5944
Ind. Eng. Chem. Res. 2006, 45, 5944-5954
Multiobjective Waste Management under Uncertainty Considering Waste Mixing L. Cavin, U. Fischer,* and K. Hungerbu1hler ETH Zurich, Institute for Chemical and Bioengineering, 8093 Zurich, Switzerland
We present a new methodology to evaluate, under uncertainty, the waste treatment options for a new product in an existing facility, considering economic and environmental objectives. The methodology highlights potential treatment problems, forecasts the required operations, and estimates costs and environmental impact. The new procedure includes a rigorous evaluation of secondary streams. The uncertainty of the waste stream composition, cost factors, and separation efficiencies are taken into consideration. The new procedure accounts for possible interaction and synergy between several waste streams. The option of stream mixing is considered to increase the efficiency of central recovery operations or to achieve dilution of critical streams in the final treatment operations, e.g., the sewage treatment plant, and, hence, to reduce the need for expensive pretreatments such as stripping. The most-efficient treatment paths for an original waste stream might be composed of complex combinations of the aforementioned options; e.g., some part of a stream is recovered directly in the production building, while the remainder is mixed with another stream for recovery in a central distillation, before being sent along with two additional streams to incineration. For each waste stream, a list of such combinations and the resulting economic and environmental assessments is obtained. 1. Introduction One of the major environmental problems of chemical production is the generation of waste. Being the superfluously yielded byproduct, rather than the nontransformed raw material or a spent solvent, waste is indeed the key to cleaner production.1 Consequently there is currently a great deal of interest in the development of methods that can be used to prevent the generation of waste during chemical production or contribute in reducing the emissions from treating it. Methodologies for waste reduction range from applying heuristic rules for the generation of cleaner process alternatives,2 over pollution balancing methods such as the waste reduction (WAR) algorithm,3,4 to the application of mathematical frameworks for creating mass exchange networks (MENs).5,6 Although many concepts for realizing more-ecological manufacturing of chemical products have been presented in the literature7 and many initiatives for cleaner production have been undertaken from the chemical industry, at the moment, abatement processes are still required at many plants, to reduce the discharge of pollutants at the end of the pipe. Some work has been published on the environmental assessment of emissions from a chemical process without assessing the impact of the waste treatment operations itself. Mallick et al.,8 Cabezas et al.,9,10 and Young and Cabezas11 extended the pollution balancing approach (WAR algorithm) that was presented by Hilaly and Sikdar,3,4 using nine potential impact indices ranging from ozone depletion to human toxicity and ecotoxicity for assessing emissions. Koller et al.12 presented an environmental, health, and safety (EHS) assessment tool that estimates and evaluates emissions to the environment after applying end-of-pipe technologies. Pistikopoulos and coworkers13-15 presented the method of minimizing environmental impact (MEI) that uses principles from life-cycle assessment (LCA) within a chemical process optimization framework. Using this latter method, in one study, the environmental impact due to the waste treatment operations itself also was included into the assessment.16 The results of this study show * To whom correspondence should be addressed. Tel.: +41 44 632 56 68. Fax:+41 44 632 11 89. E-mail:
[email protected].
that high degrees of abatement might even result in a worsening of the environmental impact through excessive energy and raw material consumption in the treatment operations itself. It is therefore important to include not only the final emissions to the environment but also the environmental performance of endof-pipe technologies into an overall environmental assessment of options for dealing with a given waste stream. Obviously, high degrees of abatement generally also imply high cost, which reduces the profitability of a chemical process.16 Thus, “overtreatment” of waste streams might be disadvantageous, from both an economic and an environmental point of view. In contrast to the modeling of chemical processes themselves, only limited work has been done so far on the development of simulators for modeling waste treatment operations. In this context, Petrides et al.17 presented a research prototype (EnviroCAD) that supports the design of new waste treatment processes. Linninger and co-workers have presented several papers on the long-term planning of waste management strategies considering uncertainty, multiple objectives, and investment decisions for new facilities or process technology.18-21 Recently, we presented a model22 in which the volume of a waste stream, as well as its composition, can be treated as uncertain and extended it for the inclusion of environmental objectives.23 The overall model automatically calculates the cost and environmental impact of all possible treatment scenarios that are available at a given plant and that are feasible and legally compliant for a given waste stream. The explicit evaluation of all treatment flowsheets is possible because, for each waste treatment problem, only the corresponding facilities available at an existing plant are considered, and because linear models were used to describe the waste treatment operations. In the present work, this model is extended to account for possible interaction and synergy between several waste streams. The option of stream mixing is considered to increase the efficiency of central recovery operations or to achieve dilution of critical streams in the final treatment operations, e.g., the sewage treatment plant, and, hence, to reduce the need for expensive pretreatments such as stripping. The most efficient treatment paths for an original waste stream might be composed of complex combinations of the aforementioned options, e.g.,
10.1021/ie051005d CCC: $33.50 © 2006 American Chemical Society Published on Web 07/26/2006
Ind. Eng. Chem. Res., Vol. 45, No. 17, 2006 5945
Figure 1. Treatment operations and system boundaries (dashed box) considered in the model.
