Format And Type Fonts CHEMICAL ENGINEERING TRANSACTIONS VOL. 45, 2015 A publication of The Italian Association of Chemical Engineering www.aidic.it/cet Guest Editors: Petar Sabev Varbanov, Jiří Jaromír Klemeš, Sharifah Rafidah Wan Alwi, Jun Yow Yong, Xia Liu Copyright © 2015, AIDIC Servizi S.r.l., ISBN 978-88-95608-36-5; ISSN 2283-9216 DOI: 10.3303/CET1545196 Please cite this article as: Xiao W., Zhou Y., Ruan X., He G., Jia X., 2015, Parameters optimisation of isopropanol purification by hybrid distillation-vapour permeation process using response surface methodology, Chemical Engineering Transactions, 45, 1171-1176 DOI:10.3303/CET1545196 1171 Parameters Optimisation of Isopropanol Purification by Hybrid Distillation-Vapour Permeation Process Using Response Surface Methodology Wu Xiao a , Yanyan Zhou a , Xuehua Ruan a , Gaohong He *,a , Xiaoping Jia b a State Key Laboratory of Fine Chemicals, R&D Center of Membrane Science and Technology, Dalian University of Technology, Dalian 116024, Liaoning, China b School of Environment and Safety Engineering, Qingdao University of Science and Technology, Qingdao 266042, Shandong, China hgaohong@dlut.edu.cn A hybrid distillation-vapour permeation (D-VP) system was selected to get Isopropanol (IPA) with high purity. In this study, a technical and economic analysis was performed comparing the hybrid D-VP systems with polymeric (PVA/PVDF) and ceramic (NaA type zeolite) membranes. Response surface methodology (RSM) was used to optimise the hybrid process parameters. First of all, the simulation of hybrid D-VP process for IPA purification was conducted with UniSim Design. Then Plackett-Burman design was employed to screen the significant parameters affecting total annual cost (TAC) from 11 variables. After that, steepest ascent method was undertaken to determine the optimal regions of the selected significant parameters. Finally, Box-Behnken design was adopted to further analyse the mutual interactions between these parameters and to identify their optimal values that would generate a minimum TAC. Optimal parameters were the vapour flow-rate to VP of 580 kg/h, VP operating temperature of 140 °C, permeate pressure of 1.5 kPa. And the TAC of optimal HDCM system is 49.06 €/t, about 19.80 % less than that of pri-optimised system. 1. Introduction Isopropanol (IPA) is widely used as one of the important solvents and cleaners in modern chemical, semiconductor, pharmaceutical and electronic industries. Recycling of IPA from IPA-water mixtures is necessary from environmental and economic point of view. However, IPA forms an azeotrope with water at 87.4 wt.% of IPA at atmospheric condition, which makes the separation of these mixtures difficult and uneconomical by conventional methods such as distillation. Vapour permeation (VP) offer a more promising and energy-efficient alternatives for azeotropic separation. Ceramic membranes are solvent and temperature stable, and zeolite NaA membranes offered significant potential for a dehydration agent with high permeation flux and separation factor as well as high chemical and thermal stability (Pina et al., 2004). However, VP itself still encounters some challenges such as the membrane productivity, the sensitivity to friction losses in the feed stream and the possibility of condensation (Cen and Lichtenthaler, 1995). Then application of VP alone is not an optimal choice. In order to reduce energy consumption, hybrid distillation-vapour permeation (D-VP) system has attracted much attention in recent years (Naidu and Malik, 2011). Szitkai et al. (2002) modelled and optimised hybrid ethanol dehydration systems using MINLP, differential equations were employed for modelling the membrane modules, computational experiences used GAMS/DICOPT. Koczka et al. (2007) considered different configurations with rigorous modelling tools based on the comparison of total annual cost (TAC), three hybrid separation processes were rigorously modelled with CHEMCAD, and optimised with the dynamic programming optimisation method for the case of the separation of ethanol-water mixture. However, they did not consider structural and operational parameter optimisation of hybrid process. 