OPTIMAL DESIGN OF INTEGRATED SEPARATION SYSTEMS Zsolt Fonyo Department of Chemical Engineering Budapest University of Technology and Economics and Research Group of Technical Chemistry Hungarian Academy of Sciences 23 March 2004, Trondheim, Norway Research activities • DISTILLATION AND ABSORPTION • • • • • Determination of Vapour-Liquid Equilibria and design of Packed Col umns. Development on distillation and absorption technologies Modelling and calculation of thermodynamic properties Modelling of batch and continuous countercurrent separation processes • EXTRACTION AND LEACHING • • Kinetics of Soxhlet-type and Supercritical Solid-Liquid Extraction of Natural Products. Mathematical modelling and optimization of the process. Supercritical fluid extraction equipment and R&D capabilities • REACTIONS • Mathematical modelling of residence time distribution and chemical reactions • MIXING OF LIQUIDS • PROCESS DESIGN AND INTEGRATION • • • • • • • • • Feasibility of distillation for non/ideal systems Hybrid separation systems Reactive distillation Design of Energy Efficient Distillation Processes Energy integrated distillation system design enhanced by heat pumping and dividing wall columns Energy recovery systems A global approach to the synthesis and preliminary design of integrated total flow-sheets Process Integration in Refineries for Energy and Environmental Management • CONTROL AND OPERABILITY • Assessing plant operability during process design • Transformation of Distillation Control Structures • ENVIRONMENTALS • Waste reduction in the Chemical Industry • • • • • CLEAN TECHNOLOGIES Membrane separations Cleaning of waste water with physico-chemical tools. Solvent recovery Synthesis of mass exchange networks with mixed integer nonlinear programming • Economic and controllability study of energy integrated separation schemes • Process synthesis of chemical plants Analysis of energy integrated separations (distillation based) Synthesis of Mass Exchange Networks Using Mathematical Programming Solvent Recovery from Non-Ideal Quaternary Mixtures with Extractive HeterogeneousAzeotropic Distillation 23 March 2004, Trondheim, Norway Integrated process design • Challenge in chemical engineering • Economical and environmental aspects • Heat integration (HEN) & mass integration (MEN) • Several synthesis strategies • The design needs CAPE Analysis of energy integrated separations (distillation based) (Economic and Controllability Analysis of Energy-Integrated Distillation Schemes) Budapest University of Technology and Economics Department of Chemical Engineering Aims of the work • Separation of ternary mixture by energyintegrated distillation schemes • Optimization of the schemes • Economic evaluation and comparison of the schemes • Optimal schemes are investigated for controllability features • Estimation of the environmental effects Case study •Mixture: (Ethanol – n-Propanol– n-Butanol) •Feed compositions: Case 1: (0.45/ 0.10/0.45) Case 2: (0.33/ 0.33/0.33) Case 3: (0.10/ 0.80/0.10) •Product purity specifications: 99 mole % C *Case 1 *Case 2 *Case 3 A B Composition Triangle Techniques and assumptions HYSYS Process Simulator Modeling of the schemes Steady-state simulation Dynamic simulation •NRTL and UNIQUAC activity models are used •The impurities in B product stream are symmetrically distributed. •Total condensers and reboilers are used. •Exchange min. approach temperature (EMAT)=8.5 oC. •Valve trays (Glitsch type) are used as column internals. Conventional distillation schemes B A L1 ABC L2 D1 D2 Direct sequence Col.2 Col.1 BC Base case for comparison Q2 C B2 A Col.2 Col.1 ABC AB C Indirect sequence B Conventional-heat integration Q2 B A L1 `AB C L2 D1 Col.1 D2 Col.2 Forward heat integration direct sequence (DQF) BC B2 C B A L1 Col.1 L2 D1 R1 =L1 /D1 D2 Backward heat integration direct sequence (DQB) Col.2 ABC V2 Q2 C BC B2 BR2 =V2 /B2 Thermo-coupling L V12 A D R=L/ D L21 Col.2 Col.1 ABC S B SR V 21 L 12 V Q C BR=V/ B B Petlyuk column (SP) Sloppy separation sequences Q L D A AB Col.2 Col.1 ABC S B BC B C Forward heat integration (SQF) A L D AB Col.2 Col.1 ABC S B BC Q C B Sloppy separation sequence, backward heat integration (SQB) The objective function is total annual cost (TAC), which includes annual operating and capital costs. Utilities cost data High utility prices (a) Low utility prices Utility LP-steam MP-steam Cooling water Electricity (b) Temperature (C) 160 184 30-45 Price ($/ton) 17.7 21.8 0.0272 Temperature (C) 160 184 30-45 Price ($/ton) 6.62 7.31 0.0067 -------- 0.1 $/kwh --------- 0.06 $/kwh (a) Based on European prices (b) Based on U.S. prices Estimation of capital cost Douglas, J. M., Conceptual design of chemical processes, McGraw-Hill Book Company Marshall & Swift index: (1056.8/280) Project life: 10 years Sizing of columns and heat exchangers are estimated by HYSYS flowsheet simulator Optimal fractional recovery of the middle component