some portion of a stream is recovered directly in the production building, while the remainder is mixed with another stream for recovery in a central distillation, before being sent along with two additional streams to incineration. For each waste stream, a list of such combinations and the resulting economic and environmental assessments is obtained. These model features are demonstrated using several case studies. 2. Model Description 2.1. Model Structure and Unit Operation Models. The aim of the model is to determine the least expensive, as well as the environmentally most benign, treatment path, out of all treatment alternatives at a given plant, for given chemical waste streams with uncertain volumes and/or compositions. In this paper, the term “stream” does not refer to a continuous flow rate but to a recurrent volume or mass of waste. The uncertainty specified might be the result of considering distributed parameters in modeling the chemical process itself, or, for a given waste stream to be treated, it might be stipulated based on heuristic knowledge. In addition to stream volume and composition, the different separation efficiencies used for modeling the treatment options also are considered to be uncertain. In the previous publications on this topic, we used a bruteforce method that calculates, for all legal and technically feasible alternatives, the treatment costs as well as the resulting environmental impact. Although the method remains similar in this paper, the exponentially larger scope of the problem (several streams with mixing options) forced the use of heuristics to reduce the size of the problem and to select only promising alternatives for computation. In the following, first the model, as presented earlier,22,23 is summarized, including some model extensions. Afterward, the extension of the method for considering mixing will be presented. Figure 1 shows the overall model structure and the treatment operations considered in the case studies presented here. The treatment operations can be separated into optional pretreatments (rectification, phase separation, stripping, ammonia recovery, decontamination) and final operations (incineration, sewage treatment plant (STP), disposal). However, the facility may be modified at runtime and further operations may be added. The
pretreatments reduce the amount of one or several of the constituents of a stream. Application of these operations might be required to achieve the input constraints of the final treatment facilities. Rectification can also be used for recycling valuable components and may be conducted both within the production line and centrally in large recovery units. Except for the phase separation and rectification/distillation, all treatments are grouped under the label “end-of-pipe”. According to the specifications of the waste stream, decisions on the selection of treatment operations are taken. Figure 1 shows which treatment paths are feasible, depending on the phase of the stream: Recycling is considered only for liquid streams. A gaseous stream can only be incinerated. When a solid stream contains at least one chemical for which a heat of combustion greater than zero has been specified, it must be incinerated; otherwise, it can be directly deposited. However, fluids are allowed to enter incineration, even when a heat of combustion less or equal to zero is specified. Final treatments are operations that lead to direct emissions from the balance region into five different compartments: air, water, sludge, ash, and slag. Therefore, for the environmental assessment, a link between the waste stream entering one of these operations and the equivalent emissions leaving it must be found. For this purpose, transfer coefficients have been implemented into the model for incinerators and sewage treatment plants as given by Zimmermann et al.,24 which originate from studies on several Swiss municipal treatment plants. For a list of input parameters, which can be molecules, atoms, or sum parameters, tables provide values for the output of the same species (“slip-through”) or for transformed ones. For solid waste, an environmental impact per kilogram of deposited waste is allocated. For the pretreatments, no transfer coefficients are included, because, for these operations, linear models are applied by calculating the output stream as a function of the incoming waste. Presently, the energy and auxiliary material requirements for the different operations are evaluated according to data provided by industry for each operation. Direct emissions from the pretreatment operations do not occur, because no waste can be released without passing a final treatment. Some operations lead to secondary streams that are also handled in the extended model. The mass balances in the different treatment operations can be described with the following generic equations:
Mi,j,k ) ki,j,kMi,j,in
(1)
∑k Mi,j,k ) Mi,j,in
(2)
∑j Mi,j,k
(3)
Mi,k )
where Mi,j,k is the mass (given in kilograms) of component j in a primary or secondary output stream k from a treatment operation i, ki,j,k the linear mass transfer coefficient (dimensionless) for component j from input stream Mi,j,in to primary or secondary output stream k, and Mi,k the total mass of an input or output stream k to or from treatment operation i. The linear mass-transfer coefficients ki,j,k represent separation efficiences, empirical enrichment constants, and transfer coefficients for incineration and STP.22,23 However, because the procedure now allows for multiple streams cohabiting in the waste treatment plants (see below), rectification and phase separation models have been modified as follows, to better handle important secondary streams.
