Prediction of adsorption efficiency for alkali activated carbon using Artificial Neural Networks

Abstract

Activated carbon is a widely used adsorbent in wastewater treatment due to its high surface area and tunable physicochemical properties. The adsorption performance of activated carbon is strongly influenced by synthesis parameters, including impregnation ratio, activation time, and activation temperature, which are studied in detail in the present work. An artificial neural network (ANN) model was developed to predict the adsorption capacity of methylene blue based on these three process variables. Activated carbons were prepared using chemical activation with potassium hydroxide (KOH) and sodium hydroxide (NaOH), and their adsorption performance was determined through experimental measurements. The ANN model adopted a multilayer perceptron (MLP) architecture with an input layer consisting of three neurons corresponding to impregnation ratio, activation time, and activation temperature. This was followed by ten hidden layers, each containing fourteen neurons, and a single output neuron representing the adsorption capacity in mg/g. The network was trained using backpropagation and optimized using the Adam optimizer. The architecture and the number of neurons were selected through successive comparison of the loss values on training and validation datasets to ensure high generalization accuracy. The trained ANN model accurately captured the nonlinear relationships between synthesis parameters and adsorption capacity. This work highlights the capability of ANN as an effective predictive tool in materials science, facilitating the design and optimization of activated carbon synthesis with reduced experimental effort. The proposed approach can serve as a foundation for developing intelligent systems for material selection and process optimization in environmental applications.