Cylindrical Battery Fault Detection under Extreme Fast Charging: A Physics-based Learning Approach

TL;DR

This paper introduces a physics-based learning framework for real-time detection of internal faults in electric vehicle batteries during extreme fast charging, combining models, observers, and uncertainty learning for early fault detection.

Contribution

It develops a novel physics-based learning approach that integrates electrochemical-thermal models, detection observers, and Gaussian Process Regression to improve fault detection accuracy and timeliness.

Findings

Simulation results confirm early fault detection capability.

Experimental case studies validate the effectiveness of the approach.

Uncertainty learning enhances detection robustness.

Abstract

High power operation in extreme fast charging significantly increases the risk of internal faults in Electric Vehicle batteries which can lead to accelerated battery failure. Early detection of these faults is crucial for battery safety and widespread deployment of fast charging. In this setting, we propose a real-time {detection} framework for battery voltage and thermal faults. A major challenge in battery fault detection arises from the effect of uncertainties originating from sensor inaccuracies, nominal aging, or unmodelled dynamics. Inspired by physics-based learning, we explore a detection paradigm that combines physics-based models, model-based detection observers, and data-driven learning techniques to address this challenge. Specifically, we construct the {detection} observers based on an experimentally identified electrochemical-thermal model, and subsequently design the observer tuning parameters following Lyapunov's stability theory. Furthermore, we utilize Gaussian Process Regression technique to learn the model and measurement uncertainties which in turn aid the {detection} observers in distinguishing faults and uncertainties. Such uncertainty learning essentially helps suppressing their effects, potentially enabling early detection of faults. We perform simulation and experimental case studies on the proposed fault {detection} scheme verifying the potential of physics-based learning in early detection of battery faults.