Yu-Sheng HuangKUO-CHING CHENChi-Jyun Ko2025-05-152025-05-152025-05-01https://www.scopus.com/record/display.uri?eid=2-s2.0-86000659663&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/729294Electrochemical impedance spectroscopy (EIS) is an experimental technique that reveals battery impedances, notably with its low-frequency components exhibiting significant correlations with battery states. However, traditional EIS (T-EIS) necessitates expensive instrumentation and extended battery relaxation periods, rendering it impractical for rapid state estimation applications. To overcome these two shortcomings, square wave EIS (Sq-EIS) in the low-frequency range, generated using a simple two-cycle square wave, emerges as a more cost-effective and time-efficient alternative, capable of achieving results comparable to low-frequency T-EIS. Even when the battery is in unrelaxed states, the total root mean square error (RMSE) between Sq-EIS and T-EIS can be <0.5 mΩ. We conduct thorough investigations into the number, amplitude, and sampling rate of 50-s period square waves, showing that a two-cycle square wave with an amplitude of 1 A and a sampling rate above 50 Hz can achieve optimal similarity between Sq-EIS and T-EIS across different scenarios, including constant current charging/discharging and dynamic discharging. Square waves of different periods, such as 30 s and 10 s, also effectively achieve this similarity. Based on these findings, by applying two 10-s square waves (for a total of 20 s) immediately after battery charging or dynamic discharging, the Sq-EIS data enables machine learning models to concurrently estimate the battery's state of charge and state of health with an RMSE of <2 % in each case.Electrochemical impedance spectroscopyLithium-ion batteryMachine learningSquare wavesUnrelaxed state[SDGs]SDG7[SDGs]SDG11journal article10.1016/j.est.2025.116229