control

parameters. The optimization block is responsible for selecting the number of samples and

generates the configurations to evaluate; that is it determines the new values for the optimization

variables. The sampling block produces the values for the uncertain parameters and passes them

to the flowsheet analysis blocks that run the flowsheet model and determines the value of the

objective function.

24
 After these last two blocks have been repeated ns times, the stochastic block generates the

statistical information, the penalty function and accepts or rejects the proposed configuration.

Finally, a control block is used to control the overall performance of the algorithm.

Based on the simple diagram in Figure 3, and on the works by Chaudhuri and Diwekar

(1996) and Painton and Diwekar (1995), the detailed algorithm for minimizing an objective

function using the stochastic annealing algorithm and the process simulator ASPEN PLUS is

presented in Figure 4.

To implement the algorithm in Figure 4, one needs to evaluate the flowsheet several

times. This implies that the user has to have control over how many times the flowsheet is

analyzed and the objective function calculated. In a typical programming language this is a

straightforward task. However, with ASPEN PLUS this is not so easily accomplished. To do so,

several dummy blocks and streams are used to force the simulator to evaluate the model for a

specified number of times. In this manner, three additional heater blocks (i.e. B1, B2, and B3)

are added to the process flowsheet (see Figure 5) and by constantly changing some of their

parameters ASPEN is tricked into performing a “DO LOOP.”

CASE STUDY: THE MANUFACTURE OF METHYL CHLORIDE

The methodology proposed was applied to the process for the manufacture of methyl

chloride through the thermal chlorination of methane (AIChE, 1966; Dantus and High, 1996).

The discussion that follows focuses mainly on the evaluation of alternatives phase. A detailed

implementation of the complete methodology is given by Dantus (1999).

25
PROCESS DESCRIPTION

The methyl chloride process consists of a reactor where four reactions take place:

CH 4 + Cl 2 ? CH 3Cl + HCl (32)

CH3 Cl + Cl 2 ? CH2 Cl 2 + HCl (33)

CH2 Cl 2 + Cl 2 ? CHCl 3 + HCl (34)

CHCl 3 + Cl2 ? CCl 4 + HCl (35)

The reactor effluent is cooled to 25 °C and it is washed with water to remove the hydrogen

chloride generated. This water becomes a waste stream that contains both HCl and small

amounts of chloromethanes. Subsequently, the water is removed from the chloromethanes

mixture through a series of dehumidification towers containing NaOH and H2SO4, thus

generating several waste streams. Finally, the gas mixture is compressed and passed through a

series of distillation columns to separate each of the products (see Figure 2).

IDENTIFICATION OF SOURCE REDUCTION ALTERNATIVES

The process given by AIChE (1966) was taken as the base case model used to represent

the performance of the existing process. An initial screening of source reduction options using

experimental design techniques identifies the following variables and parameters to consider

further in the evaluation of pollution prevention alternatives: the temperature of the original

reactor, the mole flow of chlorine, and the use of an alternative reactor and its operating

temperature. In addition, four uncertain parameters are included in the optimization phase.

In summary, the initial set of waste minimization options consists of three continuous

variables, one discrete variable, and four uncertain parameters (see Table 3). Table 3 also

26
includes the optimization ranges to be used, as well as the different types of distribution assumed

for the uncertain parameters. Consequently, the final flowsheet evaluates is obtained by

incorporating the various process alternatives into the base case model, as well as the additional

blocks necessary for the implementation of the stochastic annealing algorithm.

The parameters required for the algorithm’s implementation, including those related with

the cooling schedule, the stochastic annealing algorithm, and the neighborhood moves are given

in Table 4. The initial values used for the optimization variables corresponded to a random

permutation of the base case process’ parameters and for the number of samples its initial value

was set to 5. Once these parameters were defined the stochastic annealing algorithm was used to

estimate the best and worst performance of each criterion to build the scaling function as

describes in Equation 30 (see Table 5). For the case of minimizing the AEP, its minimum value

was set to 0.

The information in Table 5 is used to build the final objective function and was solved for

j=1, 2, and ? , assuming equal weights to each objective, that is γ1=γ2=1; obtaining the results

given in Table 6 and Figure 6. This last figure presents the location of the nondominated

solutions with respect to the ideal point z*.

