Data-Driven Fault Diagnosis for PEM Fuel Cell System Using Sensor Pre-Selection Method and Artificial Neural Network Model

Fault diagnosis is a critical process for the reliability and durability of proton exchange membrane fuel cells (PEMFCs). Due to the complexity of internal transport processes inside the PEMFCs, developing an accurate model considering various failure mechanisms is extremely difficult. In this paper, a novel data-driven approach based on sensor pre-selection and artificial neural network (ANN) are proposed. Firstly, the features of sensor data in time-domain and frequency-domain are extracted for sensitivity analysis. The sensors with poor response to the changes of system states are filtered out. Then experimental data monitored by the remaining sensors are utilized to establish the fault diagnosis model by using the ANN model. Levenberg-Marquardt (LM) algorithm, resilient propagation (RP) algorithm, and scaled conjugate gradient (SCG) algorithm are utilized in the neural network training, respectively. The diagnostic results demonstrate that the diagnostic accuracy rate reaches 99.2% and the recall rate reaches 98.3% by the proposed methods. The effectiveness of the proposed method is verified by comparing the diagnostic results in this work and that by support vector machine (SVM) and logistic regression (LR). Besides, the high computational efficiency of the proposed method supports the possibility of online diagnosis. Meanwhile, detecting the faults in the early stage can provide effective guidance for fault tolerant control of the PEMFCs system.