Life Cycle Assessment Coupled with Process Simulation under

Oct 1, 2008 - Life Cycle Assessment Coupled with Process Simulation under Uncertainty for. Reduced Environmental Impact: Application to Phosphoric Aci...
34 downloads 0 Views 2MB Size
8286

Ind. Eng. Chem. Res. 2008, 47, 8286–8300

Life Cycle Assessment Coupled with Process Simulation under Uncertainty for Reduced Environmental Impact: Application to Phosphoric Acid Production Aaro´n David Bojarski,† Gonzalo Guille´n-Gosa´lbez,‡ Laureano Jime´nez,‡ Antonio Espun˜a,† and Luis Puigjaner*,† Department of Chemical Engineering, Polytechnic UniVersity of Catalunya, 08028 Barcelona, Spain, and Department of Chemical Engineering, UniVersity of RoVira i Virgili, 43007 Tarragona, Spain

One of the most important drawbacks of life cycle assessment (LCA)-related analysis is the generation of reliable data. In the proposed methodology, this drawback is addressed using data from process simulations, based on first-principles models in the LCA calculations. Furthermore, uncertainty that arises from industrial data and a simulation hypothesis are explicitly incorporated, using Monte Carlo sampling, which allows statistical information to be translated into a set of representative scenarios for which the LCA calculations are performed. The combined use of LCA, process simulation, and sampling techniques results in a powerful environmentally conscious quantitative tool whose objective is to guide decision-makers toward the adoption of more-sustainable process alternatives. The main objective of the methodology is to show the main differences between production options. This novel methodology is applied to the specific case of phosphoric acid (PA) production. 1. Introduction With the recent trend of adopting more-sustainable technologies, there has been a growing interest for performing environmental studies in the chemical process industry. The goal of these studies is to identify the most significant environmental effects of a process and propose modifications with the objective of achieving sustainability improvements. Several approaches have been proposed, and good literature reviews of such approaches can be found.1,2 Several systematic methodologies are available for detailed characterization of the environmental impacts of chemicals, products, and processes.3 These methods include life cycle assessment (LCA), which was developed to set Environmental Management Standards through the ISO1404X series.4-7 LCA is an objective method that is used to evaluate the environmental loads associated with a product, process, or activity. The LCA framework adopts a systematic procedure that relies on quantifying the direct and indirect input and output flows of energy and materials of a given process and translating them into a set of impacts that affect the environment.8 Its application to the case of process industries has been recently reviewed in the works of several authors.9,10 Many methodologies that embody LCA principles have been proposed within the computer-aided process community, examples of such include the following: the minimum environmental impact process (MEI, or MEI methodology);11 the waste reduction (WAR) algorithm,12 which uses the pollution balance concept; the introduction of “eco-vectors” for the calculation of life cycle inventories for process industries;13 the environmental fate and risk assessment tool (EFRAT);14 and the OLCAP methodology.15 A main limitation in the application of any LCA-related technique is the lack of reliable input data. In fact, most of the LCA studies that can be found in the literature rely on estimated data.16 The most important and determining factors in the * To whom correspondence should be addressed. Tel.: +34 934 016 678. Fax: +34 934 010 979. E-mail: [email protected]. † Department of Chemical Engineering, Polytechnic University of Catalonia. ‡ Department of Chemical Engineering, University of Rovira i Virgili.

application of an LCA-related technique are the amount of information required and the reliability of available data. In this context, as in many others, the quality and validity of the decision reached at the end of the analysis are highly dependent, to a large extent, on the accuracy of the input parameters. Generating reliable data for LCA analysis can be accomplished in this methodology via process simulation, which involves the use of computer software to develop accurate and representative models that provide a better understanding of the process behavior.17 Process simulation is able to perform mass and energy balances promptly, which allows analysis of a variety of scenarios, even when there is scarce data available. Data available this way has been rigorously calculated using thermodynamical properties models and unit operation models. The use of process simulation, coupled with the calculation of environmental impacts, has been proposed by several authors,3,14,18-20 showing that the combined use of LCA and process simulation results in a robust approach that helps to overcome the lack of reliable data and ensures the quality of the recommendations given by the environmental study. Environmental impact calculations are further complicated by the high degree of uncertainty that is casued by several factors. According to Heijungs,21,22 uncertainty arises from the problem of using information or data that is unavailable, wrong, and unreliable or that shows a certain amount of variability. The same issue can be applied to relationships or choices that are used within a model. Uncertainties related to process simulation and LCA can be broadly classified into two groups: value-based uncertainty and parameter uncertainty. Value-based uncertainty results from modeling choices (i.e., the selected model, compared to other possible modeling choices); this would include, for example, the use of a given equation of state, compared to other possible thermodynamic models, to estimate phase equilibrium. Parameter uncertainty results from model parameters. According to Morgan et al.23 (Chapter 4, p 50), empirical parameters (or chance variables) are the only ones that are susceptible to description via a probabilistic measure (i.e., a probability distribution function (PDF)). All other values have an inherent bias due to decisions and should be described parametrically (using scenario analysis or scenario trees).

10.1021/ie8001149 CCC: $40.75  2008 American Chemical Society Published on Web 10/01/2008

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8287

Furthermore, empirical variables must be measurable either now or at some time in the past or future to be susceptible for description via PDF. These variables are the only type of quantity that is uncertain and can be said to have a true value as opposed to appropriate (or good) values that are subject to bias due to value decisions. Note that, within a process simulation model, several hypotheses have been made and many parameters appear as a consequence of such hypotheses. In the impact model, its uncertainty is mainly associated with dose, species toxicity estimation, and emission pathways and can be parametrized considering the impact characterization factors. Consequently, in this methodology, parameter uncertainty is modeled by means of PDFs, whereas value-based uncertainty is proposed to be treated by way of comparison of how modeling choices affect the results in different scenarios. The discussion of how uncertain parameters affect the results applied to environmental metrics has been addressed by several authors.24-26 Recently, Basson et al.27,28 presented an integrated approach for the consideration of uncertainty in decisionmaking, considering environmental metrics. Their method is based on the combined use of LCA and multiple criteria decision analysis (MCDA). Key elements in their approach are (i) the use of “distinguishability analysis” to determine whether the uncertainty in the performance information is likely to make it impossible to distinguish between the alternatives under consideration, and (ii) the use of principal components analysis (PCA) to make assessments on results. Their approach is based on three different strategies: (i) placing appropriate bounds on particular aspects, (ii) ensuring that the quality of information is such that the generated alternatives are “adequately distinguishable” between them, and (iii) propagating technical uncertainties and performing sensitivity analysis for valuation uncertainties. The objective of this paper is to propose a rigorous methodology based on LCA and process simulation, allowing for the consideration of uncertainty and with the purpose of providing a set of distinguishable options to decision-makers. The main advantage in the proposed approach is that the modeled uncertain parameters are closer to the plant real behavior, provided that the simulation model adjusts the process values. This makes it possible to actually correlate the variability of the results. For example, if a streamflow rate is uncertain and that flow passes through a pump, this pump’s power consumption will be highly correlated to the variability of the streamflow rate. The article is organized as follows: section 2 comments on the approach presented for coupling the process simulation results to LCA; in section 3, the methodology is applied to an industrial case; and, finally, the results and comments on the conclusions are discussed in section 4. 2. Methodology: Life Cycle Assessment Coupled with Process Simulation for Parameter and Value-Based Uncertainty Treatments The proposed methodology is based on ISO standards.4 Subsection 2.1 discusses the setting of goals and the scope, followed by subsection 2.2, which gives a compilation of the inventory table. Subsection 2.3 comments on impact characterization and ends with the inclusion of deterministic and stochastic impact characterization. Subsection 2.4 contains the methodological steps proposed for interpretation of the results. 2.1. Step 1: Goal Definition. The goal of this methodology is to provide the decision-makers with a set of distinguishable options. To achieve such an objective, several aspects must be

