Physics-Constrained Diagnostic Framework for Predictive Thermal Runaway and Degradation in Grid-Scale Battery Storage

Abstract

Utility-scale lithium-ion battery energy storage systems (BESS) deployed for dynamic frequency regulation and peak shaving experience nonlinear electrochemical aging and elevated risk of internal short circuits (ISCs). Purely data-driven diagnostic models fail to generalize across battery chemistries and ambient conditions, while high-fidelity electrochemical models exceed the computational budget of real-time battery management systems (BMS). This paper introduces an explainable, physics-constrained diagnostic framework that couples solid-electrolyte interphase (SEI) growth kinetics and lithium-plating physics with a real-time thermal residual observer. The observer propagates the coupled aging and thermal physics on the slow (hours-to-days) time scale associated with electrochemical degradation and flags fast (seconds-to-minutes), physically inconsistent deviations as candidate ISC events using a sequential residual test, which structurally decouples reversible thermal transients and expected capacity fade from irreversible short-circuit heating. Validated using telemetry from a 50-MWh utility BESS over 3,000 equivalent full cycles (EFCs), the physics-constrained approach identifies ISC onset 41.8 minutes earlier than industry-standard threshold alarms with a 0.3% false alarm rate, while forecasting remaining useful life (RUL) with a mean absolute percentage error (MAPE) of 1.6%–2.3%. Parametric lifetime-distribution fitting of the kind used to summarize RUL statistics is related, in a different application setting, to the four-parameter lifetime model recently proposed by Kahloul and Yahia.