Designing, Retrofitting, and Revamping Water Networks in

Designing, Retrofitting, and Revamping Water Networks in Petroleum Refineries Using Multiobjective Optimization. Shivom Sharma and ..... MS Excel work...
10 downloads 11 Views 874KB Size
Article pubs.acs.org/IECR

Designing, Retrofitting, and Revamping Water Networks in Petroleum Refineries Using Multiobjective Optimization Shivom Sharma and G. P. Rangaiah* Department of Chemical & Biomolecular Engineering, National University of Singapore, Engineering Drive 4, Singapore 117585 S Supporting Information *

ABSTRACT: In the chemical industry, water integration is beneficial from the point of both economic and environmental impacts. An optimal water network reduces fresh water consumption by efficient reuse and recycling of water within the plant. In general, water networks are designed based on a single performance criterion such as fresh water consumption or total annual cost. In this study, the performance of Premium Solver in Excel is compared with BARON (Branch-And-Reduce Optimization Navigator) in GAMS (General Algebraic Modeling System) for optimizing three industrial water networks for a single objective. Recently, water networks have been studied using multiobjective optimization (MOO) for two or more objectives simultaneously. Hence, two existing refinery water networks were retrofitted and revamped using the MOO approach, for two important scenarios: the addition of two new water-using processes in the refinery, and use of contaminated water (steam condensate) along with fresh water in water networks. In the water network retrofit, only the capacities of the regeneration units can change, and there is no change in the existing network topology. Both existing water network topology and capacities of different regeneration units can change in the water network revamping. In this, MOO problems are successfully solved using the ε-constraint method along with BARON. The obtained Pareto-optimal results are presented and discussed; they give greater insight and many optimal solutions for selection.

1. INTRODUCTION Water is used as a reactant, separation solvent, and utility in chemical process industries. Most of the existing plants use fresh water, and wastewater is treated, before discharge, to meet the prevailing environmental regulations. Although some process units in a plant require fresh water, slightly contaminated water can also be used in many other process units. Hence, wastewater from a process unit can be treated using regeneration units, and then reused in the same or other process units in the plant. Minimization of fresh water consumption is necessary to preserve this natural resource, and treated water is likely to be cheaper compared to fresh water. Water flow rate through a process depends on the amount of contaminants to be removed from that process. Further, each process unit requires water with different contaminant concentration and temperature. Fresh water is mixed with wastewater (from other processes) and treated water (from regeneration units) to fulfill water flow rate and contaminant concentration requirements of each process unit. Then, the mixed water stream to a process may exchange heat with hot/cold utility or other process streams, to achieve the target temperature. Thus, an optimal water network efficiently reuses and recycles water in a plant. In the literature, both mathematical and graphical approaches have been developed to design water networks in chemical process industries (El-Halwagi1). A comprehensive review of water network design methods can be found in Jezowski,2 and a recent review of studies on water and energy integration is available in Ahmetovic et al.3 Table 1 summarizes the optimization studies on water network design and retrofit. Most of them considered grassroots design of a water network, and only four studies have considered retrofitting of a water network. Faria and Bagajewicz4,5 have studied both © 2015 American Chemical Society

design and retrofit of industrial water networks. Sotelo-Pichardo et al.6 retrofitted a water network by minimizing the total annual cost of the retrofitted process. Recently, Sharma and Rangaiah7 studied the design, retrofit, and revamp of industrial water networks. Kirthick et al.8 combined graphical and mathematical approaches to design water networks, with the graphical approach providing an initial guess for the mathematical approach. Castro and Teles9 compared several global optimization techniques for designing water networks, and found their multiparametric disaggregation with parametrization better than other solution approaches tested. Table 1 shows that many studies on water network design and retrofit have mainly used solvers in GAMS (General Algebraic Modeling System) for solving the resulting optimization problems. Engineers and practitioners are familiar with MS Excel and use it often, but they may not have access to or used GAMS. Hence, they are more likely to prefer using MS Excel for solving water network problems. In this study, performance of Premium Solver, a more powerful version of Solver tool, available in MS Excel, is compared with BARON (Branch-And-Reduce Optimization Navigator) in GAMS on three industrial water networks of medium size. As can be seen in Table 1, water networks are mainly designed and/or retrofitted for a single performance criterion, such as fresh water consumption and total water network cost, at a time. However, in recent years, water networks are being designed and retrofitted using multiobjective optimization (MOO) for two or Received: Revised: Accepted: Published: 226