determined in this step, including the indicators to be assessed, the functional units (FUs), the system boundaries, the allocation procedure, and the appropriate model granularity. While setting those aspects, uncertainty sources should be determined and appropriate approaches for handling it are envisaged. With regard to the selection of indicators, let us note that it must be done in an iterative fashion. This is because an assessment of sustainability problems requires some knowledge of the problem and cannot be done beforehand. Specifically, in this work, it is proposed to be as exhaustive as possible, selecting as many indicators as available for calculation and calculating them accordingly (see step 3), with the objective of identifying inherent process design option tradeoffs between the various metrics. The selection of the functional unit (FU) must be performed following the ISO guidelines,4,5 which take into account the services provided by the product and not the product itself. In process industries, the former can be difficult, given that industrial facilities co-produce different products. However, in the commodities or energy production, where the products and/ or services are predefined and, in some cases, only one product is produced, this inconvenience is avoided. Commonly, the FU will be represented by a certain production flow of a given product. Conventional system boundaries are to be extended for explicit consideration of upstream processes, given that they are important contributors in the case of process industries29 (Chapter 11). However, this extension is susceptible to uncertainties related to the cutoff criteria. Scenario tree analysis is proposed to cope with this fact, which is regarded as a value consideration related to model approximation. Consideration of downstream processes must include the waste treatment of residues produced during product manufacturing and it also should take into account plant decommission. The phases of product use and final product disposal are usually difficult to model in the case of commodity products, given the wide variety of possible goods where they are used. Nevertheless, in most cases, process modifications do not alter the final product (i.e., a commodity has its properties fixed); therefore, the use and final disposal phases are usually the same for all considered options. Consequently, system boundaries are set from cradle to gate, disregarding the use and product disposal phases. In the same way, multiproduct production is common in process industry; therefore, the case where the same facility produces different commodities should be contemplated. In a regular LCA application, allocation should represent the actual environmental damage responsibility; however, because several allocation strategies and values are feasible, depending on the present and future production scenarios and environmental management policies, uncertainty that is associated with these different issues impedes the application of general assessments. Consequently, the allocation procedure will be dependent on each case. Uncertainty due to different allocation procedures must be modeled following a value-based method, by parametrizing it and generating scenarios, taking into account each modeling choice. Several allocation strategies are available, based on physical (generally mass) or economical values. The selection of any of those strategies will be dependent on the process under study. In many cases, allocation can be avoided using system boundary expansion, by considering the production of co-products from virgin raw materials. Environmental impacts associated with these production facilities are taken into account when calculating the impact of main-product production (by addition or subtraction). Note that our approach is general

8288 Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008

Figure 1. Proposed framework involving process simulation and stochastic sampling.

enough to accommodate any type of allocation method that is available in the literature.30 With regard to the model granularity and complexity, let us note that not all of the models are able to impact certain indicators. For example, measuring the impact of different processing policies could end in indistinguishability, given that all alternatives show similar values for most indicators. Thus, the model complexity should be assessed along with the selected indicators. In the case of sustainability assessment, we must select appropriate environmental, economical, and social metrics among those available in the literature.31 Following the former guidelines, uncertainty is considered through the goal definition, by assessing the sources of variability in model parameters, and also through scenario characterization by determining value-based sources of bias. 2.2. Step 2: Process Model. The second step entails the construction of a process model according to the objectives set in step 1. The process model that is created must reproduce the system under study. Special attention must be given to the information requirements of the selected indicators and, subsequently, model detail (i.e., granularity) must be defined accordingly. Most environmental indicators require mass and energy flows to calculate environmental impacts. Economic indicators use the former flows information but converted to monetary units (using prices) to estimate cash flows; they also require fixed investment estimations and several other parameters. Social indicators generally require data regarding human resources used in a factory, income distribution, and land use; in some cases, social indicators require information regarding contribution to macroeconomic indicators.20,31 Consequently, the objective of the model is to estimate the following: (i) mass flows that enter or leave the system, representing raw material extraction, production of products, or emissions; (ii) energy flows in any form, ranging from heat to electrical power must be quantified; (iii) raw material and product prices, investments, fixed costs; and (iv) employee wages and the distribution of macroeconomic data. Process simulation can provide mass and energy flows, and it also helps to design equipment that can be further quantified in terms of cost. Therefore, the selection of process simulation to build models seems natural, given the information requirements. We must emphasize that most process models are generally nonlinear, so this framework considers the inherent nonlinearity of process industries by adopting process simulation as a tool. A common example of nonlinearity is found in the estimation of emissions. This issue goes generally above

