Thermal Risk Monitoring In Electric Vehicle Batteries Using External Sensors and Machine Learning Algorithms

Thermal safety remains a critical challenge for electric vehicles, as abnormal temperature evolution in lithium ion batteries may lead to severe degradation or catastrophic failure. Conventional battery thermal monitoring is primarily performed by the Battery Management System, which relies on internal sensors, fixed thresholds, and proprietary data access. These characteristics limit adaptability to real world operating conditions and restrict deployment in retrofitted or third party scenarios. This paper presents a machine learning-based approach for battery thermal monitoring and early thermal risk awareness using exclusively external, non-intrusive temperature sensing. A low-cost sensing architecture based on an ESP32 microcontroller was developed to collect battery surface temperature, ambient temperature, and operational data during real-world riding experiments with an electric motorcycle. The collected telemetry was transmitted via a mobile gateway and processed in a backend infrastructure for temporal data analysis. Battery thermal behavior was modeled as a supervised regression problem with temporal dependence, and multiple machine learning algorithms, including K Nearest Neighbors, Decision Trees, Random Forest, Gradient Boosting, and Multilayer Perceptron, were evaluated. Results demonstrate that supervised learning models incorporating engineered temporal features can accurately capture battery surface temperature evolution, despite minor data losses and measurement noise. The findings indicate that external surface temperature measurements combined with temporally informed regression models can provide meaningful early thermal trend awareness without reliance on internal Battery Management System signals. This approach offers a practical, deployable, and non intrusive complementary safety layer for real world electric vehicle battery monitoring.