1172 Most of the studies on hybrid distillation-membrane processes involved changing one of the independent parameters at a time while maintaining the others at a fixed level. Such studies ignored the interaction effects among the important parameters affecting the separation. One possible solution is to use the response surface methodology (RSM) which is widely used to analyse the effects of multiple factors and their interactions (Conto et al., 2015). It overcomes disadvantages of conventional methods and is proved to be an effective way for process parameters analysis that uses mathematical and statistical techniques to analyse the influence of independent variables on a specific response (Dahmoune et al., 2014). In this paper, hybrid D-VP processes with ceramic membrane (HDCM) for IPA purification were simulated with UniSim Design based on the improved membrane module. Plackett-Burman (PB) design was used to screen the factors essential for TAC. Then steepest ascent method was employed to approach the vicinity of the optimal conditions. Subsequently, Box-Behnken design (BBD) for RSM was used to estimate the relationship between a response and optimal parameters for minimum TAC, and then the economic analysis was performed for the dehydration of IPA by HDCM. 2. Simulation and Experimental designs 2.1 Validation of D-VP model with UniSim Design The simulation of HDCM process for IPA purification was conducted with UniSim Design. To verify the modelling results, the simulation output was compared with the corresponding data given in literature (Hoof et al., 2004) for the same conditions. The basic parameters were given in Table 1 (Hoof et al., 2004). Table 1: Original data of literature Description Parameters Value Description Parameters Value Feed Flow rate (kg/h) 1,000 Distillation Thermodynamic model NRTL Fraction (IPA,wt%) 0.50 Number of trays 10 Temperature (°C) 20 Pressure (kPa) 100 Pressure (kPa) 110 Reflux ratio 1.35 Location 9 Top flow rate (kg/h) 601 Membrane Temperature (°C) 95 Permeate pressure (kPa) 2 Selectivity 200 Target Retentate fraction (IPA,wt%) 0.995 Figure 1: Simulation Model of IPA-water separation process in Unisim Design Model of D-VP process for IPA-water separation in UniSim Design was shown in Figure 1. The feed enters the distillation column, where the aqueous isopropanol mixture is concentrated up to nearly close to the azeotropic concentration. The top vapour stream heated up to a specific temperature enters into the vapour permeation section, where the rest of the water is removed and recycled back to the distillation column. HDCM process for IPA purification model with UniSim Design was run. For distillation process, the flow rate and the composition of the top stream were 610 kg/h and 81.92 wt.% IPA. For membrane separation process, the retentate flow rate and the target IPA fraction were 500.8 kg/h and 99.79 wt.%, respectively. Compared with the literature results (Hoof et al., 2004), the percentage errors of the flow rate and the composition of the top stream were 1.47 % and 1.30 %, and the percentage errors of the retentate flow rate and the target IPA fraction were 0.16 % and 0.14 %. Therefore, the simulation results of D-VP model with UniSim Design were available. 1173 2.2 PB Experimental designs PB design is a powerful and efficient mathematical approach when rapidly searching for key factors in a multivariable system. It offers a good and fast screening procedure and mathematically computes the significance of a large number of variables with relatively few experiments, which is time saving and gives the effect of change in more than one factors in single experiment (Singh et al., 2011). In the studies, as showed in Table 2, PB design was employed to screen the significant parameters from 11 variables (including 3 dummy variables). TAC was considered as the response in the design. These variables were investigated and 12 tests were carried out. Each independent variable was examined at two levels: -1 for low level and +1 for high level. Table 2 illustrated the variables and their corresponding levels used in the experimental design. The design matrix selected for