standards regarding LCA models that consider linearity and processes working under fixed steady-state conditions. Uncertainty considerations are taken into account by allowing the process variables and model parameters that are considered to match PDFs based on literature surveys or industrial field measurements. Tools such as analytical error propagation or sampling based tools can be used. Commercial process simulators are readily available, Aspen Plus and Aspen Hysys (www.aspentech.com), Pro II (www.simsci-esscor.com), and Prosim (www.prosim.net). Most of them do not allow for an analytical treatment of uncertainty, because the unit-operation models cannot be accessed by the user. Consequently, parameter uncertainty propagation is best-achieved by means of a sampling procedure. Furthermore, the usage of sampling-based methods can provide insightful metrics, based on variables variance32 (Chapter 2), that are global instead of local, because the information is provided by analytical error propagation procedures. The objective of this general methodology is to consider two sources of uncertainty: that from model parameters and that which is due to the choices made in building such a model. The later source can be treated parametrically by means of tree scenario analysis, where all possible modeling choices are explored. As a result of this step, an inventory of environmental interventions, as well as financial and social data, is gathered. The inventory if no uncertainty is provided (which regard to parameters or value-based decisions) is then deterministic; otherwise, the result is a set of economic, environmental, and social interventions, based on a parameter tree description. 2.3. Step 3: Metrics Calculation and Sensitivity Analysis. In this step, the metrics defined in step 1 and all process and additional process-related data from step 2 are used to calculate all criteria that measure each processing option of the scenario tree analysis. As output from this model, each process alternative is evaluated for every criteria. Dominance, contribution, break even, and other systematic analyses can be performed, with the purpose of determining the main process parameter contributors to each measure29 (Chapter 11). Sensitivity analysis techniques32 can be used during this step to determine sources of variability in the results, with the objective of increasing the model capabilities. Model validation tests are performed concerning process parameters that affect the selected indicators as expected. This issue can lead to an increase or decrease in model granularity and, as a consequence, to introduce an iteration step going back to step 2 for model

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8289

improvements. In Figure 1, the overall framework of the entire data gathering process can be observed. 2.4. Step 4: Interpretation. This final step provides decision support for the selection of the final alternative. MCDA techniques could be used to elicit preferences of the stakeholder(s) by obtaining a ranking of options, based on a single overall metric. However, the objective of this methodology is to allow the stakeholders to know how distinguishable the processing options are between them, regarding the selected criteria. Furthermore, working with different objectives is preferable, because it provides valuable insights into the problem and helps to understand the inherent tradeoffs between selected criteria. Special attention is given during this step to value-based scenarios and its possible distinguishability. A discernibility analysis previously proposed by Heijungs and Suh22 (Chapter 8), for the purpose of comparing MC results, is used. It is based on comparing products pairwise in each MC trial. The ratio (or the difference) between MC results for each is computed, and the resultant distribution is analyzed. 3. Methodology Application: Phosphoric Acid Production These next four subsections describe the application of the proposed methodology to the environmental assessment of a real phosphoric acid (PA) production process. PA is the second-largest mineral acid produced worldwide, considering its volume and value. Its production is performed via two different processing routes: a wet method and a thermal method. The thermal route involves electric-furnace smelting of the phosphate-containing mineral, using coke and silica to produce elemental phosphorus, which is then converted to PA by first burning (oxidizing) the phosphorus to P2O5 and then absorbing the obtained P2O5 in water. This process results in an expensive food-grade acid of high purity that has been proven to be overspecified for general fertilizer use.33 The wet method process is based on sulfuric acid lixiviation of apatite rock (Ca10P6O24F2 or fluoroapatite Ca10P6O24(OH)2, hydroxyapatite), followed by a filtration of the gypsum formed, which, in this industry, is known as phosphogypsum (PG), and the subsequent concentration of the solution, yielding PA in technical grade, which is also known as wet process phosphoric acid (WPPA).34,35 The wet method route is further studied throughout this paper. According to Becker34 (Chapter 2), the following reactions occur inside the apatite rock dissolution reactor: Mineral acids dispersion in the solution: H2SO4 f 2H+ + SO24 H3PO4 f H+ + H2PO4 H+ ions attack the phosphate rock: nH+ + Ca3(PO4)2(s) f 2H3PO4 + (n - 6)H+ + 3Ca2+ Ca2+ ions precipitate with SO42- as gypsum: Ca2+ + SO24 + 2H2O f CaSO4 · 2H2O(s) During the lixiviation of the mineral, by controlling the reactor temperature and the P2O5 concentration, one can select which calcium sulfate hydrate is formed: dihydrate (CaSO4 · 2H2O) at ∼70-80 °C for 26%-32% P2O5 and hemihydrate (CaSO4 · 0.5H2O) at ∼85-95 °C for 40%-52% P2O5. The WPPA obtained through this method is suitable for fertilizer production, which is the destiny of 80% of its production in Europe.36,37 According to EFMA,35 fluorine is present in most phosphate rock, to an extent of 2%-4% w/w. This element is emitted

during the reaction of the rock in acidic media, initially as hydrogen fluoride (HF). However, in the presence of silica, HF reacts to form fluorosilicic acid (H2SiF6), according to the following set of reactions: CaF2(s) + 2H+ f 2HF + Ca2+ 4HF + SiO2(s) f SiF4 + 2H2O 3SiF4 + 2H2O f 2H2SiF6 + SiO2(s) H2SiF6 f SiF4 + 2HF A certain proportion of the fluorine evolves as vapor, depending on the reaction conditions, whereas the remainder stays in the solution, leaving the process either with the product or with the process water. More-volatile fluorine compounds will appear in the vapors that are exhausted from the evaporators when the acid from the filter is concentrated. The fluorosilicic acid (H2SiF6) may decompose, under the influence of heat, to give volatile silicon tetrafluoride (SiF4) and hydrogen fluoride (HF). The PA industrial facility studied uses mineral rock from different sources, and the sulfuric acid production facility is located at the same site. Electricity generation within the plant is also considered. The plant has the following production facilities with the associated annual capacities:38 component

capacity

sulfuric acid (two facilities) 385 000 t/yr oleum (fuming sulfuric acid) 13 000 t/yr dilute phosphoric acid 110 000 t/yr concentrated phosphoric acid 40 000 t/yr anhydrous hydrofluoric acid 7500 t/yr calcium phosphate 60 000 t/yr facilities for storage, packing, and palletizing; internal distribution and loading onto trains and trucks of fertilizers and chemical products

Prior environmental studies related to the phosphorus fertilizers industry have shown that relevant environmental issues are those related to (i) greenhouse gas (GHG) emissions, (ii) process emissions (such as HF, PO43-, and SiF4 mixtures) into air and water, and (iii) management of the produced PG. It is generally accepted that the biggest environmental problem in the WPPA industry is the destiny of PG wastes35,36 and its lixiviates. The destiny of PG is usually one of three possibilities: (i) discharge it into the ocean or other water basin, (ii) store the produced PG inland into ditches and ponds, or (iii) use it as a useable product.39 In all three former cases, spent process water used to transport PG and PG itself contain trace metals that are found in the phosphate mineral and sulfuric acid used with other silicon and fluoride compounds. According to a screening LCAbased39 analysis of the Dutch fertilizer industry (two industrial sites), the overall environmental performance of the gypsum reuse scenario is better than the landfill scenario, with both of them being better than the discharge option, in the Dutch situation and in the western European situation. Regardless of this finding, in this work, PG is assumed to be stockpiled on land, given that it is the current practice of the industrial site used as a case study. The net emission of GHG from the manufacture of phosphate fertilizer is largely determined by the method of sulfuric acid production.40 GHG emissions were primarily CO2 emitted during the consumption of fossil fuels. It is also reported that the transport of raw materials, intermediates, and products comprised a considerable portion of the emissions budget, which, for some studies, were in the range of 20%-33%. Other

