Optimization of electrochemical ammonia removal using Box–Behnken design
Miao Li,Chuanping Feng,Zhenya Zhang,Xiang Liu,Weifang Ma,Qiang Xue,Norio Sugiura
DOI: https://doi.org/10.1016/j.jelechem.2011.03.012
IF: 4.598
2011-01-01
Journal of Electroanalytical Chemistry
Abstract:Electrochemical oxidation of ammonia in an undivided cell was studied in the present experiments. The optimization of the influencing factors on electrochemical removal of ammonia by response surface methodology (RSM) was also studied. Ammonia could be effectively removed by electrochemical method, and which was mainly oxidized into harmless nitrogen gas. The Box–Behnken design can be employed to develop mathematical models for predicting electrochemical ammonia removal geometry. The desired electrochemical ammonia removal can be achieved by choosing the predicted conditions using the developed models. In the present experiments, the optimal NaCl dosage, current density and electrolysis time are 0.31 g/L, 42.75 mA/cm 2 , and 100.64 min, respectively, at which the ammonia-N and nitrate-N concentration in the treated solution are 0 and 1.1 mg/L, respectively, which will meet the standards for drinking water. Keywords Optimization Box–Behnken design Ammonia Electrochemical oxidation Sodium chloride 1 Introduction Among the water pollution, ammonia contamination of water bodies is a widespread environment problem [1,2] . Ammonia promotes eutrophication, which is fatal to aquatic life and a hindrance to the disinfection of water supplies; emits an offensive smell; and aids in carcinogenesis [3,4] . Generally, physicochemical and biological treatment methods were employed for ammonia removal from domestic and industrial wastewaters, which are biological process, air stripping, ion exchange, and breakpoint chlorination. Although air stripping and ion exchange are widely used, they transfer contaminants from one medium to another. There is no destruction of the contaminant. Consequently, it needs to be further treated [5,6] . Breakpoint chlorination demands a large amount of chlorine and relatively inefficient [7] . Biological method is effective, but it requires the addition of carbon sources, large land areas and significantly impaired by low temperature in winter [8] . As electrochemical ammonia oxidation has high treatment efficiency, no sludge production, small area occupied by the plant and relatively low investment costs, a large number of researchers have focused on it recently [9–17] . During electrolysis, the ammonia is mainly reduced to nitrogen, which is further electrochemically inactive. Kim et al. [10] studied the effects of the pH and the chloride ion in the solution on the electrolytic decomposition of ammonia. They reported that the performances of the electrode were totally in the order of RuO 2 ≈ IrO 2 > Pt in both the acid and alkali conditions. Vanlangendonck et al. [9] found that chloride ion concentration and current density are key factors for ammonia removal. Although nitrogen was the most favorable reduction product, in general, during the electrochemical oxidation of ammonia, one of the problems is the generation of nitrate, which needs to be further treated. As an efficient experimentation technique, response surface methodology (RSM) has been applied in a wide range of fields including drug and food industry, chemical and biological processes, etc. [18,19] . It is well known in assessing the effect of parameters on treatment results [20] . Moreover, RSM has been successfully applied to different processes for optimization the experimental design, such as O 3 oxidation [21] and TiO 2 -coated/UV oxidation [22–24] . However, as our best knowledge, the application of RSM in electrochemical oxidation of ammonia is not yet reported. As in previous studies, the IrO 2 related anodes have been proved to be relatively efficient for ammonia removal [9,10,25] , and the combinations of Ti/IrO 2 anode and Cu–Zn cathode have never been studied, those combinations were employed in the present experiments. As our best knowledge, although