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

Area of Science:

• Materials Science
• Chemical Engineering
• Environmental Science

Background:

• Global freshwater scarcity necessitates advanced desalination technologies.
• Thin-film nanocomposite (TFN) membranes offer potential for improved water permeability and salt rejection in reverse osmosis (RO).

Purpose of the Study:

• To systematically evaluate factors influencing polyamide-based TFN-RO membrane performance.
• To develop and validate a predictive model for desalination efficiency.
• To provide data-informed guidance for TFN-RO membrane design and operation.

Main Methods:

• Integrated analytical framework using machine learning (ML), explainable AI (XAI), and particle swarm optimization (PSO).
• Evaluation of key factors: nanoparticle loading, temperature, pressure, feed concentration, contact angle, NP size, and pore size.
• Development of a novel XGBoost-GS ML model and comparison with established models.
• Application of SHAP and PDP for understanding feature-output relationships.

Main Results:

• The XGBoost-GS model achieved high prediction accuracy for water flux (0.9366) and salt rejection (0.9449).
• Identified key parameters and their influence on membrane performance through XAI techniques.
• Clarified complex relationships, including nonlinear effects and interactions between parameters.

Conclusions:

• The study provides actionable insights for optimizing TFN-RO membrane design and operation.
• Multi-objective strategies for enhancing water flux and salt rejection were identified.
• The integrated ML/XAI framework offers a powerful tool for materials discovery and process optimization in desalination.