8290 Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008

studies41 have indicated that the overseas transport of raw phosphate rock was particularly important. Along these lines, da Silva42,43 highlighted that GHG emissions are mainly caused by transportation in the case of the PA production in Brazil. Regarding fluoride emissions, which are caused by the reaction of fluor that is present in the phosphate rock impurities with an acidic reaction medium, it is reported37 that they can be reduced almost completely to zero if a water closed loop is accomplished; it is also reported35 that scrubber efficiency for their abatement is >99%. It has been found the possible generation of a byproduct H2SiF6, up to a concentration of 20%-25%, from the scrubbing liquors, which can be sold as a product that can be used for the production of aluminum fluoride.35,36 Given that the industrial site can cope with different waste treatment options, these options are further studied, following the proposed methodology. In the case of the PA production in Brazil,42,43 it has been reported that the main contributor to eutrophication are the losses of PO43- during the production of PA. 3.1. Step 1: Goal and Scope Definition. In this phase, the system boundaries and the impact categories are identified. Specifically, our analysis considers the impact of raw materials (phosphate rock and sulfuric acid) but neglects the product use and destiny (grave). Based on the former hypothesis, a cradleto-gate approach is adopted. Furthermore, it is important to highlight the following key points regarding the system boundaries: (1) For PA production, the boundary lies just after the production of concentrated phosphoric acid, considering that all produced low-concentration acid (32%) is concentrated up to 54%.35 (2) With regard to gypsum, no production of usable product is analyzed. Instead, it is considered to be stockpiled on settling ponds. The trace component lixiviates are considered to be only 10% of the trace species released to the soil,36,39 as will be discussed in section 3.2.2. With regard to the indicators used, although the methodology allows for consideration of economic and social impacts, for the sake of simplicity, we restrict our analysis to the assessment of certain environmental indicators. In the literature, environmental impact metrics that have already been developed are available. These could be separated into two different groups: (i) end- point indicators, such as Ecoindicator 99,44 and (ii) mid-point indicators, such as CML v245 or EDIP.46 Mid-point indicators are found as intermediate measures in a environmental impact chain (environmental mechanism) from emission to a final impact. To convert from emission to impact (end, mid-point), characterization factors (CFs) are used. CFs are found in the literature, tabulated for different substances and impacts in different methodologies that are available. End-point indicators rely on damage models that feature a high degree of uncertainty that cannot be easily documented. In this work, the approach used is a mid-point approach, given that end-point assessment increases uncertainty associated with the metrics, because of a more-complex impact model. Furthermore, the mid-point approach used is CML v2, given its wide use in the literature and the presence of CFs for all substances considered in this case study. In CML v2, the environmental burdens (emissions or natural resources consumed) are classified according to the relative contributions to specific potential environmental effects. It considers 10 impact categories: abiotic depletion potential (ADP), global warming potential (GWP), ozone layer depletion

potential (OLDP), human toxicity potential (HTP), photochemical oxidation potential (POP), acidification potential (AP), and eutrophication potential (EP), as well as ecotoxicity to fresh water, marine aquatic, and terrestrial ecosystems. Each one of them is expressed as a flow of a specific substances, taken as indicators of that impact. In the case of ADP, the reference flow is one kilogram of antimony per antimony equivalent; for GWP, the units are expressed in terms of kilograms of CO2 equivalent. The units used for OLD are kilograms of CFC-11 equivalent, and for PO, the substance used is expressed in terms of kilograms of C2H4 equivalent, whereas that for AP is expressed in terms of kilograms of SO2 equivalent and that for EP is measured in terms of kilograms of PO34 equivalent. The toxicity to humans and to ecosystems is measured in the CML v2 framework, considering the emission of 1,4-dichlorobenzene (1,4-DB) as a reference substance. 3.2. Step 2: Process Model. The life cycle inventory (LCI) data used in this study comes from a real industrial plant in south Europe and from a simulation of the process that is implemented in the commercial process simulator Aspen Plus.47 Figure 2 summarizes the four main processing steps considered in the inventory analysis of PA. Four main processing steps can be identified for PA production. The first two involve the raw material processing of phosphate rock, whereas the remaining two are the production of PA and the disposal of water and solid effluents. In this work, these last two processing steps are modeled in Aspen Plus. Note that the processing steps that are modeled in Aspen Plus constitute the “forward system”, for which process specific data are gathered, whereas the remaining data concerns the “background system”, for which data from average processing technologies are assumed.8 Three possible process alternatives are analyzed, which are set according to how the liquid effluent from the plant and settling ponds is treated. This effluent comes mainly from the scrubbing liquors and the gypsum filter unit. The options considered in the analysis are the following: (1) The first option considers that all wastewater is dumped into the ocean. A very small amount of the process water is recycled back to the plant (1%). A pH value of 8.2 is assumed for the calculation of the chemical species that are present in ocean water. This option is labeled as Option 1: “no wastewater treatment or ocean disposal”. (2) All wastewater is neutralized (a pH discharge of 7 is assumed) and then dumped into settling ponds. The plant recycles back part of the water required for processing; specifically, only 10% of the water consumed is disposed, whereas the remaining 90% is recycled back to the plant. The water emissions of these ponds is considered to contain the same composition of water plant effluents after neutralization. These emissions enter the groundwater compartment for the impact calculation. This option is denoted as Option 2: “neutralization only”. (3) All wastewater is treated to recover H2SiF6 (22%) and then is neutralized in a second step prior to being disposed into settling ponds. In this case, the recirculation of spent water is done in a manner similar to that described in the former option. This option is labeled as Option 3: neutralization and HF recoVery. The selection of these three options serves us to show the potential of the usage of process simulation as a tool for LCA analysis and its capabilities as a ”what-if” case generator. Note that these options by no means convey legal requirements and

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8291

Figure 2. Processing stages considered for PA production.