Comparison of theoretical and optimal fractional recovery of the middle component Schemes Case 1, Case 2, Case 3, (0.45/0.10/0.45) (0.33/0.33/0.33) (0.10/0.80/0.10) * * o o * o Petlyuk 0.36 0.36 0.35 0.37 0.33 0.33 column (SP) TAC ($/yr) 8.42E+05 8.42E+05 1.05E+06 1.04E+06 1.31E+06 1.31E+06 SQF scheme 0.39 0.32 0.37 0.33 0.35 0.35 TAC ($/yr) 1.03E+06 1.01E+06 1.01E+06 9.62E+05 9.80E+05 9.80E+05 SQB scheme 0.36 0.35 0.35 0.35 0.33 0.33 TAC ($/yr) 1.03E+06 1.02E+06 9.59E+05 9.59E+05 1.19E+06 1.19E+06 Case study Mixture: ethanol n-propanol n-butanol Equimolar feed composition (0.333, 0.333, 0,333) Product purity specification: 99 m% Detailed results of economic studies Table 3. Results of the economic optimization Description D I DQF DQB SP SQF Col.1 Col.2 Col.1 Col.2 Col.1 Col.2 Col.1 Col.2 Col. 1 Col.2 Col. 1 Col.2 Bottom temperature (oC) 102.32 117.22 117.19 96.93 153.60 117.21 102.32 134.11 102.37 117.19 150.45 117.24 Column pressure (kPa) 101.33 101.33 101.33 101.33 511.00 101.33 101.33 178.50 101.33 101.33 451.00 101.33 Column diameter (m) 0.84 0.82 0.96 0.78 0.76 0.87 0.84 0.85 0.72 0.98 0.65 0.74 Reflux ratio 2.25 1.67 0.90 1.87 2.56 2.38 2.25 2.20 0.67 2.99 0.82 1.92 Overall column efficiency 0.52 0.49 0.46 0.49 0.63 0.49 0.52 0.53 0.51 0.49 0.59 0.48 Actual number of trays 91 98 93 88 93 55 92 61 87 145 57 147 Total actual trays 189 181 148 153 232 204 Heating rate (kJ/hr) 8.01E+06 8.99E+06 5.19E+06 4.56E+06 5.28E+06 3.91E+06 Cooling rate (kJ/hr) 7.88E+06 8.85E+06 4.78E+06 4.19E+06 5.15E+06 3.78E+06 Main HX duty (kJ/hr) ………….. ………….. 3.94E+06 4.26E+06 ………….. 2.89E+06 Auxiliary heat exchanger ………….. ………….. TC,MP TR,LP LP TC,MP Steam cost ($/yr) 5.45E+05 6.11E+05 4.52E+05 3.10E+05 3.59E+05 3.41E+05 C.W cost ($/yr) 2.73E+04 3.07E+04 1.66E+04 1.45E+04 1.79E+04 1.31E+04 Operating cost ($/yr) 5.72E+05 6.41E+05 4.69E+05 3.25E+05 3.77E+05 3.54E+05 Capital cost ($/yr) 7.54E+04 7.85E+04 7.80E+04 8.14E+04 8.18E+04 7.62E+04 TAC ($/yr) 6.48E+05 7.20E+05 5.47E+05 4.06E+05 4.59E+05 4.30E+05 Capital cost saving (%) 0 -4 -3 -8 -8 -1 Operating cost saving (%) 0 -12 18 43 34 38 TAC saving (%) 0 -11 16 37 29 34 SQB Col. 1 Col.2 103.82 158.77 101.33 367.00 0.71 0.71 0.67 1.61 0.49 0.56 79 143 223 3.94E+06 3.00E+06 3.07E+06 TC,MP 3.43E+05 1.04E+04 3.54E+05 7.98E+04 4.33E+05 -6 38 33 Comparison of TAC savings (%) 40 30 % 20 10 I 0 -10 -20 D base case DQF DQB SP SQF SQB Heat loads of the studied schemes Case 1 (0.45/0.1/0.45) 3.5E+07 Case 2 (0.33/0.33/0.33) Case 3 (0.1/0.8/0.1) Heat load (kJ/h) 3.0E+07 2.5E+07 2.0E+07 1.5E+07 1.0E+07 5.0E+06 0.0E+00 D DQB SP Studied distillation schemes SQF SQB Comparison of costs ($/yr) Case (2), (0.33/0.33/0.33) 8.E+05 7.E+05 6.E+05 SP SQF SQB 5.E+05 Capital cost 4.E+05 Utility cost 3.E+05 Total annual cost 2.E+05 1.E+05 0.E+00 D I DQF DQB TAC savings of studied schemes 60 Case 1 (0.45/0.1/0.45) Case 2 (0.33/0.33/0.33) Case 3 (0.1/0.8/0.1) TAC savings (%) 50 40 30 20 10 0 D DQB SP Studied distillation schemes SQF SQB Results of the economic study Energy-integrated schemes are more economical than the best conventional schemes. Operating cost proved to be dominant on TAC. Petlyuk column is the best in TAC saving at low concentration of the middle component (case 1) with 33 % . The heat requirements for the separation increases with increasing concentration of the middle component, and heat-integrated schemes prove to be the best. The maximum TAC saving is achieved in case 3 with 53 % by sloppy sequence with backward heat integration. Controllability study •Selection of controlled variables & manipulated variables, •Degrees of freedom analysis Steady-state control indices •Niederlinski index (NI) •Morari index (MRI) •Condition number (CN) •Relative gain array (RGA) Dynamic simulations Open-loop Closed-loop Steady state controllability indices Steady state controllability indices for the optimized schemes Studied Sche mes NI MRI CN 11 22 33 D, (D1 -L2 -B2 ) 1.137 0.099 8.890 1.000 0.880 0.880 D, (L1 -D2 -Q2 ) 1.995 0.065 32.113 1.000 0.500 0.500 D, (L1 -D2 -B2 ) 1.865 0.234 4.934 0.580 0.540 0.920 DQF, (D1 -L2 -B2 ) 1.136 0.024 36.320 1.000 0.880 0.880 DQF, (L1 -D2 -Q 2 ) 1.890 0.033 21.290 1.000 0.530 0.530 DQF, (L1 -D2 -B2 ) 1.678 0.226 5.240 0.586 0.595 1.020 DQB, (D1 -L2 -B2 ) 1.093 0.023 39.660 1.000 0.910 0.910 DQB, (L1 -D2 -Q 2 ) 2.283 0.040 18.110 1.000 0.440 0.440 DQB, (L1 -D2 -B2 ) 1.540 0.246 5.040 0.647 0.645 1.000 SP, (D-S-Q) 3.515 0.182 6.890 1.000 0.320 0.280 SP, (L-S-B) 7.438 0.089 14.380 0.130 0.570 0.990 SQF, (D-S-Q) 6.470 0.010 137.400 1.000 0.250 0.150 SQF, (L-S-B) 4.030 0.008 158.100 0.250 0.250 0.998 SQB, (D-S-Q) 5.080 0.038 33.310 0.997 0.470 0.196 SQB, (L-S-B) 1.287 0.022 64.388 0.770 0.827 1.000 Evaluation of steady state indices • Base case D and heat-integrated schemes (DQF and DQB) show less interactions. • (D1-L2-B2) manipulated set proves to be better than (L1-D2-Q2) and (L1-D2-B2) for D, DQF and