Exploiting a knowledge hypergraph for modeling multi-nary relations in fault diagnosis reports

Knowledge graphs (KG) demonstrate superiority in storing and exploiting massive data and lesson-learned knowledge collected from equipment operation and maintenance management, and deliver a higher level of cognitive intelligence for fault diagnosis. However, for some indivisible multi-nary relations that connect more than two entities in complex fault diagnosis scenarios, ordinary KG has encountered many challenges in preserving their completeness and correctness with triples of < head entity, relation, tail entity >. To this end, this paper evolves the ordinary KG into Knowledge hypergraph (KHyG), based on the differentiation of schema-level and instance-level hyperedges. A pipeline for constructing KHyG using fault diagnosis reports is accordingly proposed, which adopts an end-to-end tabular information extraction manner and hyperedge-enabled schema. A two-channel KHyG embedding manner is then articulated for multi-nary relations, so as to strengthen the performance of troubleshooting solution generation in KHyG applications. To validate the proposed KHyG, real-world fault diagnosis was conducted on bridge cranes in steel factories, which demonstrated that the improvement in the modeling of multi-nary relations showed effectiveness and advances in fault investigation and troubleshooting process. It is anticipated that this study can provide several practical insights into applying more advanced KG models and techniques toward futuristic cognitive intelligence-enabled fault diagnosis.