Remaining-useful-life prediction via multiple linear regression and recurrent neural network reflecting degradation information of 20Ah LiNixMnyCo1−x−yO2 pouch cell

Abstract

This paper presents the results of various experiments and analyses pertaining to lithium‑nickel‑cobalt‑manganese oxide (NCM) batteries having a nominal capacity of 20 Ah. This pouch type battery is characterized by high power rating and high energy density. The batteries used in the experiments were manufactured by varying the design ratios of nickel, cobalt, and manganese (5:2:3 and 6:2:2) in the NCM cathode materials. An accelerated deterioration test was carried out by applying a current of 80 A at 4C-rate (C-rate is the charge-discharge rate of a battery relative to its nominal capacity). The characteristics of the differential capacity were analyzed under varying deterioration conditions. The impedance characteristics for a given state of charge (SOC) and deterioration level were analyzed through electrochemical impedance spectroscopy (EIS) tests. In addition, the battery-equivalent circuit model was designed to estimate the model parameters of the alternating current (AC) impedance. The model parameters of the direct current (DC) impedance were estimated and compared through the direct current internal resistance (DCIR) test. Machine learning was performed by using the data extracted from the accelerated deterioration test as learning data, and by applying it to the multiple linear regression and the recurrent neural network methods. Finally, a study was carried out to estimate and analyze the remaining useful life (RUL) of the NCM lithium-ion batteries with machine learning.