Thermal diagnostics of high-current EV components with an artificial neural network model

The increasing complexity of thermal management in Electric Vehicle (EV) powertrains necessitates accurate and scalable predictive models for monitoring cable and connector temperatures. This study investigates the performance of a lightweight Artificial Neural Network (ANN) for predicting main switch, connector, and motor cable temperatures using nine input features encompassing electrical and thermal parameters. Model evaluation was conducted against Support Vector Machine (SVM) and linear regression baselines using R², MAE, and RMSE metrics. Results show that the ANN consistently outperformed benchmark models, achieving R² values above 0.96 and maintaining temperature deviations within ±2.5 °C across all test scenarios. Unlike SVM and linear regression, which required separate models for each output, the ANN successfully predicted multiple outputs within a single framework, demonstrating greater scalability. Feature importance analysis identified battery temperature and post-switch cable temperature as dominant predictors, findings corroborated by high- resolution thermal imaging that confirmed localized heating in these regions. Despite the strong performance, limitations include a relatively small dataset and laboratory-based validation, which may restrict generalizability. The outcomes highlight the ANN’s potential for embedded real-time monitoring and predictive maintenance in EV thermal management, while underscoring the need for future research using larger, more diverse, and field-derived datasets.