Fault diagnosis of proton exchange membrane fuel cells based on CNN neural network

To address the limitations of traditional machine learning techniques, which have weak data analysis capabilities and struggle to process high-dimensional data, we utilized both a conventional Convolutional Neural Networks (CNN) with residual structure and a CNN optimized with Mini-batch Gradient Descent (MBGD) and Adaptive Moment Estimation (Adam) to diagnose faults in Proton Exchange Membrane Fuel Cells (PEMFCs). Empirical results demonstrate that all three methods can accurately identify membrane dry faults, flooding faults, normal states, and unknown states of fuel cells, and exhibit superior diagnostic performance compared to traditional machine learning algorithms like Support Vector Machine (SVM). Furthermore, the comparison of the optimization effect between Adam and MBGD can provide guidance for selecting the appropriate optimization algorithm for fault diagnosis in PEMFCs.