August 3, 2015 November 4, 2015 December 9, 2015 December 9, 2015 DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research Table 1. Recent Optimization Studies on Water Network Design/Retrofit ref Poplewski and Jezowski10 Faria and Bagajewicz4

examples

objective function(s)

adaptive random search DICOPT/GAMS (CONOPT/CPLEX)

total annual cost

not stated in the cited reference linear relaxation and bound contraction

DICOPT/GAMS piecewise McCormick and multiparametric disaggregation with CPLEX/CONOPT, BARON and GloMIQO (GAMS) LINGO

Faria and Bagajewicz11

3 examples, design 2 examples, design and retrofit 2 examples, design and retrofit 13 examples, design

Aviso et al.12 Kirthick et al.8

2 examples, design 1 examples, design

Ahmetovic and Grossmann13 Sotelo-Pichardo et al.6 Castro and Teles9

6 examples, design

fresh water consumption, regeneration flow rate and total annual cost fresh water consumption and total system cost fresh water consumption and number of exchangers water network cost and fresh water consumption

4 examples, retrofit 19 examples, design

total annual cost fresh water consumption

Deng et al.14

2 examples, design

Yang et al.15

3 examples, design

Boix et al.16

1 example, design

freshwater flow rate, number of connections, intercepted flow rate and mass load total water network cost Multiobjective Optimization fresh water consumption, total flow rate through RU and number of interconnections fresh water consumption and energy consumption freshwater flow rate, number of interconnections and total energy consumption fresh water consumption and total flow rate through regeneration units fresh water consumption and total flow rate through regeneration units

Faria and Bagajewicz5

Boix et al.

17

2 examples, design

Ramos et al.18

2 examples, design

Sharma and Rangaiah19

1 example, design

Sharma and Rangaiah7

1 example, design, retrofit and revamp

a

platform/solver used

Single Objective Optimizationa total annual cost net present value and return on investment

LINGO DICOPT/GAMS BARON/GAMS

DICOPT, LINDOGlobal and BARON (GAMS) COIN-BONMIN (GAMS) COIN-BONMIN (GAMS) DICOPT/GAMS (IPOPT/CPLEX) BARON/GAMS BARON/GAMS

Note that some studies on single objective optimization shown in this table have used different objectives in separate optimization problems.

Figure 1. Water (allocation) network in a chemical plant; here, subscript i′ signifies processes 1, 2, ..., NP except process i, and subscript j′ signifies regeneration units 1, 2, ..., NR except regeneration unit j.

more objectives simultaneously. Boix et al.16 optimized the water network for fresh water consumption and water flow rate at the inlet of regeneration units, simultaneously. Subsequently, Boix et al.17 simultaneously minimized the water and energy/ utility consumptions in a water network. Ramos et al.18 optimized industrial water networks for three objectives: total fresh water flow rate, number of interconnections and total

energy consumption, using goal programming. Petroleum refineries require a significant amount of water, and their water networks require retrofitting due to changes in crude quality, environmental regulations, and/or process configuration. Recently, Sharma and Rangaiah7 applied MOO to the design, retrofit, and revamp of a refinery water network (involving 11 processes, 3 regeneration units, and 4 contaminants) for the 227

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research NP

revised environmental regulations and increase in the hydrocarbon load for different process units. In the present study, two refinery water networks from the literature are analyzed for retrofitting and revamping using MOO. Both refinery water networks are retrofitted and revamped for two important scenarios not studied earlier: (1) addition of two new water using processes in the refinery, and (2) use of contaminated water (e.g., steam condensate) along with fresh water in water networks. These water networks are optimized for multiple objectives, using the ε-constraint method along with BARON. In the water network retrofit, capacities of the regeneration units and flow rates through existing interconnections can change, and there is no change in the existing water network topology. Conversely, both existing water network topology and capacities of different regeneration units can change in the water network revamping. The following section describes the mathematical model for the water network. Section 3 compares the performance of Premium Solver with that of BARON for design of three industrial water networks of medium size for a single objective. After that, section 4 presents a multiobjective problem formulation for a water network design, and also provides details on two refinery water network problems of large size. In section 5, both these refinery water networks are designed, and then retrofitted and revamped for two different scenarios, using the ε-constraint method and BARON. Finally, conclusions of this study are summarized in the last section.

