Fast charging of lithium-ion batteries generates significant internal heat that can damage cells, reduce lifespan, and create fire hazards. Yet battery core temperatures cannot be directly measured during operation, forcing engineers to rely on conservative charging protocols that sacrifice speed for safety. A new framework bridges this gap using machine learning to predict internal temperatures from external measurements, enabling smarter, safer charging algorithms.
The approach combines Kolmogorov-Arnold Networks (KAN)—a flexible neural network architecture—with control barrier functions to estimate core temperature from observable data: surface temperature, coolant temperature, coolant cooling power, and charging current. These estimates feed into an optimization algorithm that determines the optimal charging current while respecting strict thermal safety limits.
The key innovation is providing mathematical guarantees that the charging protocol remains safe even when the KAN predictions contain errors or when the battery model differs from actual physics. This "robust" design prevents the system from over-relying on imperfect estimations. The algorithm solves a convex optimization problem in real time, balancing three competing objectives: maintain safe temperatures, reach full charge quickly, and minimize wear.
Simulation results demonstrate the method achieves charging times comparable to state-of-the-art approaches while maintaining stricter thermal safety margins. Critically, competing methods that ignore temperature estimation uncertainty fail to guarantee safety across all operating conditions.
For the power and energy sector, this work has immediate relevance to electric vehicle charging infrastructure, where fast charging is commercially essential but thermal runaway remains a liability. It also applies to battery energy storage systems supporting grid services, where safe rapid cycling extends asset life and improves economics. The framework's data-driven core temperature prediction could reduce the need for expensive thermal sensors or complex cooling systems, lowering system costs while improving reliability. As EV adoption accelerates and battery cycling rates increase, robust thermal management becomes central to safe grid integration.
