A modeling study by response surface methodology (RSM) and artificial neural network (ANN) on Cu2+ adsorption optimization using light expended clay aggregate (LECA)

A modeling study by response surface methodology (RSM) and artificial neural network (ANN) on Cu2+ adsorption optimization using light expended clay aggregate (LECA)

G Model JIEC-1405; No. of Pages 11 Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx Contents lists available at SciVerse ScienceDi...

2MB Sizes 0 Downloads 2 Views

G Model

JIEC-1405; No. of Pages 11 Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Journal of Industrial and Engineering Chemistry journal homepage: www.elsevier.com/locate/jiec

A modeling study by response surface methodology (RSM) and artificial neural network (ANN) on Cu2+ adsorption optimization using light expended clay aggregate (LECA) Tahereh Shojaeimehr a, Farshad Rahimpour a,*, Mohammad Ali Khadivi a, Marzieh Sadeghi b a b

Biotechnology Research Laboratory, Chemical Engineering Department, Razi University, Kermanshah 67149-67346, Iran Analytical Chemistry Laboratory, Chemistry Department, Faculty of Chemistry, Razi University, Kermanshah, Iran

A R T I C L E I N F O

Article history: Received 16 January 2013 Accepted 9 June 2013 Available online xxx Keywords: Response surface methodology (RSM) Artificial neural network (ANN) Copper adsorption LECA

A B S T R A C T

In the present study, response surface methodology (RSM) and artificial neural network (ANN) were used to develop an approach for the evaluation of heavy metal adsorption process. LECA was used as a green and low cost adsorbent to remove Cu2+ from aqueous solution in batch system. The effect of the operational parameters such as initial pH, temperature, initial Cu2+ concentration, and sorbent dosage was studied using Central Composite Face (CCF) design. Same design was also utilized to obtain a training set for ANN. A comparison between the model results and experimental data gave a high correlation coefficient (R2ANN ¼ 0:962, R2RSM ¼ 0:941) and showed that two models were able to predict Cu2+ removal by LECA. The Langmuir and Freundlich isotherm models were applied to the equilibrium data at different temperatures. The results revealed that the Freundlich isotherm fitted better than the Langmuir isotherm. The Cu2+ adsorption kinetic was well described by the pseudo-second order kinetic model. The rate of Cu2+ removal was controlled by film diffusion and intra-particle diffusion. The thermodynamic studies proved that Cu2+ removal was physical, spontaneous, feasible, endothermic, and random process. ß 2013 The Korean Society of Industrial and Engineering Chemistry. Published by Elsevier B.V. All rights reserved.

1. Introduction The presence of heavy metal ions in water and wastewater is considered as a major problem for their toxicity, non-biodegradability, and severe damages in human health. Copper emission comes from industrial activities such as printed circuit board manufacturing, electronic plating, wire drawing, copper polishing, paint manufacturing, wood preservatives and printing processes. Cu2+ is an essential component for human. However, it may cause severe damages in unauthorized concentrations. Reproductive and developmental toxicity, neurotoxicity, acute toxicity, dizziness and diarrhea have been reported for copper [1–3]. The World Health Organization (WHO) recommended a maximum acceptable concentration of Cu2+ in drinking water less than 1.5 mg/L. It is

* Corresponding author. Tel.: +98 831 4274530; fax: +98 831 4274542. E-mail addresses: [email protected], [email protected] (F. Rahimpour).

essential that potable waters be given some treatment to remove copper ions before domestic supply [4]. There are several methods developed for removal of heavy metal ions containing precipitation, electrochemical treatment, solvent extraction, membrane filtration [5–8]. With respect to all methods, adsorption is a common and effective method for heavy metal ions removal because of low cost, high efficiency, and good operational conditions. Numerous adsorbents such as zeolite [9], resin [10], activated carbon [11], sludge [12], nano-particles [13– 15], agricultural waste materials [16], and biomass [17] have been studied for the capability of Cu2+ removal from wastewater. In recent decade, many researches were conducted to access sorbents with higher efficiency and lower cost. The cost of adsorbent is an important parameter for comparison of adsorbents. LECA is manufactured worldwide from natural materials. LECA consists of small, strong, extremely lightweight and thermally insulating particles of burnt clay in rotary kiln at around 1200 8C [18]. LECA is an environment-friendly, green and chemically natural product. The most important stages in an environmental process are modeling and optimization to improve a system and increase the

1226-086X/$ – see front matter ß 2013 The Korean Society of Industrial and Engineering Chemistry. Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jiec.2013.06.017

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 2

T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

efficiency of the process without increasing the cost [19]. The mechanism of adsorption processes is complex. This is due to the complex interaction of variables and non-linear behavior of these processes. As a result, determination of optimum experimental condition is very important to obtain maximum efficiency. The classical optimization method (single variable optimization) is not only time-consuming and tedious but also does not depict the complete effects of the parameters in the process and ignores the combined interactions between physicochemical parameters. This method can also lead to misinterpretation of results. To overcome this difficulty, some statistical methods have been used. Recently, response surface methodology (RSM) has attracted great attention as a collection of mathematical and statistical techniques useful for analyzing the effects of several independent variables. RSM assesses the relationships between the response(s) and the independent variables [19], and defines the effect of the independent variables, alone or in combination, in the processes. This method has many advantages like being more economical, needing fewer experiments number, studying interaction between parameters on response, predicting of the response, checking of method adequacy, and requiring shorter time. This process employs low-order polynomial equations in a predetermined region of the independent variables, which will be analyzed to locate the optimum values of independent variables for the best responses [19,20]. The investigation of heavy metal ions adsorption using RSM has earned special importance [4,11,20,21]. Also, the past decade has seen a host of data analysis tools based on biological phenomena develop into well-established modeling techniques, such as artificial intelligence and evolutionary computing. Indeed an artificial neural network (ANN) is a massively interconnected network structure consisting of many simple processing elements capable of performing parallel computation for data processing. This method is useful where the complexity of the mechanisms underlying process performance is high [22]. In recent years artificial neural networks (ANNs) have been widely studied to solve environmental problems because of their reliable and salient characteristics in capturing the non-linear relationships existing between variables [22]. It can be used to solve problems that are not eligible for conventional statistical methods. ANNs have been considered because of wide spread usage and their capability and ability to solve complicated problems. Process modeling and simulation especially while no analytical model exists are the applications of ANNs at chemical engineering [22–25]. Examples for application of AANs in water treatment are included Cu2+ removal by sawdust [23], arsenic removal by the modification of solid waste vegetable oil industry with Fenton reagent (FMSWVOI) [24], Pb(II) adsorption by Antep pistachio (Pistacia VeraL.) shells [25]. Some researchers tried to develop an evaluation of RSM and ANN as two useful methods to predict and simulate removal processes such as the removal of Zn(II) ions from leachate by hazelnut shell [22], and removal of lead ions by Nigella sativa seeds (black cumin) [26]. Both the ANN and RSM techniques were compared for their predictive and generalization capabilities, sensitivity analysis and optimization efficiency in heavy metal removal from wastewater [22,26]. Removal of Cu2+ from aqueous solution by different sorbents has been investigated. However, these sorbents needs to be focused and more accurately investigated. The main objective of this study is to evaluate of LECA adsorption capacity as abundance natural and low cost sorbent in copper removal through the application of RSM and ANNs. A comparison between ANN and RSM techniques was performed for estimating their performance in water treatment

