Predicting and understanding the performance of polyamide nanofiltration membrane for Li/Mg selective separation...

Area of Science:

• Materials Science
• Chemical Engineering
• Machine Learning

Background:

• Nanofiltration membranes are key for separating monovalent (Li) and multivalent (Mg) ions in lithium extraction from salt lakes.
• Optimizing polyamide membranes for selective ion separation is challenging due to complex ion-membrane interactions and ambiguous transport mechanisms.

Purpose of the Study:

• To employ machine learning (ML) to identify key features influencing nanofiltration membrane permeability and selectivity.
• To develop robust ML models for predicting membrane performance in Li/Mg separation.

Main Methods:

• Utilized a comprehensive dataset including fabrication parameters, experimental conditions, membrane properties, and salt rejection performance.
• Applied ML algorithms and the Shapley additive explanation (SHAP) method to analyze feature importance.
• Systematically compared various input variable combinations to optimize ML model configurations.

Main Results:

• ML models accurately predicted intrinsic membrane properties but struggled with permeation and selectivity using only fabrication or property data.
• Incorporating salt rejection performance significantly improved ML model accuracy for predicting permeability and selectivity.
• Substrate type and heat curing temperature primarily determined membrane permeability, while MgCl2 rejection effectively captured factors governing Li/Mg selectivity.

Conclusions:

• A multifaceted approach integrating intrinsic membrane properties and external factors is necessary for accurate performance prediction.
• MgCl2 rejection is a critical descriptor for understanding and optimizing Li/Mg ion selectivity in nanofiltration.
• ML, particularly when incorporating salt rejection data, offers a powerful tool for advancing nanofiltration membrane design for selective ion separation.