Application of response surface methodology and artificial neural network: modeling and optimization of Cr(VI) adsorption process using Dowex 1X8 anion exchange resin

It is well known that chromium is one of the major heavy metals which originates mainly from industrial wastewaters, such as paints and pigments, photography, battery, electroplating and mining. The trivalent and hexavalent oxidation states of chromium are the most stable in aqueous system. In trace amounts, Cr(III) is an essential nutrient for humans and it plays an important role in glucose metabolism (Gherasim et al. 2011). In contrast, hexavalent chromium is highly toxic, carcinogenic, and mutagenic and its tolerance limit in wastewater has been recommended as 0.05 mg/L by the US Environmental Protection Agency (De Oliveira et al. 2014). However, industrial wastewater contains a large amount of chromium, ranging from 0.5 to 270.00 mg/L (Anupam et al. 2011). In order to comply with the strict environmental standards, it is mandatory to eliminate Cr(VI) from industrial effluents to reduce the negative impacts on ecosystem and public health. Various physico-chemical techniques were focused on the treatment of chromium metal pollution, such as chemical precipitation, ion exchange, electrochemical precipitation, adsorption, membrane separation, concentration, reverse osmosis and emulsion per traction technology (Cronje et al. 2011). Among these techniques, the adsorption may be considered as one of the most useful methods for wastewater treatment due to its simple operation, higher removal efficiency, and cost effectiveness (Halder et al. 2014). In the literature, adsorption of metals using ion exchange resin is a proven technique for the purification and separation of metals from different aqueous solutions. Thus, many investigators have been devoted to discuss the adsorption mechanism, kinetics, equilibrium and thermodynamics (Pehlivan & Cetin 2009; Neagu & Mikhalovsky 2010). These studies were conducted for optimization of process parameters by changing one variable at a time while keeping all others at constant level. But this traditional approach cannot evaluate interactive effects between the variables and uses a large number of experiments (Sohrabi et al. 2014). Lately, response surface methodology (RSM), which is a combination of mathematical and statistical techniques, is one of the approaches used for developing, improving and optimizing the processes and to evaluate the relative significance of various process parameters in the presence of complex interactions (Simsek et al. 2015).

In the last decade, numerous applications of artificial neural networks (ANNs) have been massively studied to solve environmental problems because of their reliable and salient characteristics in capturing the non-linear relationships existing between variables in complex systems (Turan et al. 2011).

The purpose of the present work was to develop the performance of the anion exchange resin Dowex 1X8 to remove hexavalent chromium from polluted water. In the first step, our adsorbent Dowex 1X8 was characterized by Fourier transform infrared spectroscopy (FTIR) and X-ray diffraction (XRD) to identify the functional groups involved during the adsorption process. Afterward, a full factorial design (FFD) by RSM using STATISTICA 10 software was used to examine the combined effects of four process parameters such as initial pH, initial chromium concentration, adsorbent dose and temperature. Subsequently, this work was extended to investigate the removal of Cr(VI) from a real industrial effluent sample in order to cater for legal limits before discharging into the municipal treatment. In addition, an ANN model with three layers was developed to predict the Cr(VI) removal efficiency of Dowex 1X8 as adsorbent. The results predicted by the ANN and RSM techniques were compared with the experimental data, and advantages and further developments were discussed.

Dowex 1X8 was provided by Sigma Aldrich. The resin is a strongly basic anionic exchanger with styrene-divinyl benzene copolymers, quaternary ammonium as the functional group and chloride as the exchangeable anion. The resin shows humidity percentage of 43–48%; 8% of divinylbenzene cross-linking make this resin resistant to decomposition by chemical attack and can be used for a wide range of chemicals.

Analytical grade reagents were used in experimental studies. The stock solution of Cr(VI) used in this study was prepared by dissolving an accurate mass of in 1 L of deionized water to have an initial concentration of Cr(VI) 1 g/L. The pH of the metal solution was adjusted to its optimum value by adding sufficient amounts of 0.1 mol/L HCl and 0.1 mol/L NaOH solutions.

Response surface is a statistical method that uses a minimum number of experiments to study the interaction of the various parameters affecting the process and examine the responses of several factors by varying them simultaneously (Kalavathy et al. 2009). A FFD with two levels was investigated in this study. The FFD has 2^k runs where k is the number of independent parameters. Initial pH, initial Cr(VI) concentration, resin dose and temperature were chosen as the independent variables for Cr(VI) removal process and their levels and ranges are shown in Table 1.