process. Then, the obtained results from two models were compared with the experimental data. Finally, the adsorption kinetic, equilibrium models, and thermodynamic studies as well as LECA behavior as a sorbent in copper removal were explained. 2. Materials and methods 2.1. Materials and characterization Cu(NO3)23H2O (MW = 214.6 g/mol, Merck) was used to prepare the stock solution containing 1000 mg/L of copper ions, which was further diluted with deionized water to the desired initial concentrations. All other materials were in the analytical grade. The Cu2+ concentration in the solution was determined by a flame atomic absorption spectrometer (Analytical Jena NOVAA400 with hollow cathode lamp 4 mA, wave length 324.75 nm, Germany). pH meter (827 pH lab Metrohm, Swiss made), shaker/incubator (DKS1060, Korean) was used to conduct the batch experiments in desired mixing rate, temperature, and time. The zeta potential of LECA was measured by zeta meter (Nano ZS, Malvern, England) to determine surface charge of LECA. The morphological characterization of LECA was determined by a field emission scanning electron microscope (FESEM, Hitachi S-4160, Japan). X-ray fluorescence analysis (XRF, PW1480, Philips, Netherlands) was used to characterize the sorbent. Spectroscopic studies were conducted using FT-IR spectrometer (WQS-510, China). The measurement of the cation exchange capacity (CEC) of LECA is made by saturating the clay with BaCl2 method [27]. 2.2. The preparation of the adsorbent LECA was provided from Azarbayjani Company, Kermanshah, Iran. LECA was crashed and sieved through a 60 mesh size screen. Then it was washed with deionized water for four times to remove dust and undesirable components. LECA was dried in an oven at 120 8C for 2 days (until the weight became constant). 2.3. Batch adsorption experiments Batch adsorption studies were performed for Cu2+ removal by LECA in 250 mL glass flasks. For this purpose, 50–150 mg of LECA was added into 50 mL of Cu2+ solution with the known concentration (C0) ranged between 10 and 150 (mg/L). pH of the solution was adjusted using 1 M HNO3 and 1 M NaOH. The glass beakers were covered with plastic paraffin film. Then it was mounted on a shaker at constant temperature (20–50 8C) and agitated at a constant rate (100 rpm) for 180 min. After reach to equilibrium and separation of the sorbent from the solution, the copper ions content in residual solutions was analyzed using a flame atomic absorption spectrophotometer. The amount of adsorbed copper ions onto LECA (q) was calculated using the following equation (Eq. (1)) [4,15]: q¼

ðC 0  C e ÞV m

(1)

where q is the metal ions uptake (mg metal/g of sorbent), C0 and Cf is the initial and final of concentration Cu2+ in solution (mg/L), V is total solution volume (mL), and m is the amount of the sorbent (mg). 2.4. Response surface methodology (RSM) In the conventional method, one variable changes when all other parameters are at a specified value. The accomplishment of the experiments using conventional method and investigation of interaction between parameters is very time consuming and

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

3

impractical. RSM is a main branch of experimental design. RSM is used to evaluate the effect of several factors and their interaction on the system response. It is a combination of mathematical and statistical methods. This method is useful for developing and optimizing the independent variables and response, and providing the less number of experimental runs. In recent years, the attempts have been expanded for better understanding of adsorption processes and the effects of different parameters on adsorption behavior [28,29]. The most reported applications of RSM are in the cases where a large number of variables affect on the system response [19,28–32]. Based on the above mention, RSM contains three steps: design of experiments; response surface modeling; and optimization [19]. In this research, Central Composite Face (CCF) methodology was selected for the experimental design by MODDE software (version 8.0.2., Sweden). The correlation between response and the selected factors can been expressed by a quadratic equation that is given as [4,21]: Y ¼ b0 þ

n n n X X X bi xi þ bii x2i þ bi j xi x j þ e i¼1

i¼1

(2)

i 6¼ 1

where Y is the predicted response (predicted adsorption capacity), xi, xj are the independent variables in coded levels, bi, bii, bij are the coefficient for linear, quadratic and interaction effect, respectively, b0 is the model coefficient, n is the number of factors (independent variables), and e is the model error. 2.5. Artificial neural networks Artificial neural network (ANN) is a good inspiration of human brain and nerve systems that are known for their extreme ability to learn and classify data [23]. ANNs consist of an input and an output layer and one or more hidden layers. In input and hidden layers, each neuron receives input values. Neurons transfer input values to next layer that the strength of these connections determined by weights [33]. In the present study, different back-propagation (BP) algorithms were checked to select the best BP algorithm with a minimum mean squared error (MSE) and minimum relative error (MRE). The Levenberg–Marquardt back propagation algorithm (LMA) was applied for training of the network as the best algorithm. Also, a three-layer feed forward ANN with a linear transfer function (purelin) at output layer, a tangent sigmoid transfer function (tansig) at hidden layer was developed to predict and simulate LECA adsorption capacity for Cu2+ removal. The training parameters was 4 input nodes, 4 hidden layer neurons, 1 output node, error goal: 0.00001. 3. Results and discussion

Fig. 1. Field emission scanning electron microscope (FESEM) of powdered LECA in two scales.

the surface charge of sorbent in the studied pH range of 3–5 is negative. With respect to FTIR spectra of LECA and LECA loaded with Cu2+ depicted in Fig. 3, the peak in 1343 cm1 may be related to bending vibration frequency of Z–OH–Cu bond. The peak of 1384 cm1 can be referring to the vibration frequency of N–O bond. The N–O bond could be related to adsorbed impurities nitrate which is due to copper nitrate salt. The observed peak in 3548 cm1 could confirm the stretching vibration frequency which related to Z–OH–Cu bonds. The peak in 3548 cm1 could be related to Cu2+ adsorption on LECA surface [34].