they are modeled as such to have a grasp of the environmental profile of such “what-if” scenarios. All the processing options are compared using 1 kg of produced PA as a FU. Regarding co-product allocation, as is the case of option 3, the production of H2SiF6 is considered to prevent the environmental impacts arising from its production from virgin materials. Boundaries for options 1 and 2 do not consider the systems boundary expansion needed for H2SiF6 co-production. The use of electricity for PA production is based on the Greek power network, given that the industrial facility is located in Greece. The study assumes that a certain amount of electricity is produced onsite (20% of the total). Regarding steam consumption, which is mainly used for PA concentration, we assume that its demand is mainly covered by steam generated from the H2SO4 production facility (80% comes from site-site integration), whereas the remaining amount is considered to be obtained from natural gas combustion. The analysis also considers the use of chemicals (lime) for groundwater control in the case of options 2 and 3, in which the effluent is neutralized. On the other hand, the transportation of the rock, sulfur, and other materials is not included within the system boundaries. The emission of radionuclides is not considered either, given that no industrial information is available. Finally, the infrastructure such as buildings or mines, is included in the analysis. 3.2.1. Wet Phosphoric Acid Production Plant Simulation Model. The simulation model implemented in Aspen Plus is based on data from a real process facility, and it includes mass and energy balances associated with its main chemical compounds and reactions. The simulation of the chemistry of the process requires the use of a complex thermodynamic model to address electrolytes species in solution. Specifically, this work uses the Electrolyte-NRTL model,48,49 which allows one to

consider a great amount of the ionic species that appear in the mixture. In this case, the most important ion species considered are Ca2+-H3O+-SO42--HSO4--H2PO4--H3PO4-H2SO4H2O. The thermodynamic model relies on several hypotheses that have been considered for H3PO4 dissociation and OH- already used in the literature.50,51 Thermodynamic data for apatite rock was retrieved from the works of Bogach.52-54 Vapor-liquid (VL) equilibrium for CO2 and H3PO4, was modeled with the assumption that those species follow Henry’s solubility law. Henry’s gas constants are retrieved from the Aspen Properties data library.49 For the case of HF VL equilibrium, a special case of the ElectrolyteNRTL equation is used, taking into account HF hexamerization (6HF T HF6) in the vapor phase.55 For the sake of simplicity, the uncertainty associated with thermodynamical data was not considered. However, it could be easily covered by following the approach of Vasquez.25 Therefore, all thermodynamical parameters and properties are regarded as deterministic and are not subject to uncertainty. The phosphoric rock is modeled as a mixture of the following compounds: fluoroapatite (Ca10P6O24F2), 79.3%; calcite (CaCO3), 11.1%; anhydrite (CaSO4), 2.9%; CaF2, 4.0%; and SiO2, 2.7%.56,57 Chemical reactions that are associated with rock dissolution occur via a proton transfer mechanism. This consideration implies that all the reactions considered in the simulation are forced to attain chemical equilibrium.56 Phosphate rock dissolution is solved in a reactor that minimizes the Gibbs free energy and attains chemical equilibrium. Phosphate rock attack tanks are considered to be a combination of mixers and twophase flash units, combined in such a way that provides outlet streams results that are similar to those provided in the industry. The model reproduces a rock attack tank that is cooled by partial evaporation of its mixture, which is the technology currently

8292 Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008

implemented. The reactions that are taken into account for each reactor are as follows: Solution ion equilibrium: CO2 + 2H2O T H3O+ + HCO3 2+ HCO3 + H2O T H3O + CO3

H2SO4 + H2O T H3O+ + HSO4

made, given that no information of such type was available. The compositions (wi) of the trace species in the phosphate rock are taken from the literature.34 A partition coefficient Ri, based on the work of Seijdel,39 is considered for the split of each trace species i between gypsum and filter liquor. Equations 1 and 2 correspond to a mass balance for each species i; that, combined with eqs 3 and 4, provide a distribution of trace species between outlet flow streams for each waste treatment option j. in total_tracein i,j ) wi · rock_flowj

2+ HSO4 + H2O T H3O + SO4

∀ i, j (2)

total_waste_water_traceout i,j

H3PO4 + H2O T H3O+ + H2PO4

in gypsum_traceout i,j ) Ri · total_tracei,j

SiF26 T SiF4 + 2F

∀ i, j

(3) ∀ i, j (4)

in total_waste_water_traceout i,j ) (1 - Ri)total_tracei,j

Liquid-solid (LS) equilibrium: CaF2(s) T Ca2+ + 2FCaCO3(s) T Ca2+ + CO23 Ca10P6O24F2(s) + 12H3O+ T 2F- + 10Ca2+ + 6H2PO4 + 12H2O SiO2(s) + 4H3O+ + 6F- T 6H2O + SiF26 Gypsum crystallization is modeled using a mixed-suspension mixed-product-removal (MSMPR) crystallizer model.58,59 The solids flows were also calculated via a screen model that mimics the filter unit mass balance. The kinetic and design parameters of those models have been effectively tuned to reproduce the process plant solids mass balance. MSMPR and screen models were used from the Aspen Plus built-in library of the unitoperation models.47 In the case of the crystallizer, the former set of reactions was extended and gypsum crystallization was added: CaSO4 · 2H2O(s) T Ca

(1)

out total_tracein i,j ) gypsum_tracei,j +

HF + H2O T H3O+ + F-

2+

∀ i, j

+ SO24 + 2H2O

Fluorine air emissions of vapor effluents from rock attack reactors and the PA concentration unit are calculated considering a scrubbing efficiency of 99% on a mass basis. The efficiency value was based on BAT literature.35-37 The PA concentrator unit and the HF scrubbers are modeled as single-stage contactors that attain chemical equilibrium. H2SiF6 byproduct recovery from scrubbing liquors is calculated considering a fluorine compounds recovery of 90% (mole basis)35 and an outlet concentration of 22%. This recovery/separation unit is modeled using a component splitter. Because of the lack of available industrial data, no rigorous treatment of this recovery stage is performed. This stage is the regarded as a “black box” model (component splitter) that attains thermodynamical equilibrium. 3.2.2. Trace Species Model. For the sake of simplicity, trace species are treated separately; their chemical behavior was not taken into account in the process simulation. Only a mass balance is performed on them; the trace species considered in this model include As, Cd, Co, Cr, Cu, Hg, Mn, Ni, Pb, Ti, V, and Zn. No distinction between different oxidation states is

Allocation of the amount of traces in the outlet streams of the process such as wastewater (WW), PA product, and HF that are recovered is based on the mass flow ratios given by the simulation (βj and γj) for each option j. Equations 5 and 6 allocate trace species between the PA product and the remaining streams. out total_waste_water_traceout i,j ) product_tracesi,j +

∀ i, j (5)

out waste_water_traceout i,j + HF_tracesi,j

∀ i, j (6)

out product_tracesout i,j ) βj · total_waste_water_tracei,j

No emissions from the PA product are considered. Trace species in the other remaining streams are allocated using eqs 7 and 8. waste_water_traceout i,j ) (1 - βj) · (1 - γj) · total_waste_water_traceout i,j

