Towards Smart Sensing of Battery Degradation Modelling: Bayesian Approach

Battery degradation modelling underpins safe, reliable and cost-effective operation of electrified transport and stationary storage. We propose a smooth Bayesian changepoint model that detects the onset and rate of accelerated aging from cycle-level features commonly available in battery management systems (voltage/time stages, current, and temperature), with optional augmentation by online impedance or other embedded sensor signals. The model parameterizes pre- and post-change slopes and a logistic transition time with weakly informative priors, is fit via Hamiltonian Monte Carlo, and yields interpretable posteriors for onset time and degradation rate with credible intervals. On an open lithium-ion dataset, the approach identifies a single, well-localized change in trend and quantifies evidence for faster post-change deterioration, outperforming ad-hoc breakpoint heuristics while retaining transparency. We discuss how the estimator maps to smart-sensor/BMS deployments—including sensing modalities, sampling policies, on-device inference and calibration—and outline extensions to hierarchical, online and multi-modal formulations. By coupling richer sensor data with Bayesian uncertainty quantification, the framework supports risk-aware maintenance and real-time health management in practical battery systems. On the Battery RUL dataset, the posterior median changepoint was accompanied by a 95% highest-density interval, providing calibrated uncertainty for onset timing and slope change.