the screening of significant parameters for TAC and the corresponding responses were shown in Table 3. Table 2: Levels of the variables tested in PB design Variables Symbol Variation Levels Low (–1) High (+1) Feed tray location A 3 6 Permeate recycle location B 3 6 Reflux ratio C 1 2 Vapour flow-rate to VP (kg/h) D 600 700 VP operating temperature (°C) E 100 140 Permeate pressure (kPa) F 1 3 Feed temperature (°C) G 20 80 Column pressure (kPa) H 101.3 110 Dummy I,J,K - - Table 3: PB design matrix and response values Run order A B C D E F G H I J K TAC (€/t) 1 1 1 -1 1 1 1 -1 1 1 -1 -1 56.53 2 -1 1 1 -1 1 1 1 -1 -1 1 -1 50.06 3 1 -1 1 1 -1 1 1 -1 -1 -1 -1 77.61 4 -1 1 -1 1 1 -1 1 -1 -1 -1 1 50.25 5 -1 -1 1 -1 1 1 -1 1 1 -1 1 51.34 6 -1 -1 -1 1 -1 1 1 1 1 1 1 77.79 7 1 -1 -1 -1 1 -1 1 1 1 1 -1 55.15 8 1 1 -1 -1 -1 1 -1 -1 -1 1 1 72.32 9 1 1 1 -1 -1 -1 1 1 1 -1 1 54.36 10 -1 1 1 1 -1 -1 -1 1 1 1 -1 61.32 11 1 -1 1 1 1 -1 -1 -1 -1 1 1 59.15 12 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 59.60 A-K are symbols shown in Table 2. 2.3 Steepest Ascent Method The initial estimate of the optimal conditions was far from the actual optimum. Steepest ascent method is a simple and efficient procedure for moving the region of a response in the direction of the maximum change toward the optimum. The zero level of PB design was identified as the basic point of steepest ascent path. Experiments were conducted along with the steepest ascent path until the response showed no further increase. This point would be near the optimal point and could be selected as the center point of BBD (Zhou et al., 2011). 2.4 Response surface methodology The optimal levels of the significant factors and the interactions of these variables on TAC were analysed by BBD. In this study, a three-factor, three-level BBD with 15 runs was employed. The second-order model used to fit the response to the independent variables was shown in Eq(1): ji i j i k i k i i XXXXY    ij 2 1 ii 1 i0 ββββ (1) where Y is the predicted response; Xi and Xj are input variables that influence the response Y ; k is the number of variables; β0 is the constant term; βi is the linear coefficient; βii is the quadratic coefficient and βij is the interactive coefficient. Design-Expert 7.1.3 was used for the experimental designs as well as for regression and graphical analysis of the experimental data obtained. 1174 3. Results and Discussions 3.1 Screening of significant parameters by PB design The data listed in Table 3 indicated that for HDCM process there was a wide variation in the TAC from 50.06 €/t to 77.79 €/t in the 12 trials. This variation suggested that medium and culture condition optimisation were important for reducing TAC. Partial regression coefficients and analyses of their significance were evaluated for determination of the effects on TAC shown in Table 4. The analysis of the regression coefficients and contribution of 11 factors showed that A, D, F, G, J, K and L had positive effects on TAC, whereas B, C, E and H had negative effects on TAC for both HDCM process. Variables with significant effect were E > F > D > A > B > G > C > L > J> K> H for HDCM. The first three variables with significant effect E(VP operating temperature), F (permeate pressure) and D(vapour flow-rate to VP) were selected for further optimisation to obtain a minimum response for HDCM process. Table 4: Partial regression coefficients and analyses of their significance Factors Stdized Effects Sum of squares Contri- bution Signifi- cance Factors Stdized Effects Sum of squares Contri- bution Signifi- cance A 4.42 58.65 4.86 4 G (°C) 3.35 33.63 2.78 6 B -3.59 38.56 3.19 5 H (kPa) -0.038 0.0044 0.00037 11 C -2.63 20.67 1.71 7 J 1.18 4.16 0.34 9 D (kg/h) 7.28 159.07 13.17 3 K 0.31 0.27 0.02 10 E (°C) -12.79 490.63 40.63 1 L 2.74 22.52 0.93 8 F (kPa) 9.67 280.43 23.22 2 3.2 Optimisation by steepest ascent path experiment Based on the above regression analysis, the path of steepest ascent was started from the zero level of the PB design and moved along the direction in which E increased and D, F decreased to reduce TAC. Table 5 illustrated the changing directions of the three variables. The minimum