Although electrochemical impedance spectroscopy (EIS) is widely discussed for application-oriented battery state estimation, uncertainty remains regarding degradation’s effect on the impedance spectrum when considered alongside temperature and state of charge (SoC). To advance the resolution of this issue, this study proposes progressing from associative approaches to inferential statistics by applying selected analytical methods to a balanced, full-factorial aging study including analysis of variance, kernel-based conditional independence (KCI) testing, and do -calculus with probabilistic modeling. These methods reveal that temperature predominantly influences impedance across most frequencies, obfuscating aging effects. While KCI testing confirms significant conditional dependencies between impedance and degradation markers when controlling nuisance variables, these relationships remain highly temperature-dependent, as demonstrated through causal effect estimation. We show that degradation modes and SoH C exhibit unique, frequency-specific impedance signatures, with loss of positive active material being the most prevalent. Notably, the effect of aging on EIS is most pronounced at lower temperatures, and strong interactions between temperature and aging variables suggest that these factors cannot be separated into independent contributions. Therefore, we conclude that conventional correlation-based machine learning methods may be viable for aging quantification from impedance data when carefully augmented with explicit conditioning information regarding temperature and SoC, but likely inadequate without. • Full-factorial EIS dataset allows isolating impact of SoC, temperature, and aging. • ANOVA shows that temperature dominates variance; aging signals are partly obfuscated. • Causal analyses quantify frequency-resolved effects of SoH C and degradation modes. • Temperature heavily mediates aging effects in EIS; low temperature amplifies effects.
Natterer et al. (Tue,) studied this question.