pp = 1(pp ≠ i) NR

+

NP

Finlet, i = Foutlet, i

(3)

k k k Finlet, i·Z inlet, i + Li = Foutlet, i · Zoutlet, i

(4)

(iii) Splitters for processes (for i = 1, 2,..., NP): water balance only; no change in concentrations NP

NP

pp = 1(pp ≠ i)

pp = 1(pp ≠ i)

pr ∑ Foutlet, i + WWi

(5)

pr = 1

(iv) Mixers for regeneration units (for j = 1, 2,..., NR): water and contaminant balances (for k = 1, 2, ..., K) NP

NR

pr ∑ Foutlet, ∑ i +

Finlet, j =

rr Foutlet, j

(6)

rr = 1(rr ≠ j)

pr = 1 NP

pr k ∑ (Foutlet, i · Zoutlet, i)

k Finlet, j·Z inlet, j =

pr = 1 NR

+

k rr (Foutlet, j · Zoutlet, j)



(7)

rr = 1(rr ≠ j)

(v) Regeneration units (for j = 1, 2,..., NR): water and contaminant balances (for k = 1, 2, ..., K) Finlet, j = Foutlet, j

(8)

k k k Finlet, j·Z inlet, j(1 − ηj ) = Foutlet, j · Zoutlet, j

(9)

(vi) Splitters for regeneration units (for j = 1, 2,..., NR): water balance only; no change in concentrations NR

Foutlet, j =

NR

rp ∑ Foutlet, ∑ j +

rr Foutlet, j + WWj

(10)

rr = 1(rr ≠ j)

rp = 1

(vi) Mixers for wastewater: water and contaminant balances (for k = 1, 2, ..., K) NP

FWW =

NR

∑ WWi + ∑ WWj i=1

(11)

j=1 NP

k FWW ·Z WW =

NR

k k ∑ (WWi ·Zoutlet, i) + ∑ (WWj · Zoutlet, j) i=1

j=1

(12)

Here, Finlet,i/Foutlet,i is water flow rate through the ith process, whereas Finlet,j/Foutlet,j is water flow rate through jth regeneration pr unit. Fpp outlet,i and Foutlet,i are, respectively, water recycle flow rates from ith process to other processes and regeneration units, whereas Frroutlet,j and Frp outlet,j are respectively water flow rates from jth regeneration unit to other regeneration units and processes. Zkinlet,i and Zkoutlet,i are respectively kth contaminant concentration at the inlet and outlet of ith process, whereas Zkinlet,j and Zkoutlet,j are respectively kth contaminant concentration at the inlet and

rp ∑ Foutlet, j + FWi rp = 1

pp Foutlet, i +



Foutlet, i =

NR pp Foutlet, i +

(2)

(ii) Processes (for i = 1, 2,..., NP): water and contaminant balances (for k = 1, 2, ..., K)

Figure 1 shows a water (allocation) network in a chemical plant. In this network, water using processes and regeneration units are interconnected using mixers and splitters. Fresh water is free of contaminant, and can only be used in water using processes. Waste water from a process can be used in other processes, treated in regeneration units, and/or discharged as wastewater. Treated water from a regeneration unit can be used in any process, treated in other regeneration units, and/or discharged as wastewater. It is assumed that there are NP processes, NR regeneration units, and K contaminants; consequently, there will be NP mixers/splitters for processes, and NR mixers/ splitters for regeneration units. Note that the schematic in Figure 1 indicates a superstructure of a water network, with all possible branches/pipelines between processes and/or regeneration units, but no recycling of water within the process or regeneration unit is allowed. For example, mixer 1 (for process 1) can use wastewater from splitter 2 of process 2, splitter 3 of process 3, ... splitter NP of process NP, splitter 1 of regeneration unit 1, splitter 2 of regeneration unit 2, ... and splitter NR of regeneration unit NR. The water network model consists of mass balances around mixers, processes, regeneration units, and splitters. These model equations for the water network in Figure 1 are as follows: (i) Mixers for processes (for i = 1, 2, ..., NP): water and contaminant balances (for k = 1, 2, ..., K)



rp k ∑ (Foutlet, j · Zoutlet, j) rp = 1

2. WATER NETWORK MODEL

Finlet, i =

pp k (Foutlet, i · Zoutlet, i)



k Finlet, i·Zinlet, i =

(1) 228

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research outlet of jth regeneration unit. Lki and ηkj are load and removal of kth contaminant in ith process and jth regeneration unit, respectively.