ANN is a good inspiration of brain and nerve systems and it is known for its extreme ability to learn and classify data (Prakash et al. 2008). The ANN is widely used to approximate complex systems that are difficult to model using conventional modeling techniques, such as mathematical or physical modeling (Turan et al. 2011). The basic ANN architecture has three layers: the input layer represents the independent parameters; the output layer gives the dependent parameters and one or more intermediate nodes; and the hidden layer acts as a collection of feature detectors. In this work, the Neural Network MATLAB 7.12.0 (R2011a) mathematical software was used to develop the ANN and predict the removal percentage efficiency of hexavalent chromium by Dowex 1X8.

The same experimental design used for the RSM design was employed in designing the ANN. The input variables were as follows: initial pH (5, 10), initial Cr(VI) concentration (5, 100 g/L), resin dose (0.5, 2 g) and temperature (283, 313 K). The removal percentage of Cr(VI) was chosen as the neural network output variable.

The adsorption of Cr(VI) from aqueous solution onto Dowex 1X8 was performed in a static mode according to the FFD matrix. For this, 100 mL of chromium (VI) solutions with different pH, initial Cr(VI) concentration, adsorbent dose and temperature were placed in a 150 mL Erlenmeyer flask. The mixture was agitated at 130 rpm and at a fixed contact time.

Adsorption test was also extended with real wastewater, containing Cr(VI) ion, to investigate the efficiency Cr(VI) removal. The sample was acquired from a local wastewater (Tunis, Tunisia).

After sand filtration and microfiltration, the characteristics of the collected effluent were studied and compared with those obtained after adsorption treatment.

From Figure 2, the fundamental peaks of adsorbent were almost the same before and after use. The band around 341 cm⁻¹ indicates the presence of free bonded hydroxyl groups. The peak at 3,023 cm⁻¹, which is existing in the spectrum of Dowex 1X8 before and after chromium adsorption, was attributed to the C–H aromatic stretching. The peak observed at 2,924 cm⁻¹ can be assigned to C–H aliphatic stretching (Özdemir et al. 2011). There is a peak at about 1,620 cm⁻¹ for both spectrums. The C=C stretching vibration in aromatic ring can be responsible for this band (Kusku et al. 2014). The peak that appeared at 1,485 cm⁻¹ was due to C–N vibration of –N⁺ (CH₃)₃Cl⁻ (Zhou et al. 2011). On the other hand, the resin spectrum after Cr(VI) adsorption in Figure 2(b) demonstrated few changes in the finger print zone. In fact, the characteristic sorption peaks at 976, 890, 859 and 827 cm⁻¹ became strong after the adsorption. This could be explained by the fact that N–C vibration was strengthened after the ion exchange of Cl⁻ to Cr(VI) ions, making the corresponding bands to be strong (Wang et al. 2012). Furthermore, a weaker peak appeared at 764 cm⁻¹ after the adsorption of Cr(VI) corresponding to the stretching vibration of Cr–O in (Holman et al. 1999).

From XRD pattern, the maximum diffraction at can be associated with its crystalline region. There is a slight dissimilarity in the XRD pattern of adsorbent plotted after the chromium adsorption. Indeed, the intensity of the peak observed at in Cr(VI) loaded Dowex 1X8 becomes more intense than that for Dowex 1X8. This increase could be attributed to an increase in the size of adsorbent which indicates the presence of hexavalent chromium on the resin surface. The characteristic new peaks observed at and after adsorption process (Figure 3(b)) were in agreement with the reported value in the literature which indicates the presence of the hexavalent chromium (Kalidhasan et al. 2012). The obtained results predict the evidence of the specific adsorption of chromium ion onto Dowex 1X8.

RSM was used to identify the influence of pH, initial Cr(VI) concentration, resin dose and temperature on chromium adsorption process by the anion exchange resin Dowex 1X8.

The main objective of RSM is to develop a regression model for the removal process. A total of 16 experiments were carried out according to a 2⁴ FFD. The total number and sequence of experimental data were determined using STATISTICA 10 software and the obtained results are listed in Table 2.

As shown in Equation (8), pH of solution exhibited a significant negative effect, while the resin dose had the most positive effect.

