Prediction of Electrochemical Characteristics of Metallic Bipolar Plates for PEMFCs Using Machine Learning and Deep Learning Model

Abstract

In this research, machine learning and deep learning models were employed to develop a predictive model for the electrochemical characteristics of 316L stainless steel used in metallic bipolar plates of PEM fuel cells (PEMFCs). The results indicated that hydrogen ion concentration (pH) exhibited the strongest correlation with electrochemical characteristics, as determined through SHAP analysis, Spearman correlation analysis, and the random forest model. Furthermore, the interaction between hydrogen ion concentration and hydrofluoric acid significantly decreased corrosion resistance, while hydrogen peroxide enhanced corrosion resistance by forming an oxide film on the surface. When comparing the performance of the MLP, DNN_Adam, and ResNet_Nadam prediction models, the DNN and ResNet models outperformed the MLP model. This superior performance is attributed to the use of appropriate optimizers (Adam and Nadam) andBayesian optimization techniques to identify the optimal hyperparameters for each model. Notably, the ResNet_Nadam model emerged as the best predictor of electrochemical characteristics, achieving the highest R2 value and showing a difference of less than 10% between training and validation data loss in the learning curve. These findings suggest that a thorough understanding of the independent variable data and the application of optimal parameters can enhance the predictive performance of deep learning models.