AI-Based Expert System Design for Optimisation of Heavy Metal Adsorption on Waste-Derived Adsorbents

Abstract Heavy metal contamination in wastewater is a major global problem that requires effective, automated, and sustainable treatment systems capable of linking laboratory-scale adsorption processes to intelligent large-scale operation. This work introduces a novel Expert System ( ES ) that combines a Genetic Algorithm ( GA ), Artificial Neural Networks ( ANN ), and Monte Carlo ( MC ) analysis for the automated removal of heavy metals using adsorbents obtained from Low-Density Concrete waste. This research is significant because adsorption systems are complex to automate due to their nonlinear behaviour and sensitivity to fluctuating operating conditions. By facilitating intelligent process control and optimisation, the system overcomes the main drawback of adsorption technology, which is its lack of automation. Experimental datasets for Pb 2+ , Co 2+ and Mn 2+ removal were used to train the ANN , which predicts removal from inputs (pH, adsorbent mass, contact time, initial concentration). GA then used this trained ANN to optimise these same factors. According to the sensitivity analysis, the most important control variables in heavy metal adsorption were heavy metal type, pH, and adsorbent mass. The ANN model demonstrated strong predictive performance ( R 2 = 0.908, RMSE = 7.43 %), enabling GA to optimise process conditions and achieve removal efficiencies of 100 % (Pb 2+ ), 87.8 % (Co 2+ ), and 71.62 % (Mn 2+ ) under identified optimal operating parameters. Operational reliability measured by MC simulations showed Co 2+ had a higher variability ( CV : Coefficient of Variation = 5.39 %) than Mn 2+ ( CV = 3.29 %) and Pb 2+ ( CV = 2.13 %), which had minimal uncertainty. Using resources from the circular economy, the ES provides a framework for a digital twin of an Industry 4.0-compliant water treatment system that enables translation of lab-based procedures into practical applications.