DQB. • Serious interactions can be expected for the sloppy schemes (SQF, SQB and SP). • (L-S-B) manipulated set proves to better than (D-S-Q) for SP, SQF and SQB schemes. Dynamic simulations •Feed rate disturbance: 100 100.5 kmol/h •Feed composition disturbance: (0.33/0.33/0.33) (0.30/0.40/0.30) 1. Open composition control loops: Composition control loops are not installed Results of open loop dynamic simulation Ethanol Propanol Butanol 0.993 105.5 0.989 103.5 102.5 0.985 101.5 Feed rate (kmol/h) Product mole fraction 104.5 0.981 100.5 Feed rate disturbance 0.977 99.5 0 10 20 30 40 50 Time (unit) 60 70 80 90 Heat integrated (DQB) column, open loop, feed rate disturbance 0.995 Ethanol Propanol Butanol 0.43 0.4 0.985 0.37 0.98 0.34 Feed composition disturbance 0.975 Feed mole fraction Product mole fraction 0.99 0.97 0.31 0 10 20 30 40 Time (unit) 50 60 70 Heat integrated (DQB) column, open loop, feed composition disturbance 0.995 Ethanol Propanol Butanol 104.5 103.5 102.5 0.985 101.5 Feed rate (kmol/hr) Product mole fraction 0.990 0.980 100.5 Feed rate disturbance 0.975 99.5 0 10 20 30 40 Time (unit) 50 60 70 Petlyuk column, open loop, feed rate disturbance Ethanol Propanol Butanol 0.994 0.43 0.4 0.986 0.982 0.37 0.978 0.34 Feed composition disturbance 0.974 0.97 0.31 0 10 20 30 40 50 60 70 80 Time (unit) Petlyuk column, open loop, feed composition disturbance Feed mole fraction Product mole fraction 0.99 Summary of open-loop simulation results Open loop performance for feed rate disturbance Ethanol (XA) n-Propanol (XB ) n-Butanol (XC ) Average time Studied Time constant Time constant Time constant constant schemes (time unit) (time unit) (time unit) (time unit) D 16 8 3 9 DQF 20 11 2 11 DQB 14 9 6 10 SP 16 5 6 9 SQF 16 3 5 8 SQB 23 13 11 16 •quite similar dynamic behaviour but • sloppy backward heat integrated (SQB) is the slowest scheme 2. Closed composition control loops: Composition and level controller are installed P-controller PI-controller For level control For composition control Controllers tuning by Tyerus-Luyben cycling method Overshoot, settling time, and their product are evaluated. Results of closed-loop dynamic simulation Ethanol (A) Propanol (B) Butanol (C) 105.5 0.9900 103.5 0.9898 101.5 Feed rate (kmol/h) Product mole fraction 0.9903 Feed rate disturbance 0.9895 99.5 0 5 10 15 20 25 30 35 40 45 50 55 Time (unit) Heat integrated (DQB) column, closed loop (D1-L2-B2), feed rate disturbance Product mole fraction 0.43 0.9900 0.4 0.9898 0.37 0.9896 Feed composition disturbance Feed mole fraction Ethanol (A) Propanol (B) Butanol (C) 0.9902 0.34 0.9894 0.31 0 5 10 15 20 25 30 35 Time (unit) Heat integrated (DQB) column, closed loop (D1-L2-B2), feed composition disturbance 0.9902 Ethanol(A) Propanol(B) Butanol(C) 108 0.9900 105 0.9899 102 Feed rate (kmol/hr) Product mole fraction 0.9901 0.9898 Feed rate disturbance 0.9897 99 0 5 10 15 20 25 30 35 40 Time (unit) Petlyuk column, closed loop (L-S-B), feed rate disturbance Ethanol(A) Propanol(B) Butanol(C) 0.9901 0.43 0.4 0.9899 0.37 0.9897 Feed mole fraction Product mole fraction 0.9903 0.34 Feed composition disturbance 0.9895 0.31 0 5 10 15 20 25 30 35 40 Time (unit) Petlyuk column, closed loop (L-S-B), feed composition disturbance Summary of closed-loop simulation results Closed-loop performance for feed rate disturbance ST OS PSO FBL D (D1 -L2 -B2 ) 10.580 0.010 0.075 1.3 D (L1 -D2 -B2 ) 32.740 0.011 0.227 1.0 DQF (D1 -L2 -B2 ) 11.620 0.006 0.051 1.3 DQF (L1 -D2 -Q 2 ) 20.100 0.012 0.178 1.0 DQF (L1 -D2 -B2 ) 36.550 0.012 0.346 1.0 DQB (D1 -L2 -B2 ) 23.330 0.024 0.310 3.0 DQB (L1 -D2 -Q 2 ) 26.750 0.033 0.751 1.0 DQB (L1 -D2 -B2 ) 34.250 0.034 0.677 1.0 SP (L-S-B) 30.610 0.023 0.238 2.0 SP (D-S-Q) 39.440 0.020 0.262 5.0 SQF (L-S-B) 40.500 0.057 0.813 2.0 SQF (D-S-Q) 59.960 0.042 0.903 3.0 SQB (L-S-B) 70.150 0.030 0.799 6.0 SQB (D-S-Q) 150.000 0.031 1.564 20 Studied schemes Closed-loop performance for feed composition disturbance Studied schemes FBLT ST OS PSO D (D1 -L2 -B2 ) 14.200 0.006 0.044 1.3 D (L1 -D2 -B2 ) 21.200 0.021 0.264 1.0 DQF (D1 -L2 -B2 ) 13.160 0.003 0.025 1.3 DQF (L1 -D2 -Q 2 ) 23.340 0.041 0.631 1.0 DQF (L1 -D2 -B2 ) 31.950 0.081 1.951 1.0 DQB (D1 -L2 -B2 ) 16.530 0.019 0.133 3.0 DQB (L1 -D2 -Q 2 ) 37.850 0.130 3.790 1.0 DQB (L1 -D2 -B2 ) 44.840 0.130 3.512 1.0 SP (L-S-B) 28.390 0.043 0.441 2.0 SP (D-S-Q) 42.680 0.064 0.907 5.0 SQF (L-S-B) 41.600 0.017 0.261 2.0 SQF (D-S-Q) 68.290 0.030 0.789 3.0 SQB (L-S-B) 104.850 0.076 3.398 6.0 SQB (D-S-Q) 107.310 0.083 4.654 20 Conclusions of closed-loop dynamic simulations •Simple energy integration (heat integration) doesn’t influence dynamic behaviour compared to the non-integrated base case •Higher detuning factor is needed due to stronger interactions in complex distillation systems (they became slower in closed loop) •The cases, where material and energy flows (energy integration) go into the same direction (DQF, SQF), are better