From State-of-Health Uncertainty to Maintenance Decisions: Risk-Aware Predictive Maintenance of Lithium-Ion Batteries under Supply Uncertainty

Abstract

To bridge the gap between prognostic uncertainty and maintenance-resource allocation, this paper proposes an uncertainty-driven, risk-aware predictive maintenance framework for lithium-ion batteries. Future state-of -health (SOH) uncertainty scenarios are generated through conformal calibration and historical multi-step residual sampling, and are mapped into empirical remaining useful life (RUL) distributions and pre-maintenance end of life (EOL) probabilities, thereby propagating uncertainty from the health-state domain to the lifetime-risk domain. The planned maintenance time, spare-ordering time, and advance spare-preparation decision are then jointly optimized under stochastic delivery delays. A risk-economic maintenance–spare model is formulated by integrating preventive/corrective replacement, procurement, inventory-holding, downtime, and remaining-life loss costs, together with reliability constraints and conditional value-at-risk (CvaR)-based tail-risk control. Validation on CALCE cells CS2-36 and CS2-37 and NASA cells B0005 and B0006 shows that all maintenance decisions occur before the actual EOL, with margins of 2, 6, 2, and 3 cycles, respectively. Compared with deterministic SOH prediction, the risk-adjusted objective is reduced by 14.5%, 11.2%, 13.2%, and 10.6%, while the average reduction relative to immediate ordering reaches approximately 24.5%. The resulting battery-specific maintenance and ordering strategies demonstrate the framework’s ability to adaptively balance remaining-life utilization, inventory cost, and tail failure risk across different degradation patterns.