the application of RSM in electrochemical treatment landfill of leachate in a three-dimensional electrochemical reactor has been reported [26] , the application of RSM only focusing on the removal of ammonia by electrochemical method in an undivided cell has never been studied. Moreover, the combination of Ti/IrO 2 as the anode and Cu–Zn as the cathode for this application has not yet studied. Therefore, the aim of this work is to completely remove ammonia and removal of by-products in an undivided cell; and to find a proper condition to perform both anodic oxidation of ammonia and cathodic reduction of the formed nitrate, and to optimize the influencing factors on electrochemical oxidation of ammonia by RSM. A laboratory scale electrochemical reactor was employed for an artificial ammonia contaminated wastewater treatment. As current density, chloride ion concentration (NaCl dosage), and electrolysis time was the most important factors for electrochemical ammonia oxidation [9,11] , those factors were selected as three influencing variables in Box–Behnken design (BBD). RSM was used to determine the optimal condition and an empirical model correlating the ammonia removal efficiency to the three variables was then developed. 2 Materials and methods 2.1 Design of experiments The RSM is efficient for optimization of multiple variables with minimum number of experiments by using a set of designed experiments to obtain an optimal response [27–30] . Box–Behnken design is an independent, rotatable quadratic design with no embedded factorial or fractional factorial points where the variable combinations are at the midpoints of the edges of the variable space and at the center. In the present experiments, by employing the Box–Behnken statistical experiment design and the RSM [31] , the effects of the three independent variables on the response functions were investigated and to determine the optimal conditions maximizing the percent ammonia removal and minimizing the percent nitrate production and power consumption. Among all the RSM designs, Box–Behnken design requires fewer runs (e.g., 17 runs for a three-parameter experimental design) and allows calculations of the response function at intermediate levels and enables estimation of the system performance at any experimental point within the range studied through careful design and analysis of experiments [32] . The independent variables were the dose of NaCl ( X 1 ), current density ( X 2 ), and time ( X 3 ). The low, center and high levels of each variable are designated as −1, 0, and +1, respectively, as shown in Table 1 . The dependent variables (or objective functions) were the ammonia ( Y 1 ) and nitrate ( Y 2 ) concentrations. The optimization procedure involves studying the response of the statistically designed combinations, estimating the coefficients by fitting the experimental data to the response functions, predicting the response of the fitted model and checking the adequacy of the model. 2.2 Batch electrolysis The cylindrical undivided electrochemical cell ( Fig. 1 ) with a net working of 400 mL volume was designed. For the cell, Ti/IrO 2 (TohoTech company, Japan) plate of 75 cm 2 (15 cm × 5 cm) was used as the anode and Cu–Zn electrode with the same area as the cathode, a distance of 8 mm between the cathode and anode was set. The immersed areas of the anode and cathode in the treated solution were the same at 40 cm 2 . A DC power supply with a voltage range of 0–240 V and a current range of 0–25 A was employed. In the present study, synthetic ammonia solutions with concentration of 100.0 mg/L ammonia-N ((NH 4 ) 2 SO 4 ) was prepared for the electrolysis experiments. To investigate the effect of sodium chloride (NaCl) dosage on the complete ammonia oxidation, the NaCl of 1 to 5.0 g/L (w/v) (the exactly concentration of NaCl was described in each figure) were added into the synthetic ammonia solutions, respectively. 