Table 3. Problem Characteristics and Optimal Results Obtained for the Three Industrial Water Networks, Using BARON (in GAMS) and Premium Solver (in Excel); DVDecision Variables

3. PERFORMANCE COMPARISON OF BARON AND PREMIUM SOLVER To compare the performance of BARON and Premium Solver, three medium size industrial water networks (IWN) are chosen and optimized. Tables S1 to S3 in the Supporting Information present important data for these three IWN, taken from Faria and Bagajewicz;11 these include limiting data for water using processes and operating data for regeneration units. Here, water flow rate through each process is fixed. Table 2 presents the single objective optimization (SOO) problem formulation for minimization of total flow rate through

IWN IWN1 IWN2 IWN3 IWN1 IWN2 IWN3

objective function Min. total flow rate through regeneration F2 = ∑NR j = 1Finlet,j units decision variables and constraintsBARON FWi pr rr rp Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j Zkinlet,i, Zkoutlet,i, Zkinlet,j, Zkoutlet,j FWi ≥ 0 pr rr rp Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j ≥ 0 Zkinlet,i, Zkinlet,j ≥ 0, Zkinlet,i ≤ Zk,max inlet,i ∑NP i = 1FWi ≤ 40 Finlet,i or Foutlet,i = Fi WWi ≥ 0, WWj ≥ 0, ZkWW ≤ 10 ppm

total fresh water consumption water flow rate through each process waste water from each process and regeneration unit and contaminant concentration in wastewater water and contaminant balances eqs 1−12 decision variables and constraintsPremium Solver water flow rate through different interconnections contaminant concentration at the inlet of processes and regeneration units bounds on the decision variables

Zkinlet,i, Zkinlet,j

fresh water consumption in each process and total fresh water consumption water flow rate through each process waste water from each process and regeneration unit and contaminant concentration in wastewater water and contaminant balances

Finlet,i or Foutlet,i = Fi WWi ≥ 0, WWj ≥ 0, ZkWW ≤ 10 ppm

no. of DV 89 89 150 FW (t/h) 40 40 40

no. of constraints 62 62 98 FRU (t/h) 57.85 134.23 222.21

Premium Solver no. of DV 42 42 80 FW (t/h) 40 40 40

no. of constraints 33 36 60 FRU (t/h) 57.68 140.91 160.26

method (along with default search termination criteria using precision = 10−6 and convergence = 10−4) is used for solving IWN problems. MS Excel worksheet cells are used for decision variables, to calculate objective function and constraints, and to satisfy explicitly water and contaminant balances. Here, for pr rr rp optimization by Premium Solver, Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j, k k Zinlet,i and Zinlet,j are the decision variables. Fresh water consumption in each process (FWi), water flow rate through each regeneration unit (Finlet,j/Foutlet,j) and wastewater from all processes (WWi) and regeneration units (WWj) are calculated, using decision variables and model eqs 1, 3, 5, 6, 8, 10 and 11. For each process, outlet contaminant concentration (Zkoutlet,i) is then calculated using flow rate (Finlet,i), inlet contaminant concentration (Zkinlet,i) and contaminant load (Lki ) (via eq 4). Similarly, for each regeneration unit, outlet contaminant concentration (Zkoutlet,j) is calculated using inlet contaminant concentration (Zkinlet,j) and removal ratio (ηkj ) (via eq 9). After that, inlet k contaminant concentration for each process (Zinlet,i ′ ) is recalculated using fresh water consumption (FWi), wastewater k from other processes (Fpp outlet,i, Zoutlet,i) and treated water from rp k regeneration units (Foutlet,j, Zoutlet,j). Also, inlet contaminant concentration for each regeneration unit (Zkinlet,j ′ ) is recalculated k using wastewater from processes (Fpr outlet,i, Zoutlet,i) and treated water from other regeneration units (Frroutlet,j, Zkoutlet,j). To satisfy contaminant balances for mixers of both processes and regeneration units (eqs 2 and 7), Zkinlet,i − Zkinlet,i ′ and Zkinlet,j − k Zinlet,j ′ should be very small (0.001 to 0.005 compared to contaminant concentration of 5 to 50 ppm); these requirements are included as additional inequality constraints (not shown in Table 2) in the water network optimization problem for solution by the Premium Solver. The SOO problem for water network is solved using BARON (V23.6) in GAMS, which is allowed to run for 1 h. All the water network model equations are written in GAMS, and then the objective function and constraints are defined. BARON can handle exponential, logarithmic, and power functions in the optimization problem formulation. The water allocation network problem has bilinear terms, which arise from the multiplication of flow rate and contaminant concentration. To achieve faster convergence, bounds for all variables in the water network model are provided. BARON terminates after 10 and 14 min respectively for IWN1 and IWN2, and shows normal completion message (i.e., termination based on the default criterion). In the case of IWN3, it terminates based on maximum allowable CPU time (i.e., 1 h). On the other hand, Premium Solver takes less than 5 min to solve each of these three water network problems.

