A Novel Dynamic Health Index Based on Core Graph Model of Multichannel Signals for Machine Health Monitoring

Constructing an accurate dynamic health index (HI) is essential to monitor the operational status of machines in real time. Multichannel signals contain richer machine state information than single-channel signals, and graph signal processing shows great potential for dynamic HI construction. This makes the construction of dynamic HIs from multichannel signals combined with graph signal processing a promising application. For the construction of dynamic HIs based on multichannel signals, adopting appropriate feature extraction and HI construction methods are the key issues. Therefore, a novel dynamic HI based on the core graph model of multichannel signals is proposed for machine health monitoring. First, the path graph signal of multichannel envelope signals is subjected to the graph Fourier transform (GFT) to get the envelope graph spectrum. The localized domains are used to amplify the frequency intervals related to fault features in the envelope graph spectrum, constructing a multichannel feature graph model. Second, the tensor data structure of the multichannel feature graph model is turned to a tensor graph model. High-order singular value decomposition is used as a feature layer fusion method to reconstruct the tensor graph model by selecting the number of valid singular values and removing the redundant information to obtain the core graph model. Third, the Frobenius norm (F-norm) is used as a graph model-to-HI mapping method to calculate the proposed dynamic HI. Then, the dynamic 3-sigma criterion is combined with the Mann–Kendall (M–K) as a different fault stage alarming technique. Finally, two case studies and an engineering case study demonstrate the excellent performance of the proposed dynamic HI for health monitoring.