than the opposite •Since the sloppy schemes show similar economic parameters, their controllability features make the decision to the favour of SQF ! •Petlyuk columns controllability parameters are between the ones of the heat-integrated and the sloppy schemes Estimation of the flue gas emissions The main gaseous pollutants that are considered in this work are: CO2, SO2, and NOx The flue gas emissions of Case 1 Schemes Heating rate (MW) Fuel type CO2 emissions (kg/hr) SO x emissions (kg/hr) NOx emissions (kg/hr) Total emissions (kg/hr) Emissions saving % D DQB SP SQF SQB 5.5 3.7 3.5 3.4 3.5 Natural Oil Natural Oil Natural Oil Natural Oil Natural Oil gas gas gas gas gas 1030 1447 691 971 659 926 689 968 716 1006 0.75 3.57 0.50 2.39 0.48 2.28 0.50 2.39 0.52 2.48 0.16 0.66 0.11 0.44 0.11 0.42 0.17 0.47 0.18 0.49 1031 1451 692 974 660 928 689 971 717 1009 0 33 36 33 31 Final conclusions with energy-integration about 53 % TAC saving can be realised in case of SQF scheme Petlyuk column has a limited TAC saving of 30-33 % in all the three feed composition cases conventional heat-integration shows the best economic and controllability features considering all the three feed compositions sloppy schemes show good economic features but the selection is made according to their different controllability features (SQF has better features than SQB) economic and controllability features are to be handled simultaneously during process design. Closed loop dynamic simulations •Simple energy integration (heat integration) doesn’t influence dynamic behaviour compared to the non-integrated base case •more complicated systems: higher detuning factor is needed due to stronger interactions (they became slower in closed loop) •The cases, where material and energy flows (energy integration) go into the same direction (DQF, SQF), are better than the opposite •Since the sloppy schemes show similar economic parameters, their controllability features make the decision to the favour of SQF (!) •Petlyuk (dividing wall) column‘s controllability parameters are between the ones of the heat integrated and the sloppy structures. Synthesis of Mass Exchange Networks Using Mathematical Programming Budapest University of Technology and Economics Department of Chemical Engineering Outline I. Mass Exchange Network Synthesis (MENS) A Extension of the MINLP model of Papalexandri et al. (1994) B Comparison of the advanced pinch method of Hallale and Fraser (2000) and the extended model of Papalexandri et al. C New, fairly linear MINLP model for MENS II. Rigorous MINLP model for the design of distillation-pervaporation systems III. Rigorous MINLP model for the design of wastewater strippers Approach: Mixed Integer Nonlinear Programming (MINLP) optimisation software: GAMS / DICOPT I. Mass Exchange Network Synthesis El-Halwagi and Manousiouthakis, AIChE Journal, Vol 35, No.8, pp. 1233-1244 RICH STREAMS LEAN STREAMS (MSAs) xsj NS=NSE+NSP 1 2 Mass integration for the analogy of the concept of heat integration. Absorber, extractor etc. network synthesis . . . Gi 1 2 ysi . . . MASS EXCHANGE NETWORK yti NR xtj yi=f(xj) The synthesis task: Stream data + equipment data + equilibrium data + costing Network structure lean stream flow rates min (Total Annual Cost, TAC) Previous work: early pinch methods (no supertargeting) Water pinch: Wang & Smith (1994, 1995), Kuo & Smith (1998) El-Halwagi & Manousiouthakis (1989a) El-Halwagi (1997) advanced pinch method (includes supertargeting) Hallale & Fraser (1998, 2000) sequential mathematical programming methods El-Halwagi (1997), Garrison et al. (1995) Alva-Argaez et al. (1999) simultaneous mathematical programming models Papalexandri et al. (1994) Papalexandri & Pistikopoulos (1995, 1996) Comeaux (2000); Wastewater: Benkő, Rév & Fonyó (2000) I/A Extensions of the MINLP model of Papalexandri et al. (1994) • Integer stage numbers • Generation of feasible initial values • Kremser equation: 1 y m x bij 1 ln 1 A y m x bij A N A1 ln A in i out i in ij j in ij j f(x) f x if x c f x 1 not defined if x c x LO x xUP yiin yiout N A1 out yi mij x inj bij f1(x) xLO yi*=mijxj*+bij A Lj mij Ri c xUP x Removable discontinuity at A=1 • Previous mathematical programming models for MENS assumed that A is always greater than 1 • Numerical difficulty when using GAMS Handling of removable discontinuities in MINLP models: Example: The Kremser-equation ifIf A1 ifIf A=1 NTPA1 NTPA1 yiin m j x inj b j 1 ln 1 out y m x in b A j j j i ln A yiin yiout out yi m j x inj b j 1 A A Lj m jGi • The usual form of the Kremser equation has a removable discontinuity at A=1 • Using only the first form of the Kremser equation in MINLP models leads to a division by zero error or gives solutions that have no physical meaning • Restricting all the values of A under or over 1 very likely excludes the real optimal solution from the search space Different possibilities for handling the discontinuity of the