A METHOD FOR DATA-DRIVEN RISK ANALYSIS OF LITHIUM-ION BATTERY ACCIDENTS BY CONSIDERING UNCERTAINTY

With the expanding scope of applications, safety concerns related to lithium-ion batteries (LIBs) have garnered increasing attention, particularly in relation to thermal runaway incidents. Risk analysis of such events provides essential insights for the development of targeted risk mitigation strategies. Leveraging the availability of accident data, data-driven methodologies offer a means to reduce reliance on expert judgment. Copula functions, known for their ability to capture nonlinear dependencies, are employed to address model uncertainty. This study presents a data-driven copula Bayesian network model for the risk assessment of LIB-related accidents. Initially, thermal runaway accident reports are collected and preprocessed to extract relevant risk-influencing factors (RIFs). Copula functions are subsequently applied to model the dependencies among these variables. Following this, data-driven techniques are utilized to determine the causal relationships between the identified RIFs. A Bayesian network model is then constructed, incorporating correlation analysis, predictive modeling, and diagnostic evaluation to estimate the probability distributions of failure modes and their corresponding impact on risk levels. A case study based on LIB data from the aviation sector demonstrates that the proposed model effectively supports the assessment of accident-related risks and informs risk management strategies.