Prognosis of Li-ion batteries under large load variations using hybrid physics-informed neural networks

Abstract

The development of new modes of transportation such as electric vertical takeoff and landing aircraft and the use of drones for package and medical delivery have increased the demand for reliable batteries. Capacity degradation and discharge behavior can vary from battery to battery and can also be influenced by changes in load due to internal thermal stress. Therefore, predicting the degradation of a battery's state-of-health (SOH) and state-of-charge (SOC) is a crucial task to ensure high reliability standards and prevent failures during operation. At the same time, recent advanced in physics-informed machine learning models have demonstrated potential to model both SOC and SOH, merging physics-derived equations and data-driven kernels in a hybrid model trained with back-propagation. In this paper, we enhance a hybrid physics-informed machine learning version of a Li-ion battery model we presented in previous works. The enhanced model captures the effect of wide variation of load levels, in the form of input current, which causes large thermal stress cycles. The cell temperature buildup during a discharge cycle is used to identify temperature-sensitive model parameters. We also extend the aging model built upon cumulative energy drawn by introducing the effect of load levels. We then map cumulative energy and load level to battery capacity with Gaussian process regression. To validate our approach we use a battery aging dataset collected on a self-developed testbed, where we used a wide current level range to age battery packs in an accelerated fashion. Prediction results show that our model can be successfully calibrated and generalizes across all applied load levels.