Enhancing Electric Vehicle Safety: AI-Driven Multiphysics Approach for Predicting Thermal Failures in Li-Ion Batteries

Abstract

The focus of this dissertation is to address the critical issue of safety in lithium-ion batteries (LIBs), particularly focusing on thermal runaway (TR), a significant threat to their reliability and widespread adoption. With the surge in demand for energy storage driven by electric vehicles (EVs) and renewable energy uptake, ensuring LIB safety is paramount. The research innovatively combines advanced machine learning (ML) techniques with multiphysics modeling to predict and prevent TR incidents. Initially, I applied convolutional neural networks (CNNs) and the “you look only once” (YOLO) object detection model to forecast TR in single LIBs. This phase leveraged simulated thermal images generated by a comprehensive multiphysics model that incorporates thermal, electrochemical, and degradation processes, particularly focusing on solid electrolyte interface (SEI) dynamics. The models demonstrated high accuracy in predicting TR stages and pinpointing heat sources within a battery. Expanding this approach, the I, then developed a novel ML framework to address TR in LIB modules, which is crucial for EV safety. This framework integrates graph neural networks (GNN) for spatial analysis and Long Short-Term Memory (LSTM) networks for temporal predictions, which are trained on sensor-derived temperature data. The advanced model effectively identified spatio-temporal temperature variations and potential hotspots within battery modules, offering real-time insights to avert TR. This research significantly advances the safety and reliability of LIBs by integrating cutting-edge ML with detailed multiphysics modeling. It marks a substantial step towards safer, more sustainable energy storage solutions, contributing to the broader goal of a low-carbon future by enhancing LIB integration in diverse applications, from portable electronics to EVs and renewable energy systems.