Battery Modeling Techniques for State-of-Health and Remaining Useful Life Estimation: A Comparative Review

Abstract

Lithium-ion batteries are central to electric transport and grid-scale energy storage, yet they degrade continuously under varying load, temperature, and duty-cycle conditions. Battery management systems therefore require reliable estimates of state-of-health (SOH) and remaining useful life (RUL) to support safe operation, warranty management, and lifecycle planning. Meeting this requirement in the field is challenging: degradation arises from multiple interacting mechanisms, on-board measurements are limited to voltage, current, and temperature, and operating conditions shift unpredictably over time. This paper presents a deployment-focused comparative review of SOH and RUL modelling approaches for lithium-ion batteries, drawing on a corpus of 48 peer-reviewed studies selected through a structured screening process. A standardised extraction framework captures input signals, training mode, validation protocol, performance metrics, uncertainty handling, and deployment constraints for each study, and an evidence-strength label is assigned on the basis of leakage control, dataset diversity, and reporting completeness. From this foundation, the review develops a five-family taxonomy covering physics-based models, probabilistic state-space filtering, classical machine learning, deep-learning architectures (including recurrent networks, convolutional models, and attention-based transformers), and hybrid fusion pipelines. A central finding is that no single modelling family is universally superior: physics-based and filtering methods offer interpretability and online tractability but carry a calibration burden, while deep learning achieves lower error on benchmark datasets but can produce unreliable predictions when operating conditions shift. To support practical adoption, the paper provides deployment artefacts covering on-board constraint mapping, feature feasibility checklists, online update strategies, failure-mode mitigations, and model-governance checklists for BMS integration.