0.50 g/L Na 2 SO 4 was added into all the experiments in order to enhance the conductivity. Three hundred milliliter of synthetic ammonia solution prepared as above was poured into the electrochemical cell, the reaction started with the application of specified current density. At different intervals, 1.5 mL of sample was drawn from the electrochemical cell by a chromatography syringe for analysis. The electrolysis was carried out at the room temperature (25 ± 2 °C) and uncontrolled. Box–Behnken statistical experiment design and the RSM [31] were used to investigate the effects of the three independent variables on the response functions and to determine the optimal conditions maximizing the percent ammonia removal and minimizing the percent nitrate production and power consumption. The optimization procedure involves studying the response of the statistically designed combinations, estimating the coefficients by fitting the experimental data to the response functions, predicting the response of the fitted model and checking the adequacy of the model. The independent variables were the dose of NaCl ( X 1 ), current density ( X 2 ), and time ( X 3 ). The low, center and high levels of each variable are designated as −1, 0, and +1, respectively, as shown in Table 1 . The dependent variables (or objective functions) were the ammonia ( Y 1 ) and nitrate ( Y 2 ) concentrations. 2.3 Analysis All analyses were done according to standard methods [33] . The determination of ammonia was performed by Ion meter (Ti 9001, Toyo chemical laboratories Co., Ltd.). Nitrate was determined by standard colorimetric method using spectrophotometer (DR/4000U Spectrophotometer, USA) and ion chromatography (Yokogawa IC7000, AS9-HC column). Nitrite was analyzed by ion chromatography (Yokogawa IC7000, AS9-HC column). Total nitrogen (T-N) was determined with a T-N, T-P auto analyzer (Auto Analyzer 3, Bran + Luebbe). 3 Results and discussion Table 2 shows the values of the dependent and independent variables and the observed and predicted experimental data for ammonia-N and nitrate-N concentration for every experiment. The center point (0, 0, 0) was repeated five times and nearly the same results were obtained indicating the reproducibility of the data. 3.1 The response function coefficients The mathematical relationship between the response function ( Y ) and the independent variables ( X ) can be approximated by a quadratic polynomial equation as follows: (1) Y = b 0 + b 1 X 1 + b 2 X 2 + b 3 X 3 + b 12 X 1 X 2 + b 13 X 1 X 3 + b 23 X 2 X 3 + b 11 X 21 + b 22 X 22 + b 33 X 2 The application of RSM offers an empirical relationship between the response function and the independent variables. The coefficients of the response functions for different dependent variables were determined correlating the experimental results with the response functions by using a Design-expert (Stat-Ease, trial version) regression program. Eqs. (2) and (3) shows the response functions with the determined coefficients for ammonia removal and nitrate production (2) Y ammonia-N = 175.11042 - 28.59375 X 1 - 3.73125 X 2 - 1.33111 X 3 + 0.30187 X 1 X 2 + 0.10250 X 1 X 3 + 0.014000 X 2 X 3 + 1.08750 X 1 2 + 0.019313 X 2 2 + 0.002407 X 3 2 ( R 2 = 0.9923 ) (3) Y nitrate-N = - 27.38611 + 0.79458 X 1 + 1.21196 X 2 + 0.41796 X 3 - 0.079375 X 1 X 2 - 0.030556 X 1 X 3 - 0.002778 X 2 X 3 - 0.23313 X 1 2 - 0.017206 X 2 2 - 0.002498 X 3 2 ( R 2 = 0.9356 ) It can be seen from the coefficients in Eqs. (2) and (3) that ammonia oxidation rate increases with the NaCl dosage ( X 1 ), current density ( X 2 ) and time ( X 3 ). The NaCl dosage has a more profound effect on ammonia oxidation rate as compared to current density ( X 2 ) and time ( X 3 ). This is in agreement with previous studies [9,11] . According to the study by Kim et al. [11] , the ammonia oxidation rate increased with the concentration of the chloride ion in the solution, especially the chloride ion concentration lower than 10.0 g/L. The nitrate-N formation rate increases with the NaCl dosage ( X 1 ), the current density ( X 2 ) and time ( X 3 ), in which a relatively more profound effect by the current density can be observed. It can be seen