Gagana S, Pavan K, Harsha A P, Sharon M , Robin Rohit Vincent , Prashanth Kannadaguli "Machine Learning on IoT Telemetry in a Digital Twin-Driven Predictive Maintenance Framework for Electric Vehicle Batteries "

Paper_id: 15_2

Abstract

Electric​‍​‌‍​‍‌​‍​‌‍​‍‌ vehicles (EVs) rely on lithium-ion batteries that eventually wear out, which is a primary contributor to their performance, safety, and lifespan issues. Here, we propose a Digital Twin–based predictive maintenance system that visualizes operational data, machine learning (ML), and cloud pipelines to localize, quantify, and forecast the most important health indicators such as State of Health (SoH), Battery Temperature, Remaining Useful Life (RUL), and Failure Probability. The system leverages the EVIoT-PredictiveMaint Dataset, which is a multi-modal IoT telemetry resource covering five years of EV operational data. A Random Forest Regressor on these data is able to short-term degradation trends of SoH with high precision. The system coordinates streaming through Kafka/MQTT, TimescaleDB for time-series storage, and Flask–Grafana for communication, making it real-time inference, and visualization of results. The SoH (R² = 0.85) and temperature (R² = 0.78) prediction accuracies along with the inference latency of fewer than 5 ms are demonstrated by the experiments. The integration of digital twin and predictive analytics the framework discussed is an excellent instrument of the implementation of preventative maintenance, enhanced reliability, and enhanced sustainability of EV battery systems.