Vision-Based Early Detection of Thermal Runaway in Battery Packs Using Physics-Guided Deep Learning - [Work in Progress]

Date:

1. Motivation

  • Lithium-ion batteries are used in EVs, grid-scale storage, consumer electronics, and aerospace. Across all domains, thermal runaway (TR) is a leading safety concern, capable of causing fires and explosions.

Problem: Existing warning systems rely on embedded sensors (temperature, voltage, pressure) that are slow to respond or provide signals only after failure is underway.

Gap: There is little to no research demonstrating a vision-based, physics-informed approach to detect TR earlier and more reliably.

2. Objective

  • Develop a computer-vision + physics hybrid method using infrared (IR) thermal imagery to:

  • Detect incipient hotspots across battery modules.

  • Forecast heat propagation using a heat-equation–constrained predictor.

  • Issue alarms 30–50% earlier than simple threshold or sensor-based methods.