<p>Lithium-ion batteries, as the core component of energy storage systems, although their occurrence rates of internal short circuits (ISCs) and thermal runaway are relatively low, they may trigger catastrophic accidents and threaten the safety and economy of energy storage. The safety status assessment method based on fault diagnosis and warning mechanism has been widely concerned. Based on the coupling mechanism of electrochemical, thermal and mechanical multi-abuse scenarios, this paper reveals the cross-scale evolution laws of battery faults. Regarding fault diagnosis technologies, the progress and bottlenecks of four types of methods, namely model-driven, data-driven, hybrid-driven and new technologies, are reviewed and their commercial potential is evaluated: Model-driven methods have the ability to predict comprehensively, but their adaptability to dynamic working conditions is limited; data-driven methods have high sensitivity, but their physical mechanisms are ambiguous; hybrid-driven methods have strong ability to suppress false alarms and generalize, but the integration of models and data is difficult; multi-modal sensing technology provides a new path for non-invasive detection, but its accuracy and cost need to be optimized. Current research faces challenges such as the fragmentation of cross-scale modeling and insufficient lightweight deployment. In the future, efforts should focus on building digital twin models that integrate micro- and macro-evolution, developing adaptive hybrid diagnosis algorithms, and establishing a dynamic coupling framework between faults and aging, promoting the transformation of battery safety management from passive response to proactive prevention.</p>