Researchers from the University of Cambridge and Newcastle University developed a machine learning method that predicts the health and lifespan of a lithium battery. The method could further the development of safer and more reliable lithium-ion batteries for electric vehicles and consumer electronics.
Source: Pixabay
Predicting battery health and its remaining useful lifespan limits the widespread adoption of electric vehicles because lithium-ion battery performance degrades over time. Tracking a battery's health is also important. Battery performance degrades over time due to many chemical processes working together in the battery. Individually, these processes do not affect the battery much, but when they happen together they can seriously shorten the battery performance and life span.
Current methods for tracking battery life and performance are based on tracking a battery's current and voltage during charging and discharging. This often misses important features that indicate battery health but tracking these require new ways of probing batteries. Algorithms that can detect subtle performance signals as the batteries are charged and discharged.
The new method works by sending electrical pulses into the battery and measuring the battery's response. A machine-learning algorithm analyzes the measurements and predicts the battery's health and lifespan. The algorithm finds specific features in the electrical response that indicates battery aging. This non-invasive system is a simple add on to any existing battery system.
To train the algorithm, researchers performed over 20,000 experiments and measurements to train the model. The model learned how to distinguish important signals from irrelevant noise.
The machine learning model can give insights into the physical mechanism behind degradation and inform which electrical signals are more correlated with battery aging. This allows researchers to design specific experiments to probe into why and how different batteries degrade.
Signals are identified by the machine learning model can be a starting point for theoretical and experimental studies. Developing optimal battery charging protocols will enable fast charging and minimize degradation.
A paper on this research was published in Nature Communications.