3.1. Characteristics of the adsorbent The morphology of LECA surface in two different magnifications is indicated in Fig. 1. It can be implicated a porous structure with abundant porosity that is constituted by very small aggregated components. The results of XRF analysis for determination of LECA characterization are listed in Table 1. From the data tabulated in Table 1. LECA contained significant amount of SiO2 (64.83%), Al2O3 (15.05%), and Fe2O3 (7.45). Non-bonding electron of oxygen atom in SiO2, Al2O3, and Fe2O3 structure can constitute complex with copper ions in the solution. The cation exchange capacity (CEC) of LECA was determined 76.3 meq/100 g using saturating of the clay with BaCl2 method [27]. The surface charge of the solid particles can been determine by zeta potential analysis. Fig. 2 shows the zeta potential of LECA as function of pH which was measured across pH range of 2–10. The isoelectric point of LECA was observed at 2.3. As indicated in Fig. 2,

Table 1 The sorbent characterization. Constituent

wt%

SiO2 Al2O3 Fe2O3 CaO Na2O K2O MgO TiO2 MnO P2O5 SO3 L.O.I

64.83 15.05 7.45 2.98 1.10 2.55 3.67 0.63 0.13 0.13 0.11 1.37

L.O.I, loss of ignition.

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

4 5

pH 0 2

Zeta potentiol (mV)

0

4

6

8

10

-5

-10

-15

-20

-25

Fig. 2. Zeta potential of LECA as a function of pH. Fig. 3. The FT-IR spectra for LECA before and after Cu2+ adsorption.

3.2. The experimental design for adsorption studies The initial copper ions concentration, temperature, initial pH, and adsorbent dosage were chosen as the independent variables based on previous experiments [2–5,12,15,20,21,35,36,39,40]. Cu2+ adsorption capacity (q) was selected as the experiments response. The experimental runs order was conducted in three levels randomly. Twenty seven runs were determined as the experimental design runs. To consider the pure error, the central point was repeated three times. All experiments were conducted in

duplicate. The Cu2+ adsorption performance was investigated with analysis of the response. The coded levels of the variables as the low (1), middle (0) and high (+1) levels of the factors as well as the experimental design and the obtained results are reported in Table 2. 3.2.1. The statistical analysis Since the full quadratic model could not satisfy the necessary requirements, the model was modified by eliminating of additional terms. The resulted approximating function of

Table 2 Independent variables, their domain, and experimental data for CCF. Variables

Symbol

Factor code

pH T C0 m

pH Temperature (8C) Initial concentration (mg/L) Adsorbent dosage (mg) Run

Coded levels XI

XII

XIII

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 0 0 0 0 0 0 0 0 0

1 1 +1 +1 1 1 +1 +1 1 1 +1 +1 1 1 +1 +1 0 0 1 +1 0 0 0 0 0 0 0

0 0 0 0 +1 +1 +1 +1 0 0 0 0 +1 +1 +1 +1 0 0 0 0 1 +1 0 0 0 0 0

XI XII XIII XIV

Level of factors 1

0

+1

3 20 10 50

4 35 80 100

5 50 150 150

Observed adsorption capacity

Predicted adsorption capacity

XIV

q (mg/g)

q (mg/g)

1 1 1 1 1 1 1 1 +1 +1 +1 +1 +1 +1 +1 +1 0 0 0 0 0 0 1 +1 0 0 0

3.780 6.834 6.135 9.604 60.093 84.126 71.978 103.261 2.383 2.854 2.620 3.136 19.043 23.440 28.480 34.153 10.335 23.680 27.036 35.687 4.349 53.754 47.569 16.030 29.058 26.178 25.431

1.418 10.113 3.001 11.697 61.880 85.045 74.845 98.009 3.612 0.388 5.195 1.195 14.328 24.797 27.293 37.762 13.277 22.859 28.785 36.059 3.048 51.563 47.373 18.346 27.306 27.306 27.306

Residual

2.362 3.279 3.134 2.093 1.787 0.919 2.867 5.252 1.229 3.243 2.575 1.941 4.715 1.357 1.187 3.609 2.942 0.821 1.749 0.372 1.301 2.191 0.196 2.316 1.752 1.128 0.125

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

5

Table 3 Analysis of variance for quadratic model in copper ions adsorption. Sources of variation

Sum of squares

Degree of freedom

Mean square

F value

Probability>F

Model Lack of fit Pure error Residual Total

18,302 155.902 4.171 160.072 18,462

11 13 2 15 26

1663.810 11.992 2.085 10.672 710.078

155.912 5.751

<0.0001 0.158

response as coded factors was explained according to Eq. (3) [4,19]: Y ¼ 27306 þ 4:791 X I þ 3:637 X II þ 24:257 X III 2 14:513 X IV  9:238 XI2 þ 5:116 XII2 þ 5:554 XIV (3) þ3:617 X I X III  3:174 X I X IV þ 2:845 X II X III  12:436 X III X IV Eq. (3) makes a good visualization of the effects of each parameter and their interactions on the response. The comparison between the observed and predicted values in Table 2 can express a good correlation between the observed and predicted values of Cu2+ adsorption capacity with the high value of the determination coefficient (R2 = 0.991). This value indicates that only 0.9% of the total variation is not explained by the regression model. Analysis of variance (ANOVA) was applied to examine the model presented Table 3. The high value of the adjusted determination coefficient (R2adj ¼ 0:985) implicates to the significance of the model parameters. The low probability (<0.05) with Fvalue (155.912) implied that the model was accurate. Also, the validity of the model demonstrated by the value of lack of fit (0.158, more than 0.05) [35]. The significance of the parameter coefficients, the associated standard error, and the effect of each terms in Eq. (3) are presented in Table 4. According to p values (<0.05 is significant) and t ratios, it can be shown that all the firstorder main effects (XI) are highly significant. The initial concentration effect (XIII) has the highest importance between the main effects. This reality was also seen by Cojocaru et al. in Cu2+ adsorption by dried yeast biomass [17]. The second-order effect of pH (XI2 ) on copper ions adsorbed has the more significant effect among the other second-order effects. Meanwhile, the effect of the interaction between the initial concentration and adsorption dosage (XIIIXIV) shows a high significant effect. The negative value of the main effect coefficient demonstrates that Cu2+ adsorption capacity decreased with increasing the effect. Also, the negative coefficient of the second order parameters shows a maximum value in response within selected ranges of the parameters. Other 2 variables such as XIII , XIXII and XIIXIV had non-significant effect on Cu2+ uptake due to p value were more than 0.05.

