Towards Smart Sensing of Battery Degradation Modelling: Bayesian Approach

Reliable detection of the onset of accelerated degradation is essential for safe and cost-effective operation of lithium-ion batteries. This paper proposes a Bayesian changepoint model designed for ``smart sensing'' in battery management systems (BMS), where intelligence is applied to standard voltage–current–temperature (V–I–T) telemetry rather than new sensing hardware. We define a simple, interpretable health indicator (HI)--the ratio of charge time to discharge time--computed directly from cycle-level BMS-compatible features. A probabilistic, piecewise-linear model with a single changepoint is fitted using Hamiltonian Monte Carlo, yielding posterior distributions for the onset of accelerated degradation and for pre/post-change slope behaviour. The method provides calibrated uncertainty intervals that can be integrated into risk-aware BMS decision processes, addressing the limitations of deterministic breakpoint heuristics that supply only point estimates. Using an open dataset of 18650 lithium-ion cells, the model consistently identifies a mid-life transition in the HI trajectory and demonstrates good predictive adequacy through posterior predictive checks. The approach is computationally lightweight, transparent, and directly compatible with embedded implementations. By transforming standard BMS telemetry into an uncertainty-aware degradation signal, the proposed framework supports the development of intelligent and deployable sensing strategies for next-generation battery systems.