Cambridge University uses AI to improve battery health and safety

Predicting the health and remaining useful life of lithium-ion batteries is a major challenge limiting the widespread adoption of electric vehicles . Over time, battery performance degrades through a complex series of subtle chemical processes. Individually, these processes do not have much impact on battery performance, but together, they can severely reduce the performance and life of the battery.

Researchers from the Universities of Cambridge and Newcastle have devised a new method to monitor batteries by sending electrical pulses into them and measuring their response. They then process the data using a machine learning algorithm to predict the battery's health and lifespan.

"Safety and reliability are the most important design criteria as we develop batteries that can store a lot of energy in a small space," said Dr. Alpha Lee of Cambridge's Cavendish Laboratory. "By improving the software that monitors charging and discharging, and using data-driven software to control the charging process, I believe we can significantly improve battery performance."

The researchers devised a method to monitor batteries by sending electrical pulses to them and measuring their reactions. A machine learning model was then used to identify specific features of the electrical reactions that are signs of battery aging. The researchers took more than 20,000 experimental measurements to train the model. Importantly, the model learned how to distinguish important signals from irrelevant noise. Their method is non-invasive and a simple add-on system.

The researchers also found that the machine learning model could provide clues to the physical mechanisms of degradation. The model informed which electrical signals were most associated with aging, allowing them to design specific experiments to explore why and how the battery degraded.

"Machine learning complements and enhances our understanding of physics," said co-first author Dr. Yunwei Zhang, also from the Cavendish Laboratory. "The interpretable signals identified by our machine learning model are the starting point for future theoretical and experimental studies."

The findings were published in the journal Nature Communications.