Work

Cross cell evaluation of electrochemical impedance features for lithium ion battery remaining useful life prediction

Year: 2026
Type: article
Abstract: Electrochemical impedance spectroscopy is widely proposed as an early indicator of lithium-ion cell degradation, yet the claim that impedance features transfer across cells is rarely tested. We evaluate that claim on four NASA 18​650 cells (B0005, B0006, B0007, B0018) using electrolyte resistance Re and charge-transfer resistance Rct read directly from the recorded impedance fields without circuit fitting. Two questions are separated: whether impedance discriminates between nominally identical cells, and whether it predicts remaining useful life on a cell the model has not seen. For the first, a linear mixed-effects model with a random intercept per cell gives an intraclass correlation of 0.899 for Re and 0.854 for Rct against 0.646 for capacity, with η² of 0.600, 0.480 and 0.038 respectively; at the cell level the coefficient of variation is 13.6%, 9.6% and 2.8%. Impedance therefore separates cells far more sharply than capacity does. For the second, a four-fold leave-one-battery-out random forest gives a mean RMSE of 19.05 cycles (median 19.41, range 5.13 to 32.24) and a mean R² of 0.668. A feature ablation shows this performance is carried by the capacity trajectory: a model using Re and Rct alone reaches 33.20 cycles with a mean R² of − 0.020, while a training-free linear extrapolation of capacity reaches 22.19 cycles. Re and Rct are strongly coupled within three of the four cells (R² = 0.894 to 0.966) but only weakly in B0018 (R² = 0.146), and the pooled coupling of R² = 0.937 overstates a relationship that is heterogeneous across cells. We conclude that impedance features encode cell identity more than they encode a degradation trajectory that transfers to an unseen cell, and that this explains why they discriminate well and predict poorly.
Language: English
Open Access status: diamond
OpenAlex ID: W7235668292
FWCI:
Cites: 38
Cited by:
Related to: 10