To find the significant main and interaction effects of factors affecting the removal efficiency analysis of variance (ANOVA) was followed and the obtained results are listed in Table 3.

The P-value less than 0.05 indicates the statistical significance of an effect at 95% confidence level and F-test was used to estimate the statistical significance of all terms in the polynomial equation within 95% confidence interval. The F-value for 95% confidence interval, 1 degree of freedom and 16 factorial runs (F_0.05,1.16) is 4.49 (Krishnaiah et al. 2012). The smaller P-value of model (<0.0001) and the higher F-value (194.283) implied that the model was statistically significant. All effects with F higher than 4.49 are statistically significant. By observing the F and P values from Table 3, it is shown that the resin dose and pH have the greatest effect on the Cr(VI) removal percentage.

As in Figure 4, the data points were well distributed close to a straight line (R²= 0.932, adjusted R²= 0.927), which suggested an excellent relationship between the experimental and predicted response.

Figures 5(a)–5(c) give the interactions of initial pH with resin dose, initial Cr(VI) concentration and temperature, respectively. As it can be seen, the maximum adsorption of chromium was reached at a pH ranging from 4 to 6.5. At low pH (pH <6), the dominant specie is ⁠, which leads to electrostatic attraction between positively charged adsorbent surface and negatively charged chromium species (Jain et al. 2011). Beyond pH 6.5, the removal percentage of Cr(VI) decreased significantly at any fixed resin dose, initial chromium concentration and temperature. This could be explained by the dominance of at higher pH and their competition with the hydroxide anions (Kumar et al. 2011). Therefore, the adsorption sites could lead to a reduction in the percentage adsorption of Cr(VI).

The effect of initial Cr(VI) concentration and its interaction with other factors are presented in Figures 5(b), 5(d) and 5(e). As it can be seen from these plots, chromium removal % was decreased slightly with increasing Cr(VI) initial concentration. This was due to larger amount of Cr(VI) ions competing for the available adsorption sites of ion exchange resin.

The combined effects of resin dose with pH, initial Cr(VI) concentration and temperature are visualized in Figures 5(a), 5(d) and 5(f) respectively. As shown in these figures, the increase in the resin dose from 0.5 to 2 g led to an increase in chromium removal percentage. This can be explained by the increase in the number of available sorption sites with the increase in the resin amount.

In adsorption process, it was also observed that the temperature played a moderate role in Cr(VI) removal by the ion exchange resin Dowex 1X8. The effect of temperature can be inferred from Figures 5(c), 5(e) and 5(f). It was seen that the adsorption of Cr(VI) increased slightly with the decreasing of temperature ranging from 313 to 283 K. This result indicates that the adsorption of Cr(VI) on Dowex 1X8 is an exothermic process and is favored at low temperatures.

The thermodynamic parameters such as standard enthalpy change⁠, standard free energy and standard entropy change were evaluated to confirm the effect of temperature.

The calculated thermodynamic parameters are listed in Table 4.

The negative value of suggests that the adsorption process of Cr(VI) was exothermic. The values were found to be negative, which indicated the feasibility and spontaneity of the adsorption. The negative value of the standard entropy showed the decreased randomness at the solid/solution interface during the adsorption process.

The response desirability conception implicates specifying the DF for chromium (VI) removal percentage by attribution of predicted values. The range 0.0 (undesirable) to 1.0 (very desirable) is used to obtain the optimum design. The factorial matrix (Table 2) shows the maximum (95.96%) and minimum (67.08%) for Cr(VI) removal percentage. According to this result, the DF values for the dependent variables of removal percentage (Figure 7) indicate that the desirability of 1.0 was assigned to maximum removal (95.96%), 0.0 for minimum (67.08%) and 0.5 for the middle (81.52%). The individual desirability scores used to calculate the removal percentage in the desirability of 1.0 are illustrated at the bottom of Figure 7. For desirability of 1.0, the global response obtained from these plots with the actual level of each variable is presented at the top (left) of Figure 7. In overview of the obtained plots, we can remark that variables affect simultaneously the response and its desirability. Based on this calculation and desirability score of 1.0, optimum conditions are defined in Table 5.

The adsorption potential of Dowex 1X8 in Cr(VI) removal process was evaluated with industrial effluent. The optimum values of variables obtained from RSM (Table 5) were applied. The physico-chemical characteristics of the effluent before and after adsorption process were studied. The results are summarized in Table 6.