3.2.2. The optimization of Cu2+ adsorption process To gain the better comprehension of copper ions adsorption process, the three dimensional response surface plots were analyzed. In each plot, the effect of two factors on Cu2+ adsorption capacity when the other factors were maintained in the optimum value was investigated. The response surface plots for Cu2+ adsorption capacity are shown in Figs. 4 and 5. Temperature and pH are the most important environmental parameters that can impress on the adsorption capacity. Fig. 4 shows the simultaneous effect of temperature and initial pH on Cu2+ adsorption capacity. Also, in constant pH, Cu2+ adsorption capacity increases with temperature enhancement. This is likely the result of the more activity of the surface sites, mobility of metal ions, and the change in sorbent pores size [29]. Also, the decrease of the fluid viscosity and then decrease of the resistance against Cu2+ diffusion into sorbent pores may cause to increase the Cu2+ adsorption capacity. These results were also reported by other researchers in copper ions adsorption [17,36–38]. In a specified temperature, the increase of the initial pH causes to the increment of Cu2+ adsorption capacity up to 4.6 and then the value slightly declines. According to zeta potential of LECA surface, in pH range of 3–5, the sorbent surface had a negative charge. With increasing of pH value and the increment of surface negative charge, Cu2+ adsorption capacity increases, gradually. The first-order effect of pH (XI) has a positive effect and the second-order effect of pH (XI2 ) has a negative effect that implies a maximum point in the selected range. The Cu2+ uptake in the pH value of 3.0 was low. Because, in the acidic region, the competition between H+ and Cu2+ causes to decrease in the occupation of the active sites on sorbet surface by Cu2+. This case causes to reduce of Cu2+ adsorption capacity. As mentioned by Elliot and Huang, the predominant species of copper in pH 3–5 are Cu2+ and CuOH+. However, at pH > 6.3 precipitate of Cu(OH)2 is formed [39,40]. Therefore, it can be demonstrated that any precipitate is not created in the studied range. The effect of the initial Cu2+ concentration and sorbent dosage is shown in Fig. 5. The Cu2+ adsorption capacity increased with increasing the initial concentration. The

Table 4 Regression analysis and significance of the components in the quadratic model. Factor (coded)

Coefficient

Standard error (SE)

Effect

p value

t ratio =(effect/SE)

Constant XI XII XIII XIV XI2 XII2 2 XIII 2 XIV XIXII XIXIII XIXIV XIIXIII XIIXIV XIIIXIV

27.3058 4.791 3.637 24.257 14.513 9.238 5.116  5.554  3.617 3.174 2.845 – 12.436

1.188 0.769 0.769 0.769 0.769 1.952 1.952  1.952  0.817 0.817 0.817 – 0.817

9.582 7.274 48.515 29.027 18.476 10.232  11.108  7.235 6.348 5.691 – 24.873

4.17E13 1.63E05 0.000272 4.03E15 7.43E12 0.000267 0.019294  0.012293  0.000488 0.001461 0.003331 – 1.57E10

12.460 9.459 63.088 37.746 9.465 5.242  5.691  8.856 7.769 6.966 – 30.444

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 6

T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

Fig. 6. Effect of the number of neurons in hidden layer on the performance of the neural network.

Fig. 4. The effect of temperature and pH on Cu2+ adsorption capacity of LECA (m/V: 1, (g/L), C0:150 (mg/L)).

enhancement of the initial concentration leads to the increment of the driving force (the concentration gradient) which overcomes the mass transfer resistance against Cu2+ diffusion from bulk to surface active sites [41]. Likewise, the copper ions adsorption capacity decreases with increasing of the adsorbent dosage that previously has been reported for Cu2+ removal by Enteromorpha prolifera [4]. This verity can be for the unsaturated surface sites. Another reason likely is the interaction between the sorbent particles. Namely, aggregation phenomenon can causes to the decrease of the total surface active sites and the increase of diffusion path length [42]. By using of the response surface design and optimization, the optimum adsorption capacity (99.289 mg/g) was obtained at pH 4.6, temperature of 50 8C, initial concentration 150 mg/L, and sorbent mass of 50 mg.

3.3. Neural network training The required input–output data for network training were obtained from adsorption experiments and were planned through CCF. The used deviations for selecting the best ANN architecture are the mean square errors (MSE), mean relative errors (MRE) and absolute fraction of variance (R2) which can be defined as follows [22,25,26,33]: MSE ¼

N 1X ðt  yi Þ2 N i¼1 i

(4)

MRE ¼

 N  100 X jt i  yi j N i¼1 ti

(5)

R2 ¼

Pi¼1 N

P 2 ðt i  t m 2  i¼1 N ðt i  yi Þ Pi¼1 2 N ðt i  yi Þ

(6)

where N is the number of data points, t is the target (experimental) data, and y is the predicted value. The Levenberg–Marquardt back propagation algorithm (LMA) was applied for the network training as the best algorithm. The optimization of network is very important step in network training. Therefore, the number of neurons in hidden layer should be optimized. For this purpose, the different number of neurons, in the range of 1–12 was tested in hidden layer. According to Fig. 6, the optimum number of neurons at hidden layer is equal to 4 as the best case with the minimum value of MSE and MRE (0.000357 and 0.056115, respectively). As a result, in this study three layered feed forward back propagation neural network (4:4:1) was used for modeling of adsorption process.

Fig. 5. The effect of the initial concentration and sorbent dosage on Cu2+ uptake capacity of LECA (T: 50 8C, pH: 4.6).