∀ i, j (7)

HF_tracesout i,j ) (1 - βj) · γj · total_waste_water_traceout i,j

∀ i, j (8)

Table 1 summarizes the βj, γj values that have been calculated. When HF is not recovered, γj is considered to be zero. Equations 1-8 account for the distribution of the trace species between all the streams that leave the process. The emission to soil and water of the ith trace species is modeled by considering an emission constant, depending on the sink, as follows (see eqs 9 and 10). soil_emissioni,j ) kGE · gypsum_traceout i,j

∀ i, j

(9)

water_emissioni,j ) kWE · pond_tracesout i,j

∀ i, j

(10)

The former formulation is a rigorous attempt to model trace species flow rates without considering the complex chemistry that is involved in such a chemical system. The model presented contains species in very low concentrations (ppms and ppbs) and in different possible states of oxidation, and it is specially suited to the industrial data available. The estimation of kGE is

Table 1. Trace Species Allocation Coefficients, Calculated from Aspen Plus Results Option 1

a

Option 2

Option 3

variable

mean

stdev

mean

stdev

mean

stdev

βj γj

5.14E-01 N/Aa

9.62E-04 N/Aa

5.14E-01 N/Aa

9.78E-04 N/Aa

5.14E-01 4.44E-02

9.34E-04 7.37E-04

Not available.

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8293 Table 2. Variables Ranges Used for Monte Carlo Simulation Feed to Aspen Plus Range variable

distribution

min

max

water inlet temperature air inlet temperature reactor 1 flash vessel temperature reactor 2 flash vessel temperature reactor 1 flash vessel pressure reactor 2 flash vessel pressure scrubber 1 pressure scrubber 2 pressure scrubber 3 pressure flash concentration unit pressure scrubber 4 pressure

uniform uniform uniform uniform uniform uniform uniform uniform uniform uniform uniform

25 °C 20 °C 63 °C 63 °C 640 mmHg 640 mmHg 700 mmHg 580 mmHg 700 mmHg 560 mmHg 680 mmHg

33 °C 30 °C 73 °C 73 °C 720 mmHg 720 mmHg 780 mmHg 660 mmHg 780 mmHg 640 mmHg 760 mmHg

set to 10% for all the trace species that flow with gypsum in all the waste treatment options.39 With regard to kWE, it was considered that all the trace species in the water effluent are water emissions (100%). 3.2.3. Model Uncertainty Sources. The uncertainty of the entire model results from the industrial and literature data that are used. These data have a specific degree of accuracy and variability. Specifically, the uncertain parameters considered in this study can be separated into three different groups: (1) Parameters of the process simulation model: Simulation variables values and distribution functions are based on modeling hypothesis and the available industrial data. Unitoperation temperatures and pressures, as well as inlet stream temperatures, are assumed to follow uniform probability functions (see Table 2). (2) Trace species model parameters: Distribution coefficients are taken from the literature39 (wi, Ri, kGE, kWE). The uncertainty of the former variables is taken into account using uniform PDFs. Others variables that are calculated from simulation results (rock_flowjin, βj, γj) are modeled considering normal or log-normal distributions. (3) Parameters from other echelons of the production supply chain: These parameters are associated with the production of sulfuric acid, phosphate rock, and lime, as well as electricity and heat generation. The PDFs used to represent variability in these parameters are various normal or log-normal distributions, and they are dependent on the information available in the Ecoinvent database. The first two items represent data that correspond to foreground processes and are directly affected by the definition of the FU taken and the selected option. On the other hand, the third item represents data that remain in the background system, where individual plants and operations cannot be identified. A sampling procedure scheme (Monte Carlo (MC)) was used to address the uncertainty that results from the model parameters. The MC sampling was implemented in two consecutive stages. The first stage involves the first group of uncertain parameters and is implemented in Matlab,60 which generates several equally probable scenarios (i.e., all of them have the same probability of occurrence), based on the statistical information available. These scenarios are then fed into the Aspen Plus program,47 using the COM interface. Here, a simulation is run for every scenario and the associated results are compiled for the calculation of a partial life cycle inventory (LCI). This LCI corresponds to the simulated echelon of PA production associated with a waste treatment option. In a second stage, the foreground information compiled in the first stage is used in combination with information that results from the trace model and Ecoinvent databases to calculate

the complete LCI, which covers the entire supply chain of the PA (see section 3.3.2). The selection of the process simulation variables that are regarded as stochastic comes from a sensitivity analysis that identifies which variables have the greatest influence on the emissions of HF into air and water. The stochastic variables and their selected uncertainty distribution functions can be seen in Table 2. A uniform distribution function was used, given that industrial information regarding the most-probable value was not available and that the usage of uniform distributions permits the selection of operating parameters that allow the process simulation model to converge on all scenarios. Table 3 summarizes the results obtained by following the previously discussed procedure. In particular, it shows the mean value and standard deviation of the Aspen Plus simulation results for each option generated by MC sampling. These results are expressed per kilogram of PA produced. The number of simulation runs is equal to the number of scenarios, which is set to 1000. The number of scenarios was fixed by gradually increasing it and stopping whenever no significant changes were detected in the simulation results. Upon analysis of these results, the first thing to notice is that the three options lead to similar outcomes in most of the calculated ratios. Nevertheless, the following differences in mean values can be still observed: (1) Lime Consumption: Option 2 leads to a higher consumption, compared to option 3. This is due to the fact that, in option 3, the amount of acids being dumped to ponds is less. (2) HF Emissions to Water: Options 1 and 2 give similar values, whereas the results from option 3 are in an order of magnitude lower. This is attributed to the recovery of HF as a byproduct. (3) Steam Consumption: In option 3, the amount of steam consumed is slightly greater than in the other two options. This is mainly due to the steam consumption associated with the recovery of HF. From these results, we conclude that the steam production related to the impacts of option 3 is greater than that which corresponds to the other options, whereas the water and air emission impacts of options 1 and 2 are greater than those in option 3. 3.3. Step 3: Metrics Calculation and Sensitivity Analysis. The quantification of the environmental performance of the PA production process requires the use of an impact model that allows for the translation of the process environmental interventions (i.e., emissions or raw material consumptions) into the corresponding environmental impacts. The calculation of these impacts is performed using SimaPro.61 SimaPro is a software tool that allows one to collect, analyze, and monitor the environmental performance of different products or processes by following a life cycle thinking approach. To clarify, note that the simulation model provides all the required information (regarding the foreground system), while all the background information is retrieved from different sources. This model, which contains partial LCI data (i.e., data from the foreground system), is complemented with information from other sources, such as available industrial data and also data from the Ecoinvent database.62 The latter source provides the inventory of emissions associated with the most widely used manufacturing technologies found in Europe. For the sake of clarity, the trace species model is embedded into the simulation model. Specifically, the environmental impacts analyzed in this work are those which correspond to the CML v2 method with the normalization and weighting coefficients set for western Europe

