Machine Learning for Predicting Thermal Runaway in Lithium-Ion Batteries With External Heat and Force

The current study aims to predict the thermal runaway in lithium-ion batteries using five artificial intelligence algorithms, considering the environmental factors and various design parameters. Multiple linear regression, k-nearest neighbors, decision tree, and random forest are used as machine learning algorithms, while artificial neural networks are used as deep learning algorithms. Nineteen experimental datasets are used to train the models. First, Pearson's correlation matrix is used to investigate the effects of input parameters on the thermal runaway onset time. The dataset is then updated to include only tests with thermal runaway produced by an external heat source. As a result of comparison among model performance prediction, it is determined that the decision tree model is the best-performing model with a coefficient of determination (R²) score of 0.9881, followed by random forest, k-nearest neighbors, artificial neural networks, and multiple linear regression models. The dataset is modified when the thermal runaway is triggered by external heating and compression forces. Results show that in this case, the performance of the decision tree model has an R² of 0.9742. Finally, the force range in which the model has the best performance is predicted, which is helpful in conducting tests to obtain reliable results.