from Table 2 that higher concentration of NaCl and higher current density result in more amount of nitrate to be produced. In order to reduce the generation of by-product nitrate, the appropriate NaCl concentration and current density should be selected. During electrochemical oxidation of ammonia, chlorine gas can be generated in the presence of chloride ion [34] . (4) 2 Cl - → Cl 2 + 2 e - Then the hypochlorous acid (HClO) will be formed due to the generated molecular chlorine is easily hydrolyzed, which is consecutively changed to hypochlorite ion (ClO − ) [9,10,34] . (5) Cl 2 + H 2 O → HClO + H + + Cl - (6) HClO → ClO - + H + The hypochlorous acid and hypochlorite ion have high oxidation ability which can oxidize the ammonia to nitrogen gas or nitrate [1,9,10] . (7) 2 NH 4 + + 3 HClO → N 2 + 3 H 2 O + 5 H + + 3 Cl - (8) 2 NH 4 + + 3 ClO - → N 2 + 3 H 2 O + 2 H + + 3 Cl - (9) NH 4 + + 4 HClO → NO 3 - + H 2 O + 6 H + + 4 Cl - (10) NH 4 + + 4 ClO - → NO 3 - + H 2 O + 2 H + + 4 Cl - If the nitrate ions are formed during the oxidation of ammonia, they can be re-reduced to nitrite, nitrogen, and even to ammonia at the cathode [25] , but because of the presence of hypochlorous acid and hypochlorite ion, the nitrite or ammonia could be quickly oxidized again. Therefore, in the present experiments, no nitrite was detected in the solution and the amount of total nitrogen was equal to that of the ammonia nitrogen and nitrate nitrogen. (11) NO 3 - + H 2 O + 2 e - → NO 2 - + 2 OH - (12) NO 3 - + 3 H 2 O + 5 e - → 1 / 2 N 2 + 6 OH - (13) NO 3 - + 6 H 2 O + 8 e - → NH 3 + 9 OH - In all, during the electrolysis, the large amount of hypochlorite ions produced around the Ti/IrO 2 anode would oxidize the ammonia to nitrogen or nitrate, while around the cathode, nitrate would be reduced to nitrite or ammonia. The response function predictions were in good agreement with the experimental data. The analysis of variance (ANOVA) tests were also conducted for each response indicating the fact that the predictability of the model is at 95% confidence level. The F -test in the ANOVA analysis is used for comparing model variance with residual (error) variance. If the variances are almost the same, the ratio will be nearly one indicating that it is less likely any of the factors have a significant effect on the response. If the Prob > F -value is very small (less than 0.05) then lack of fit is significant. p -Values in the test indicate probability of seeing the observed F -value if the null hypothesis is true. Small probability values call for rejection of the null hypothesis. The data obtained from Eqs. (2) and (3) are significant. It is verified by F -value and the analysis of variance (ANOVA) by fitting the data of all independent observations in response surface quadratic model. Significance of each coefficient of Eq. (2) was determined by applying t -test and p -values of each are listed in Table 3 . Since the p -values of the all the coefficients are p < 0.05, it implies that these are significant. The linear effect of coefficients X 1 , X 2 , and X 3 , i.e., NaCl dosage ( p < 0.0001), current density ( p < 0.0001) and time ( p < 0.0001) is significant. Similarly, the interactive effects of NaCl dosage and current density ( p < 0.0001), NaCl dosage and time ( p < 0.0001), and current density and time ( p < 0.0001) are also significant. p -Values of the quadratic terms, i.e., NaCl dosage ( X 1 2 ) ( p < 0.0071), current density ( X 2 2 ) ( p < 0.0003) and time ( X 2 2 ) ( p < 0.0040) are significant. Thus, statistical analysis of data shows that small variations in the values of the selected variables alter the ammonia removal efficiency. In this model X 1 ; X 2 ; X 3 ; X 1 X 3 ; X 1 X 2 ; X 2 X 3 ; X 1 2 ; X 2 2 ; and X 3 2 are significant model terms. Analysis of variance (ANOVA) for response surface quadratic model gave F -value 100.49, R 2 value 0.9923, probability < 0.0001 and coefficient of variation (CV = 29.81%) signifying that model is highly significant and experiments are highly accurate and reliable. Significance of