Fast charging has become essential for modern EV adoption, especially for long-distance travel and high-utilisation fleets such as taxis and delivery vehicles. But the convenience comes with a serious hidden cost: pushing high currents into lithium-ion cells accelerates internal wear through a damaging process called lithium plating.

Lithium plating occurs when metallic lithium deposits on the anode electrode rather than being properly stored within the battery's structure. This reduces capacity, increases internal resistance, and in severe cases can create safety risks. Critically, the risk grows more pronounced as the battery ages — yet today's standard charging protocols apply the same fixed voltage and current limits to both a brand-new pack and one that has been in service for five years.

As researcher Meng Yuan puts it simply: "Batteries change over time. But charging strategies typically do not."

The Chalmers team embedded reinforcement learning — a type of machine learning in which an algorithm improves its decisions through trial and error — directly into the battery management system. Rather than using static charging profiles, the AI system learns to dynamically adapt charging current in real time based on two key inputs: the battery's instantaneous state of charge (SoC) and its accumulated state of health (SoH).

The AI is rewarded for achieving good long-term outcomes — maintaining charging speed while minimising the harmful degradation mechanisms that standard protocols ignore. As the battery ages and its electrochemical state evolves, the system continuously refines its strategy, modifying cut-off voltage thresholds to protect battery health at every stage of the pack's life.

The result: battery lifetime in the simulation reached 703 equivalent full cycles — the number of full charge-discharge cycles before capacity drops to 80% of its original value — compared to the standard baseline, representing a 22.9% improvement. Average charge time was 24.12 minutes versus 24.15 minutes for the conventional method, a difference of just three seconds.

The current results are simulation-based, validated using a digital twin of one of the most common EV batteries on the market. The research team acknowledges that because the relationship between charging voltage and battery health depends on temperature and cell chemistry, the model must be characterised for different battery types before broad deployment.

To address this, the team is exploring transfer learning — a technique that allows a trained AI model to adapt more quickly to new battery chemistries, reducing the experimental workload required for recalibration. The critical next step is validation on physical battery cells, moving the research from simulation into the real world.

Key Takeaways

• →Chalmers University researchers have developed an AI-based fast-charging strategy that extends EV battery life by 22.9% — reaching 703 equivalent full cycles before capacity drops to 80%.
• →The method uses reinforcement learning to adapt charging current in real time based on the battery's state of charge and state of health, directly targeting the root cause of lithium plating degradation.
• →Charging speed is virtually unaffected — average charge time increases by just three seconds (24.12 min vs. 24.15 min), making the approach transparent to drivers.
• →A 23% battery life boost could unlock an additional 70,000 to over 100,000 miles of range per vehicle, with major implications for total cost of ownership and fleet economics.
• →The strategy could be deployed as a software update to existing battery management systems — no new hardware required, though chemistry-specific calibration is needed.
• →Results are currently simulation-based; the team's next step is physical battery validation, with transfer learning being explored to accelerate adaptation across different cell chemistries.