Real-time data-driven fault diagnosis of proton exchange membrane fuel cell system based on binary encoding convolutional neural network

Abstract

The performance of proton exchange Membrane fuel cell (PEMFC) fault diagnosis system plays an important role in normal operation of PEMFC. Therefore, a new fault diagnosis algorithm based on binary matrix encoding neural network called BinE-CNN is proposed. In BinE-CNN, high-dimensional features are extracted through binary encoding, and the feature maps are transferred to a convolutional neural network (CNN) to realize seven-category fault classification. For development of BinE-CNN, a PEMFC model is modeled to generate simulative datasets. Simulative test precision and Frames per second (FPS) of BinE-CNN have reached respectively 0.973 and 999.8 (better than support vector machines (SVM), long short-term memory neural network (LSTM), etc.). In experimental verification section, fault datasets are collected during bench test. After that, BinE-CNN is deployed on vehicle control unit (VCU) to verify its engineering value (real-time and precision). The result meet both requirements, with time cost of 96.15 ms and precision of 0.931 . • Binary encoding (BinE) is used as data preprocessing to extract abstract details. • A multi-classification multi-level fault diagnosis algorithm: BinE-CNN is designed. • AVL CRUISE M is used to generate simulation fault datasets. • BinE-CNN is translated into C language and deployed on a VCU (Infineon TC275). • An over-120kW FCS to used verify the real-time and precision of BinE-CNN.