Data-driven copula Bayesian network for risk analysis of lithium-ion battery accidents

Abstract: Owing to their high energy density, low self-discharge rate and long service life, lithium-ion batteries (LIBs) are extensively employed. With the surge of application scope, the safety issues of LIBs have received increasing attention, especially thermal runaway accidents. Risk analysis on thermal runaway accidents is beneficial to develop the risk control measures of batteries. With the availability of accident statistics, data-driven approach can reduce the reliance on expert experience. The copula function can capture nonlinear features to characterize model uncertainty. This article proposes a data-driven copula Bayesian network (DCBN) model for risk analysis of LIB accidents. First, we collect records of battery thermal runaway accidents and process the data containing multiple failure modes and risk-influencing factors (RIFs). Second, we use copula functions to model the correlations between variables, particularly the nonlinear relationships between variables. Subsequently, we introduce the data-driven method to depict causal relationships between RIFs. Eventually, we establish a Bayesian network (BN) model with correlation analysis, predictive analysis, and diagnostic analysis to evaluate the probability distribution of fault modes and their impact on risk status. We conduct a case study using LIB data from the aviation domain. The results show that this model can assess risk on LIB accidents and support for battery risk control.