Exploring the Knowledge Attained by Machine Learning on Ion Transport across Polyamide Membranes Using Explainable...

Area of Science:

• Membrane Science and Technology
• Materials Science
• Computational Chemistry

Background:

• Machine learning (ML) is increasingly used for membrane separation performance and material design.
• The ability of ML to capture fundamental principles of membrane science with limited data remains uncertain.
• Understanding ML's learned knowledge is crucial for reliable applications in membrane technology.

Purpose of the Study:

• To investigate the knowledge learned by ML regarding ion transport mechanisms in polyamide membranes.
• To validate if ML can capture fundamental principles of membrane science using explainable AI (XAI).
• To provide a framework for evaluating ML model interpretability in membrane applications.

Main Methods:

• Applied explainable artificial intelligence (XAI) to analyze ML models trained on 1,585 data points from 26 membrane types.
• Utilized the Shapley additive explanation (SHAP) method to determine feature importance for ML predictions.
• Investigated ion transport mechanisms across polyamide reverse osmosis (RO) and nanofiltration (NF) membranes.

Main Results:

• XAI confirmed that ML models correctly identified the roles of size exclusion and electrostatic interactions in membrane separation.
• ML models demonstrated distinct mechanisms for cation and anion rejections in RO and NF processes.
• The study provides a robust framework for assessing the interpretability of ML models in membrane science.

Conclusions:

• Machine learning models are capable of learning and representing fundamental mechanisms of ion transport in polyamide membranes.
• Explainable AI is essential for validating and understanding the knowledge acquired by ML in membrane science.
• Enhanced model interpretability will lead to more reliable and explainable ML applications for membrane selection and design.