Battery Prognostics and Health Management (BPHM) is essential for safe, reliable and cost‑effective operation of electric vehicles, grid storage and consumer electronics. Conventional approaches such as physics‑based models and task‑oriented deep learning suffer from computational inefficiency, heavy parameter tuning, limited cross‑domain generalization, reliance on extensive run‑to‑failure labels, and poor interpretability. Recent large models (LMs) built on Transformer architectures and self‑supervised pre‑training offer a new paradigm.
This review first outlines the foundational technologies behind LMs: Transformer architecture, self‑supervised learning, large‑scale multimodal datasets, and parameter‑efficient fine‑tuning (PEFT). It then surveys progress along four critical dimensions: mitigating data scarcity, enhancing generalization and robustness, integrating domain knowledge for interpretability, and enabling system‑level automation. Despite promising early results, challenges remain in data accessibility, intelligence validation, trustworthiness, and deployment feasibility. Accordingly, we propose a future roadmap: building collaborative data ecosystems, validating intelligence for industrial use, embedding physics constraints to improve trustworthiness, and achieving efficient on‑device deployment. The review provides a systematic perspective for advancing LM‑driven BPHM, guiding researchers and practitioners toward next‑generation battery management systems that are safe, reliable and autonomous throughout the battery lifecycle.
Blogger's Review: The article offers a comprehensive synthesis of how large models can address longstanding bottlenecks in battery health management, highlighting the role of self‑supervised pre‑training and PEFT in reducing data dependence and boosting cross‑domain adaptability. Realizing open data platforms and tightly coupling physical constraints with LMs will be key to unlocking their full potential as the core intelligence for future battery management solutions.