each coefficient of Eq. (3) was determined by applying t -test and p -values of each ( Table 4 ). Since the p -values of the all the coefficients, except for X 3 , X 1 X 2 , X 1 X 3 , X 2 X 3 , are p < 0.05, it implies that these are significant. The linear effect of coefficients NaCl dosage X 1 ( p < 0.0002) and current density X 2 ( p < 0.0096) are significant, while time X 3 ( p < 0.0907) is insignificant, which means that the production of nitrate is more related to the NaCl dosage and current density. The coefficients of the qlinear term X 1 had negative effects. The interactive effects of NaCl dosage and current density ( p < 0.1002), NaCl dosage and time ( p < 0.1450), and current density and time ( p < 0.1796) are insignificant. p -Values of the quadratic terms, i.e., NaCl dosage( X 1 2 ) ( p < 0.5862) is insignificant, but current density ( X 2 2 ) ( p < 0.0040) and time ( X 2 2 ) ( p < 0.0175) are significant. Analysis of variance (ANOVA) for response surface quadratic model gave F -value 11.29, R 2 value 0.9356, probability < 0.0021 and coefficient of variation (CV = 26.24%) signifying that model is significant and experiments are accurate and reliable. The shape of the standard error plot fits on the design points, and polynomial being fit showed low and flat error exhibiting circular contours and symmetrical shape around the centroid, which is the ideal condition ( Fig. 1 ). A base standard deviation of 1.0 is used to generate the standard error plot for design evaluation. The actual magnitude of the plot will be a function of the standard deviation, which depends on the response data. The standard error value around the centroid is 0.5882, which is the best value. Standard error value increases at the centroid as well as away from optimization point. Fig. 2 show response surface plots for the relationship between NaCl dosage, current density and time on the removal of ammonia. Fig. 3 show response surface plots for the relationship between NaCl dosage, current density and time on the generation of nitrate. 3.2 Optimization using desirability functions The main objective of optimization was to minimize the final ammonia-N and nitrate-N to the allowing concentrations with recalculating all responsible factors by using desirability functions. The numerical optimization of the software has been chosen in order to find the specific point that maximizes the desirability function. The desired goal was selected by adjusting the weight or importance that might alter the characteristics of a goal. The goal fields for response have five options: none, maximum, minimum, target and within range. The goal for NaCl dosage, current density and time is assigned as the same of ‘minimize’ with corresponding ‘weight’ 0.1. The goal for ammonia-N and nitrate-N concentrations is assigned as the same of ‘minimize’ with corresponding ‘weight’ 1.0. The lower limit values of NaCl dosage, current density and time is assigned as the same of 0. The lower limit values of ammonia-N and nitrate-N concentrations are assigned as the same of 0. The optimization procedure was conducted under these settings and boundaries. Our main objective was to maximize the removal% with recalculating all responsible factors by using desirability functions The individual desirability functions ( di ) for each of the responses, and the calculated geometric mean as maximum over all desirability ( D = 0.958) is represented in Fig. 4 . For the optimization of the process parameters, the point prediction option in the software is used. For their validation, duplicate confirmatory experiments were conducted using the optimized parameters. It is found that the optimal NaCl dosage, current density and electrolysis time are 0.31 g/L, 42.75 mA/cm 2 , and 100.64 min, min, respectively, at which the predicted ammonia-N and nitrate-N concentration are 0 and 0 mg/L, respectively, as against the actual ammonia-N and nitrate-N concentration of 0and 1.1 mg/L. Thus, the single electrochemical treatment of ammonia will meet the standards for drinking water. The results are closely related with the data obtained from optimization analysis using desirability functions, indicating Box–Behnken design in corporate with desirability functions could be effectively used to optimize the design for experiments on the electrochemical ammonia oxidation. 