Water Management Fault Diagnosis of Polymer Electrolyte Membrane Fuel Cells Based on Complementary Ensemble Empirical Mode Decomposition and Sensitivity Analysis

Polymer electrolyte membrane fuel cells (PEMFCs) experience significant performance deterioration due to flooding and membrane dehydration. In order to maintain optimal PEMFC operation, it is crucial to effectively and accurately diagnose issues related to water management faults. Cell voltage has been employed widely for diagnosing PEMFC faults in existing research. However, because various PEMFC faults exhibit a similar drop in cell voltage, it is challenging to identify distinguishable features for flooding and membrane dehydration. In this study, an easily adopted and efficient water management fault diagnosis method using complementary ensemble empirical mode decomposition (CEEMD) and sensitivity analysis is proposed, in which the representative features are extracted by decomposing PEMFC voltage into intrinsic mode functions (IMFs) using CEEMD, whereas sensitivity analysis is applied for identifying the most representative features of PEMFC faults, thereby facilitating effective fault classification. Finally, a support vector machine (SVM) is employed to distinguish different PEMFC states with selected features. The generalizability of the proposed method was investigated using experimental data from two different membrane electrolyte assemblies (MEAs) under various states. The results demonstrated that the proposed method outperforms the previously used signal decomposition and features selection techniques. The proposed method had accuracy of 99.8% in distinguishing between normal, flooding, and dehydration conditions.