Residual-Hypergraph Convolution Network: A Model-Based and Data-Driven Integrated Approach for Fault Diagnosis in Complex Equipment

Timely and accurate fault diagnosis plays a critical role in today's smart manufacturing practices, saving invaluable time and expenditure on maintenance process. To date, numerous data-driven approaches have been introduced for equipment fault diagnosis, and part of them attempt to involve equipment knowledge in their data-driven models. However, those combinations mainly concentrate on feature engineering and superposition of their separate results without considering or leveraging the relationship between equipment knowledge and collecting sensor data. To fill this gap, this research proposes a residual-hypergraph convolution network (Res-HGCN) approach that holistically embeds equipment's structure and operational mechanisms as a hypergraph form into data-driven model, considering the reaction among equipment's components. The generic model-based hypergraph construction framework is first introduced, which represents a synergetic mechanism of complex equipment. Then, a multisensory data-driven Res-HGCN approach, combining residual block and hypergraph convolution network (HGCN), is presented for fault diagnosis based on predefined hypergraph. Lastly, a case study of turbofan engine is conducted and compared with other typical methods to reveal the superiority of the proposed approach. This work establishes the association of different sensing variables through equipment's structure and operational mechanisms, thus integrating the advantages of model-based and data-driven-based approaches holistically. It is envisioned that this research can provide insightful knowledge for many other model-based and data-driven integrated manufacturing scenarios.