4 Conclusions The following conclusions are drawn from this investigation: (1) The Box–Behnken design can be employed to develop mathematical models for predicting electrochemical ammonia removal geometry. (2) The desired electrochemical ammonia removal can be achieved by choosing the predicted conditions using the developed models. In the present experiments, it is found that the optimal NaCl dosage, current density and electrolysis time are 0.31 g/L, 42.75 mA/cm 2 , and 100.64 min, respectively. (3) Parameters interacting together can be identified in such a typical process electrochemical removal. The ammonia removal is sensitive to the NaCl dosage and current density in the present study. Acknowledgment The authors thank the Water Pollution Control and Management Project of China ( 2008ZX07313-001 ) and the Fundamental Research Funds for the Central Universities for the financial support of this work. References [1] S.K. Bae S.C. Park J. Korean Soc., Environ., Eng. 6 1984 44 55 [2] E.J. Bouwer P.B. Crowe J. AWWA 80 1988 82 93 [3] A.T. Palin Water Water Eng. 54 1950 151 160 [4] S.H. Lin C.L. Wu Water Res. 30 1996 715 721 [5] C. Feng N. Sugiura S. Shimada T. Maekawa J. Hazard. Mater. B 103 2003 65 78 [6] T.L. Huang J.M. Macinnes K.R. Cliffe Water Res. 35 2001 2113 2120 [7] J.N. Jensen J.D. Johnson J. AWWA 81 1989 59 64 [8] V. Mateju S. Cizinska J. Krejci J. Tomas Enzyme Microb. Technol. 14 1992 170 183 [9] Y. Vanlangendonck D. Corbisier A. Lierde Water Res. 39 2005 3028 3034 [10] K.-W. Kim Y.-J. Kim I.-T. Kim G.-I. Park E.-H. Park Electrochim. Acta 50 2005 4356 4364 [11] K.-W. Kim Y.-J. Kim I.-T. Kim G.-I. Park E.-H. Lee Water Res. 40 2006 1431 1441 [12] L. Li Y. Liu J. Hazard. Mater. 161 2009 1010 1016 [13] F. Tanaka C.P. Feng N. Sugiura T. Maekawa Jpn. J. Water Treat. Biol. 39 2003 75 84 [14] K. Bouzek M. Paidar A. Sadilkova H. Bergmann J. Appl. Electrochem. 31 2001 501 507 [15] M. Li C. Feng Z. Zhang Z. Shen N. Sugiura Electrochem. Commun. 11 2009 1853 1856 [16] T.A. Pressley D.F. Bishop S.G. Roan Environ. Sci. Technol. 6 1972 622 628 [17] M. Li C. Feng W. Hu Z. Zhang N. Sugiura J. Hazard. Mater. 162 2009 455 462 [18] M. Meilgaard G.V. Civille B.T. Carr Advanced Statistical Methods, Sensory Evaluation Techniques second ed. 1991 CRC Press Boca Raton, FL pp. 275–304 [19] M. Otto Statistics and Computer Applications in Analytical Chemistry 1999 Wiley-VCH Chichester [20] D.C. Montgomery Design and Analysis of Experiments third ed. 1991 Wiley New York pp. 270–569 [21] W. Zhao Z. Wu D. Wang J. Hazard. Mater. 137 2006 1859 1865 [22] A. Danion C. Bordes J. Disdier J.V. Gauvrit C. Guillard P. Lanteri J.-R. Nicole J. Photochem. Photobiol. A: Chem. 168 2004 161 167 [23] N. Daneshvar M.H. Rasoulifard A.R. Khataee F. Hosseinzadeh J. Hazard. Mater. 143 2007 95 101 [24] A. Aguedach S. Brosillon J. Morvan E.K. Lhadi J. Hazard. Mater. 150 2008 250 256 [25] M. Li C. Feng W. Hu Z. Zhang N. Sugiura Electrochim. Acta 54 2009 4600 4606 [26] H. Zhang Y. Li X. Wu Y. Zhang D. Zhang Waste Manage. 30 2010 2096 2102 [27] S.H. Krishna B. Manohar S. Divakar S.G. Prapulla N.G. Karanth Enzyme Microb. Technol. 26 2000 131 136 [28] A. Vohra T. Satyanarayana Process Biochem. 37 2002 999 1004 [29] F. Francis A. Sabu K.M. Nampoothiri S. Ramachandran S. Ghosh G. Szakacs A. Pandey Biochem. Eng. J. 15 2003 107 115 [30] F. Ay E.C. Cokay F. Kargi J. Hazard. Mater. 162 2009 230 236 [31] R.H. Charles V.T. Kennneth Fundamental Concepts in the Design of Experiments 1999 University Press Oxford [32] E. Hamed A. Sakr J. Control. Release 73 2001 329 338 [33] APHA, AWWA, WPCF, Standard Methods for the Examination of Water and. Wastewater, twentieth ed., American Public Health Association, Washington, DC, USA, 1998. [34] K. Rajeswar J.G. Ibanez 1997 Academic Press San Diego