Maintenance decision-making for wind turbines serves as a critical link that translates quantitative risk assessment results into actionable inspection schedules, troubleshooting workflows and on-site resource allocation. In view of heavy reliance on empirical knowledge in existing wind-farm operation-and-maintenance (O&M) practices, this paper proposes a risk-guided intelligent maintenance decision architecture coupling deterministic knowledge rules and large-language-model generation. A deterministic knowledge layer retrieves safety constraints and O&M resource requirements from structured databases and extracts empirical maintenance experience from historical fault cases. On this basis, risk classification results and matched domain facts drive the generation of maintenance recommendations. A risk-aware prompting strategy is designed to fix response priority and time constraints, while an independent assembly module assembles model-generated descriptive texts and structured database fields into standardized JSON-formatted maintenance work orders. Furthermore, an execution-to-knowledge feedback loop is embedded into the overall architecture to support iterative knowledge updating.

wind turbines,maintenance knowledge,structured work orders,maintenance decision-making