Constructing battery impedance spectroscopy using partial current in constant-voltage charging or partial relaxation voltage

Electrochemical impedance spectroscopy (EIS) is an important technique to measure the impedance of lithium-ion batteries. However, practical applications of this technique are hindered by various factors such as expensive equipment costs, prolonged battery relaxation time, and lengthy measurement time, posing significant challenges and limitations. Without the need of an impedance analyzer, this study presents a machine learning (ML) approach by utilizing the current signals in constant voltage (CV) charging or the relaxation voltage (RV) data after charging as the input to construct the complete impedance spectrum of a battery at its full capacity. To validate the robustness and reliability of this approach, various scenarios, including the changes in the data length, the sampling interval, and the ML model, are discussed. We demonstrate that with 600 s of input data, using the CV current yields a root mean square error (RMSE) of 0.84 mΩ, while the RV achieves an even lower RMSE of 0.69 mΩ. With an input data of as short as 30 s, the two respective RMSEs simply increase to 1.94 and 0.82 mΩ. Incorporating the voltage curve in constant current (CC) charging into estimation analysis shows that, with the same data length, both CC and RV inputs yield even more accurate predictions than CV data.