Kremser equation: • For all mass exchangers in the superstructure both formulas are used to calculate the theoretical number of stages • A binary variable Y has to be generated to be able to choose between the two calculated stage numbers Y=1 A=1 Y=0 A1 then NTP Y NTPA1 1 Y NTPA1 The binary variable Y can be generated in the following ways: Methods for calculating Y taken from the literature: Three (or two, y3=1-y1-y2) binary variables are used to denote the interval in which the actual value of A lies y1 0.01 y2 0.99 y1=1 when 0.99A0.01 etc. y3 1.01 100 y1+y2+ y3=1 Big-M method: A 0.99 M 1 (1 y1 ) ; A 1.01 M 2 (1 y 2 ) ; A 100 M 3 (1 y 3 ) A 0.01 M 1 (1 y1 ) ; A 0.99 M 2 (1 y 2 ) ; A 0.01 M 3 (1 y 3 ) M 1 99.01 ; M 2 98.99 ; M 3 1 0.01 A 100 ; y1 y 2 y 3 1 ; yi 0 or 1 ; Y y2 Multi-M method: A 0.99 M 1, 2 (1 y1 ) ; A 1.01 M 2,1 (1 y 2 ) ; A 100 M 3,1 (1 y 3 ) A 0.01 M 1,3 (1 y1 ) ; A 0.99 M 2,3 (1 y 2 ) ; A 0.01 M 3, 2 (1 y 3 ) M 1, 2 99.01 ; M 1,3 0 ; M 2,1 98.99; M 2,3 0.98; M 3,1 0; M 3,2 1 0.01 A 100 ; Y y2 y1 y 2 y 3 1 ; yi 0 or 1 ; Convex hull: A1 0.99 y1 ; A2 1.01 y 2 ; A3 100 y 3 A1 0.01 y1 ; A2 0.99 y 2 ; A3 1.01 y 3 0 Ai 100 ; 0.01 A 100 ; A1 A2 A3 A y1 y 2 y 3 1 ; yi 0 or 1 ; Y y2 A formulation of Raman & Grossmann: 99 z1 0.01 A 1 z1 z 2 y 3 1 A 1 991 z1 z1 y3 0 11 z 2 A 1 z2 y3 0 A 1 99 z 2 0.01 Y y3 A simple logical formulation (L-formulation): 0.01 y1 0.99 y 2 1.01 y 3 A 0.99 y1 1.01 y 2 100 y 3 y1 y 2 y 3 1 ; y i 0 or 1 ; Y y2 The main drawback of the methods taken from the literature is, that they use three binary variables for each removable discontinuity or mass exchanger. Solving MEN synthesis problems this may mean that in case of large superstructures the problem size exceeds the practical solvability limit (approx. 80-100 binary variables in an MINLP model). New formulation for handling removable discontinuities: ( A c) V y V is continuous y is a binary variable n=2 n For the Kremser equation: (A-1)2=Vy y=1 y=0 (A-c)2 y=1 0.01 A 100 2 VLO 104 V 9810 Y=1-y A=c A This method uses only one binary variable for calculating the binary variable Y. Several mass exchange network synthesis problems were solved using our method. It proved to be fast and well applicable. N Y N A1 1 Y N A1 Adopted literature methods • Big-M formulation • Multi-M formulation • a Convex-hull like formulation • Raman & Grossmann (1991) • Simple logic formulation y1 0 0.01 y2 0.99 y3 1.01 100 are linear but use 3 binary variables New method ( A c) 2 V y ALO A AUP VLO V VUP Advantages: 1. faster 2. larger problems can be solved nonlinear but uses 1 binary variable only (the models are nonconvex anyway) Large nonconvex MINLP problems solved by DICOPT++: There exists a critical upper limit of the number of binary variables I/B Comparison of the advanced pinch method of Hallale and Fraser (2000) and the extended model of Papalexandri et al. (1994) Example Objective Pinch solution of MINLP (CMINLP-CPinch) / function Nick Hallale Solution CMINLP *100 Target / Design 3.1 CAP 830 000 / 860 000 1 044 285 +17.6 % 3.2 CAP 448 000 / 455 000 453 302 -0.4 % 3.3 CAP 819 000 / 751 000 637 280 -17.8 % 3.4 CAP 591 760 / 637 000 637 000 0.0 % 4.1 CAP 296 000 / 298 000 255 068 -16.8 % 5.1 TAC 226 000 / 228 000 226 000 -0.9 % 5.2 TAC 226 000 / 228 000 226 000 -0.9 % 5.3 TAC 226 000 / 228 000 226 000 -0.9 % 5.4 TAC 49 000 / 49 000 50 279 +2.5 % 5.5 TAC 524 000 / 526 000 527 000 +0.2 % 6.1 TAC 692 000 / 706 000 720 000 +1.9 % 6.2 TAC 28 000 / 28 000 32 000 +12.5 % 6.3 CAP 591 000 / 539 000 536 000 -0.6 % TAC-total annual cost in USD/yr, CAP–annualised capital cost in USD, 13 example problems have been solved C - cost The two methods perform more or less the same. Why are the MINLP solutions not always better? The MINLP model is nonconvex. I/C New, fairly linear MINLP model for MENS concentration location 1 x 1,T y 1,S R1 x 1,1 k=1 R1-L1 R1-L1 R2-L1 R2-L1 x 2,S y 2,3 y 2,2 R2-L2 x 1,S x 2,3 x 2,2 y 2,1 x 1,3 L1 y 1,3 R1-L2 R2 concentration location 3 y 1,T R1-L2 x 2,1 y 2,S x 1,2 k=2 y 1,2 y 1,1 x 2,T concentration location 2 L2 y 2,T R2-L2 Similar to the HEN superstructure of Yee & Grossmann (1990) The stagewise superstructure enables almost linear mass balance formulation Model equations mass balances me R y y i i, k i, k 1 i, j , k j R y s y me i i i, last i, j, st j, st me L x x j j, k j , k 1 i, j , k i L x x s me j j, first j i, j, st i, st concentration constraints y y i, k i, k 1 x j, k x j, k 1 y YT i, last i XT j, first j y Ys i, first i x x j, last Xs j minimise c L TAC f mass i, j, k j j i, j , k j s.t. big-M constraints for the existenxe of the units me z 0 i, j , k i, j i, j , k driving force constraints 1 z dy y m x b i, j , k i, k i, j j , k i, j i, j, k i, j, k 1 z dy y m x b i, j , k 1 i, k 1 i, j j , k 1 i, j i, j , k i, j, k constraints on the number of existing units max zi, j , k U i, j , k U min z i, j , k i, j , k Chen’s approximation for the log mean conc