Lithium-ion Battery Risk Assessment for New Energy Vehicles Based on Bayesian Network

Accurate alarms for Lithium-ion battery faults are essential to ensure the safety of New Energy Vehicles(NEVs). Related research shows that the change characteristics of the battery are important parameters reflecting the fault of NEVs. In this study, the ferrous lithium phosphate batteries data of 30 NEVs for 9 months in the National Monitoring and Management Center for New Energy Vehicles (NMMC-NEV) was analyzed. In addition to the measurement defined in GB/T 32960–2016, new indicators that can reflect the characteristics of battery changes are also analyzed, which are: Euro per volt (Ω/V), the maximum cell voltage difference, and the maximum cell temperature difference. Binary Logistic regression analysis was used to extract characteristic indicators related to three fault alarms: UnderVoltage_Alarm, OverVoltage_Alarm, and Insulation_Alarm, and the definition of the fault level is defined according to GB/T 32960-2016: satisfied, acceptable, tolerable, and unacceptable. A novel quantitative evaluation method for battery risk assessment was proposed by using Bayesian networks. Then, the robustness and reliability of the model was verified based on historical operating data, when some kind of fault alarm occurs (the probability of the fault alarm is 100%), the risk assessment result is consistent with the risk status interval divided according to the fault level. Verification results show that this study can assess potential fault risk timely and accurately.