Fig. 7: SOH estimation results from the linear regression model (LRM) using power autocorrelation (\({{{\rm{LRM}}}}_{{P}_{{{\rm{Autocorr}}}}}\)), and energy during charging and discharging (\({{{\rm{LRM}}}}_{{E}_{{{\rm{ch}}}},{E}_{{{\rm{dis}}}}}\)) as input features versus the aging cycle number (Cycle).

Profiles of capacity loss and estimation error for cells V4 (a), W5 (b), W7 (c), and W9 (d). Augmented capacity points (obtained as discussed in Sec. Data augmentation approach) are shown in red. SOH estimation using power autocorrelation as input is shown in brown (with training data from cell W8) and yellow (with training data from all cells except the test cell). The dark blue and light blue lines show SOH estimation using energy features as input, with training data from cell W8 (dark blue) and from all available cells except the test cell (light blue). Gaps in the capacity curves for cells W5 (b) and W7 (c) are due to voltage measurements anomalies affecting the reliability of feature values (see Sec. Cell cycling and experimental dataset and Supplementary Note 5). The capacity drop for cell W8 (d) results from issues with the aging protocol implementation.