Large Language Models for Wind Turbine Fault Diagnosis: A Review of Enabling Technologies, Component-Level Applications, and Future Perspectives

Wind turbine fault diagnosis (WTFD) requires combining heterogeneous measurements with engineering knowledge under changing operating conditions. Large language models (LLMs) are being investigated for numerical diagnosis, inspection interpretation, and diagnostic decision support, but these roles have different evidence requirements. This structured narrative review examines enabling technologies and component-level applications for hub and blades, gearboxes, bearings, and generators. Explicit review questions guide a consolidated evidence map linking components, tasks, data modalities, model roles, and validation maturity. Individual study summaries distinguish direct research evidences, related-domain demonstrations, and proposed implementation pathways. Comparisons show why results from feature selection, adapted numerical prediction, knowledge retrieval, and report generation cannot be interpreted as interchangeable diagnostic capabilities. Furthermore, the discussion addresses unsuccessful retrieval configurations, possible temporal leakage, unverified explanations, and the gap between retrospective field-data evaluation and operational deployment. Prioritised research directions connect these uncertainties to matched ablations, validation, and prospective field studies. This review offers researchers a conceptual roadmap and practitioners a reference for evaluating the readiness of LLM-based diagnostic tools in operational wind energy environments.