Machine learning – based on analysis of EV battery thermal runaway simulation
摘要
Thermal runaway prevention of electric vehicle batteries was simulated according to five different thermal runaway prevention films (mica, aerogel, glass fiber, carbon composite, and polyurethane) and internal short circuit locations (front type, side type). Random forest and long short-term memory (LSTM) models were developed to analyze the patterns and trends of the simulation data based on the data. The results of the study show that when the internal short circuit is of the side type, the onset time of thermal runaway of the battery is delayed and the maximum temperature is reduced. In addition, when carbon composite and mica film are applied, the temperature gradient (∇T) inside the battery is reduced due to the high thermal conductivity compared to other materials, which reduces the formation of hot spots at specific points and delays the chain reaction of thermal runaway. The results of the machine learning model evaluation show that the average mean absolute error (MAE) of the random forest is 4.07×10−2 and the average mean relative error (MRE) was 1.01×10−2 %, which is higher than that of the LSTM model. This study provides an efficient analytical approach that combines simulation and machine learning, which can be utilized in future data-driven battery safety evaluation processes and is expected to contribute to the research of CFD-ML-based hybrid physical models.