Data-driven battery aging diagnostics and prognostics

Abstract: Lithium-ion (Li-ion) batteries play a pivotal role in transforming the transportation sector from heavily relying on fossil fuels to a low-carbon solution. But, as an electrochemical device, a battery will inevitably undergo irreversible degradation over time. Therefore, accurate and reliable aging diagnostics and prognostics become indispensable for safe and efficient battery usage during operation. However, diverse aging mechanisms, stochastic usage patterns, and cell-to-cell variations impose significant challenges. With the ever-increasing awareness of the importance of vehicle operating data, more and more automotive companies have started to collect field data. Meanwhile, the rapid advancement in computational power has drawn tremendous attention to using machine learning algorithms to solve complex and challenging tasks. In this thesis, recent data-driven modeling techniques, using both field data collected during vehicle operation and laboratory cycling data, are applied to improve the overall performance of battery aging diagnostics and prognostics. A series of data-driven methods are proposed ranging from battery state of health estimation, future aging trajectory prediction, and remaining useful life prediction. The algorithms are extensively evaluated with various data sources of different battery kinds. The evaluation results indicate that the developed methods are accurate and robust, but more importantly, they are applicable to the harsh conditions encountered in real-world vehicle operations.