Fast charging without shortening battery lifetime is one of the key challenges in the electrification of transport. Researchers at Chalmers University of Technology now show that it is possible to combine rapid charging with significantly reduced battery wear — using artificial intelligence.
Photographer: Victoria University of Wellington
In a new study, the researchers present an AI-based charging strategy that extends battery lifetime by 22.9 percent compared to today’s standard method — without increasing charging time.
“We demonstrate that it is possible to charge just as fast as today, but with substantially less long-term degradation,” says Meng Yuan, researcher at the Department of Electrical Engineering, Chalmers.
Increased cycle numbers
Battery lifetime was measured in so-called equivalent full cycles (EFCs) — meaning how many full charge and discharge cycles the battery can undergo before its capacity drops to 80 percent of the original value. This threshold is commonly considered the end of life for electric vehicle applications.
Using the new method, the battery achieved a 22.9 precent extension of its lifetime in EFCs compared with the conventional charging approach. At the same time, charging time remained virtually unchanged: 24.12 minutes on average, compared to 24.15 minutes for the standard method.
Why Does Fast Charging Damage Batteries?
When a battery is charged rapidly, high currents are pushed into the cell. This can trigger unwanted side reactions. One of the most critical is lithium plating, where metallic lithium deposits on the electrode instead of being properly stored within the battery’s structure.
Lithium plating reduces capacity, increases internal resistance, and in severe cases can affect safety. The risk becomes more pronounced as the battery ages.
Conventional charging strategies use fixed voltage and current limits, regardless of whether the battery is new or has already been used for several years.
“Batteries change over time. But charging strategies typically do not, Meng Yuan explains.”
Photographer: Chalmers tekniska högskola
“This work shows that the true bottleneck of fast charging is not simply current limits, but the evolving electrochemical state inside the battery. By integrating AI with physics-based understanding, we move closer to health-aware charging strategies that maximize both performance and lifetime,” says Changfu Zou, Professor at the Department of Electrical Engineering, Chalmers.
What Is Reinforcement Learning?
The new method is based on reinforcement learning, a type of machine learning in which an algorithm learns by interacting with an environment and gradually improving its decisions.
In this case, the “environment” is the battery. The AI system is trained to find a charging strategy that keeps charging time short while minimizing harmful degradation mechanisms. By being rewarded for good long-term outcomes, the algorithm learns how to adapt charging current dynamically.
The result is a flexible charging profile that adjusts according to the battery’s current health condition.
What Does SoH Mean?
A key concept in the study is State of Health (SoH), which describes how much of the battery’s original capacity remains. A new battery has 100 percent SoH. When capacity falls to 80 percent, it is often considered to have reached the end of its useful life in EV applications.
The researchers developed a method linking the charging cut-off voltage to the battery’s health. This mapping was experimentally derived and validated using three-electrode cells in a temperature-controlled environment.
The overall life-cycle evaluation was then conducted using a high-fidelity simulation environment.
Potential for Software-Based Implementation
An important advantage of the approach is that the trained AI model does not require specialized laboratory sensors during operation. In principle, the strategy could therefore be implemented through software updates in existing battery management systems.
For vehicle manufacturers, a nearly 23 percent lifetime extension could translate into reduced warranty risks, improved residual value, and more efficient use of critical raw materials.
Next Steps in the Research
Because the relationship between charging voltage and battery health depends on temperature and cell chemistry, the method must be characterized for different battery types. The research team is therefore exploring the use of transfer learning to accelerate adaptation across chemistries and reduce experimental workload.
The next step is to test the trained AI controller directly on physical batteries.
Lifelong Reinforcement Learning for Health-Aware Fast Charging of Lithium-ion Batteries
For more information, please contact:
Meng Yuan, researcher, at the Department of Electrical Engineering, Chalmers
Changfu Zou, Professor, at the Department of Electrical Engineering, Chalmers
+46317723392
For more information, contact:
- Professor, Systems and Control, Electrical Engineering