TAC was at the third order and no further improvement could be achieved in the response. The parameters were vapour flow-rate to VP of 620 kg/h, VP’s operating temperature of 120 °C, permeate pressure of 1.5 kPa. It suggested that these points were near the optimal points and then were chosen for further optimisation, taken as the 0 level in RSM. Table 5: Experimental design of steepest ascent and experimental data Run order D (kg/h) F (kPa) E (°C) TAC (€/t) Run order D (kg/h) F (kPa) E (°C) TAC (€/t) 1 580 2.5 100 65.39 4 640 1 130 53.86 2 600 2 110 55.85 5 660 0.5 140 55.04 3 620 1.5 120 53.79 3.3 Optimisation by Box-Behnken design Based on the PB design and the method of the steepest ascent, three variables (vapour flow-rate to VP, VP operating temperature and permeate pressure) were used to determine the optimum levels of these parameters. The design matrix of the variables was given in Table 6, along with the experimental values of the response. The mathematical model for HDCM process representing the TAC as a function of the independent variables within the region under investigation was expressed by the following Eq(2) based on the simulation data of Table 6: 222 35.10047.000012.013.00063.000032.094.1633.118.032.53 FEDEFDFDEFEDTAC  (2) The statistical significance of Eq(2) was confirmed by an F-test, and the analysis of variance (ANOVA) for response surface quadratic models were summarized in Table 7. As can be seen from Table 7, the F-value of 24.63 implied that the model was significant. There is only a 0.13 % chance that a “Model F-value” could occur due to noise. The P-values were used to check the significance of each variable, and the smaller P-value is, the bigger the significance of the corresponding variable is. Values of “Prob>F” less than 0.05 indicated that the model fitness was significant. According to the P-values, the model terms D, E, F, EF and E 2 were found significant. The model determination coefficient R 2 suggested that the fitted model could explain 97.79 % of the total variation. Those implied the regression model was very reliable for IPA purification in the present study from the above analysis. By solving the Eq(2), the optimal values of the parameters were listed in Table 8. It was found that the minimum TAC was 48.54 €/t IPA for HDCM. 1175 Table 6: BBD and simulation values Run order D (kg/h) F (kPa) E (°C) TAC (€/t) Run order D (kg/h) F (kPa) E (°C) TAC (€/t) 1 580 1.5 100 56.65 9 620 0.5 100 56.39 2 660 1.5 100 61.37 10 620 0.5 140 52.50 3 580 1.5 140 49.06 11 620 2.5 100 66.60 4 660 1.5 140 54.80 12 620 2.5 140 52.54 5 580 0.5 120 50.67 13 620 1.5 120 53.78 6 660 0.5 120 56.92 14 620 1.5 120 53.78 7 580 2.5 120 53.47 15 620 1.5 120 53.78 8 660 2.5 120 58.72 Table 7: ANOVA for the regression quadratic model equation of BBD Source Sum of Squares df Mean Square F Value P Value (Prob>F) Comments Model 262.12 9 29.12 24.63 0.0013 significant D (kg/h) 60.34 1 60.34 51.02 0.0008 significant E (°C) 128.87 1 128.87 108.96 0.0001 significant F (kPa) 27.56 1 27.56 23.3 0.0048 significant DE (°C·kg/h) 0.26 1 0.26 0.22 0.6598 not significant DF (kPa·kg/h) 0.25 1 0.25 0.21 0.6638 not significant EF (°C·kPa) 25.83 1 25.83 21.84 0.0055 significant D 2 (kg 2 /h 2 ) 0.13 1 0.13 0.11 0.7561 not significant E 2 (°C 2 ) 12.97 1 12.97 10.96 0.0212 significant F 2 (kPa 2 ) 6.73 1 6.73 5.69 0.0627 not significant Residual 5.91 5 1.18 Lack of Fit 5.91 3 1.97 Pure Error 0 2 0 Cor Total 268.03 14 R 2 =97.79% Table 8: Optimal values for D-VP process with different membranes Item vapour flow-rate to VP (kg/h) VP operating temperature (°C) permeate pressure (kPa) TAC (€/t IPA) Data 580 140 1.5 48.54 3.4 Economic evaluations The TAC of the HDCM system was divided up in operation cost, investment cost and maintenance cost. Table 9 showed the costs for the process investigated. When the operation costs of pre-optimised and optimised D-VP processes with different membranes are compared, the systems optimised have the higher operation costs for HDCM. This may be because the VP process of pre-optimised was performed at operationg temperature of 95 °C and permeate pressure of 2 kPa while the system optimised was performed at operationg temperature of 140 °C and permeate pressure of 1.5 kPa. Due to the increased requirement of heating steam as the desired temperature to VP increased, and more requirement of cooling energy as permeate pressure decreased, the operation costs increased. For the HDCM pre- optimised, membrane area of 25 m 2 was needed, and for the system optimised, membrane area of 12 m 2 was needed. The different investment and maintenance costs between HDCM pre-optimised and optimised is due to the fact that more membrane area is required with VP systems pre-optimised than that optimised. 