Lithium-ion Battery Diagnostic and State of Health Estimation for Electric Vehicles Applications

Li-ion batteries have become the main technology for the electric mobility. One of the most urgent challenges concerns the development of reliable methods for their state-of-health (SOH) diagnosis and consequent estimation of remaining useful life. For the electric mobility, capacity degradation prediction is highly critical to ensure service availability (km range) and life duration. In most of the existing studies, batteries cells go through standard charge/discharge patterns in laboratory conditions to study the ageing processes. However, this approach does not reflect the real world conditions of electric vehicles (EV) batteries. The goal of this work is to develop a data-driven approach suitable for on-board SOH estimation of electric vehicles batteries, based on data acquired at different levels, from single cells to full battery packs. In the first part, literature and state of the art exploration led to the selection of a set of different Neural Networks (namely LSTM, GRU, CNN, CNN-LSTM, IndRNN) that is later trained on a NASA dataset on single Lithium-ion cells tested in dynamic conditions, representative of their use in electric vehicles. Then, the same architectures are tested on data of different cars collected by a monitoring company, to assess the suitability of models based on laboratory data to real-world cases. Some of the NNs trained and tested on single cells showed good performance in SOH estimation, basing on the chosen metrics. However, their application to EVs battery data pointed out the necessity of further refinement works to be able to better catch battery packs dynamics, quite different from simple cells ones.