3.3.1. Test and validation of the model The samples were split into training and test subsets. The training data consist of 27 designed runs by CCF. The regression analysis was used to find the relation between targets (data set) with outputs (predicted values) training network (Fig. 7). This plot consist of two lines, the dashed line is the perfect fit (y = x, i.e. predicted data = experimental data) and the solid line is the best fit on training data set (y = 0.994x + 0.15) with acceptable correlation coefficient R2 (0.994), MSE (10.89787) and MRE (0.000063). For the validation purpose and evaluation of the power models, experiments were conducted for 9 new trials consisting combinations of experimental factors, which do not belong to the training data set. The actual and predicted values by RSM and ANN are

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

Predicted adsorpon capacity (mg/g)

120

100

7

as the equilibrium time. It can be implied that during the initial stage, numerous vacant surface sites were available. By occupying the vacant surface sites by metal ions, the changes of adsorption capacity were decreased. The adsorption capacity value and the initial copper ions concentration at the end of 90 min were considered as the equilibrium values (qeq (mg/g), Ceq (mg/L)).

y = 0.994 x + 0.150 R² = 0.994

80

60

3.5. Adsorption isotherm

40 Train data of ANN Test data of ANN

20 Test data of RSM Y=X

0 20

0

40

60

80

100

120

Experimental adsorpon capacity (mg/g)

The equilibrium behavior of adsorbent is a basic necessity for the adsorption system design and the feasibility study of adsorption systems [43]. The correlation between the equilibrium adsorption capacity on the sorbent surface and the equilibrium solute concentration in aqueous solution is explained by adsorption isotherm plots. For this aim, two common equilibrium models, i.e., Langmuir [44] and Freundlich [45] isotherms were studied. Common expression of Langmuir theory is expressed by:

Fig. 7. Comparison of the experimental and predicted results via ANN and RSM.

qe ¼

qm K L C e 1 þ K LCe

(7)

presented in Fig. 7. R2 values for ANN and RSM model was found to be 0.962 and 0.941, respectively. Although, the above values is lower ratio to correlation coefficient of training data but an acceptable difference exists between deviation value of training and testing data that can verify the suggested models. Two models were well fitted to experimental data. Both of models have their own advantages. RSM can show the effect of each main factor and the interaction between independence parameters on response in comparison with ANN. Meanwhile, ANN can develop and simulate process behavior as any form of non-linearity and overcome RSM limitation (quadratic non-linear correlation assumption). Also, ANN gives a simulation of process behavior without any standard experimental design.

where qe (mg/g) is the equilibrium amount of Cu2+ adsorbed per unit mass of sorbent, Ce (mg/L) is the equilibrium Cu2+ concentration in the solution, and qmax (mg/g) is the maximum copper ions adsorption capacity. qmax demonstrates the maximum amount of the adsorption capacity for the sorbent based on a monolayer coverage of metal ions that is fully covered the sorbent homogeneous surface. KL (L/mg) is the Langmuir constant related to adsorption energy. The Freundlich model has an exponential form:

3.4. The contact time effect

qe ¼ K f Ce

And in its linear form is written as: Ce 1 1 ¼ Ce þ ka qmax qe qmax

(8)

1=n

The contact time is one of the effective parameters in adsorption capacity. Fig. 8 shows the variation of copper ions adsorption capacity value with the contact time at the various initial concentrations of copper ions. It is obvious that Cu2+ adsorption on LECA was fast during the first 15 min. After 75 min, the copper ions adsorption capacity tends to get slowly and reached the equilibrium at 90 min, gradually. With passing the time, the adsorption capacity in all concentrations remained unchanged within the test duration. Then, 90 min was considered

100

80

(9)

The linear form is expressed as: Ln qe ¼ Ln k f þ

1 Ln C e nf

(10)

where kf and 1/nf are the Freundlich constants that indicates adsorption capacity and adsorption intensity, respectively. The Freundlich isotherm indicates the heterogeneous surface. To investigate the equilibrium isotherms, the experiments were conducted in the range of 10–150 (mg/L) of the initial Cu2+ concentration in different temperatures and the optimum condition. According to Table 5, the maximum monolayer coverage capacity of LECA for Cu2+ adsorption was obtained 113.636 (mg/g) at optimum temperature of 50 8C. The Langmuir constant (kL) increased with the increasing of temperature from 20 to 50 8C. This means that at high temperature, the tendency of the binding sites

q (mg/g)

C0= 150 mg/L

60

C0=115 mg/L C0=80 mg/L

Table 5 The Langmuir and Freundlich model parameters at different temperature.

C0= 45 mg/L

Temperature (8C)

40 C0=10 mg/L

Langmuir isotherm qmax (mg/g) KL (L/mg) R2

20

0 0

30

60

90 120 time(min)

Fig. 8. The effect of contact time on Cu 1 (g/L)).

2+

150

180

adsorption capacity (T: 50 8C; pH: 4.6; m/V:

Freundlich isotherm kf nf R2

20

35

50

107.527 0.044 0.978

111.111 0.056 0.988

113.636 0.086 0.929

6.706 1.610 0.988

8.471 1.676 0.990

14.524 2.060 0.999

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

8

100

120

100

80

C0=150 mg/L

60

q (mg/g)

qeq (mg/g)

80

T=50 C

40

C0=115 mg/L C0=80 mg/L

60

C0=45 mg/L C0=10 mg/L

T=35 C

40

T=20 C 20 Langmuir

20

Feriundlich 0

0

0

10

20

30

40 Ceq(mg/L)

50

60

70

2

4

6

8

10

12

14

16

t ^0.5 (min^0.5)

Fig. 9. Experimental data and calculated isotherms at different temperatures.

Fig. 10. Weber–Morris model plot (T: 50 8C, m/V: 1 g/l, pH: 4.6).

