Machine learning in state of health and remaining useful life estimation: Theoretical and technological development in battery degradation modelling

Designing and deployment of state-of-the-art electric vehicles (EVs) in terms of low cost and high driving range with appropriate reliability and security are identified as the key towards decarbonization of the transportation sector. Nevertheless, the utilization of lithium-ion batteries face a core difficulty associated with environmental degradation factors, capacity fade, aging-induced degradation, and end-of-life repurposing. These factors play a pivotal role in the field of EVs. In this regard, state-of-health (SOH) and remaining useful life (RUL) estimation outlines the efficacy of the batteries as well as facilitate in the development and testing of numerous EV optimizations with identification of parameters that will enhance and further improve their efficiency. Both indices give an accurate estimation of the battery performance, maintenance, prognostics, and health management. Accordingly, machine learning (ML) techniques provide a significant developmental scope as best parameters and approaches cannot be identified for these estimations. ML strategies comparatively provide a non-invasive approach with low computation and high accuracy considering the scalability and timescale issues of battery degradation. This paper objectively provides an inclusively extensive review on these topics based on the research conducted over the past decade. An in-depth introductory is provided for SOH and RUL estimation highlighting their process and significance. Furthermore, numerous ML techniques are thoroughly and independently investigated based on each category and sub-category implemented for SOH and RUL measurement. Finally, applications-oriented discussion that explicates the advantages in terms of accuracy and computation is presented that targets to provide an insight for further development in this field of research.