Analyzing the results given in Table 6, the decision variable y1 should probably be set to

0, that is changing the reactor type generates a better performance on the process both from an

environmental impact and economical perspective. The latter is improved even though

additional capital investment is required.

Regarding the optimum values for the continuous variables, as for the case of the discrete

ones, the final decision should be left to the decision maker that might select the best option

based on his/her particular preferences. Given that equal weights to each objective were

27
assigned, the best compromise solution is given when j = 1 in Table 6. That is, the best

compromise solution is to operate an alternative reactor (y1= 0) at a temperature of 487.7 °C (x3)

using a chlorine flow of 159.7 kgmol/h, obtaining an expected annual equivalent profit of

896,700 ? 36,900 $/yr and an environmental impact of 1090 ? 62.3 EIU/kg of CH3Cl.

CONCLUSIONS

The main objective of this work was to develop a comprehensive methodology that takes into

account the uncertainties present when evaluating process alternatives that seek to reduce the

waste generated in a chemical process. The procedure proposed consisted of six steps:

characterization of waste streams, evaluation of environmental impacts, development of the

process model, identification of pollution prevention alternatives, evaluation of pollution

prevention alternatives, and their implementation.

The methodology incorporated the use of multiple criteria decision making to evaluate

possible investment projects using two competing objectives: maximize profit and minimize the

environmental impact. The former is measured using the annual equivalent profit (AEP) tool and

the latter using an environmental impact index. On one hand, the AEP included the usual costs

associated with the process, as well as the various waste related costs, for which a detailed

discussion was given including the different ways available to estimate them. On the other hand,

the environmental impact index included toxicological characteristics of each chemical present in

a process stream and its release potential.

Multiple objective optimization techniques and stochastic programming methods were

successfully incorporated in the methodology to evaluate the uncertainty in optimizing the two

competing objectives. This was accomplished using the process simulator ASPEN PLUS and

28
combining the compromise programming approach and the stochastic annealing algorithm.

However, the main obstacle encountered was the large computational requirements, in particular

of the stochastic annealing algorithm. Even though a careful selection of its parameters was

made, the method still required a large number of iterations to reach a solution. Each iteration

involves the solution and evaluation of the complete process flowsheet, resulting in a

computationally intensive analysis.

The production of methyl chloride through the thermal chlorination of methane was selected

to evaluate the methodology. This process was used mainly as a case and did not intend to

represent an actual process in operation. Therefore, the options analyzed represent only a small

subset of a large number of possible source reduction alternatives.

As shown in the methyl chloride process, the uncertainty increases the cost associated with

the analysis of alternatives. So, is the increase in cost justified? The answer to this question

depends on the type of process being evaluated and on the particular uncertainties considered.

For example, in the methyl chloride process minimizing the environmental impact leads to the

selection of an alternative reactor operated at 498 °C and a chlorine flowrate of 137.4 kgmol/h

with an environmental impact of 986+77 EIU/kg of CH3Cl. However, if no uncertainties are

included in the optimization process the alternative reactor should be operated at 353 °C and a

chlorine flowrate of 154.6 kgmol/hr with an environmental impact of 569 EIU/kg of CH3Cl.

ACKNOWLEDGMENTS

The present work was made possible through the support of the Consejo Nacional de

Ciencia y Tecnología (CONACYT) and the Fulbright program.

29
NOMENCLATURE

AEP Annual equivalent profit

b0 Weighting function constant

cj Cost of component j

cL Percentage of the claims that require compensation payments

Dj Disposal cost

dj Distance to waste treatment or disposal facility

E(x) Expected value of x

EIU Environmental impact units

f Flowrate (kg/hr)

