Identification of electric vehicle susceptible to thermal runaway based on risk accumulation

Accurate fault diagnosis of power batteries is significant to ensure safe operation of electric vehicles. Most of existing methods rely heavily on real-time collection of battery status parameters, including voltage, current, and temperature, by the on-board battery management system to facilitate diagnosis. Nevertheless, these approaches suffer from inherent latency issues, lacking precision in risk predictions, and unable to accurately provide insights on vehicles that do not conform to the diagnostic criteria. To overcome these limitations, an innovative fault diagnosis method is proposed in this paper, where faulty vehicle diagnosis is done based on risk accumulation throughout usage history instead of current states of batteries. As the first step, various harmful usage behaviors and harsh environmental conditions that may cause potential damage on batteries are identified, and cumulative risk measure can be calculated to establish quantitative representations of multiple types of risks. On top of this, Logistic Regression algorithm with discarded deviation terms is used to generates quantitative mathematical expressions for classification. Validation is conducted on data of two types of electric vehicles, where an accuracy of 87.5 % for vehicle type A and accuracy of 84.8 % for vehicle type B can be achieved even before optimization of the algorithm. And higher accuracy can be achieved through recursive optimization. Benefit from the fundamental logic of risk accumulation through usage history, the proposed method can be performed offline with much lower computational complex compared with most existing big databased diagnosis methods, thus suitable for large-scale analysis. The proposed method can also be used for vehicle design optimization through sensitivity analysis of risks and comparative study over multiple types of vehicles, opening up a new way for vehicle upgrade.