8294 Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 Table 3. Monte Carlo Aspen Plus Simulation Resultsa Option 1

a

Option 2

Option 3

variable

mean

stdev

mean

stdev

mean

stdev

rock consumption H2SO4 consumption lime consumption H2SiF6 recovered steam consumption HF air CO2 air HF water H2SO4 water H3PO4 water

1.35E+00 1.83E+00 N/A N/A 5.51E-01 1.00E-05 6.15E-02 6.49E-02 9.78E-01 3.34E-03

6.87E-06 9.32E-06 N/A N/A 9.62E-04 2.86E-07 9.72E-05 1.15E-04 5.68E-03 1.70E-08

1.35E+00 1.83E+00 4.45E-01 N/A 5.51E-01 1.00E-05 6.15E-02 6.15E-02 9.78E-01 3.34E-03

6.82E-06 9.25E-06 6.05E-05 N/A 7.66E-03 2.89E-07 9.74E-05 1.15E-04 5.72E-03 1.69E-08

1.35E+00 1.83E+00 4.10E-01 1.95E-01 5.65E-01 1.00E-05 6.15E-02 6.80E-03 3.86E+01 3.34E-03

7.14E-06 9.69E-06 1.82E-05 3.55E-04 7.26E-03 2.84E-07 9.69E-05 4.08E-05 2.27E-01 1.77E-08

Mean values are expressed in units of kg/kg.

Figure 3. Comparison of normalized environmental impacts for different options and processing possibilities (expressed in units of yr-1). Table 4. Deterministic Environmental Impact Assessment Results Option 1

Option 2

Option 3

impact category

units

Value

Position

Value

Position

Value

Position

ADP AP EP FWEP GWP HTP MAEP OLPD POP TEP

kg Sb equiv kg SO2 equiv kg PO43- equiv kg 1,4-DB equiv kg CO2 equiv kg 1,4-DB equiv kg 1,4-DB equiv kg CFC-11 equiv kg C2H4 kg 1,4-DB equiv

3.53E-03 3.03E-02 4.03E-03 2.71E-02 5.07E-01 1.64E-01 7.99E+02 5.24E-08 1.19E-03 2.28E-03

first second third first first third third first second third

4.51E-03 3.07E-02 9.31E-04 3.73E-02 9.43E-01 1.39E-01 6.23E+02 8.23E-08 1.27E-03 1.35E-03

third third second third third second second third third second

3.62E-03 2.37E-02 8.73E-04 3.38E-02 7.89E-01 9.04E-02 1.79E+02 6.86E-08 9.84E-04 1.18E-03

second first first second second first first second first first

in 1995. The impact assessment requires the definition of a LCI in SimaPro, which is built using the results provided by Aspen Plus. The consumption of raw materials (inlets) and outlet flows (emissions) of the PA process are taken from the simulation

results, and they are fitted to a given distribution curve inside SimaPro. The selection of the curve shape is made based on minimum square differences criteria. Data from generic PA production units pre-existent in the Ecoinvent database62 are also used. The combined use of process simulation and standard

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8295 Table 5. Deterministic Impact Assessment Results (Normalization Results) Normalization Result (yr-1) impact category

option 1

option 2

option 3

ADP AP EP FWEP GWP HTP MAEP OLDP POP TEP

2.38E-13 1.11E-12 3.23E-13 5.37E-14 1.05E-13 2.16E-14 7.04E-12 6.29E-16 1.45E-13 4.84E-14

3.04E-13 1.12E-12 7.46E-14 7.39E-14 1.96E-13 1.83E-14 5.49E-12 9.88E-16 1.53E-13 2.87E-14

2.44E-13 8.68E-13 7.00E-14 6.70E-14 1.64E-13 1.19E-14 1.57E-12 8.23E-16 1.19E-13 2.51E-14

Table 6. Uncertainty Sources for Different Versions of Waste Treatment Option 1 Remark/Value source of uncertainty

Figure 4. Comparison of normalized environmental impacts, normalized to the maximum value.

data from the environmental database allows for the calculation of the inventory of emissions required to determine the environmental impact of every process alternative being analyzed. With regard to the Ecoinvent data used, it is considered that the sulfuric acid is produced from BAT in Europe. Other consumption data, such as electricity and heating, are taken from other unit operations and LCIs available in the Ecoinvent databases. Phosphate rock processing is considered to be conducted in a manner similar to that done in the United States. These processes take into account the following activities: mining process, transport to the beneficiation plant, and wet processing (including screening, washing, and flotation). It also considers land use for mining and reclamation. It does not take into account drying or calcination, and it does consider energy consumption data that are related to the mass of rock moved. Regarding H2SiF6 byproduct recovery, its impact calculation has been performed using an Ecoinvent LCI, which provided the environmental gain that is achieved when the product is recovered instead of being produced from virgin material. The data for the production of fluosilicic acid were taken from the literature.35 In this case, it is produced from apatite rock treated with H2SO4. The most important issue in the LCI data is the estimation of water and air fluoride emissions, which are both rigorously calculated using the previously described Aspen Plus simulation model. 3.3.1. Deterministic Approach. As a first step, the environmental impact calculations were performed under the assumption that no dispersion in the input LCI data exists and using the mean value of the fitted distribution from the Aspen Plus simulation results. Data consistency was checked through comparison with built-in process units that are present in the Ecoinvent datasbase. Thus, PA US/U and PA MA/U process units are retrieved from the Ecoinvent database and are taken as reference for comparison purposes. The term “PA US/U” represents data of the production of PA in the United States, while the term “PA MA/U” considers the production of PA in Morocco. The main difference between both processes lies in the way phosphate rock is processed. The goal of this consistency step is to check whether a similar environmental profile is found in all the cases. Specifically, this environmental impact profile is characterized by large impacts in marine aquatic

uncertainty in simulation LCI? uncertainty in database LCI? total number of variables number of variables with uncertain data number of variables without uncertain data percentage of uncertain variables

