Deep learning-based fault diagnosis and Electrochemical Impedance Spectroscopy frequency selection method for Proton Exchange Membrane Fuel Cell

Abstract

Electrochemical Impedance Spectroscopy (EIS) serves as a valuable tool for analyzing the health of Proton Exchange Membrane Fuel Cell (PEMFC). However, the practical application of EIS-based fault diagnosis algorithms continues to face challenges, including time-consuming EIS measurements, inefficient Equivalent Circuit Model (ECM) parameter estimation, and limited generalization capability of fault diagnosis models. To enhance the utility of the EIS-based fault diagnosis methods, this paper develops a new deep-learning-based PEMFC fault diagnosis framework, along with a frequency selection method. Specifically, a Parameter Estimation Unit (PEU) is introduced to derive the ECM parameters directly from EIS measurements. A Frequency Selection Unit (FSU) based on the Gumbel-Softmax trick is proposed to enable the identification of the optimal frequency solution to maximize fault diagnosis performance with a predetermined number of measured frequency points. Furthermore, a Feature Augmentation Unit (FAU) is proposed to generate robust diagnosis features based on ECM parameters, thus reducing the influence of non-fault operations on fault diagnosis results. Experiments based on real EIS data demonstrate the effectiveness of the proposed algorithm. With these innovations, the proposed framework could significantly improve the efficiency and accuracy of fault diagnosis in PEMFC.