A Magnetic Field-Based Fault Localization Method for Proton Exchange Membrane Fuel Cell Stacks

Existing proton exchange membrane fuel cell (PEMFC) stack fault diagnosis mainly focuses on fault identification, whereas effective and practical approaches for accurately localizing faults within the stack remain limited. Magnetic field offers the promising in-situ and nondestructive approach, while fault localization in PEMFC stack using magnetic field information still faces critical challenges. In this article, the multiphysical PEMFC stack model and magnetic array strategy are developed as theoretical foundation and hardware support for PEMFC stack fault localization. On such basis, an enhanced dual-feature selection random forest method is proposed, where dual-feature selection captures both global and local features with physical interpretability. Results show this method achieves high localization accuracy of 92.67% on simulation dataset and fault diagnosis accuracy of 96.66% on experiment dataset under simulated industrial electromagnetic interference conditions, which demonstrate strong robustness and applicability, and outperform conventional methods in accuracy and computational cost. The findings enable cell-level fault localization within stack, and offers the promising technique for in-situ and on-line PEMFC fault localization in practical applications.