DECODING PHYSICOCHEMICAL AND OPERATIONAL DRIVERS OF ADSORPTION CAPACITY FOR PROCESS OPTIMIZATION

Abstract

The adsorption performance depends on the complex interplay of material properties, solution chemistry and processing conditions, and prediction is important for process optimization. In this study, 6,172 observations were analyzed to examine the physicochemical and operational factors affecting the adsorption capacity and included specific surface area, pore volume, material properties, pH, temperature, initial concentration, flow rate, and column height. The statistical association analysis was combined with linear regression and nonlinear machine-learning models. Initial concentration had the highest correlation with adsorption capacity, and contributed to 82.98% of the feature importance for the random-forest, suggesting that it was the most important factor. Random forest has the greatest predictive accuracy (R² = 0.9725, RMSE = 13.65 mg/g, MAE = 3.55 mg/g) than gradient boosting and linear regression. Nonlinear effects and interactions between the main adsorption determinants were also found from the response patterns. The results show that combining the physicochemical interpretation with the nonlinear predictive modeling approach is an effective way to identify the main factors that control the adsorption process and offers a practical way to optimize the adsorption process and remove contaminants.