differences 1/ 3 dy / 2 lmcd dy dy dy i, j, k i, j, k i, j , k 1 i, j, k i, j, k 1 calculation of the mass of the exchangers mass K lmcd me i, j , k W i, j , k i, j , k Only the lean stream mass balances are bilinear Example problems Example 4.1 (Hallale, 1998) 3.66e-3 2.5e-3 4.05e-3 1.059e-2 298,000 USD Solution obtained by using the R2 3.5kg/s 0.01 network 5e-3 Pinch solution of Nick Hallale (1998) R1 4 kg/s 5e-3 2.5e-3 Capital cost of the 2 kg/s 5e-3 3.26e-3 1e-3 2.5e-3 9.16e-3 1e-2 R4 0.5kg/s 8e-3 4.08e-3 R3 1.5kg/s R5 255,068 USD 0 MINLP model of Papalexandri et al. (1994) 3.77e-3 1.64e-3 S3 1.7e-2 7.79e-3 2.48 kg/s Solution obtained by the 284,440 USD 1 kg/s S2 new MINLP model S1 2.5e-3 8.48e-3 5.82e-3 3.86e-3 3.63e-3 1.7e-3 1.8 kg/s 7.1e-3 Capital cost, based on exchanger mass: 284,440 USD Extensions: stagewise exchangers, multiple components 0.9 kg/s Authors Total annual cost of the MEN MINLP solution of 917,880 USD/yr. Advanced pinch solution of N=4.93 N=2.88 N=2.73 N=3.25 R2 427,000 USD/yr. 1.752 kg/s Hallale and Fraser (2001) Solution obtained using the 0.062 kg/s R1 0.1 kg/s Papalexandri et al. (1994) N=4.23 0.022 kg/s S1 436,289 USD/yr new MINLP model 2.169 kg/s 0.487 kg/s S2 The new model is most suitable for solving single component MENS problems, where packed columns are used exclusively. In this case, no special initialisation is needed. TAC=436,289 USD/yr Two component example 0.566 kg/s II. Rigorous MINLP model for the design of distillation-pervaporation systems Distillation column The synthesis task is to determine: Vacuum vessel Pervaporation unit Inlet ethanol ~80 m/m% EtOH retentate (dehydrated ethanol) • Nth of the column • feed tray position • reflux ratio • membrane structure • reflux scheme permeate (mainly water) Rigorous modelling: Dist. Column: 1 bar, MESH equations, tray by tray, Margules activity coeff. for the liquid phase, ideal vapour phase, latent heat enthalpy Membrane unit: transport calculation is based on experimental data 1/3 m2 flat membranes, costing - industrial practice Adequate costing equations, utility prices Superstructure max n pieces of membrane modules 1 N N-1 column feed mixer feed RF F retentate distillate from the column P1 permeate feed pump refi imin recycled permeate ifeed permeate consplitter denser to the next section of membranes 2 n heat exchanger P3 pump i=1…m ibmax to the vacuum pump permeate condensate bui 1/3 m2 flat PVA membranes in blocks ethanol product max m sections like this The blocks (or modules) can be connected in both series or parallel 2 1 P4 recycled permeate P2 column bottom product Distillation column superstructure: Viswanathan & Grossmann (1993) Membrane superstructure: new Multiple level optimisation (successive refinement) enables reducing the number of binary model variables Modelling of the membranes is based on experimental data Industrial example 12 x 81 pieces of 1/3 m2 flat membranes =324 m2 total (fixed industrial configuration) feed 80 mass% EtOH total permeate recycling D=0.875 m min=97.5% 12 x 107 pieces of 1/3 m2 flat membranes = 428 m2 total 84 1046.3 kg/hr theor. 91.44 mass% stages total permeate recycling D=0.679 m 1175 kg/hr 1175 kg/hr 7 4 membrane capital investment : 69,058 USD membrane replacement : 110,758 USD column capital investment : 13,931 USD column operating cost : 134,377 USD membrane capital investment : 52,362 USD membrane replacement : 83,936 USD column capital investment : 18,05 USD column operating cost : 219,472 USD 1 1 recycled permeate 72 kg/hr 28.96 mass% EtOH bottom product 254.3 kg/hr 0.087 mass% EtOH bottom product 254.3 kg/hr 0.087 mass% EtOH recycled permeate 125.6 kg/hr 30.86 mass% EtOH TAC=373,820 USD/yr Optimised Base case 12% savings in the TAC base case optimally designed system 400 400 3,5 membrane capital investment membrane replacement column capital investment column operational cost TAC reflux ratio 350 350 3 300 300 plant membrane cost plant TAC optimised membrane cost optimised TAC optimised column cost plant column cost 250 200 TAC (thousand USD/yr) TAC (thousand USD/yr) Other calculations using the MINLP model TAC=328,124 USD/yr 2,5 250 200 2 reflux ratio feed 80 mass% EtOH min=97.5% 80 992.7 kg/hr theor. 94.56 mass% stages retentate (product): 920.7 kg/hr 99.7 mass % EtOH reflux ratio: 1.38 retentate (product): 920.7 kg/hr 99.7 mass % EtOH reflux ratio: 3.262 150 1,5 100 150 1 50 100 94,5 95 95,5 96 96,5 97 97,5 98 98,5 99 specified ethanol yield (%) 99,5 0 300 0,5 350 400 450 500 overall membrane surface in square meters Ethanol yield - TAC Membrane surface - TAC OPTIMISATION OF HYBRID ETHANOL DEHYDRATION SYSTEM Z. Fonyo, Z. Lelkes, Z. Szitkai, E. Rev • Introduction & problem statement • MINLP model and superstructure • Membrane model • Industrial case study • Conclusions Department of Chemical Engineering, H-1521 Budapest, Hungary retentate (abs. EtOH product) inlet stream 1000 kg/hr 94 mass% EtOH plant membrane configuration 12 sections in series each consisted of 81 pieces of 1/3 m2 flat membranes in parallel measured: 940 kg/hr 99.6-99.7 mass% EtOH calculated: 921.5 kg/hr 99.6 mass% EtOH permeate measured: 60 kg/hr 15 mass% EtOH calculated: 78.5 kg/hr 28 mass% EtOH Calculated and measured output stream properties for the fixed industrial inlet stream and membrane configuration retentate (product): 920.7 kg/hr 99.7 mass % EtOH reflux ratio: 3.262 feed 80 mass% EtOH min=97.5% 80 992.7 kg/hr theor. 