Deep learning modeling and column experiments for Cu(II) adsorption on amino-functionalized carboxymethylcellulose beads in aqueous solutions

Abstract

In this study, we investigated the adsorption properties of Cu(II) on amino-functionalized carboxymethyl cellulose (Am@CA) beads (diameter = 1.86 ± 0.31 mm; point of zero charge = 6.17). Adsorption tests indicated that the modified intraparticle diffusion model presented the most precise fit for the kinetic data, while the Redlich–Peterson isotherm yielded the most accurate description of the isotherm data. The Am@CA beads exhibited a theoretical maximum adsorption capacity of 136.6 mg/g for Cu(II), and the adsorption process was characterized as exothermic. Instrumental analyses along with pH experiments confirmed the involvement of chelation in the adsorption. Reusability experiments with 0.1 M HCl revealed the ability to regenerate and reuse the Am@CA beads for Cu(II) adsorption. Experimental conditions for column experiments were determined utilizing central composite design (CCD) to generate Cu(II) breakthrough curves (n = 17). Within the CCD methodology, input variables encompassed influent Cu(II) concentration, flow rate, and column length, while output variables comprised breakthrough time, exhaustion time, and Cu(II) removal rate. Employing breakthrough curve data, deep learning modeling was performed to construct the artificial neural network (ANN) model with a topology of 3:20:20:20:3, comprising 3 input variables, 20 neurons in each of the first, second, and third hidden layers, and 3 output variables. Using data from the ANN model, 3-D plots of the response surface methodology were created to illustrate the relationship between input variables and response (output) variables. An additional column experiment was conducted to further assess the developed ANN model, revealing that the model exhibited reliable predictability for the breakthrough curve. The relative importance of the input variables indicated that flow rate was the most influential factor affecting the output variables. © 2024 Elsevier Ltd