5946
Ind. Eng. Chem. Res., Vol. 45, No. 17, 2006
The phase separation is based on the polarity of the different chemicals; hence, organic-organic phase separations can be conducted. Two phases are considered: polar and apolar. Each chemical j has a polarity Pj (dimensionless) between 0 and 1 (water is 1), which indicates the proportion of it that will migrate/reside in the polar phase. A polarity of 0.5 indicates an even split between the two phases. Efficiency kPS,j,k indicates the proportion for each chemical j that had no time to migrate to the correct output stream k. Equations 3 and 4 summarize the model:
MPS,j,pol ) MPS,j,inPjkPS,j,k
(4)
MPS,j,apol ) MPS,j,in - MPS,j,pol
(5)
where MPS,j,pol, MPS,j,apol, and MPS,j,in are the mass of chemical j in the polar (pol) phases, apolar (apol) phases, and the input mixture (in), respectively. The two output streams that result from the phase separation can both undergo all other treatments before passing the final treatment. The distillation/rectification model is straightforward: The distillation is batch, and, hence, any product can be recycled with a high purity. Two or three streams are produced: a bottom stream that contains all the products with a higher boiling point than the target chemical to be recycled, a recycled stream that contains the target chemical, and an overhead stream that contains chemicals with a lower boiling point. A single efficiency indicates the percentage of the target chemical that will be recycled. The recycled stream is directly taken into account as a financial and ecological bonus (see below), whereas the two other streams can both undergo all other treatments before passing the final treatment. No overhead stream is produced if the target chemical has the lowest boiling point. There is, however, always a bottom stream, because some impurities are always present. The nonrecycled portion of the target chemical is split evenly between the bottom and overhead streams. If there is no overhead stream, the recycled stream is augmented by the amount that would be in the overhead stream (i.e., the efficiency is comparatively higher). It is considered that any chemical can be recycled via a batch distillation; however, of course, it is more or less intricate. To model this, the energetic requirement U is computed as follows. The empirical approximation is based on the vaporization of the components j that go to the overhead and recycled streams, respectively, and their masses Mj,vap, and is calculated as the mass-weighted sum of the corresponding vaporization enthalpies ∆Hvap,j (expressed in units of kJ/kg). If the difference of boiling points ∆Tb (given in units of °C) between the target chemical and the two neighboring chemicals in the sorted list of boiling points is small, the energy requirement is augmented to account for a higher reflux ratio. Equation 6 shows the overall calculation of the energy requirement including a base requirement Ubase:
U ) Ubase +
(x ) (x )
∑j (Mj,vap∆Hvap,j) max 1,
A
∆Tb,j-1,j
max 1,
×
A
∆Tb,j,j+1
(6)
Parameter A is an empirical value that is chosen to best fit experimental measurements, if available. Below, we will introduce the mixing options considered in this work. The mass Mi,j,in of component j in a mixed input
stream to unit operation i is calculated from the massess Mi,j,km of the streams k to be mixed:
Mi,j,in )
Mi,j,km ∑ km
(7)
2.2. Cost Calculation. The treatment costs of a stream k in treatment operation i (Ci,k, expressed in terms of dollars) are calculated based on the quantity Qs,k (expressed in different units) of certain specifications of a stream (e.g., content of halogenated compounds, heat of combustion, or just the mass or volume) and corresponding unit-specific cost factors cfi,s (expressed in terms of monetary units per unit of Qs,k):22
Ci,k ) Ci,b +
∑s cfi,sQs,k
(8)
where Ci,b is a unit-operation specific base cost, independent of the mass and nature of the stream, which is added to discriminate toward simpler solutions. The empirical cost factors cfi,s were obtained from industry, and some are not constant but are dependent on the quantity Qs,k (for example, the content of halogenated compounds). The overall costs of a treatment sequence for a given stream k (Ck) are obtained by summation over all unit operations i included in the sequence:
Ck )
∑i Ci,k
(9)
Monetary benefits CBk are obtained for recycled solvents, generated energy, and recovered NH3 that originate from the treatment of stream k, i.e., for example, the purchase cost of fresh solvent or the saved utility cost are subtracted from the overall treatment cost of a stream to obtain net cost CNk:
CNk ) Ck - CBk
(10)
2.3. Environmental Assessment. The functional unit chosen in the environmental assessment is the treatment of a given waste stream, which is defined by a certain mass, density, and composition. Therefore, the implemented methodology assesses the remaining emissions of waste stream k from a treatment operation i into the environment, if any, and also considers the auxiliary materials and energy used during this operation:23
Ei,k )
∑j ufjQUi,j,k + ∑s efsQEi,s,k
(11)
where Ei,k is the environmental impact resulting from treatment of stream k in operation i (for the units of E, see the later discussion), QUi,j,k and QEi,s,k are the quantities of utility j used in operation i and emissions s (e.g., the amount of halogenated compounds) emitted from operation i, respectively, and ufj and efs are the environmental impact factors (see later discussion) per quantity of utility j and emission s, respectively. The overall environmental impact Ek of a treatment sequence for a given stream k are again obtained by summation over all unit operations i included in the sequence:
