A Data-Driven Particle Filter For Lithium-Ion Batteries State-Of-Life Prognosis

Abstract

In recent years, Lithium-Ion rechargeable batteries have quickly become the most popular portable power sources, with applications ranging from consumer electronics (smartphones, laptops, etc.) to electric vehicles and unmanned aerial/space vehicles. Indeed, Li-ion batteries are subject to degradation over time, in particular due to the irreversibility of the electro-chemical processes driving their functioning. The deterioration of the performances is further worsened by the operational and environmental boundary conditions in which they operate (e.g. discharge rates, usage and storage temperatures, etc.). The consequences of unexpected failures due to degradation may range from mild, for example in consumer electronics, to very severe, if not catastrophic, in particular in aerospace applications, both from the economical and the safety points of view. In this context, the prediction of future degradation performances of the batteries plays a fundamental role. In this work, we exploit a recently introduced prognostic algorithmic scheme, which combines the real-time prediction capabilities of particle filters with the flexibility and simplicity of feed-forward neural networks, for adaptively predicting the State-of-Life (SOL), i.e. the capacity, of Li-Ion batteries, on the basis of past and current capacity observations. The major advantage of the proposed method lies in the fact that the algorithm automatically adapts to different degradation dynamics, without the need to derive and/or calibrate any physics-based model. The method is demonstrated with reference to actual Li-Ion battery discharge data taken both from the prognostics data repository of the NASA Ames Research Center database and from literature.