for the copper ions adsorption is stronger. Also, it can be seen from Table 5, the Freundlich constant (kf) increased with the temperature enhancement. Also, at all temperatures, nf value was more than unity confirming the favorable adsorption conditions [46]. The equilibrium data and calculated isotherms are shown in Fig. 9. Comparison of R2 values in Table 5 and Fig. 9 reveals that the Cu2+ adsorption data on LECA was very well fitted with both of Freundlich and Langmuir isotherm. However, Freundlich isotherm seems to be better than Langmuir isotherm. The specific surface area as the monolayer coverage of the sorbent surface by copper ions can be calculated by the following equation:

where Ki is the intra-particle rate constant (mg/g min1/2), qt is the adsorbed value of the metal ions (mg/g) at time t, and t is the contact time (min). The Webber and Morris plot for Cu2+ adsorption process by LECA is given in Fig. 10. It is obvious that the plot is divided to three regions. The first region that is the sharper portion is related to the external surface adsorption. The second portion is the moderate adsorption stage that belongs to intra-particle diffusion process. And, the third section is attributed to the final equilibrium stage. In this step, the intra-particle diffusion extremely declines due to decrease of solute concentration. Therefore, the adsorption process achieves to the equilibrium condition, gradually. The intra-particle diffusion constant for the different concentrations at 50 8C temperature is presented in Table 6. According to the data tabulated in Table 6, the intra-particle diffusion constant (Ki) increased from 0.165 to 2.455 (mg/g min1/2) with the enhancement of the copper ions concentration from 10 to 150 (mg/L). This result shows that the increment of the Cu2+ concentration leads to the decrease of the resistance against the diffusion of the copper ions into LECA pores [4]. The high regression coefficient values showed a good fitting between Cu2+ adsorption data with Weber–Morris model. According to this model, if the plot of q versus t1/2 passes from the origin the intra-particle diffusion is considered as the rate controlling stage. According to Fig. 10, the linear section does not pass through the origin revealing the mechanism of the copper ions adsorption by LECA as the mixture of the intra-particle diffusion and film diffusion. The external diffusion model can be characterized by the following equation:

S ¼ qmax

NA Aw

(11)

where S is the specific surface area (m2/g sorbent); qmax is the maximum monolayer adsorption capacity (g Cu/g sorbent); N is the Avogadro number (6.02  1023); A is the cross-sectional area of the metal ion (m2) that for Cu2+ is reported 1.58 A˚2 in the articles; Mw is the atomic weight of the metal ion (63.5 for Cu2+) [42]. The specific surface area of LECA for Cu2+ adsorption was obtained 17.021 (m2/g sorbent). Compression of the specific surface area LECA (17.021 m2/g) with E. prolifera (8.558 m2/g) and the dehydrated wheat bran (DWB) (7.720 m2/g) for Cu2+ adsorption showed that LECA possesses more specific surface area [4,47]. 3.6. Adsorption kinetics

C ¼ expðbL StÞ C0

(13)

The various models might be used in order to study the adsorption mechanism and determination of the rate controlling steps. Ignoring metal ions diffusion from the fluid bulk to the liquid film (the boundary layer), two steps can take place during adsorption process. The diffusion stage consists of the transportation of the metal ions from the boundary layer to the external surface of the sorbent (film diffusion or external diffusion model) and the diffusion of ions from the sorbent surface to the intra-particle active sites (intra-particle diffusion model). Finally, metal ions adsorption by active sites was occurred. The third stage is fast and almost instantaneous. Hence, this stage cannot be considered as the rate-limiting step [48]. Webber and Morris proposed a model to explain the intraparticle diffusion during the adsorption process presented by the following equation [49]:

where C0 is the initial metal ions concentration (mg/L), S is the specific surface area and bL is the external mass transfer coefficient (cm/min). The value of bLs was calculated from the slope of the external diffusion plot at time = 0 depicted in Fig. 11. Also, the external mass transfer coefficient decreases with the increment of Cu2+ concentration (Table 6). The higher bL value presents lower external mass transfer resistance against the Cu2+ adsorption and the higher mass transfer velocity of copper ions from the boundary layer to active sites [48]. The kinetic model commonly is used for batch adsorption process of full scale from the system design point of view. The pseudo-first-order and pseudo-second-order kinetic models were used with Eqs. (14) and (15), respectively [50,51]:

qt ¼ K i t 1=2

logðqe  qt Þ ¼ logðqe Þ 

(12)

K1  t 2:303

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

(14)

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

9

Table 6 The kinetic parameters at various Cu2+ concentrations. Kinetic model

pseudo-first-order model

pseudo-second-order model

C0 (mg/L)

qe,exp (mg/g)

qe,cal (mg/g)

K1 (min1)

R2

10 45 80 115 150

9.584 37.642 60.563 80.225 100.111

4.141 25.316 34.970 49.023 56.572

0.041 0.037 0.034 0.032 0.031

0.792 0.877 0.818 0.879 0.899

Webber–Morris model R2

qe,cal (mg/g)

K2 (g/mg.min)

h

9.794 40.984 63.291 84.746 105.263

0.0305 0.0020 0.0015 0.0014 0.0013

2.926 3.460 6.329 10.055 14.404

0.998 0.989 0.981 0.994 0.999

Ki (mg/g min0.5)

R2

0.165 1.694 1.959 2.449 2.454

0.865 0.741 0.981 0.982 0.971

Boundary layer model R2

bLS (1/min) 0.464 0.037 0.036 0.034 0.033

0.947 0.891 0.980 0.893 0.886

2.5

t t 1 ¼ þ qt qe K 2  q2e

(15)

C0=150 mg/L C0=115 mg/L

2

C0=80 mg/L

1.5

C0=45 mg/L C0=10 mg/L

log (qe-qt)

where K1 is the rate constant of pseudo-first-order adsorption process (1/min), K2 is the pseudo-second-order rate constant of adsorption (g/mg min). qe and qt are the amount of metal ions adsorbed (mg/g) at equilibrium and at time t; respectively. Meanwhile, h = k2qe2 is regarded as the initial adsorption rate. Figs. 12 and 13 show the plots of the kinetics model equations. The rate constant and the equilibrium Cu2+ uptake of the pseudo-firstorder kinetic model were obtained from the slope and intercept of the plot log(qe  qt) versus t (Table 6). Based on Table 6, the R2 value of pseudo-first-order model is low and the calculated copper ions adsorption capacity has a high different with the experimental data. However, the regression coefficient of pseudo-second-order kinetic model for all Cu2+ concentrations is high and near the unity. Also, the calculated Cu2+ adsorption capacity (qe,cal) is in agreement with experimental value (qe,exp) for each of the initial copper ions concentration. The slope and intercept of the straight lines t/qe versus t give the equilibrium copper ions adsorption capacity and the rate constant of the pseudo-second-order kinetic model. From Table 6, the values of K2 decreases with increasing of the initial copper ions concentration which suggests the enhancement of the diffusion in the solid phase and adsorption rate [4]. The experimental results were achieved from kinetic data presented a good fitting with pseudo-second-order rate equation. Similar result is previously reported by other researchers in Cu2+ adsorption process onto the different sorbents [36,37].

