Critical Evaluation of Electrochemical Impedance Spectroscopy for State-of-Health Estimation of Lithium-Ion Cells Against Battery Management System Baselines

Machine-learning models that estimate the state of health (SOH) of lithium-ion cells from electrochemical impedance spectroscopy (EIS) are rarely compared with models using only data a battery management system (BMS) records. Of 30 audited studies using three public datasets, one reported such a baseline. With whole cells held out, impedance brought no detectable gain over the strongest baseline on two exploratory datasets. A test on 88 independent coin cells, registered before the data were downloaded, was inconclusive (19 cells; +0.79 SOH points with impedance, 95% CI −0.07 to +1.66; smallest effect of interest 1.32). The capacity label varied strongly with cycling currents. Post hoc, with the label corrected for these currents without the test cell, impedance changed the full-life error of all 88 cells by −1.95 points (−2.86 to −1.03). With features of the previous discharge curve in the baseline, the change was −0.39 (−1.10 to +0.33). The comparison of impedance with these features, repeated as originally published, lost most of its gain within sets once the baseline received the previous discharge current, but not when models trained at 23 °C were tested at 35 °C. Impedance-based SOH studies should compare impedance with the strongest baseline a BMS can build.