Table 2. SOO Problem Formulation for Industrial Water Network

fresh water consumption in ith process water flow rate through different interconnections contaminant concentration at the inlet and outlet of processes and regeneration units bounds on decision variables

BARON/GAMS

pr rr rp Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j

pr rr rp Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j ≥ 0 Zkinlet,i, Zkinlet,j ≥ 0, Zkinlet,i ≤ Zk,max inlet,i FWi ≥ 0, ∑NP i = 1FWi ≤ 40

explicitly satisfied (discussed in text)

regeneration units using BARON and Premium Solver. Water and contaminant balances (eqs 1 to 12) are explicitly satisfied/ solved in the problem formulation for Premium Solver because they are bilinear equations and can be easily solved for some decision variables; details are given in the next paragraph. This formulation has fewer decision variables compared to that for BARON (Table 3); this reduction in number of decision variables is likely to result in easier and faster solution of the optimization problem. However, water and contaminant balances could not be explicitly satisfied/solved for solution by BARON; hence, they are included in constraints. Premium Solver (V12.5.3.0) has five optimization methods. In the present study, large size generalized reduced gradient 229

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research Table 3 compares the optimal results obtained using Premium Solver and BARON on three IWN problems. It can be seen that both the solution methods gave comparable optimal results for the IWN1 problem, whereas BARON is slightly better compared to Premium Solver for IWN2 problem. Premium Solver gave a significantly better optimum for the IWN3 problem compared to BARON. To improve the solution, BARON was allowed to run for 2 h but the solution obtained was the same as that in Table 3. These results indicate that Premium Solver in MS Excel is better for solving water network problems of medium size. However, our tests showed that Premium Solver is not able to solve larger water network problems, and so only BARON is used for MOO study on two large refinery water network problems. Table A1 in the Appendix provides water network branches and associated flow rates in the optimal solutions obtained for IWN problems using BARON and Premium Solver. These results show that solutions obtained using BARON have fewer interconnections compared to those obtained using Premium Solver, and that a number of interconnections have very small flow rates. For example, solutions of IWN3 using BARON and Premium Solver have 3 and 11 interconnections respectively, with flow rates less than 0.1 t/h. Thus, the optimal solutions obtained by BARON and Premium Solver are different even though the solutions by BARON and Premium Solver for IWN1 have practically the same objective function values (Table 3).

Table 4. MOO Problem for Refinery Water Networks objective functions min fresh water consumption F1 = ∑NP i = 1FWi min total flow rate through regeneration F2 = ∑NR j = 1FWinlet,j units decision variables and their bounds fresh water consumption in ith process water flow rate through different interconnections contaminant concentration at the inlet and outlet of both processes and regeneration units bounds on the decision variables

FWi pr rr rp Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j

waste water from each process and regeneration unit, contaminant concentration in wastewater water and contaminant balances

