Li-ion Battery Prognostics with Statistical Model and RNN Trained with EIS-Based Features

The mass adoption of battery electric vehicles introduces new challenges in the automotive industry, such as designing high-performance Li-ion batteries, finding the optimal operating condition for the vehicle, and proposing a sustainable recycling solution for retired Li-ion batteries. This paper addresses the longevity and performance aspects of Li-ion batteries by proposing a method to calculate the Li-ion battery cells' remaining useful life (RUL) and state of health (SOH). The methodology of the prognostics algorithm is based on the cycling history of the first life of the battery and on assuming a certain load profile in the second life of the battery. RUL forecasting is done using an AutoRegressive Integrated Moving Average (ARIMA) statistical analysis model combined with a Recurrent Neural Network (RNN) with long short-term memory (LSTM) layers. The RNN prognostics model confirms the expectation that the batteries with the highest C-rate and highest depth of discharge (ΔDOD) loose the most capacity over a certain number of cycles. The prognostics model furthermore shows that high DOD's lead to significantly faster degradation than high C-rates. This finding can assist EV users in choosing the optimal driving profile aimed at minimising battery degradation.