F Cash flow

fD Depreciation factor

g( ) Set of equality constraints

h( ) Set of inequality constraints

HVx Hazard value of endpoint x

if Annual inflation rate

ir Interest rate

ID Depreciable investment

j Compromise index

kc Compensation costs incurred in the event of a claim given a release

kf Fines and/or penalties for non-compliance of an environmental regulation

kN Natural resource damage expenses to be paid in the event of a release

30
kr Remediation costs incurred in the event of a release

Lc Legal defense costs

Lj Distance from the ideal point

Mi Transportation cost to waste treatment or disposal facility

mj,i Mass fraction of component j in waste stream i

ML Markov chain length

MOOP Multiple objective optimization problem

ns Sample size

Ny Project’s lifetime

P Product flowrate (kg/hr)

r Release factor
3
RfC Chronic reference concentration (mg/m )

RfD Chronic reference dose (mg/kg-day)

SC Sample size for continuous variable

SF Oral slope factor (risk per mg/kg-day)

Ss Number of samples step size

SOOP Single objective optimization problem

T Temperature

Tri Treatment cost

Tx Tax rate
3
UR Inhalation unit risk (risk per mg/m )

W Direct waste cost

31
 wi Flowrate of waste stream i (kg/hr)

x Vector of continuous variables

y Vector of discrete variables

z Objective function

αΤ Temperature function constant

Φ Environmental impact index of chemical j (EIU/kg)

γ Preference weight

θ Environmental impact (EIU/kg)

σ Standard deviation

Ω Vector of uncertain variables

? (x) Probability of x

Subscripts

A/B Chemical is known or is probable to be a human carcinogen

C Chemical is possible to be a human carcinogen

BCF Aquatic bioconcentration factor

BOD Biological oxygen demand

REFERENCES

AIChE (1966). 1964 Problem: Meeting Market Demands--Chloromethanes Plant. In Student

Contest Problems and First-Prize-Winning Solutions 1959-1965, New York: American
Institute of Chemical Engineers.

32
Alliet Gaubert, M. and X. Joula (1997). Tools for Computer Aided Analysis and Interpretation of
Process Simulation Results. Computers Chem. Engng. 21, S205.

Alvaargaez, A., Kokossis, A. C. and Smith, R. (1998). Waste-water Minimization of Industrial-

Systems Using an Integrated Approach. Computers Chem. Engng. 22, S741.

Anderson, D. R., Sweeney, D. J. and Williams, T. A. (1991). An Introduction to Management

Science: Quantitative Approaches to Decision Making. 6th ed., New York: West Publishing.

Andricioeai, I. and Straub, J. E. (1996). Generalized Simulated Annealing Algorithms Using

Tsallis Statistics Application to Conformational Optimization of a Tetrapeptide. Physical
Review E 53, 4, R3055.

Bohachevsky, I. O., M. E. Johnson, and M. L. Stein (1986). Generalized Simulated Annealing for
Function Optimization. Technometrics 28, 209.

Bohachevsky, I. O., Johnson, M. E. and Stein, M. L. (1995). Simulated Annealing and

Generalizations. In Adaption of Simulated Annealing to Chemical Optimization Problems, J.
H. Kalivas, ed. Amsterdam: Elseiver.

Bouwes, N. W. and Hassur, S. M. (1997). Toxic Release Inventory Relative Risk-Based

Environmental Indicators Methodology. Washington DC: United States Environmental
Protection Agency.

Cabezas, H., Bare, J. C. and Mallick, S. K. (1997). Pollution Prevention with Chemical Process
Simulators: The Generalized Waste Reduction (WAR) Algorithm. Computers Chem. Engng.
21, S303.

Chadha, N. (1994). Develop Multimedia Pollution Prevention Strategies. Chem. Engng. Prog.
90, 11, 32.

Chankong, V. and Haimes, Y. Y. (1983). Multiobjective Decision Making: Theory and

Methodology. North-Holland Series in System Science and Engineering, A. P. Sage, ed. New
York: North-Holland.

Chaudhuri, P. D. and Diwekar, U. M. (1996). Process Synthesis Under Uncertainty: A Penalty

Function Approach. AIChE J. 42, 742.

Chaudhuri, P. D. and Diwekar, U. M. (1997). Synthesis under Uncertainty with Simulators.

Computers Chem. Engng. 21, 733.

Ciric, A. R. and Huchette, S. G. (1993). Multiobjective Optimization Approach to Sensitivity

Analysis: Waste Treatment Cost in Discrete Process Synthesis and Optimization Problems.
Ind. Engng. Chem. Res. 32, 2636.

