Machine learning-guided predictive modeling and optimization of polyamide-based thin-film nanocomposite reverse...

The growing scarcity of global freshwater resources, coupled with steady advancements in seawater desalination technology, has made the development of high-performance reverse osmosis (RO) membranes a critical priority. Among various membrane types, the thin-film nanocomposite (TFN) membranes represent a promising route, owing to their potential to improve both water permeability and salt rejection simultaneously. In this study, an integrated analytical framework-incorporating machine learning (ML), explainable artificial intelligence (XAI), and particle swarm optimization (PSO)-is utilized to systematically evaluate how key factors-including nanoparticle (NP) loading, operating temperature, transmembrane pressure, feed concentration, membrane water contact angle (WCA), NP size, and membrane pore size-impact the desalination performance of polyamide-based TFN-RO membranes, using a carefully assembled dataset from literature. A novel ML model, XGBoost-GS, is proposed and compared to established ML models, achieving top prediction performance w.r.t. two key performance indicators (KPIs) - Water Flux and Salt Rejection, reaching test-set R² values of 0.9366 and 0.9449, respectively. To unpack the black-box relationships between input features and membrane outputs, Pearson correlation analysis, SHapley Additive exPlanations (SHAP)-based feature importance ranking, and univariate/bivariate Partial Dependence Plot (PDP) analysis are applied. This multifaceted approach clarifies the pathways, relative importance, nonlinear marginal effects, and interaction effects through which membrane properties and operating conditions influence Water Flux and Salt Rejection. From this integrated perspective, practical sensitivity ranges and synergistic tuning directions are identified for key parameters, proposing multi-objective strategies to achieve high water flux and salt rejection. Ultimately, the obtained results provide data-informed, actionable guidance for the rational design and operational matching of polyamide-based TFN-RO membranes.