Modeling and prediction of copper removal from aqueous solutions by nZVI/rGO magnetic nanocomposites using ANN-GA and ANN-PSO

Reduced graphene oxide-supported nanoscale zero-valent iron (nZVI/rGO) magnetic nanocomposites were prepared and then applied in the Cu(II) removal from aqueous solutions. Scanning electron microscopy, transmission electron microscopy, X-ray photoelectron spectroscopy and superconduction quantum interference device magnetometer were performed to characterize the nZVI/rGO nanocomposites. In order to reduce the number of experiments and the economic cost, response surface methodology (RSM) combined with artificial intelligence (AI) techniques, such as artificial neural network (ANN), genetic algorithm (GA) and particle swarm optimization (PSO), has been utilized as a major tool that can model and optimize the removal processes, because a tremendous advance has recently been made on AI that may result in extensive applications. Based on RSM, ANN-GA and ANN-PSO were employed to model the Cu(II) removal process and optimize the operating parameters, e.g., operating temperature, initial pH, initial concentration and contact time. The ANN-PSO model was proven to be an effective tool for modeling and optimizing the Cu(II) removal with a low absolute error and a high removal efficiency. Furthermore, the isotherm, kinetic, thermodynamic studies and the XPS analysis were performed to explore the mechanisms of Cu(II) removal process.

Reduced graphene oxide-supported nanoscale zero-valent iron preparation details. The preparation of graphene oxide (GO) was carried out using the improved Hummers method, and reduced graphene oxide (rGO) was synthesized by the reduction of GO with the NaBH 4 solution 1,2 . The nZVI/rGO composites were prepared by the reduction of FeSO 4 ·6H 2 O and GO with NaBH 4 (mass ratio, carbon:iron = 1:2). 1g of GO was dispersed in deionized water (300 mL) by 2 hours ultrasonication. 10g of FeSO 4 .7H 2 O was added into the GO suspension, which was mixed for 12 hours with magnetic stirring to reach the adsorption equilibrium. NaBH 4 solution (5.46 g/50 mL) was added into the above solution at room temperature with magnetic stirring for 30 min. The prepared nZVI/rGO composites were separated from the liquid phase via centrifugation and dried at 50 °C in the vacuum drying oven for 24 hours.
RSM modeling and optimization. RSM is a commonly utilized statistical method for optimizing the process parameters and providing the statistical relationship among the variables. Namely, operating temperature (X 1 ), initial pH (X 2 ), initial concentration (X 3 ) and contact time (X 4 ) were selected as independent variables, and the Cu(II) removal efficiency (Y) was considered as the dependent variable. The factor levels were coded as -1 (low), 0 (central point) and 1 (high). The number of experiments needed to investigate the optimization of removal process are 81((3) 4 ), which were reduced to 29 by using a Box-Behnken experimental design. The experimental data were fitted to a second-order multiple regression analysis and analysis of variance (ANOVA) to investigate the behavior of the system using the least squares regression methodology. The quadratic model can be represented as below: where Y stands for the removal efficiencies of Cu(II), c 0 represents the constant coefficient, c i , c ii and c ij are the coefficient for linear, quadratic and interaction effect, respectively, x i and x j are the values of the independent variables; a is the residual error. All experimental ranges and levels of independent variables chosen are collected in Table S2, which are based on the single factor experiments. The single factor experiments were carried out to provide a reasonable range for the independent variables of response surface experiments.
Fitting of second order polynomial equations and statistical analysis. The quadratic model with evaluated coefficients for the Cu(II) removal is given as follows: The result of ANOVA for the quadratic model is shown in Table S1, which implies that this model was significant with a low probability value with F value of 22.35. There is only a 0.01% chance that a large "Model F-value" could occur due to the noise. Statistical analysis of the data indicated the values of 0.9572 and 0.9144 for R 2 and adjusted R 2 .
In addition, the initial concentration, contact time and operating temperature are quite significant for the Cu(II) removal except for initial pH. Based on the F-values of independent variables, the order for these parameters influencing the Removal kinetics. The four kinetic models, such as the pseudo-first order, pseudo-second order, intraparticle diffusion and Elovich models, can be described as follows: 1 t e e e t t q k q q where q e and q t represent the quantity of Cu(II) removal at equilibrium and at time t, k 1 (1/min), k 2 (g/mg/min) and k 3 (mg/g min 0.5 ) stand for the rate constant of pseudo-first-order kinetic model, pseudo-second-order kinetic model and intraparticle diffusion model, B is the intercept that is associated to the boundary layer thickness, α (mg/g/min) represents an initial removal rate and β (g/mg) is a desorption constant. The pseudo-first-order kinetic model describes the rate of removal to be proportional to the number of unoccupied active sites by the solutes, while the pseudo-second-order kinetic model assumes that the removal rate of Cu(II) is proportional with the square of difference between the quantity of Cu(II) remoed with time and the amount of Cu(II) absorbed at equilibrium 3,4 . The intraparticle diffusion model hypothesizes that the solute uptake changes proportionally with t 0.5 rather than with the contact time t 5 .
The Elovich model was generally applied in chemisorption kinetics, which was found in overlapping a broad range of slow adsorption rate 6 .
Adsorption isotherms. The Langumuir isotherm assumes that the adsorption sites are monolayer and identically homogeneous adsorption on the outer surface of adsorbent with a finite number of adsorption sites 7 , which can be expressed as follows: where k L represents the Langmuir adsorption equilibrium constant (L/g), q max stands for the maximum adsorption (mg/g). In addition, the essential feature of the Langmuir isotherm can be evaluated by means of R L , which is given as follows: where K F and n stand for the Freundlich constants related to adsorption intensity and adsorption capacity, respectively.
The 1/n value between 0 and 1 indicated that the adsorption process of heavy metals was favorable. The Temkin isotherm supposed that the heat of adsorption process decreases linearly as the degree of adsorption increases, which can be represented by the following equation: where RT/a t and b t are the constants related to the heat (J/mol) and binding energy (L/g) of the adsorption process. The experimental data were also fitted to the D-R isotherm to evaluate the nature of the adsorption process as physical or chemical, which can be expressed as follows: where α represents the activity coefficient related to adsorption mean free energy (mol 2 /J 2 ), ε stands for the Polanyi potential and E (kJ/mol) is the mean energy of adsorption.