Estimating state of charge of cylindrical lithium-ion cells using multiple random convolutional kernel transform and low-frequency stress waves

Abstract

Lithium-ion batteries (LIBs) represent an energy storage technology increasingly utilized across various fields. To maintain the safety and reliability of a LIB, monitoring the LIB state of charge (SoC) is essential. Currently, there are limited studies on SoC estimation of cylindrical LIBs using the low-frequency stress wave technique due to its limited sensitivity to internal changes of LIBs during the charging/discharging process. Here, we present the first demonstration of SoC estimation of cylindrical LIBs using low-frequency stress wave (5-50kHz). We adopt Pearson correlation coefficients (PCCs) to select sampling points relevant (>0.5) to the SoC from the response stress wave signals and then apply the multiple random convolutional kernel transform (Multi-Rocket) model to extract four types of features for SoC estimation. The SoC estimation performance of the developed method achieves similar accuracy in both original and down-sampled (low-resolution) signals, with estimation root mean square error of about 0.7%, 1.4%, 3.5%, and 5.3%, respectively in four scenarios. This new method based on low-frequency stress waves enables a drastic cost reduction compared to expensive high-frequency ultrasound methods, paving the way for low-cost, real-time SoC monitoring with excellent accuracy.