1

0.5

0 0

10

20

30

40

50

60

70

80

-0.5

-1

time (min)

Fig. 12. The pseudo-first-order reaction plot for Cu2+ adsorption capacity onto LECA at different concentrations.

temperature on Cu2+ adsorption by LECA was investigated in this study. The thermodynamic parameters, i.e. enthalpy change (DH8), entropy change (DS8) and free energy change (DG8) can be estimated according to [37]:

DG ¼ RT Ln K D

(16)

DG ¼ DH  T DS

(17)

3.7. Thermodynamics of the adsorption process Ln K D ¼ The thermodynamic parameters are studied to evaluate the inherent energetic changes of the system. The effect of the

DS R



DH

(18)

RT

12

1.2 10

C0=150 mg/L

1

C0=115 mg/L

c0=150 mg/L

8 t/q (min.g/mg)

C0=115 mg/L

0.8

C0=80 mg/L

C/C0

C0=45 mg/L

0.6

C0=10 mg/L

C0=80 mg/L C0=45 mg/L

6

C0=10 mg/L

4

0.4 2

0.2 0

0

0

0

50

100 time (min)

150

Fig. 11. Boundary layer model plot (T: 50 8C, m/V: 1 g/l, pH: 4.6).

200

20

40

60

80

100

time (min) Fig. 13. The pseudo-second-order reaction plot for Cu2+ adsorption capacity onto LECA at different concentrations.

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx

10

Table 7 Thermodynamic parameters for Cu2+ adsorption onto LECA. T (K)

KD

Ln KD

DG8 (kJ/mol)

293 298 308 323

5.789 6.811 10.698 20.497

1.756 1.919 2.370 3.020

4.278 4.753 6.069 8.111

DH8 (kJ/mol)

DS8 (kJ/mol K)

33.672

0.129

Table 8 The comparison of the adsorption capacity between LECA with other sorbents. Sorbent

qmax;Cu2þ ðmg=gÞ

Reference

Dye loaded groundnut shells Dye loaded sawdust Powdered waste sludge Spent activated clay Dithiocarbamated-sporopollenin Natural zeolite tuff Fired coal fly ash Enteromorpha prolifera P. yezoensis Ueda Rice bran Wheat bran Walnut hull Chestnut shell Trametes versicolor LECA

7.6 8.07 156 10.9 46.609 23 20.92 57.14 5.04 10.41 6.85 3.52 5.48 60.98 113.636

Shukla et al. [38] Shukla et al. [38] Pamukoglu et al. [11] Weng et al. [31] Ersoz et al. [32] Wang et al. [8] Papandreou et al. [2] ¨ zer et al. [37] O Wang et al. [33] Wang et al. [33] Wang et al. [33] Wang et al. [33] Va´zquez et al. [25] Sahan et al. [17] This work

- The optimum conditions were found as initial pH of 4.6, temperature of 50 8C, initial Cu2+ concentration of 150 mg/L, sorbent dosage of 50 mg, and the optimum adsorption capacity of 99.289 mg/g. - The results of two RSM and ANN methodologies based on validation data showed that RSM (R2 = 0.941) and ANN (R2 = 0.962) are useful and accurate methods to predict adsorption process. - Freundlich isotherm showed the best agreement with the equilibrium data than Langmuir isotherm. - The thermodynamic studies proved that Cu2+ removal using LECA was a spontaneous, feasible, endothermic and random process with mechanism of physical adsorption. - The pseudo-second-order rate model accurately described the kinetics of adsorption. - the good efficiency of copper removal makes LECA as an abundant and cheaply available adsorbent in nature for the treatment of copper contaminated wastewater. As a result LECA may be an alternative to expensive adsorbents.

Acknowledgements The authors are thankful to Research vice Department of Razi University for partial financial support to this work (grant no. 46473). References

where KD is the distribution coefficient (L/g), T is the absolute temperature (K), and R is gas constant (kJ/mol K). DH8 and DS8 were calculated from the slope and intercept of the Van’t Hoff plot of Ln(KD) versus 1000/T. The best fit on Van’t Hoff plot is y = 4.050x + 15.543 with high correlation coefficient (R2 = 0.997). The calculated thermodynamic parameters are listed in Table 7. Accordingly, the positive value of the enthalpy change (33.672 kJ/mol) confirms the endothermic nature of adsorption process. This reveals the effect of temperature on distribution coefficient KD which increases with the increment of temperature. DS8 value was found to be positive showing the tendency of LECA to adsorb copper ions and the randomness in the liquid/solid interface during the adsorption process. The negative value of the free energy DG8 proves the spontaneous nature and feasibility of adsorption process. A more negative DG8 with the increment of temperature implies a favorable reaction at higher temperature. The DG8 value up to 3.82 kcal/mol (20 kJ/mol) is consistent with the physical adsorption (electrostatic adsorption) and the values greater than 7.62 kcal/mol (40 kJ/mol) demonstrates the chemical adsorption as a result of charge sharing and the formation of coordinate bond [36]. According to DG8 obtained values in Table 7, the Cu2+ adsorption process onto LECA is a physical adsorption process. Many reports have also presented similar thermodynamic behavior for copper ions adsorption on the other sorbents [4,36]. The maximum sorption capacity of LECA was compared with other natural sorbents for Cu2+ adsorption capacity in Table 8. As can be seen, LECA has a good and acceptable sorption capacity relative to other sorbents for copper ions removal from aqueous phase.

4. Conclusion The capability of LECA in Cu2+ removal from aqueous solution and the effects of operational parameters on adsorption capacity in the batch system were investigated using of RSM and ANNs. The most important conclusions from this work are summarized as follows:

