AI will increase the accuracy of determining battery life by 10 times. Developed by researchers from the Universities of Cambridge and Newcastle, the neural network will also improve their safety and reliability. The job description is published in the journal Nature Communications.
For several decades, the development speed of new, more energy-intensive and efficient batteries was limited by one serious problem – the amount of time that researchers had to spend on estimating battery life. To do this, scientists conducted several cycles of charging and discharging, and then measured the time during which the battery can produce the required voltage.
This process takes from several months to several years. Modern methods for predicting battery life are based on tracking current and voltage during charging and discharging. This method is ineffective due to the fact that it will not allow you to track specific characteristics that indicate the state of the battery.
In the course of the new work, the researchers created a method for monitoring the condition of batteries, which involves sending electric pulses to it and measuring the reaction of the battery to them. After that, the neural network searches for the specific characteristics of this response, which are a clear sign of the aging of the battery.
During the training of the neural network, researchers performed more than 20 thousand experimental measurements – according to them, this is the largest data set in this area. As a result, AI learned not only to perform its main function but also to distinguish important signals from insignificant noise. According to the researchers, this method will increase the accuracy of determining the battery life by about 10 times.
“Machine learning complements and enhances the physical understanding of the processes that occur inside the battery. The interpreted signals defined by our machine learning model are the starting point for future theoretical and experimental research”.
Yunwei Zhang, lead author of the study