33
Clark, P. A. and Westerberg, A. W. (1983). Optimization for Design Problems Having More
Than One Objective. Computers Chem. Engng. 7, 259.

CMA (1993). Designing Pollution Prevention into the Process: Research, Development and
Engineering. Washington, D.C.: Chemical Manufacturers Association.

Collins, N. E., Eglese, R. W. and Golden, B. L. (1988). Simulated Annealing - An Annotated

Bibliography. American J. Math. Mgmnt. Sciences 8, 209.

Dantus, M. M. (1999) Methodology for the Design of Economical and Environmental Friendly
Processes: An Uncertainty Approach. Ph.D. Dissertation, Oklahoma State University.

Dantus, M. M. and High, K. A. (1996). Economic Evaluation for the Retrofit of Chemical
Processes Through Waste Minimization and Process Integration. Ind. Engng. Chem. Res.
35, 12, 4566.

Davis, G. A., L. Kincaid, M. Swanson, T. Schultz, J. Bartmess, B. Griffith, and S. Jones (1994).
Chemical Hazard Evaluation for Management Strategies: A Method for Ranking and Scoring
Chemicals by Potential Human Health and Environmental Impacts. US Environmental
Protection Agency, EPA/600/R-94/177.

DOE (1997). Office of the Industrial Technologies: Enhancing the Competitiveness, Efficiency,
and Environmental Quality of American Industry Through Technology Partnerships. US
Department of Energy, DOE/GO-10097-334.

Doerr, W. W. (1993). Plan for the Future with Pollution Prevention. Chem. Engng. Prog. 89, 1,
25.

Douglas, J. M. (1988). Conceptual Design of Chemical Processes. New York: McGraw-Hill.

Douglas, J. M. (1992). Process Synthesis for Waste Minimization. Ind. Engng. Chem. Res. 31,
238.

Dyer, J. and Mulholland, K. (1998). Prevent Pollution via Better Reactor Design and Operation.
Chem. Engng. Prog. 94, 2, 61.

Edgar, T. F. and Himmelblau, D. M. (1988). Optimization of Chemical Processes. New York:

McGraw-Hill.

El-Halwagi, M. M. and Manousiouthakis, V. (1989). Synthesis of Mass Exchange Networks.

AIChE J. 35, 1233.

EPA (1988). Waste Minimization Opportunity Assesment Manual. US Environmental Protection

Agency, EPA/625/7-88/003.

34
EPA (1989). Pollution Prevention Benefits Manual. US Environmental Protection Agency, EPA
230-R-89-100.

EPA (1992). Total Cost Assessment: Accelerating Industrial Pollution Prevention through
Innovative Project Financial Analysis with Application to the Pulp and Paper Industry. US
Environmental Protection Agency, EPA 741-R-92-002.

EPA (1996). Valuing Potential Environmental Liabilities for Managerial Decision-Making: A

Review of Available Techniques. US Environmental Protection Agency, EPA 742-R-96-003.

Fonyo, Z., Kurum, S. and Rippin, D. W. T. (1994). Process Development for Waste
Minimization: The retrofitting Problem. Computers Chem. Engng. 18, S591.

Federal Register (1990). Hazard Ranking System Final Rule. Federal Register 55, 241
(December 14), 51532.

Goicoechea, A., Hansen, D. R. and Duckstein, L. (1982). Multiobjective Decision Analysis with
Engineering and Business Applications. New York: John Wiley & Sons.

Grossmann, I. E. and Kravanja, Z. (1995). Mixed-Integer Nonlinear Programming Techniques

for Process Systems Engineering. Computers Chem. Engng. 19, S189.

Hopper, J. R., Yaws, C. L., Ho, T. C., Vichailak, M. and Muninnimit, A. (1992). Waste
Minimization by Process Modification. In Industrial Environmental Chemistry: Waste
Minimization in Industrial Processes and Remediation of Hazardous Waste, D. T. Sawyer
and A. E. Martell, eds. New York: Plenum Press.

Ierapetritou, M. G., Acevedo, J. and Pistikopulos, E. N. (1996). An Optimization Approach for

Process Engineering Problems Under Uncertainty. Computers Chem. Engng. 20, 703.

