Chalmers AI extends EV battery life 23%

Chalmers University of Technology said on May 12 that its researchers had developed an artificial-intelligence charging strategy that extended electric-vehicle battery life by 22.9% in simulations without increasing charging time. The work was led by Meng Yuan and Changfu Zou at the university’s Department of Electrical Engineering, according to a Chalmers news release and the paper record. The study describes a reinforcement-learning system designed to adapt fast charging as a battery ages, rather than relying on fixed charging limits. The paper appears in *IEEE Transactions on Transportation Electrification*, according to the version of record listed by Chalmers. ### How much longer did the batteries last in the reported tests? Chalmers said the battery lifetime increased by 22.9% when measured in equivalent full cycles, a standard count of how many full charge-discharge cycles a battery can complete before capacity falls to 80% of its original level. The university said that 80% threshold is commonly treated as end of life for electric-vehicle use. (chalmers.se) The same release said charging time was virtually unchanged. Chalmers reported an average charging time of 24.12 minutes with the new method, compared with 24.15 minutes for the standard approach. ### What exactly did the AI system do differently? The paper by Yuan and Zou said the method used lifelong reinforcement learning to build a health-aware fast-charging strategy for lithium-ion batteries. (chalmers.se) The authors wrote that the system maps a battery’s state of health to a charging-voltage constraint and uses a twin delayed deep deterministic policy gradient, or TD3, framework to balance charging speed against long-term degradation. Chalmers said the practical target was battery wear that changes over time. “Batteries change over time. But charging strategies typically do not,” Yuan said in the university release. Zou said the bottleneck in fast charging was “the evolving electrochemical state inside the battery,” according to the same release. ### Why does fast charging damage batteries in the first place? (research.chalmers.se) Chalmers said rapid charging pushes high currents into the cell and can trigger side reactions. The release identified lithium plating as one of the most critical of those reactions, describing it as metallic lithium depositing on the electrode instead of being stored properly in the battery structure. (chalmers.se) The university said lithium plating reduces capacity, raises internal resistance and, in severe cases, can affect safety. Chalmers also said the risk becomes more pronounced as the battery ages, which is why the researchers focused on a charging method that changes with battery condition over its service life. ### Were these results shown in real cars or in simulation? (chalmers.se) The paper said the validation used a high-fidelity single-particle model with electrolyte implemented in the PyBaMM simulation platform. The authors wrote that the model captured degradation phenomena at realistic scales and compared the reinforcement-learning controller with conventional constant-current/constant-voltage charging, its variants and constant current-constant overpotential methods. (chalmers.se) Chalmers described the result as a study showing that rapid charging and lower battery wear could be combined using artificial intelligence. The university release did not say the method had yet been deployed in commercial vehicles or public charging networks. ### Where could this show up next? Chalmers said the work points toward health-aware charging strategies for advanced battery-management systems. (research.chalmers.se) The paper said the results demonstrated the “practical viability” of deep reinforcement learning for that use and set up future work on performance-optimized charging strategies. The next concrete reference point is the published study itself. Chalmers lists the article by Meng Yuan and Changfu Zou as appearing in *IEEE Transactions on Transportation Electrification*, with DOI 10.1109/TTE.2025.3625421. (chalmers.se) (research.chalmers.se)