4. Conclusions In the present study, RSM was used to optimise the hybrid D-VP process parameters for IPA purification with HDCM process, TAC as the response value. Optimal parameters were the vapour flow-rate to VP of 580 kg/h, VP operating temperature of 140 °C, permeate pressure of 1.5 kPa, the minimum TAC predicted was 48.54 €/t. The HDCM pre-optimised and optimised were simulated in Unisim Design, it can be concluded that the TAC of optimal HDCM system is 49.06 €/t, about 19.80 % less than that of pri- optimised system. 1176 Table 9: Economic evaluation for D-VP process Costs pre-optimised optimised Cooling water costs (€/y) 6,017 7,267 Steam costs (€/y) 45,867 50,844 Permeate condensing costs (€/y) 255 404 Total operation costs (€/y) 52,139 58,515 Specific operation costs (€/t IPA) 14.47 16.23 Distillation columns (€/y) 60,851 60,851 VP unit (€/y) 15,924 7,292 Total investment costs (€/y) 76,775 68,087 Specific investment costs (€/t IPA) 21.31 18.88 Membrane replacement (€/y) 71,238 32,368 Maintenance installations (€/y) 20,204 17,917 Total maintenance costs (€/y) 91,442 50,285 Specific maintenance costs (€/t IPA) 25.39 13.95 Specific TAC (€/t IPA) 61.17 49.06 Acknowledgements The authors acknowledge the financial support from the National Natural Science Foundation of China (21206014 and 21125628), the Fundamental Research Funds for the Central Universities (DUT14LAB14) and Funded Project of the China Petroleum and Chemical Corporation (X514001). References Cen Y., Lichtenthaler R.N., 1995, Membrane Separations Technology-Principles and Applications, Membr. Sci. Technol., 2, 85-112. Conto L.C., De Souza J., Veeck A.P.L., Ponce G.H.S.F., Schmiele M., 2015, Sensory properties evaluation of pine nut (araucaria angustifolia) cereal bars using response surface methodology, Chemical Engineering Transactions, 44, 115-120. Dahmoune F., Moussi K., Remini H., Belbahi A., Aoun O., Spigno G., Madani K., 2014, Optimization of Ultrasound-Assisted Extraction of Phenolic Compounds from Citrus sinensis L. Peels using Response Surface Methodology, Chemical Engineering Transactions, 37, 889-894. Hoof V., Abeele V.D., Buekenhoudt A., Dotremont, C., Leysen, R., 2004, Economic comparison between azeotropic distillation and different hybrid systems combining distillation with pervaporation for the dehydration of isopropanol, Sep. Purif. Technol., 37(1), 33-49. Koczka K., Mizsey P., Fonyo Z., 2007, Rigorous modeling and optimisation of hybrid separation processes based on pervaporation, Central Eur. J. Chem., 5, 1124-1147. Pina M.P., Arruebo M., Arruebo M., Felipe M., Fleta F., Bernal M.P., Coronas J., Menéndez M., Santamaría J., 2004, A semi-continuous method for the synthesis of NaA zeolite membranes on tubular supports, J. Membr. Sci., 244(l-2), 141-150. Szitkai Z., Lelkes Z., Fonyo E., Rev Z., 2002, Optimisation of hybrid ethanol dehydration systems, Chem. Eng. Process., 41(7), 631-646. Singh A., Tuteja S., Singh N., Bishnoi N.R., 2011, Enhanced saccharification of rice straw and hull by microwave-alkali pretreatment and lignocellulolytic enzyme production, Bioresource Technology, 102, 1773-1782. Naidu Y., Ranjan K.M., 2011, A generalized methodology for optimal configurations of hybrid distillation– pervaporation processes, Chem. Eng. Res. Des., 89(8), 1348-1361. Zhou J., Yu X., Ding C., Wang Z., Zhou Q., Pao H., Cai W., 2011, Optimisation of phenol degradation by Candida tropicalis Z-04 using Plackett-Burman design and response surface methodology, J. Environ. Sci., 23 (1), 22-30. http://www.sciencedirect.com/science/journal/02552701/41/7