Ensemble Learning-Based Fault Diagnosis of Electric Vehicle Lithium-Ion Batteries Using Operational Data

Abstract

In this study, Lithium-ion batteries, while being the fundamental energy component of electric vehicles, can experience critical failures such as internal short circuits and over-discharge, triggering thermal runaway risks. This study proposes a high-accuracy, ensemble learning-based fault diagnosis system to improve operational safety in battery management systems (BMS). To this end, the model was trained using a numerical dataset of lithium-ion battery systems. Within the proposed methodology, the performance of eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), Random Forest (RF), Categorical Boosting (CatBoost), Extremely Randomized Trees (ExtraTrees), and Adaptive Boosting (AdaBoost) algorithms was analyzed in both binary and multi classification. The effectiveness of the models was evaluated using accuracy, sensitivity, recall, and F1 score. Experimental results show that ensemble learning-based methods exhibit consistent and balanced performance in both classification processes. In the binary classification process, the ExtraTrees model achieved the highest performance with 80.14% accuracy, while in the multi-classification process, the CatBoost model stood out with 92.64% accuracy and 0.9448 AUC. These findings demonstrate that ensemble learning-based approaches offer a viable and reliable framework for lithium-ion battery fault diagnosis problems.