WWi ≥ 0, WWj ≥ 0, ZkWW ≤ 400 ppm

Zkinlet,i, Zkoutlet,i, Zkinlet,j, Zkoutlet,j

FWi ≥ 0 pr rr rp min Fpp outlet,i, Foutlet,i, Foutlet,j, Foutlet,j ≥ FIC pr rr rp max Fpp , F , F , F ≤ F outlet,i outlet,i outlet,j outlet,j IC Zkinlet,i, Zkoutlet,i, Zkinlet,j, Zkoutlet,j ≥ 0 k k,max Zkinlet,i ≤ Zk,max inlet,i , Zoutlet,i ≤ Zoutlet,i constraints

eqs 1−12

find the Pareto-optimal front for many Chemical Engineering applications (Masuduzzaman and Rangaiah;21 Sharma and Rangaiah;22 Rangaiah et al.23). In the present study, the converted SOO problem is solved repeatedly for minimization of fresh water consumption, each time with a different limit on F2. The SOO problem is a mixed integer nonlinear programming (MINLP) problem, and it is solved using BARON (V23.6) in GAMS, which is allowed to run for 1 h to obtain one Paretooptimal solution at a time. In this study, an interface between GAMS and Excel has been developed and used for water network design, retrofit, and revamp optimization. In general, engineers and practitioners are familiar with Excel for data handling and processing. Hence, the water network model is implemented in GAMS, whereas Excel is used to input water network data and also to display the optimal solution of the water network problem. Figure 2 shows features

4. MOO PROBLEM FORMULATION FOR REFINERY WATER NETWORKS (RWN) In a water allocation network, fresh water consumption conflicts with the total flow rate through regeneration units, and these two objectives affect operating and investment costs of a water network. Further, minimization of fresh water consumption is necessary to preserve this natural resource. Hence, each refinery water network (RWN) is optimized for minimizing both fresh water consumption and total flow rate through regenerator units, simultaneously. Data for two RWN are presented in Tables S4 and S5 in the Supporting Information. Maximum concentrations of contaminants at the process outlet are fixed for these networks. RWN1 and RWN2 have 6 and 11 water using processes, respectively. Further, three regenerations units are available for the treatment of wastewater in both RWN. They are RU1, a reverse osmosis unit that reduces salts to 20 ppm; RU2, an API separator followed by ammoniacal copper arsenate (ACA) that reduces organics to 50 ppm; and RU3, a Chevron wastewater treatment that reduces H2S and NH3 to 5 and 30 ppm, respectively (Faria and Bagajewicz11). Table 4 presents the MOO problem for RWN. In this, fresh water consumption in the ith process, water flow rate through different interconnections between processes and/or regeneration units, and contaminant concentrations at the inlet and outlet of both processes and regeneration units are the decision variables. Through each of the interconnections between processes and/or regeneration units, minimum (Fmin IC ) and maximum (Fmax IC ) flow rates of 1 and 200 t/h are specified for RWN 1, and they are 1 and 100 t/h for RWN 2. Contaminant concentration at the inlet and outlet of each process should k,max be below permissible limits (i.e., Zk,max inlet,i and Zoutlet,i). Further, contaminant concentration in the wastewater stream leaving the water network should be below the discharge limit (i.e., Zk,max WW ). The biobjective optimization problem can be transformed into a SOO problem by the ε-constraint method via making F2 as an additional constraint (Deb20). This method has been used to

Figure 2. Interface between GAMS and Excel for solving water network problem.

of the interface between GAMS and Excel. Here, both files are placed in the same directory, and then the user needs to run GAMS. This interface is useful for solving the water network problem repeatedly for different ε values, which can be provided 230

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research

between 9 and 16, whereas it varies significantly (between 6 to 28) in alternate designs for RWN2. Some interconnections in each water network design have a flow rate of 1.0 t/h (Table A2), due to the limit on the minimum flow rate through interconnections. In Figure 3 for RWN1, solution “J” has the lowest fresh water consumption; compared to this, solution “I” has fewer interconnections and significantly lower total flow through regeneration units. In Figure 4, solution “R” for RWN2 has the lowest fresh water consumption, and solution “P” has about 50% fewer interconnections and 60% reduction in total flow rate through regeneration units. For RWN1 and RWN2, solutions “I” and “P” in Figures 3 and 4 have relatively lower values of both objectives (i.e., fresh water consumption and total flow rate through regeneration units), and so they are assumed to be selected and implemented. Therefore, solutions “I” and “P” are employed in the following retrofit/revamp studies on RWN1 and RWN2, respectively. In water network retrofitting, there is no change in the network topology, and flow rate through existing interconnections and capacities of regeneration units can change. Conversely, flow rate through existing interconnections, network topology, and capacities of regeneration units can change in water network revamping. Hence, water network revamping could be complex and costly, but it may give more operational benefits compared to water network retrofitting. In the following subsections, water networks corresponding to solutions “I” and “P” (Figures 3 and 4) for RWN1 and RWN2 are retrofitted via MOO for two scenarios: (A) the addition of two new processes (requiring water) to each water network, and (B) use of contaminated water (e.g., steam condensate) along with fresh water. Limiting data for the two new processes and quality of steam condensate are provided in Table S6 in the Supporting Information. The number of continuous variables, integer variables, and constraints for design, retrofitting and revamping of two RWN problems are summarized in Table 5. As expected, retrofit problems are smaller

