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

History-independent identification of two mechanistically distinct degradation pathways (LLI/SEI and LAM) in NMC/C-SiO lithium-ion cells, from a single 10-minute EIS measurement at five states of charge — without cycling history, charge/discharge sweeps, or cell provenance. Validated on 228 NMC/C-SiO cells (LG INR18650HG2, 3.0 Ah; Luh & Blank 2024) across eight ageing protocols spanning 0–40 °C real temperature. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ VERSION HISTORY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ v1 (March 2026) - T_nominal used for Arrhenius. Eₐ = 0.235 eV (R² = 0.883). - LAM cohort: n = 8 (confirmed collapse, CYCLIC_25C only). - LLI: 80/177. Single retrospective guard. v3 (March 2026) - Corrected to T_real (3.3 / 10.5 / 25.8 / 40.4 °C experimental). - Eₐ = 0.226 eV (R² = 0.998, LOO ≤ 6.5%). - LAM cohort: n = 28 (all cycling protocols). - Added: Mslow prognostic model (LOO MAE = 3.3 ckp), DCIR ratio analysis, cross-validation vs experimental record, state-space/bifurcation framing, 8 figures, 14 references. v4 (May 2026) — THIS VERSION - Recalibrated on clean T_real means (1.5 / 10.6 / 24.9 / 40.3 °C). - Protocol names updated to reflect T_real ranges: CYCLIC_0C / 10C / 25C / 40C + PROFILE_0C / 10C / 25C / 40C. - Eₐ = 0.2027 eV (R² = 0.978, LOO max 15.8%; larger than v3 due to protocol regrouping — coldest group n = 33 vs n = 9 in v3). - LAM cohort: n = 26 (three-guard τ-jump criterion, stricter than v3 single guard). - LLI extended: 103/177 (added CYCLIC_0C + all PROFILE protocols). - NEW: No-BoL LAM detection (inter_soc_std_mean, AUROC = 0.951, LOPO F1 = 0.818). - NEW: BoL-anchored LAM mode (d10_max, AUROC = 0.998, FP = 1/151). - NEW: Stage-0 screen (M_slow, 157/177 cells at SoH = 103.5%). - NEW: γ(τ) mechanism analysis — new SEI arc at τ ≈ 716 μs (f ≈ 222 Hz). - NEW: LOPO prospective validation (zero leakage, F1 = 0.818 LAM, 0.653 LLI). - NEW: Section 6.5 Limitations. - Mslow prognostic model and DCIR analysis deferred to journal submission. - Dataset DOI corrected: 10.35097/1969 → 10.35097/krk531nmj4bsshha. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ KEY RESULTS (v4) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ LLI/SEI: 103/177 (58.2%) at median SoH = 103.3%, +7.5 ckp lead to SoH < 80%. Eₐ = 0.2027 eV (CYCLIC, R² = 0.978); Eₐ = 0.1464 eV (PROFILE). AUROC = 0.747. LOPO F1 = 0.653. LAM no-BoL: AUROC = 0.951, FP = 7/151, LOPO F1 = 0.818. No prior data needed. LAM BoL: AUROC = 0.998, FP = 1/151, TP = 24/26. Stage-2: 26/26 TP, 0 FP in 151 non-LAM. SNR = 70–90×. Permutation p < 10⁻⁴. Mechanism: New SEI arc at τ_new ≈ 716 μs (f ≈ 222 Hz) confirmed by γ(τ) analysis. BMS path: Im(Z)@147 Hz + Re(Z)@1 kHz, 5 SoC, ~10 min. No DRT inversion required. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ RELATED WORKS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Preprint: DOI 10.5281/zenodo.18356226 τ₆₃ in PEM fuel cells (concurrent preprint, same author): DOI: 10.1149/osf.io/hpu7y Dataset companion paper (Luh & Blank, Scientific Data 2024): DOI: 10.1038/s41597-024-03831-x ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ REPRODUCIBILITY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Full algorithmic details withheld pending IP review. Dataset: DOI 10.35097/krk531nmj4bsshha (Note: DOI 10.35097/1969 used in v1/v3 resolves to an earlier dataset version with different protocol metadata; use the current DOI for reproduction.) Competing interests: None. Funding: None declared.