A sampling fault diagnosis method for power battery data in cloud platforms based on a ResNet–BiLSTM neural network

Abstract

As the basis for many functions of the battery management system (BMS) such as state estimation and thermal runaway warning, stable sampling data are crucial for the safe operation of electric vehicles (EVs). In this paper, a sampling fault diagnosis method for power battery data in cloud platforms is proposed based on a residual network (ResNet) and bi-directional long short-term memory (BiLSTM) neural network, which can effectively identify the abnormalities of the battery sampling data and recognize the failure modes. Firstly, through the analysis of fault data and sampling circuits for real EVs, four typical failure modes are selected to complete the fault injection experiments. The physical simulation model of the fault circuit is established, and the corresponding mathematical empirical model is condensed. Then, based on the understanding of the abnormal data distribution pattern, the fault diagnosis algorithms based on a threshold and the ResNet–BiLSTM neural network are developed, respectively. Finally, the algorithms are introduced into the simulation dataset and real-vehicle dataset for testing. The results show that both algorithms have high effectiveness and accuracy, with the latter exhibiting strong fault diagnosis capability. In summary, the proposed sampling fault diagnosis method is feasible and provides a theoretical basis for future multi-type fault diagnosis of BMSs.