Ek )
∑i Ei,k
(12)
Similar to the economic assessment, the environmental benefits EBk that result from the recycling of materials and
Ind. Eng. Chem. Res., Vol. 45, No. 17, 2006 5947
recovery of energy originating from the treatment of stream k are also considered:
ENk ) Ek - EBk
(13)
The environmental benefit for recycled material includes the emissions that would result from the production of the same amount of recycled substance. In this manner, the method balances emissions from the synthesis of new material with the emissions resulting from the recycling of the same amount of substance. The balance region for the environmental assessment is shown in Figure 1. As discussed previously, the balance region explicitly takes into account the auxiliaries (e.g., steam, electricity, supporting fuel for incinerating material with low heat of combustion, precipitating agents), which are implicitly taken into account in the cost factors for the different operations. In Figure 1, the arrows that cross the system boundary represent streams for which environmental impact is regarded. The waste stream is deliberately drawn inside the balance region, because it is not evaluated itself, which is a point also stressed by the interrupted line leading from the process to the waste treatment region; only the environmental impact that is due to the applied treatment, as well as the remaining emissions, are assessed. For this task we use a set of indicators, rather than a single parameter. Emissions, energy consumption, auxiliaries, and recycled compounds are evaluated via the method of Ecological Scarcity.25,26 The concept of this method has been summarized (and compared to other methods) by Baumann and Rydberg27 and Hertwich et al.28 The method of Ecological Scarcity allows a comparative weighting and aggregation of various environmental interventions by the use of so-called ecofactors (units: UBP/kg, i.e., environmental impact points (Umwelt Belastungs Punkte) per unit of mass emitted). Compartment-specific emissions for most of the considered compounds (atoms, single molecules, and sum parameters) can be assessed. It is also possible to evaluate disposal in this fashion, because ecofactors for landfills (in UBP/ kg deposited material) are available, depending on the class of deposit needed (normal versus underground deposits). Moreover, the method of Ecological Scarcity takes the energy demand or gain into account and transforms it to UBP: 1 MJ of energys regardless of its formsis set equal to 1 UBP. Data for measuring emissions in UBP were observed in BUWAL,26 whereas the SimaPro software,29 in which the environmental impact of a large inventory of substances is expressed in UBP, was used to assess the gray input of auxiliaries, as well as to compute a bonus for each recycled compound. More details on this method, its implementation, and the hypothesis and extensions used in this project are given in Jankowitsch et al.23 In addition to these four indicators, in terms of UBP, which can also be aggregated, because they are expressed in the same units, 11 parameters from the European Chemical Industries Responsible Care program complete the environmental assessment.23 2.4. Uncertainty Propagation and Multiobjective Comparison. In the model, parameter uncertainty is considered by Monte Carlo simulation, because this approach allows propagating any type of probability distribution and the computational burden of a single evaluation of the presented model is low. The model was implemented in MATLAB so that the Monte Carlo simulation could be realized by vector and matrix operations. This means that, first, all uncertain parameters are generated according to the stipulated probability distributions, using a random number generator. These parameters are stored
in separate vectors or matrixes. In the latter, each column entry represents one set of parameters used in one Monte Carlo simulation. Afterward, in one single run, the model calculations are performed using the matrixes instead of single-parameter values.22 A Pareto approach is used in the multiobjective optimization. The objective of the Pareto approach is the elimination of alternatives that are clearly dominated (i.e., other feasible alternatives exist, which are better, in regard to both objective functions (i.e., cost and environmental impact)). As long as a deterministic approach is considered, the Pareto approach is simple to implement. However, if uncertainties must be incorporated, the situation becomes more complex. The uncertainty transforms each point into a range in both dimensions: from a best-case scenario (lowest costs and environmental impact) to a worst-case scenario (highest costs and environmental impact). In this perspective, point A dominates point B only if the worst-case scenario of point A dominates the bestcase scenario of point B. However, one problem with this definition is that synchronized effects on two alternatives are not taken into account at all; often, two alternatives are different only for a single operation. In such cases, it is not surprising that a change in one parameter affects both alternatives similarly. If