Research on high-precision fault identification of proton exchange membrane fuel cell experiment based on multiple correlation analysis and deep learning

Abstract

The rapid application of Proton exchange membrane fuel cell (PEMFC) in transportation and energy sectors, driven by advancements in hydrogen energy technology, underscores the critical importance of ensuring operational safety for widespread adoption. Intelligent diagnosis with high precision and robustness is imperative to address the primary challenge of performance and longevity. This study introduces an intelligent diagnostic framework tailored for identifying flooding faults in PEMFC, integrating feature optimization, sample enhancement, and model refinement. Initially, a feature selection approach leveraging Pearson and Spearman weighted fusion is devised to identify key physical parameters highly correlated with flooding by considering both linear and nonlinear relationships. Subsequently, a sliding window sample amplification strategy is implemented to enrich the local dynamic features of time series data, enhancing the model's ability to perceive fault evolution. Lastly, a weighted pooling convolutional neural network (CNN) model with adaptable channel weights is proposed, achieving a fault recognition accuracy of 99.9 % on the test dataset and demonstrating robust generalization on an independent dataset. This methodology offers a novel avenue for reliably identifying PEMFC flooding faults, crucial for ensuring system safety and enabling intelligent operational maintenance practices.