Machine learning analysis and prediction models of alkaline anion exchange membranes for fuel cells

Abstract

The degradation of anion exchange membranes (AEMs) hindered the practical applications of alkaline membrane fuel cells. This issue has inspired a large number of both experimental and theoretical studies. However, it is highly difficult to draw universal laws from the resulting data. Here, for the first time, artificial intelligence (AI) technology was presented to forecast the chemical stability of AEMs for fuel cells. The chemical stability of AEMs was quantified by Hammett substituent constants based on a materials genomics strategy, and then classified by a decision tree. Among five machine learning algorithms applied, the artificial neural network (ANN) showed the highest accuracy in predicting the chemical stability of AEMs (R² = 0.9978). Combined with the computational works, long-term chemical stability experiments were conducted to demonstrate the robustness and prediction accuracy of the proposed approach. This study highlights the potential of data-driven modelling for predicting the alkaline stability of AEMs, and thus unnecessary experiments can be avoided for the development of alkaline membrane fuel cells.