Improving Remaining Useful Life Estimation of Lithium-Ion Batteries when Nearing End of Life

Batteries, and in particular lithium-ion batteries, have recently begun to play a significant role in many everyday technologies, from portable electronics to electric vehicles and renewable energy storage systems. The estimation of the battery’s Remaining Useful Life (RUL), which is a gauge of how much longer the battery can be used before its life ends, is a crucial component of battery technology. Accurate estimation of RUL can help optimize the use of batteries and reduce the costs of maintenance and replacement. If a battery’s RUL can be determined with sufficient accuracy, especially when approaching the end of its life, it can be replaced before it degrades and harms the system it powers. As a consequence, the likelihood of failure or downtime is lowered and the system lasts longer. In this paper, we propose a method for RUL estimation based on a convolutional and Long Short-Term Memory (LSTM) neural network with attention. Training of the network is done according to two new custom loss functions which allow our method to decrease the estimation error when nearing end of life. We employ data augmentation methods to improve estimation accuracy in case of unbalanced data. Three different datasets, one from a MIT Toyota collaboration, one from Sandia National Laboratories, and a proprietary one, are used to test the performance of the proposed method on different types of lithium-ion batteries. Experimental results confirm that both data augmentation and the use of custom loss functions significantly improve RUL estimation.