Battery system reliability prediction method using surrogate models based on Operator Learning

Abstract

Toshiba Corporation, Corporate Research and Development Center, Advanced Intelligent systems laboratories The lifetime of a battery varies depending on its operational history, such as charge-discharge waveforms and temperature conditions. For improving the reliability of battery systems, it is important to have probabilistic reliability prediction methods that can consider the impact of uncertainties, such as load history and cooling performance, on lifetime distribution. In this study, we propose a method for rapidly executing probabilistic lifetime predictions by modeling the irregular waveforms of charge-discharge of lithium-ion batteries with an Evolutional spectrum, and utilizing a surrogate model based on Operator Learning, which can learn the operators of differential equations related to electrical circuit analysis and thermal analysis.