A failure knowledge graph learning framework for offshore wind turbines with incomplete knowledge

This study presents a novel framework for Failure Knowledge Graph (FKG) construction tailored for the safe operation and maintenance of offshore wind turbines. Specifically, Bidirectional Encoder Representations from Transformers (BERT) and Conditional Random Field (CRF) are combined for failure extraction, enhanced by iterative learning for failure data transfer from onshore to offshore wind turbines. Additionally, this framework incorporates a rule-based pseudo-label module and an innovative replacement-based pseudo-sample module to mitigate the impact of label errors and failure data imbalance during the iterative learning process. With the failure events extracted, the affiliate components and corresponding failure modes are identified to construct a tree-structured FKG automatically for offshore wind turbines. The feasibility and effectiveness of the proposed framework are validated by the presentation of an FKG regarding 313 offshore wind turbines recorded in the LGS-offshore dataset. Overall, the study provides the offshore wind sector with an intelligent framework for failure data analysis, presentation, and understanding and contributes to the safe operation of offshore wind turbines and wind farms. • A novel failure data analysis and presentation framework of the wind energy sector. • A novel Failure Knowledge Graph construction method of offshore wind turbines. • A Knowledge Transfer concept to learn failure information of offshore wind turbines from onshore devices.