A Knowledge-Augmented Multistage Reasoning Approach for Wind Turbine Fault Cause Analysis

作者：Yaping Hu, Peihan Wen, Yu-Jie Dai · 发表于：IEEE Transactions on Industrial Informatics · 年份：2026 · DOI：10.1109/tii.2026.3658534 · 被引用次数：3 · 研究领域：Computer Science

Addressing challenges of insufficient knowledge retrieval and weak event causal structure modeling in the existing fusion methods of knowledge graph and large language model (LLM) for complex wind turbine chain faults, in this article, a knowledge-augmented multistage reasoning approach is proposed. First, an LLM-based knowledge extraction strategy is designed by simulating expert cognitive pathways. Second, a funnel-based subgraph retrieval method that integrates semantic embedding and topological logic is proposed to progressively refine semantic alignment and ensure hierarchical logical consistency for precise retrieval. Third, five representative causal structure patterns’ characteristics of wind turbine chain faults are defined and utilized to uncover deep causal relationships. Finally, an intelligent agent mechanism for intent-driven module scheduling is developed to support flexible, multiturn conversational analysis. The experimental results demonstrate that the proposed methods outperform state-of-the-art in knowledge extraction, multievent subgraph retrieval, and response generation tasks of fault cause analysis. Particularly, it exhibits a significant advantage when handling multievent queries and effectively enhances domain knowledge understanding and reasoning capabilities by providing LLMs with precisely extracted causal relationship chains.