Towards Intelligent Operations and Maintenance: A Novel Failure Knowledge Graph Learning Framework

Description

Operations and Maintenance (O&M) optimization involves the analysis of historical failure data and corresponding maintenance actions to enhance the performance of engineering systems. This process fundamentally relies on the quantity and quality of failure data gathered over time. Knowledge Graph—a tool that graphically represents semantics by articulating entities and their relationships—has been recently suggested to encapsulate failure information and associated maintenance actions derived from the failure data, thereby creating Failure Knowledge Graphs (FKGs) for engineering systems. FKGs serve as the foundation of O&M investigation because the breadth and depth of understanding of engineering systems’ failures and maintenance hinge on the thoroughness of failure data analysis and the precision of failure information extraction. However, the rapid accumulation of failure data, particularly textual maintenance logs pertaining to failures and maintenance, presents a significant challenge to the traditional approaches for manual construction of the FKG.To this end, the project aims to propose an intelligent FKG learning framework powered by Natural Language Processing (NLP) techniques for decision support during routine O&M activities of engineering systems. The proposed framework will be able to extract critical failure and maintenance information automatically and intelligently from fieldcollected failure data, accompanied by an effective online updating mechanism. A series of intelligent NLP-based algorithms will be first proposed to extract failure and maintenance information automatically. They will be able to fulfill the full demand for element identification and relationship reasoning, including FKG initialization, extension, and updating. Subsequently, a mirrored Bayesian Network (BN) mapping concept and its solving algorithms will be presented, including the mapping algorithm from FKG to BN and the causal-based inference algorithm for BN reasoning. The proposed mapping concept and its corresponding algorithms enable information inferences, updating, and transformation to the FKG, recommending the most appropriate maintenance actions, in response to failures.The feasibility and effectiveness of the proposed intelligent FKG-based learning framework will be validated by failure data collected from operating wind farms. A systematic FKG for both onshore and offshore wind turbines will be issued from this project. The outcomes of this project will benefit both industry and academia, providing a novel systematic framework and operational procedure for failure data analysis and FKG construction. The proposed framework exhibits high-level intelligence, which will further advance the field of O&M optimization.