version 1

version 2

version 3

version 4

no

yes

no

yes

no

no

yes

yes

4065

4065

52589

52589

0

29

38597

38626

4065

4036

13992

13963

0.00%

0.71%

73.39%

73.45%

ecotoxicity potential (MAEP), AP, EP, and ADP categories. In our case, we observed impacts that were lower than those reported in the database. This can be due to the differences in the boundary sets for each of the systems (recall transport and energy integration), as shown in Figure 3. The characterization results are normalized according to the work of Huijbregts et al.,63 which makes use of the cumulative environmental impacts that take into account all of western Europe. The use of such a normalization scheme allows for comparison of emissions into the context of Europe. For comparison purposes between alternatives, the results of the different options were normalized by taking the maximum value for each environmental category as a reference. By following this procedure, the impact values fall in the interval [0, 1]. These normalized impact characterization results are summarized in Table 4 and are shown in Figure 4. In Table 4, the position called “first” refers to the best (lesspolluting or less-resource-depleting) option, while ther term “third” refers to the least environmentally friendly (morepolluting or more-resource-depleting) option. The options that are in the first place are neutralization with HF recovery and ocean dump (i.e., options 3 and 1, respectively). Neutralization (option 2) always occupies the second- or third-best positions. Table 5 shows the normalized results (in units of yr-1)63 associated with the three waste treatment options. It can be observed how the highest impact corresponds to the MAEP damage category. In that table, the first five most important environmental interventions considering its normalized contribution are: MAEP, AP, EP, ADP and GWP. Further analysis of the results shows that, in options 2 and 1, MAEP is mainly due to the PA production process itself (in the range of 66%-74% of the total impact, depending on the waste treatment option). In the case of neutralization and HF recovery (option 3), one observes how the recovery of HF, which is coupled to PA production itself, leads to a reduction in MAEP. The second-largest process contributing to MAEP is

8296 Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008

Figure 5. Comparison of confidence intervals for different sources of uncertainty, for the same treatment option (option 1). Table 7. Stochastic Environmental Impact Assessment Results (Normalized) Normalized Results [yr-1] Option 1

Option 2

Option 3

impact category

mean

standard deviation

coefficient of variation (%)

mean

standard deviation

coefficient of variation (%)

mean

standard deviation

coefficient of variation (%)

ADP AP EP FWEP GWP HTP MAEP OLDP POP TEP

2.78E-13 1.12E-12 3.28E-13 1.60E-13 1.22E-13 5.46E-14 8.00E-12 6.89E-16 1.46E-13 7.47E-14

6.64E-14 1.32E-12 1.03E-14 1.29E-13 2.27E-14 3.87E-14 9.54E-13 1.95E-16 1.74E-13 2.21E-14

23.9 117.9 3.1 80.5 18.7 70.9 11.9 28.3 118.9 29.6

3.47E-13 1.12E-12 7.93E-14 1.84E-13 2.13E-13 5.19E-14 6.35E-12 1.05E-15 1.54E-13 6.07E-14

6.21E-14 1.20E-12 9.17E-15 1.14E-13 2.20E-14 3.71E-14 8.34E-13 2.93E-16 1.59E-13 2.05E-14

17.9 107.0 11.6 62.0 10.3 71.5 13.1 27.9 103.1 33.8

2.78E-13 8.63E-13 7.37E-14 1.58E-13 1.78E-13 3.81E-14 2.29E-12 8.71E-16 1.19E-13 5.21E-14

5.21E-14 6.92E-13 8.39E-15 8.81E-14 1.79E-14 2.19E-14 2.05E-12 2.47E-16 9.15E-14 1.53E-14

18.7 80.1 11.4 55.7 10.1 57.3 89.5 28.4 76.7 29.4

determined to be the burning of lignite and coal, which are both used as raw materials for electricity generation. In all three options, the substance flow with the highest contribution to MAEP is HF that is released into the air, followed by trace species flows to water (Be, V, and others). Regarding AP, all options exhibit the same tendencies. AP is mainly due to the sulfuric acid and sulfur production processes and, consequently, to the emission of SO2 to air. All options show a similar mean consumption of H2SO4 (see Table 3). EP in all process options is mainly attributed to the PA production and the phosphoric rock enrichment process. These processes contribute to the emission of phosphates and phosphorus to water. These results are consistent with those shown by da Silva et al.43 ADP for all three options is caused by the consumption of natural resources (such as oil and coal) as well and, consequently, to the processes that use them. The consumption of phosphate rock is among the least significant contributions (specifically, 1%-2% of the total).

3.3.2. Stochastic Approach. We focus our attention on two sources of uncertainty: (i) the uncertainty associated with parameters of the simulation model, and (ii) variability of the LCIs results given by the Ecoinvent database. To compare these uncertainty sources, three MC sampling runs over different versions of the same waste treatment option were conducted. In each one of the MC runs, certain sets of variables were fixed to its mean value while the others were regarded as being stochastic. Table 6 summarizes this information. Note that, when using no uncertain information from the database, the number of variables is drastically reduced; this is an inherent issue to how the Ecoinvent database information is structured. The results of the MC simulation runs can be seen in Figure 5. Figure 5 clearly shows that most of the variability in the results is a consequence of uncertain information that comes from LCI data stored in the Ecoinvent database. Version 2 shows the smallest confidence intervals, given that it considers only simulation model uncertainty (i.e., the foreground system variables), whereas versions 3 and 4 of the same waste treatment

Ind. Eng. Chem. Res., Vol. 47, No. 21, 2008 8297

Figure 6. Comparison of normalized environmental impacts resulting from stochastic simulation for all options. Table 8. Comparison of Stochastic and Deterministic Ranking of Options Option 1 impact category

Det value

Det pos.

ADP AP EP FWEP GWP HTP MAEP OLDP POP TEP

2,38E-13 1,11E-12 3,23E-13 5,37E-14 1,05E-13 2,16E-14 7,04E-12 6,29E-16 1,45E-13 4,84E-14

first first third second second first first third third third

Option 2

MC Mean

MC position

Det value

Det pos.

2,78E-13 1,12E-12 3,28E-13 1,60E-13 1,22E-13 5,46E-14 8,00E-12 6,89E-16 1,46E-13 7,47E-14

first third third second first third third first second third

3,04E-13 1,12E-12 7,46E-14 7,39E-14 1,96E-13 1,83E-14 5,49E-12 9,88E-16 1,53E-13 2,87E-14

third third second third third third third second second second

option show the largest confidence intervals (in this case, ocean disposal (option 1) is selected). The probability level of the confidence interval (CI) to calculate the error bars was fixed at 95%. An uncertainty analysis of the impact assessment for each of the waste treatment options was performed next. All the variables sets (simulation and database) were considered to be stochastic. A MC simulation was performed in SimaPro, using 1000 equally probable scenarios. The number of scenarios was set by gradually increasing it and stopping whenever no significant changes in the environmental interventions can be appreciated (i.e., the value of the impact was