[1] T.G. Chuah, A. Jumasiah, I. Azni, S. Katayon, S.Y., Desalination 175 (2005) 305–316. [2] A. Papandreou, C.J. Stournaras, D. Panias, Journal of Hazardous Materials 148 (2007) 538–547. [3] B. Yu, Y. Zhang, A. Shukla, S.S. Shukla, K.L. Dorris, Journal of Hazardous Materials B 80 (2000) 33–42. ¨ zer, G. Gu¨rbu¨z, A. C¸alimli, B.K. Ko¨rbahti, Chemical Engineering Journal 146 [4] A. O (2009) 377–387. [5] J. Khosravi, A. Alamdari, Journal of Hazardous Materials 166 (2009) 695–700. [6] D. Voglar, D. Lestan, Journal of Hazardous Materials 180 (2010) 152–157. [7] B. Topuz, M. Macit, Environmental Monitoring and Assessment 173 (2011) 709– 722. [8] R. Molinari, P. Argurio, T. Poerio, Separation and Purification Technology 70 (2009) 166–172. [9] S. Wang, T. Terdkiatburana, M.O. Tade´, Separation and Purification Technology 62 (2008) 64–70. [10] A. Ayar, B. Mercimek, H. Kara, Y. C¸engelog˘lu, Separation and Purification Technology 28 (2002) 25–28. ¨ zcimen, A. Ersoy-Mericboyu, Journal of Hazardous Materials 168 (2009) [11] D. O 1118–1125. [12] M.Y. Pamukoglu, F. Kargi, Process Biochemistry 41 (2006) 1047–1054. [13] M.E. Mahmoud, H.M. Al-Bishri, Chemical Engineering Journal 166 (2011) 157– 167. [14] M.E. Mahmoud, Desalination 266 (2011) 119–127. [15] M.E. Mahmoud, A.A. Yakout, H. Abdel-Aal, M.M. Osman, Bioresource Technology 106 (2012) 125–132. [16] C. Saka, O. Sahin, M.M. Ku¨cu¨k, International journal of Environmental Science and Technology 9 (2012) 79–94. [17] C. Cojocaru, M. Diaconu, I. Cretescu, J. Savi’c, V. Vasi’c, Colloids and Surfaces A: Physicochemical and Engineering Aspects 335 (2009) 181–188. [18] M. Malakootian, J. Nouri, H. Hossaini, International Journal of Environmental Science and Technology 6 (2) (2009) 183–190. [19] D.C. Montgomery, Design and Analysis of Experiments, 7th ed., John Wiley and Sons, New York, 2008. [20] T. S¸ahan, H. Ceylan, N. S¸ahiner, Nahit AktaS¸, Bioresource Technology 101 (2010) 4520–4526. [21] B. Preetha, T. Viruthagiri, Journal of Hazardous Materials 143 (2007) 506–510. [22] N.G. Turan, B. Mesci, O. Ozgonenel, Chemical Engineering Journal 173 (2011) 98– 105. [23] N. Prakash, S.A. Manikandan, L. Govindarajan, V. Vijayagopal, Journal of Hazardous Materials 152 (2008) 1268–1275. [24] N. Jaafarzadeh, M. Ahmadi, H. Amiri, M.H. Yassin, S.S. Martinez, Journal of the Taiwan Institute of Chemical Engineers 43 (2012) 873–878. [25] K. Yetilmezsoy, S. Demirel, Journal of Hazardous Materials 153 (2008) 1288– 1300. [26] D. Bingo¨l, M. Hercan, S. Elevli, E. Kılıc¸, Bioresource Technology 112 (2012) 111– 115. [27] G.P. Gillman, E.A. Sumpter, Australian Journal of Soil Research 24 (1986) 61–66. [28] K. Yetilmezsoy, S. Demirel, R.J. Vanderbei, Journal of Hazardous Materials 171 (2009) 551–562.

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017

G Model

JIEC-1405; No. of Pages 11 T. Shojaeimehr et al. / Journal of Industrial and Engineering Chemistry xxx (2013) xxx–xxx [29] G. Va´zquez, M. Calvo, M.S. Freire, J. Gonza´lez-Alvarez, G. Antorren, Journal of Hazardous Materials 172 (2009) 1402–1414. [30] J.A. Cornell, How to Apply Response Surface Methodology, 2nd ed., American Society for Quality Control, Wisconsin, 1990. [31] E. Bayraktar, Process Biochemistry 37 (2001) 169–175. [32] R.H. Myers, D.C. Montgomery, Response Surface Methodology: Process and Product Optimization Using Designed Experiments, 2nd ed., John Wiley & Sons, USA, 2002. [33] A. Khataee, A. Khani, International Journal of Chemical Reactor Engineering 7 (2009) 1–16. [34] R.A. Nyquist, R.A. Kagel (Eds.), IR Spectra of Inorganic Compounds, Academic Press, New York, 1971. [35] D.R. Hamsaveni, S.G. Prapulla, S. Divakar, Process Biochemistry 36 (2001) 1103– 1109. [36] C.H. Weng, C.Z. Tsai, S.H. Chu, Y.C. Sharma, Separation and Purification Technology 54 (2007) 187–197. ¨ nlu¨, M. Ersoz, Separation and Purification Technology 52 (2007) 461–469. [37] N. U [38] X.S. Wang, Z.Z. Li, C. Sun, Desalination 235 (2009) 146–159. [39] H.A. Elliot, C.P. Huang, Water Research 15 (1981) 849–855.

11

[40] M.F. Talhi, A. Cheriti, N. Belboukharia, L. Agha, C. Roussel, Desalination and Water Treatment 21 (2010) 323–327. [41] B. Zhu, T. Fan, D. Zhang, Journal of Hazardous Materials 153 (2008) 300–308. [42] S.R. Shukla, R.S. Pai, Separation and Purification Technology 43 (2005) 1–8. [43] O. Altin, H.O. Ozbelge, T. Dogu, Journal of Colloid and Interface Science 198 (1998) 30–140. [44] I. Langmuir, Journal of the American Chemical Society 38 (1916) 2221–2295. [45] H.M.F. Freundlich, Journal of Physical Chemistry 57 (1906) 385–471. [46] V.B.H. Dang, H.D. Doan, T. Dang-Vu, A. Lohi, Bioresource Technology 100 (2009) 211–219. ¨ zer, D. O ¨ zer, Process Biochemistry 39 (2004) 2183–2191. [47] A. O [48] M. Sarkar, P.K. Acharya, B. Bhattacharya, Journal of Colloid and Interface Science 266 (2003) 28–32. [49] J.W. Weber, J.C. Morris, Journal of the Sanitary Engineering Division, ASCE 89 (1963) 31–60. [50] M. Kousha, E. Daneshvar, H. Dopeikar, D. Taghavi, A. Bhatnagar, Chemical Engineering Journal 179 (2012) 158–168. [51] Y.S. Ho, G. McKay, Process Biochemistry 34 (1999) 451–765.

Please cite this article in press as: T. Shojaeimehr, et al., J. Ind. Eng. Chem. (2013), http://dx.doi.org/10.1016/j.jiec.2013.06.017