KAN-rCBF framework uses AI to estimate battery core temp for safer fast charging

Researchers Faysal Ahamed and Tanushree Roy from the arXiv paper 2608.12638 propose a new framework that uses Kolmogorov-Arnold Networks (KAN) combined with robust control barrier functions (rCBF) to enforce thermal safety during lithium-ion battery fast charging. The problem: core temperature cannot be measured directly, so charging must be controlled conservatively or risk overheating. Their approach, called KAN-rCBF, estimates the core temperature in real time from surface temperature, coolant temperature, coolant power, and charging current, then uses that estimate to solve a quadratic programming problem that respects safety constraints. This enables aggressive charging while maintaining a safe thermal envelope.

The key innovation is that the framework provides analytical safety guarantees even when the KAN estimation has errors or model uncertainty exists. In simulations, the method kept battery temperature within safe limits while achieving charging times comparable to state-of-the-art charging policies—which, the authors note, cannot guarantee the same level of thermal safety. This means the KAN-rCBF approach could allow EV batteries to charge faster without sacrificing safety, potentially reducing charging times while preventing thermal degradation or runaway. The 14-page paper includes 3 figures and is available on arXiv with DOI 10.48550/arXiv.2608.12638.