Machine Learning Models for Predictive Maintenance of EV Thermal Systems Reducing Catastrophic Failure Risk

Electric Vehicles (EVs) use complicated thermal management systems to maximize the performance of the battery and power electronics. Predictive maintenance based in Machine Learning (ML) can change the way we identify and preemptively mitigate thermal system issues before they become undesirable or potentially catastrophic failures. This paper proposes a data-driven framework incorporating supervised learning models - XGBoost, Random Forest, and LSTM networks - to predict thermal system failures in EVs. Historical operational data pertaining to coolant temperature, compressor load, ambient conditions, and battery temperature are utilized to engineer features to construct the models. We plan to evaluate the models using Root Mean Square Error (RMSE), Area Under the Curve (AUC), and F1-score metrics. Our results show that ensemble methods, in combination with temporal models, have the highest predictive accuracy and act as an early warning system for intervention. This early warning system will not only improve vehicle reliability and safety but also reduce costs for warranty claims and unplanned service events.