this is not taken into account and only the extreme values are compared, one might set against each other two figures resulting from inconsistent simulations that have different underlying parameter values. Thus, in this study for the determination of the Pareto optimal alternatives, instead of examining deviations in two directions from a mean value, the discrete points resulting from two series of Monte Carlo runs were compared point by point. This procedure guarantees that the comparison is performed on points that originate from the same parameter values. To implement this procedure, the Pareto analysis routine was repeated for every Monte Carlo run individually. Afterward, an alternative was eliminated if it was dominated for all runs. 3. Multiple Streams and Mixing Even when a single waste stream is considered, separation operations create new streams. Each of those streams might benefit from different pretreatment, and require a different final treatment. Thus, for all those streams, the full treatment tree should be considered. Moreover, such streams might benefit from being mixed. Assume, for instance (see Figure 2), a phase separation to separate the organic phase. Afterward, this phase is distilled for the recovery of one valuable solvent. The bottom that contains organic remains and some water might be mixed with the initial aqueous phase for processing in the STP, so that the organics are diluted for sufficient degradation. If several waste streams are present on-site, which is usually the case, an even larger benefit might be gained by mixing and combining the streams: certainly not only dilution effects, as discussed previously, might be beneficial, but recycling might also be improved. Indeed, the amount of a chemical in a stream might be too low for a financially or ecologically interesting recovery. However, if several streams share the same component, the sum of all streams might lead to attractive recovery solutions. This is due in part to the fixed costs that are associated with operations: preparing, monitoring, and staffing a column for small amounts of a substance is not profitable. To leverage the mixing potential in the computation, we decided to allow mixing at two levels in the operation tree for liquid waste: after the local (in-process) recovery operations (i.e., right before the central recovery operations (MR, Mixing
5948
Ind. Eng. Chem. Res., Vol. 45, No. 17, 2006
Figure 2. Separation operations generate multiple streams that might be mixed (with other streams) for further treatment.
Figure 3. Schematic procedure and position of the mixing options within the treatment tree. The width of the lines on the left of the figure corresponds to the number of possibilities at this point (see text for further explanation).
for Recovery)), as well as after all recovery operations (i.e., before end-of-pipe treatments (ME, Mixing for End-of-pipe)). Figure 3 summarizes the two mixing options for a simplified treatment tree (see Figure 1 for the complete tree). The single stream separation and recovery and the first mixing option, as well as one central recovery scheme, are each labeled “Mixing and Recovery Scenario” (for brevity, this is labeled as “scenario” in the results plots given later in this work). Each stream in the system must be treated. For instance, a scenario might consider the aqueous phase of waste stream 1 only. However, the organic phase must also be treated. This means that several scenarios must be combined for a complete treatment of the waste; this is labeled a “combination”. In addition, the wastes must be treated up to a release into the environment. End-of-pipe operations must be conducted for all streams from all scenarios that constitute a combination; these streams can also be mixed. The combination, the end-of-pipe treatments, and possible mixing of streams before these represent a “complete solution” to the overall problem. Each combination may lead to different complete solutions when different endof-pipe mixes and treatments are selected.
The width of the lines on the left side of Figure 3 schematically shows the number of alternatives that are possible at different points in the procedure. It can be seen that mixing increases the problem size exponentially. The fact that partial solutions must be combined to obtain complete solutions again increases the number of possibilities. Therefore, heuristics have been added to reduce the number of potential solutions. These heuristics are of two different types: (i) chemical knowledge and user-defined limits (type 1), and (ii) computational limits (type 2). The first category of heuristics (type 1) is applied mostly for the first mixing (MR), but those that still apply remain enforced in the second mixing (ME). These heuristics are a collection of rules defined by experts and target the elimination of uninteresting (from the perspective of minimized cost and environmental impact), impossible, or illegal alternatives early in the computation. They range from basic rules, such as (i) two streams obtained by a phase separation should not be directly remixed, (ii) two streams can be mixed for central recovery only if they share common chemicals, and (iii) do not send plain streams without any water to the STP (if necessary, mix them earlier with water-containing streams), up to more experience-related rules, such as (i) do not try to recover a chemical that is too diluted (for instance,