Janssen, R. (1992). Multiobjective Decision Support for Environmental Management.

Dordrecht: Kluwer Academic.

Kirkwood, C. W. (1997). Strategic Decision Making: Multiobjective Decision Analysis with

Spreadsheets. Belmont, CA: Duxbury Press.

Kolluru, R. V. (1995). Minimize EHS Risks and Improve the Bottom Line. Chem. Engng. Prog.
91, 6, 44.

Linnhoff, B. (1994). Use Pinch Analysis to Knock Down Capital Cost and Emissions. Chem.
Engng. Prog. 90, 8, 32.

Maffioli, F. (1987). Randomized Heuristics for NP-Hard Problems. In Stochastics in

Combinatorial Optimization, ed. G. Andreatta, F. Mason and P. Serafini, eds, 76. Singapore:
World Scientific.

35
Mallick, S. K., Cabezas, H., Bare, J. C. and Sikdar, S. K. (1996). A Pollution Reduction
Methodology for Chemical Process Simulators. Ind. Engng. Chem. Res. 35, 4128.

Manousiouthakis, V. and Allen, D (1995). Process Design for Waste Minimization. In

Proceedings of the Foundations of Computer Aided Process Design, New York: American
Institute of Chemical Engineers.

Munda, G., Nijkamp, P. and Rietveld, P. (1994). Multicriteria Evaluation in Environmental

Management: Why and How?. In Applying Multiple Criteria Aid for Decision to
Environmental Management, M. Paruccini, ed. Dordrecht: Kluwer Academic.

Nelson, K. E. (1990). Use These Ideas to Cut Waste. Hydrocarbon Processing 69, 3, 93.

Painton, L. and Diwekar, U. (1995). Stochastic Annealing for Synthesis Under Uncertainty. Eur.
J. Oper. Res. 83, 489.

Peters, M. S. and K. D. Timmerhaus (1991). Plant Design and Economics for Chemical
Engineers. New York: McGraw-Hill.

Pohjola, V. J., Alha, M. K. and Ainassaari, J. (1994). Methodology of Process Design.

Computers Chem. Engng. 18, S307.

Rakic, A. D., Elazar, J. M. and Djurisic, A. B. (1995). Acceptance-Probability-Controlled

Simulated Annealing a Method for Modeling the Optical-Constants of Solids. Physical
Review E 52, 6, 6862.

Sawaragi, Y., Nakagama, H. and Tanino, T. (1985). Theory of Multiobjective Optimization.

Orlando, FL: Academic Press.

Siegell, J. H. (1996). Exploring VOC Control Options. Chem. Engng. 103, 6, 92.

Spriggs, H. D. (1994). Design for Pollution Prevention. In Pollution Prevention via Process and
Product Modification, M. M. El-Halwagi and D. P. Petrides, eds. No. 303, Vol. 90. AIChE
Symposium Series. New York: American Institute of Chemical Engineers.

Stancu-Minasian, I. M. (1990). Overview of Different Approaches for Solving Stochastic

Programming Problems with Multiple Objective Functions. In Stochastic Versus Fuzzy
Approaches to Multiobjective Mathematical Programming Under Uncertainty, R. Slowinski
and J. Teghem, eds. Dordrecht: Kluwer Academic.

Stefanis, S. K. and Pistikopoulos, E. N. (1997). Methodology for Environmental Risk

Assessment of Industrial Nonroutine Releases. Ind. Engng. Chem. Res. 36, 3694.

36
Tovey, C. A. (1988). Simulated Simulated Annealing. American J. Math. Mgmnt Sciences 8,
389.

van Laarhoven, P. J. M. and Aarts, E. H. L. (1987). Simulated Annealing: Theory and

Applications. Dordrecht, Holland: D. Reidel Publishing Company.

Vanderbilt, D. and S. G. Louie (1984). A Monte Carlo Simulated Annealing Approach to

Optimization over Continuous Variables. J. Computational Physics, 56, 259.