Table 6

Characteristics of the wastewater before and after treatment by Dowex 1X8

Parameters .	Wastewater before treatment .	Wastewater after treatment .	Tunisian standards liquid discharges into the network ‘National Office for Sanitation’ (O.N.A.S.) (NT 106-02) .
pH	5.02	5.02	–
Salinity (mg/L)	1,251.35	1,237.02	–
Turbidity (NTU)	0.17	0.17	–
Cr(VI) (mg/L)	14.95	0.42	0.5
(mg/L)	535.66	593.04	700
(mg/L)	22.38	< LOD*	90
(mg/L)	496.09	423.29	400
(mg/L)	266.36	266.95	1,000
(mg/L)	19.99	19.46	50
(mg/L)	3.822	3.915	5
(mg/L)	13.257	13.011	0.1
(mg/L)	0.139	0.137	1

Parameters .	Wastewater before treatment .	Wastewater after treatment .	Tunisian standards liquid discharges into the network ‘National Office for Sanitation’ (O.N.A.S.) (NT 106-02) .
pH	5.02	5.02	–
Salinity (mg/L)	1,251.35	1,237.02	–
Turbidity (NTU)	0.17	0.17	–
Cr(VI) (mg/L)	14.95	0.42	0.5
(mg/L)	535.66	593.04	700
(mg/L)	22.38	< LOD*	90
(mg/L)	496.09	423.29	400
(mg/L)	266.36	266.95	1,000
(mg/L)	19.99	19.46	50
(mg/L)	3.822	3.915	5
(mg/L)	13.257	13.011	0.1
(mg/L)	0.139	0.137	1

*Limit of detection = 0.333 mg/L.

View Large

The hexavalent chromium in the sample was the subjected of adsorption process and its removal percentage was found to be 97.20%. Moreover, it is necessary to mention that the adsorption process on Dowex 1X8 can be used for removal of other pollutants from real wastewaters. These results suggested that the Dowex 1X8 resin has an excellent potential application for the removal of Cr(VI) from wastewater highly and rapidly.

Taking into account the high determination coefficient of this plot (Figure 8), linear regression shows an excellent compatibility between ANN predicted data and the experimental data results.

The RMSE for RSM and ANN was found to be 2.464 and 0.582, respectively. These results revealed that the RSM prediction has a greater deviation than the ANN prediction. The AAD for the RSM model was calculated to be 0.870%, whilst that of the ANN model was 0.704%. R² values were found to be 0.932 and 0.996 for RSM and ANN, respectively. The two models gave good predictions. However, the ANN model is more powerful than RSM in predicting removal percentage of Cr(VI) due to the higher value of R² and smaller value of AAD. Both models have their own advantages. RSM has the favor to give a regression equation for prediction, evaluate the effect of main factors and their interactions and identify the optimal conditions of adsorption process in comparison with ANN. Meanwhile, ANN can develop and simulate process behavior as any form of non-linearity without any standard experimental design. Also, this methodology is flexible and permits adding to new experimental data to build a trustable ANN model (Geyikçi et al. 2012). These results showed that RSM and ANN are very impressive methods to predict chromium removal percentage by Dowex 1X8.

The objective of this study was to estimate, optimize, and model the removal of Cr(VI) from aqueous solution using strong anionic resin Dowex 1X8. The adsorbent was characterized by FTIR and XRD. Then the RSM was successfully applied to examine the influences of initial pH, initial Cr(VI) concentration, resin dose and temperature on chromium removal. Polynomial model was developed for process optimization. The significant main and interaction effects of factors affecting the removal efficiency were tested with ANOVA. Resin dose (P = 0.002, F = 33.568) was found to be the most significant parameter for chromium removal followed by the initial pH (P = 0.003, F = 26.335).

Response surface contour plots were investigated to understand the combined effect of process variables on percentage removal. The DF indicated that 95.96% removal of chromium can be possible by using the optimal conditions of initial pH 5, initial Cr(VI) concentration 100 mg/L, resin dose 2 g and temperature 283 K. Subsequently, the experiments demonstrated that strong anionic resin had a good potential in the efficient removal of Cr(VI) from wastewater. The same experimental design was also used to obtain ANN model. RSM and ANN methodologies were statistically compared by the RMSE, determination coefficient (R²) and AAD and the results showed that both of them provided good quality predictions. However, the ANN showed better predictive performance as compared to RSM.