in the Excel worksheet. Interested readers can obtain the interface files from the corresponding author of this paper.

5. RESULTS AND DISCUSSION 5.1. Design of Refinery Water Networks. In this study, 20 different ε values are used for design, retrofitting, and revamping of RWN. For each ε value, we found that BARON gives a nondominated solution, dominated solution, or no solution at all, probably because of difficulties in finding the global optimum of large, nonlinear, and constrained optimization problems. Here, only nondominated solutions for water network design, retrofit, and revamp are presented and discussed. Figures 3 and 4 show the Pareto-optimal fronts obtained for

Figure 3. Pareto-optimal results for simultaneous minimization of fresh water consumption and total flow rate through regeneration units in RWN1; the two values shown for each solution are, respectively, total flow rate through generation units and fresh water consumption.

Table 5. Number of Continuous Variables (CV), Integer Variables (IV) and Constraints for Design/Retrofit/Revamp of RWN1 and RWN2 RWN1

Figure 4. Pareto-optimal results for simultaneous minimization of fresh water consumption and total flow rate through regeneration units in RWN2; the two values shown for each solution are, respectively, total flow rate through generation units and fresh water consumption.

design retrofit scenario A revamp scenario A retrofit scenario B revamp scenario B

simultaneous minimization of both fresh water consumption and total flow rate through regeneration units, for RWN1 and RWN2, respectively. In these figures, each nondominated solution has a different water network topology, fresh water consumption, and total flow rate through regeneration units. For RWN1, Faria and Bagajewicz11 have reported a minimum fresh water consumption of 33.571 t/h (and corresponding water flow rate through regeneration units = 550.06 t/h). Pareto-optimal solution “J” in Figure 3 has fresh water consumption of 33.571 t/h and total flow through regeneration units of 230 t/h. Hence, solution “J” obtained here is significantly better than the optimal solution in Faria and Bagajewicz.11 Table A2 in the Appendix presents the interconnections (with flow rates) between processes and/or regeneration units for the nondominated solutions in Figures 3 and 4. The number of interconnections in alternate designs for RWN1 varies

RWN2

CV

IV

constraints

CV

IV

constraints

211 279 279 220 220

72 52 110 13 72

291 397 397 291 291

396 484 484 410 410

182 73 240 15 182

586 732 732 586 586

than the design problems, whereas revamp problems are comparable with the design problems. 5.2. Retrofitting and Revamping of RWN1 for Scenario A. Figures 5 and 6 show the Pareto-optimal fronts obtained for RWN1 retrofitting and revamping for two new water using processes (with limiting data in Table S6 in the Supporting Information), respectively. Here, most of the nondominated solutions have significantly large fresh water consumption, and so only the nondominated solutions with fresh water consumption below 50 t/h are considered as potential for retrofitting and revamping of water network. Table A3 in Appendix presents the interconnections (with flow rates) between processes and/or regeneration units for selected retrofit and revamp solutions in Figures 5 and 6. Two retrofit solutions (RET-AP 1 and RET-AP 2) for scenario A have 231

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research

Figure 7. Pareto-optimal solutions for retrofit of existing water network for RWN1 (solution I): use of contaminated water (e.g., steam condensate) along with fresh water.

Figure 5. Pareto-optimal solutions for retrofit of existing water network for RWN1 (solution I): addition of two new processes requiring water.

Figure 6. Pareto-optimal solutions for revamp of existing water network for RWN1 (solution I): addition of two new processes requiring water.

