Diagnosis of electrode-level degradation modes from field EV data without in-life reference cycles

초록

Mechanistic open-circuit voltage (OCV) reconstruction enables interpretable diagnosis of battery degradation, but existing methods require a reference capacity rarely available in production fleets. We propose a reference-free pipeline-requiring only a one-time beginning-of-life characterization at manufacture-that propagates a single field-data modality through state-of-health (SOH) estimation, pseudo-OCV reconstruction, and electrode-level degradation-mode identification. A preliminary SOH trajectory, obtained by mapping charge increments to a reference OCV curve with Gaussian process smoothing, anchors the Delta Q-method, whose stoichiometric parameters are fitted with a voltage-bin-weighted loss that corrects the high-SOC skew of field data. Applied to multi-year records from three production electric vehicles, the method distinguished normal aging from an anomalous rapid-fade event. Positive-electrode active-material loss and lithium-inventory loss were the primary degradation modes, while negative-electrode active-material loss was buffered by the surplus anode capacity. The voltage-bin weighting reduced reconstruction errors by up to 45%. Tested against an independent single-particle model with no fitting to the driving data, the reconstructed modes predicted the aged-stage driving voltage within a 9-13 mV band per vehicle. Requiring only onboard relaxed-voltage and charge logs, the framework suits fleet-wide health monitoring.