Vanderpooten, D. (1990). Multiobjective Programming: Basic Concepts and Approaches. In

Stochastic Versus Fuzzy Approaches to Multiobjective Mathematical Programming Under
Uncertainty, R. Slowinski and J. Teghem, eds. Dordrecht: Kluwer Academic.

Wilkendorf, F., Espuna, A. and Puigjaner, L. (1998). Minimization of the Annual Cost for
Complete Utility Systems. Chem. Engng. Res. Design 76, 239.

Yamane, T., Inoue, Y. and Sakurai, M. (1998). A New Simulated Annealing Method Combined
with a Polarizable Continuum Model-Based on the Boundary-Element Method. Chem.
Physics Letters 291, 1-2, 137.

Yoon, K. P. and Hwang, C. (1995). Multiple Decision Making: An Introduction. Thousand Oaks,
CA: Sage Publications.

Zeleny, M. (1973). Compromise Programming. In Multiple Criteria Decision Making, J. L.

Cochrane and M. Zeleny, eds. Columbia, SC: University of South Carolina Press.

37
 LIST OF FIGURES

1. Proposed methodology’s general diagram

2. Carbon tetrachloride SFC for the methyl chloride process

3. Stochastic annealing algorithm using ASPEN PLUS

4. Stochastic annealing algorithm and ASPEN PLUS: A detailed description

5. Generalized flowsheet model

6. Multiobjective optimization results

LIST OF TABLES

1. Uncertainty sources in analysis of waste minimization projects

2. Guidelines for estimating the release factor r

3. Definition of optimization variables and uncertain parameters

4. Case study’s optimization parameters

5. Ideal point and worst case scenario

6. Optimization results for the methyl chloride process

38
 Need to change, modify or implement a new system

DEVELOPMENT OF Design Retrofit CHARACTERIZATION

PROCESS MODEL OR
OF PROCESS STREAMS

CHARACTERIZATION EVALUATION OF
OF PROCESS STREAMS ENVIRONMENTAL IMPACTS

EVALUATION OF DEVELOPMENT OF
ENVIRONMENTAL IMPACTS PROCESS MODEL
IDENTIFICATION OF
POLLUTION PREVENTION
ALTERNATIVES

EVALUATION OF POLLUTION PREVENTION ALTERNATIVES

Statement of the decision problem (to be followed in strict order)

1. Identification of the decision making unit
2. Definition of the decision environment
3. Identification of objectives
4. Definition of attributes
5. Definition of decision rule

Optimization of process alternatives (to be followed in strict order)

1. Definition of a. Identification phase
optimization b. Screening phase
variables c. Definition of ranges

2. Definition of a. Cooling schedule

optimization b. Stochastic annealing
parameters c. Neighborhood moves / sampling procedures

a. Determine maximum and Stochastic annealing

minimum point algorithm with ASPEN
3. Multiple objective
stochastic b. Assign preference weights
optimization Interact with decision maker
c. Transform MOOP into
Stochastic annealing
SOOP: Solve for j=1, 2, ?
algorithm with ASPEN

Yes More No Decision No

alternatives? maker
satisfied?
Yes

IMPLEMENTATION OF POLLUTION PREVENTION ALTERNATIVES

Figure 1

39
Figure 2

40
Initialization Optimization Sampling Flowsheet
block block block analysis

Stochastic Control
block block

Figure 3

41
 Initialize variables
T, Tlev, α, stopping criteria,
and initial configuration S

T=T*α Stopping Yes

TL=TL+1 criteria Stop
Yes met?
No
Equilibrium
criteria met No
Select number of samples ns
(Loop2>ML or
ACC<Nacc)

Generate a new configuration S’

- Select new xi
- Select new yi

Set Loop=Loop + 1
Generate sample of uncertain
parameter Ui No

- Run flowsheet model

Loop = ns
- Calculate objective function

Generate statistics Yes

Loop2=Loop2+1 E(S’) and σ(S’)

Accept Generate penalized objective function

configuration - Calculate weighting function b(TL)
ACC = ACC +1 No - Calculate modified objective
function OBJ(S’)

Yes
W > rdm( ) ∆ = OBJ(S’) - OBJ(S)

No ∆<0
W=exp (-∆ / T)

