Predicting the discharge capacity of a lithium-ion battery after nail puncture using a Gaussian process regression with incremental capacity analysis

Abstract

This work uses a Gaussian process regression to predict the discharge capacity of small Lithium-ion pouch cells after a nail puncture. Previous studies have shown that cells can operate at a reduced capacity after experiencing abuse similar to what can be seen during extreme field operation, where the ability to predict cell functionality can be critical to safety. Other studies have shown that different features of cell incremental capacity curves can be used to determine the extent of cell degradation during normal operation, which can be used to predict future operation. For this work, 15 cells are punctured with a nail and allowed to continue operating for 100 total cycles to collect data. The incremental capacity curves are calculated, then the magnitude and corresponding voltage of the highest peak are determined. A Gaussian process regression is used to predict the discharge capacity during operation after the nail punctures. The results show a mean coefficient of determination of 0.923 with a median value of 0.95, a mean root mean square error of 0.013 and median value of 0.09, and a mean absolute error of 0.011 with a median value of 0.08, indicating the regression can be useful in predicting discharge capacity.