History-Independent Identification of Two Distinct Degradation Mechanisms in NMC/C-SiO Lithium-Ion Cells via Electrochemical Impedance Spectroscopy (Preprint v3)

We present a history-independent, BoL-anchored EIS/DRT framework identifying two mechanistically distinct degradation pathways from a standard 10-minute impedance measurement at five states of charge (SoC = 10, 30, 50, 70, 90%), requiring only two initial BoL reference spectra. Framed in state-space terms: δ measures the cell's distance from its BoL attractor; LLI manifests as continuous drift; LAM as a bifurcation-like critical transition. Validated on 228 NMC/C-SiO cells (Luh 46 right-censored; Wilcoxon p = 1.8×10⁻⁷). Detection thresholds follow Arrhenius kinetics: Ea = 0.226 eV, R² = 0.998 (measured temperatures), LOO threshold prediction error 0.7–6.5%. Loss of Active Material (LAM): 28/177 cells (15.8%) via two-tier detection. Stage-1 alert (δ at SoC = 30%) at median SoH = 102.5% (range 97.4–106.3%); Stage-2 confirmation (abrupt τ-jump ≥ 1.0 decade in τ₆₃) with zero false alarms in 149 non-LAM cells. AUROC = 0.821 95% CI: 0.745–0.888; bootstrap n = 2,000; Mann-Whitney p = 3.6×10⁻⁸. Stage-1 precedes terminal state by a median of ~40 impedance-SoH percentage points (median SoHfinal = 63.9%). Cross-validated: 22/28 formal failure criteria; 6 survivors confirm structural collapse. Prognostic model: Mslow post-inflection rate predicts time to τ-jump; temperature-adjusted LOO MAE = 2.9 checkups (CYCLIC protocols, n = 23; Spearman r = 0.85, n = 28 full dataset). v1→v3 changes: LAM set 8→28; detection SoH 94.7%→102.5%; threshold formalised as Arrhenius equation (Ea = 0.226 eV, R² = 0.998, LOO ≤ 6.5%); two-tier LAM system; T-adjusted prognostic model; cross-validation vs experimental record; state-space/critical-transition framing.