Yes

Figure 4

42
B1 B2 B3

Flowsheet model

Figure 5

43
 950

j =1
900
z*
AEP
850
x 1,000
($/yr)
800
j =2

j=?
750
900 950 1000 1050 1100 1150 1200
θ (EIU / kg of CH3Cl)

Figure 6

44
 Table 1

Type Example

Process model uncertainty Kinetic constants, physical properties,

and transfer coefficients

Process uncertainty Flowrate and temperature variations,

stream quality fluctuations

Economic model and environmental Capital costs, manufacturing costs, direct

impact model uncertainty costs, release factors, hazard values,
hidden costs, liability costs, and less
tangible costs

External uncertainty Product demand, prices, feedstream

availability, feed composition

Discrete uncertainty Equipment availability and other discrete

random events.

Regulatory uncertainty MACT standards, modified emission

standards, and new environmental
regulations

Time uncertainty Investment delays (i.e., the project might

have a better performance in the future)

45
 Table 2

Frequency r

Constant: Stream characterized as waste stream 1

Frequent: Release expected to occur several times a 0.3 - 1.0

year

Ocassional: Release expected to occur several 0.1 - 0.3

times during the facility lifetime

Remote: Release expected to occur about once 0.01 - 0.1

during the facility lifetime

Not expected: Release highly unlikely to occur < 0.01

during the facility lifetime

46
 Table 3

Variable Description Variable Base case Optimization Distribution type

type value Range
Temperature of CSTR Continuous 525 °C 350 - 550 °C
(x1)

Mole flow of Cl2 Continuous 145 kgmol/h 130 -160 kgmol/h

(x2)

Alternative reactor’s Continuous -189,900 350 - 550 °C

operating temperature
(x3)

Type of reactor Discrete 1 0-1

(y1)

Release factor for Uncertain 0.1 Triangular

non-waste streams distribution with:
Mo=0.1,
Low = 0.05,
High=0.3

High pressure steam Uncertain 0.01 $/kg Normal distribution

price with:
µ=0.01 $/kg
σ = 0.002

Pre-exponential factor Uncertain 2.56 x 108 Normal distribution

for reaction 32 with:
(m3 / kgmol – s) µ=2.56 x 108
σ = 8 x 106

Environmental impact Uncertain 34.7 Uniform random

index for CH3Cl distribution between
16.1 and 34.7

47
 Table 4

Parameter Value

Cooling schedule

Initial “temperature” For Max AEP T0=200

For Min θ T0=25
For Max θ T0=25
For Min L? T0=10

Stopping criterion Small change in 5 continuous Markov

Chains

“Temperature” function Ti +1 = 0.9 Ti

Markov chain length Continue until number of accepted

configurations = 10 or M Lmax = 10n

Stochastic annealing parameters

Weighting function constant (b0) 0.01

Rate of increase constant (k) 0.9

Neighborhood moves

Binary variables ??1 if rand [0,1] > 0.5

yi1 = ?
?? 0 if rand [0,1] < 0.5

Continuos variables xi1 = xi0 + ( 2 ? rand [ 0,1] − 1) Sc ,i

where: Sc,1 = 34
Sc2 = 7
Sc,3 = 34

Number of samples step size (Ss) Ss = 5

48
 Table 5

Objective AEP θ x1 x2 x3 y1
function EIU / kg of °C kgmol/h °C
$ / yr CH3Cl
Max AEP 869,300 ? 2,500 1089 ? 74.8 469.6 159.6 494.1 0

Min θ -1,351,500 ? 2,700 986 ? 77.0 496.9 137.4 498.3 0

Max θ -1,889,950 ? 2,500 1272 ? 70.4 477.5 138.8 481.1 1

49
 Table 6

j Lj AEP θ x1 x2 x3 y1
$ / yr EIU/ kg of °C kgmol/hr °C
CH3Cl
1 0.353 ? 0.14 896,700 ? 36,900 1090 ? 62.2 388.5 159.7 487.7 0

2 0.444 ? 0.15 781,300 ? 1,900 1106 ? 56.8 435.6 159.4 506.7 0

? 0.519 ? 0.075 774,150 ? 17,400 1097 ? 48.2 411.6 159.0 474.4 0

50