94.56 mass% stages 12 x 81 pieces of 1/3 m2 flat membranes =324 m2 total (fixed industrial configuration) total permeate recycling D=0.875 m 1175 kg/hr 4 membrane capital investment : 52,362 USD membrane replacement : 83,936 USD column capital investment : 18,05 USD column operational cost : 219,472 USD 1 recycled permeate 72 kg/hr 28.96 mass% EtOH bottom product 254.3 kg/hr 0.087 mass% EtOH TAC=373,82 USD/yr Base case: optimised hybrid ethanol dehydration plant with fixed industrial membrane structure retentate (product): 920.7 kg/hr 99.7 mass % EtOH reflux ratio: 1.38 feed 80 mass% EtOH min=97.5% 84 1046.3 kg/hr theor. 91.44 mass% stages 12 x 107 pieces of 1/3 m2 flat membranes = 428 m2 total total permeate recycling D=0.679 m 1175 kg/hr 7 membrane capital investment : 69,058 USD membrane replacement : 110,758 USD column capital investment : 13,931 USD column operational cost : 134,377 USD 1 recycled permeate 125.6 kg/hr 30.86 mass% EtOH bottom product 254.3 kg/hr 0.087 mass% EtOH TAC=328,124 USD/yr Optimised hybrid ethanol dehydration plant with optimised membrane structure 400 TAC, industrial TAC (thousand USD/yr) 350 TAC, optimised 300 plant membrane cost plant TAC optimised membrane cost optimised TAC optimised column cost plant column cost 250 200 150 100 94,5 95 95,5 96 96,5 97 97,5 98 98,5 99 99,5 specified ethanol yield (%) Influence of the specified ethanol yield on the TAC optimised system vs. plant existing in the industry 350 300 TAC TAC (thousand USD/yr) membrane capital investment membrane replacement 250 column capital investment column operational cost 200 TAC 150 100 50 0 94,5 95 95,5 96 96,5 97 97,5 98 98,5 99 Specified ethanol yield in % Influence of the specified ethanol yield on the TAC, optimised systems only 99,5 industrial case optimised 400 3,5 membrane capital investment membrane replacement column capital investment column operational cost TAC reflux ratio 350 3 2,5 250 200 2 reflux ratio TAC (thousand USD/yr) 300 150 1,5 100 1 50 0 300 0,5 350 400 450 500 overall membrane surface in square meters Dependence of the TAC and the reflux ratio on the overall membrane surface Shepard’s metric interpolation: c ( P) r ( P) rj ( P ) ( cT )i 4 i j i Sheppard's metric interpolation alpha=2 3,5 T 3 j i j depending on the value of alpha: • local minima • step function • peaks 2,5 cT ,mass % i cT calculated by differential equations 2 cT, metric interpolated 1,5 1 0,5 0 Power function: 0 5 10 15 20 25 30 j0, kg/hr (c0=4.06 mass%) 0,145 CT 0,55 C0 J 0 J T 0,999 J 0 0,031 C0 parameter fitting: method of least squares 35 40 45 50 334 TAC (thousand USD/yr) 332 330 328 326 324 TAC, optimised 322 320 318 316 314 94 95 96 97 98 99 Specified ethanol yield in % Influence of the specified ethanol yield on the TAC 100 Results: • Design tool to optimise the hybrid ethanol dehydration process • Large, but solvable MINLP model • In case of an industrial dehydration plant: 12% saving in TAC is possible by addition of 32% more membrane surface • Sensitivity analysis on membrane replacement cost, membrane surface and ethanol yield III. Rigorous MINLP model for the design of wastewater strippers Wastewater cleaning by stripping Minor quantities of acetone, methanol, and ethanol in water total condenser top product feed 20 5 mol/s xacetone = 0.05 xmethanol= 0.04 xwater = 0.90 xethanol = 0.01 Superstructure 19 18 17 . . . Nth=? boil-up vapour VLE calculation 3 2 1 bottom product Similar to the distillation column superstructure of Viswanathan & Grossmann (1993) Xwater0.999 water85% Wilson binary interactions Ideal vapour phase Theoretical stages 1 bar Latent heat enthalpy Antoine vapour pressure Conclusions: Complex evaluation of distillation based heat integrated separation schemes is presented. New sloppy structures proved to be competitive. New, fairly linear, MINLP modell for MENS is developed and succesfully tested for literature examples and industrial case studies. . Utility prices Utility Temperature Price level (ºC) ($/ton, kWh) Low pressure steam 160 17.7 Middle pressure steam 184 21.8 Cooling water 30-45 0.0272 Electricity -------- 0.1 $ Controllability investigations, design – interactive and challenging part of process design or development. Control structure synthesis * control targets are defined, * the sets of controlled variables and possible manipulated variables are determined (degrees of freedom) * pairing of the controlled and manipulated