<p>The degradation of amine-based solvents, such as methyldiethanolamine (MDEA) and diethanolamine (DEA), due to the accumulation of heat-stable salts (HSS), poses a significant challenge to the efficiency, safety, and sustainability of acid gas removal systems. Ion exchange using strong-base anion resins has been widely adopted as a practical method for HSS removal; however, the optimization and control of this process remain challenging due to its nonlinear and multivariable nature. In this study, a predictive model based on artificial neural networks was developed to estimate the residual HSS concentration and final solution pH following ion exchange purification. A comprehensive dataset <br /> representing industrially relevant variations in key process parameters-initial HSS concentration, amine strength, flow rate, initial pH, and temperature-was generated and used for model training in MATLAB. The ANN architecture consisted of a two-hidden-layer feedforward network trained using Bayesian regularization, enabling robust learning without overfitting. The model achieved high correlation coefficients of R = 0.9586 for overall prediction, R = 0.9173 for HSS concentration, and R = 0.9898 for pH prediction. Error histograms demonstrated low and symmetrically distributed residuals, confirming the model’s accuracy and generalization capabilities. These results confirm that ANNs can serve as a reliable surrogate model for real-time monitoring and predictive control in solvent purification systems. The proposed methodology contributes to the development of intelligent process optimization in chemical engineering, enhancing operational efficiency and reduced environmental impact.</p>