State of health estimation of second-life batteries through electrochemical impedance spectroscopy and dimensionality reduction

Abstract

When lithium-ion batteries (LiBs) reach the end of their useful life in electric vehicles (EVs), they can still be used in less demanding applications, a process known as second-life (SL). Nevertheless, to ensure a safe, efficient and reliable operation of the retired LiBs in a SL application, many tests and procedures should be done to determine their internal conditions, and naturally, one of the most important parameters to be determined in these procedures is the state of health (SoH). However, the tests to obtain this parameter are intrusive, requiring taking the LiBs out of service and performing successive discharge and charge cycles in laboratory benches with special equipment, which is costly, have a high energy demand, and is time-consuming. These limitations hinder the SL market since it is estimated that in the next few years, a significant amount of LiBs will be available for SL purposes. This work suggests an approach that employs experimental electrochemical impedance spectroscopy (EIS) data and neural networks (NN) to assess the SoH of lithium-ion cells subjected to SL conditions. Also, a novel approach to perform the dimensionality reduction of the EIS measurements – reducing the time needed to perform the EIS measurement without the loss of significant data – is proposed, where the measurement time can be reduced to 19.10% of its original value, and the number of assets generated by the measurements can be reduced from 81 to 4. The proposed methodology was validated offline using two datasets of lithium-ion cells under SL conditions, the first one considers the EIS data of 395 Lithium Iron Phosphate (LFP) high-capacity cells, and the second one the EIS data of 36 Lithium Nickel Manganese Cobalt (NMC) medium-capacity cells. Root mean square errors of 1.18% and 0.95% in the SoH estimation were obtained for the LFP and NMC cells, respectively. Therefore, the novel methodology that has been proposed in this work offers significant advantages over the conventional capacity test, including non-invasiveness, fast processing time and low energy consumption.