variables: steady state control indices, dynamic behaviours in the cases of open and closed control loops of the promising control structures. Demonstration of interaction between design and control • comprehensive design of five energy integrated separation schemes • three-component-alcohol-mixture is separated in five distillation based energy integrated two-column separation systems: – two heat integrated distillation schemes – fully thermally coupled distillation column (Petlyuk, Kaibel) – sloppy separation sequences Solvent Recovery from Non-Ideal Quaternary Mixtures with Extractive Heterogeneous-Azeotropic Distillation Budapest University of Technology and Economics Chemical Engineering Department Motivation • Industrial companies ( printing, pharmaceutical) have waste streams of solvents (quaternary mixtures) • 4 groups of solvents with different VLLE, azeotropes • Separation of non-ideal quaternary mixtures is less studied Goal Guideline for the design of separation schemes for non-ideal quaternary mixtures Heterogeneous-azeotropic distillation Extractive distillation Extractive heterogeneous – azeotropic distillation D1 F2 Extr. agent Feed W1 Group 1 Acetone, ETOH, MEK, Water Acetone, ETAC, ETOH, Water One volatile component forming no azeotropes Acetone Binary azeotropes Binary azeotropes Water-ETOH Water-ETOH Water-MEK Water-ETAC ETOH-MEK ETOH-ETAC Ternary azeotrope Ternary azeotrope Water-ETOH-MEK Water-ETOH-ETAC Investigated separation schemes A Aceton D2 F2 C2 D1 Water mix F B2 C1 Feed Group 1 ETOH 95 w% W1 H2O MEK (ETAC) Representation of separation in Column 1 Water addition F W1 Separation in C1 Hypothethical feed D1 Water – Acetone - MEK Ternary mixture Acetone 1 D2 0.9 0.8 0.7 0.6 0.5 F2 0.4 0.3 0.2 0.1 B2 MEK MEK 0 R Phase sep. 0 BB2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Water 1 Investigated separation schemes B D2 Group 1 MEK (ETAC) Water ACETONE C2 Feed C1 D1 B1 Water ETOH MEK(ETAC) B2 ETOH 95 w% H2O Water – ETOH – MEK Ternary mixture ETOH 1 0.9 0.8 0.7 0.6 0.5 0.4 F2 Water addition 0.3 0.2 0.1 MEK D2 Phase sep. 0.4 0.5 B2 C3 0 MEK 0 0.1 0.2 0.3 0.6 0.7 0.8 0.9 1Water Economic comparison of structures A and B 6 Total Annual Cost [1e5 €] 5 A 4 3 B 2 0 0.1 0.2 0.3 0.4 mole fraction Acetone in Feed 0.5 0.6 Group 2 ETAC, ETOH, IPAC, Water ETOH, MEK, IPAC, Water Binary azeotropes Binary azeotropes Water-ETOH Water-ETOH Water- ETAC Water-MEK Water-IPAC Water-IPAC ETOH-ETAC ETOH-MEK ETOH-IPAC ETOH-IPAC Ternary azeotropes Ternary azeotropes Water-ETOH-IPAC Water-ETOH-MEK Water-ETOH-IPAC Water-ETOH-IPAC Investigated separation schemes for the mixtures of Group 2 mix D3 Water F1 D2 B2 IPAC ETOH 95 w% W1 H2O C3 F2 C2 D1 Group 2 C1 Feed B3 ETAC (MEK) The VLLE Data and representation of separation Hypothetical feed Water B1 R Water feed addition Separation in C1 Operating line of phase separator F F1 IPAC D1 R1 F2 ETOH ETAC Water – ETAC – IPAC Ternary mixture ETAC 1 B3 0.9 D2(F3) 0.8 0.7 D3(R1) F2 0.6 0.5 0.4 0.3 Immiscibility region 0.2 0.1 B2 0 IPAC 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Water Group 3 ETAC, ETOH, MEK, Water ETAC, IPOH, MEK, Water Binary azeotropes Binary azeotropes Water-ETAC Water-ETAC ETOH-ETAC Water-MEK Water-MEK Water-IPOH ETOH-MEK IPOH-ETAC ETAC-MEK IPOH-MEK Water - ETOH ETAC-MEK Ternary azeotropes Ternary azeotropes Water-ETOH-ETAC Water-IPOH-MEK Water-ETOH-MEK Water-ETAC-MEK Water-ETAC-MEK MEK-IPOH-ETAC Separation schemes for the mixtures of Group 3 R1 C1 Water Feed F F2 ETAC MEK Water R2 B2 ETAC 95 w% D3 C3 V1 D2 C2 Water Group 3 B1 ETOH 95 w% (IPOH 85 w%) Water Water MEK 93 w% Representation of extractive heterogeneous-azeotropic distillation for the separation of mixtures of Group 3 ETOH Water addition MEK F F2 D1 B1 R ETAC Water Operating line of Separation in C1 Hypothetical phase separator feed Water – ETAC – MEK Ternary mixture MEK 1 0.9 MEK 0.8 Phase sep. 0.7 D3 0.6 0.5 C3 F2 0.4 0.3 0.2 B2 0.1 ETAC D2 R3 C2 Phase sep. B3 0 ETAC 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 R2 1 Water Group 4 ETOH, MEK, N-Heptane, Water Total possible combination Binary azeotropes Water-ETOH Water-MEK Water- N-Heptane ETOH-MEK ETOH- N-Heptane MEK – N-Heptane Ternary azeotropes Water-ETOH-MEK Water-ETOH- N-Heptane Water – MEK – N-Heptane ETOH – MEK – N-Heptane Separation schemes for mixture of Group 4 C3 Water R1 C1 Water Feed F F2 B1 N-Heptane MEK Water ETOH 95 w% Water MEK Water N-Heptane C4 B3 C2 V1 MEK 93 w% Water The VLLE Data ETOH N-Heptane MEK Water Water – N-Heptane – MEK Ternary mixture MEK 1 0.9 MEK 0.8 Phase sep. D2 0.7 R3 0.6 Water addition F2 0.5 D3 D1 0.4 C4 C2 C3 0.3 0.2 0.1 0 N-Heptane 0 B3 B2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Water Classification of processes Group 2 Group 1 P1 R1 P2 Water R Water F C2 C1 F F1 D2 C2 C3 C1 P3 P2 B2 B1 Group 3 Group 4 Water P2 Water R Water F C2 C1 B2 B1 P3 C3 B3 C3 R Water F P3 C2 C4 C1 P2 B1 P4 Thank you for your attention!