Uncovering Structure-Conductivity Relationships in Anion Exchange Membranes (AEMs) Using Interpretable Machine...

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry
• Machine Learning

Background:

• Anion exchange membranes (AEMs) are critical components in electrochemical energy conversion devices like fuel cells and water electrolyzers.
• The complex relationship between AEM structure and ionic conductivity hinders efficient material discovery and optimization.
• Data-driven approaches are needed to accelerate the design of advanced AEMs.

Purpose of the Study:

• To develop and apply a machine learning framework for predicting and interpreting ionic conductivity in AEMs.
• To identify key descriptors governing ionic conductivity using descriptor-based and graph-based machine learning models.
• To accelerate the data-driven design of high-performance AEMs.

Main Methods:

• Utilized a machine learning framework combining conditional graph neural networks (cGNNs), descriptor-based models, and a Hybrid Graph Autoencoder-Regressor Ensemble (HGARE).
• Employed Principal Component Analysis (PCA), ablation studies, and SHAP analysis for descriptor identification in the descriptor-based pipeline.
• Applied dimensionality reduction (t-SNE, SOM) and clustering (KMeans) for membrane analysis, alongside Graph Convolutional Networks (GCN) and HGARE for predictive modeling.

Main Results:

• Descriptor-based analysis identified electronic, topological, and compositional factors as crucial for anion conductivity.
• Dimensionality reduction and clustering revealed distinct membrane groups, some exhibiting high ionic conductivity.
• The HGARE model achieved the highest predictive accuracy for ionic conductivity, outperforming other graph-based methods like GCN.
• GCN atom-level saliency maps highlighted the importance of polarizable and flexible regions for conductivity.

Conclusions:

• The developed machine learning framework effectively predicts and interprets ionic conductivity in AEMs.
• Key material descriptors influencing conductivity were identified, guiding future AEM design.
• This work demonstrates a significant advancement in the accelerated, data-driven discovery of high-performance AEMs for energy applications.