Figure 8. Pareto-optimal solutions for revamp of existing water network for RWN1 (solution I): use of contaminated water (e.g., steam condensate) along with fresh water.

respectively 12 and 13 interconnections between processes and/or regeneration units, which are comparable/equal to 13 interconnections in the existing water network (solution “I”). For RET-AP 1 solution, two new interconnections (p7-p5 and p8-p4) are required (along with 10 interconnections in the existing water network) due to addition of new processes p7 and p8. In RET-AP 2 solution, 12 interconnections persist from the existing water network “I”, one new interconnection between process 8 and process 4 is required, and water leaving process 7 directly mixes with the wastewater from other processes and regeneration units. Many new interconnections are required in revamped solutions: REV-AP 1 and REV-AP 2. As expected, REV-AP 2 solution for revamping requires lower amount of fresh water compared to those for retrofitting, but it has significantly larger total flow rate through regeneration units (see Figures 5 and 6). 5.3. Retrofitting and Revamping of RWN1 for Scenario B. The existing RWN1 (solution I) is retrofitted and revamped via MOO for the use of contaminated water (steam condensate with quality data in Table S6 in the Supporting Information) along with fresh water. The Pareto-optimal fronts obtained for both retrofitting and revamping are presented in Figures 7 and 8, respectively. For easier comparison, existing RWN1 is also included in these figures. Table A4 in the Appendix presents the information on interconnections (with flow rates) between processes and/or regeneration units for the selected retrofit and revamp water networks (i.e., fresh water consumption < 50 t/h). In the case of water network retrofitting for the use of steam condensate, no new interconnection is required (because there are no new processes consuming water), but interconnection from process 3 to regenerator 2 is no longer needed due to low flow rate through it (Table A4). RET-SC 1 and RET-SC 2 do not use any steam condensate. Note that steam condensate

has a significant amount of contaminants (Table S6 in the Supporting Information), and so retrofit and revamp water networks may not use it. In general, the number of interconnections in revamped water networks (Table A4) is comparable with that in the existing solution “I”. Rev-SC 1 has four new interconnections and it uses 0.063 t/h of steam condensate in process 1. There is no new interconnection for solution REV-SC 2, and it does not require steam condensate. 5.4. Retrofitting and Revamping of RWN2 for Scenario A. In this study, retrofitting and revamping of the existing RWN2 (solution P) is also studied for the addition of two water using processes (Scenario A) and use of steam condensate in the water network (Scenario B). The limiting data for the two new processes and quality of steam condensate are presented in Table S6 in the Supporting Information. Figures 9 and 10 show

Figure 9. Pareto-optimal solutions for retrofit of existing water network for RWN2 (solution P): addition of two new processes to the water network. 232

DOI: 10.1021/acs.iecr.5b02840 Ind. Eng. Chem. Res. 2016, 55, 226−236

Article

Industrial & Engineering Chemistry Research

Figure 10. Pareto-optimal solutions for revamp of existing water network for RWN2 (solution P): addition of two new processes to the water network.

Figure 12. Pareto-optimal solutions for revamp of existing water network for RWN2 (solution P): use of contaminated water (e.g., steam condensate) along with fresh water.

the Pareto-optimal fronts obtained for RWN2 retrofitting and revamping for scenario A, respectively. In these figures, most of the nondominated solutions have significantly large fresh water consumption. To conserve fresh water, only nondominated solutions with fresh water consumption below 30 t/h are considered as potential retrofit and revamp water networks. REV-AP 3 requires less fresh water compared to RET-AP 1 and existing solution “P”, but has significantly larger total flow rate through regeneration units. Table A5 in the Appendix presents the interconnections (with flow rates) between processes and/or regeneration units for selected retrofit and revamp solutions in Figures 9 and 10. The retrofit solution RET-AP 1 (Table A5) has all 15 interconnections from the existing RWN2 (solution P), and it has two new interconnections (p12-p1 and p13-p3) due to new processes added to the water network. All the revamp water networks in Table A5 have 15 or more interconnections, and only about 30% interconnections from the existing/operating water network are reused. Hence, many changes in water network topology are required for implementing the revamp solutions in Table A5. 5.5. Retrofitting and Revamping of RWN2 for Scenario B. For retrofitting and revamping RWN2 network to use steam condensate, Figures 11 and 12 respectively show the Pareto-optimal

solution “P”, but it has significantly larger total flow rate through regeneration units. Table A6 in the Appendix presents details on interconnections (with flow rates) between processes and/or regeneration units for the selected retrofit and revamp solutions of RWN2 (i.e., fresh water consumption