An Integrated Framework via Spectrum Sparsity Measure and Dynamic Alarm Thresholds for Online Fault Detection

Online fault detection through continuous vibration monitoring is crucial to prevent potentially catastrophic accidents throughout the service life of bearings. However, the high complexity and dynamics of actual vibration environments make it challenging, especially for incipient faults. This article develops an integrated framework for this issue via spectrum sparsity measure and dynamic alarm thresholds. First, a new spectrum sparsity measure is proposed to establish a spectrum sparsity index (SSI), which is modeled as a noisy sequence to obtain denoised SSI (DSSI) using the generalized total variation (GTV) algorithm. SSI is achieved by extracting the enhanced envelope spectrum (EES), and further, 1.5-dimensional EES (1.5DEES) is defined for explicit fault identification. Then, a dynamic alarm threshold setting strategy is designed using the generalized extreme value distribution (GEVD) to fit real-time updated SSIs. Finally, an integrated framework is constructed using DSSI as the observation line, SSI-GEVD-based dynamic alarm thresholds for alarming anomalies, and 1.5DEES for immediately identifying fault type. Two life-cycle experimental signals of rolling bearings and one real-world monitoring signal from a high-speed shaft bearing are studied to validate the proposed framework. Results demonstrate that it can effectively achieve online fault detection with superior performance in the lowest false and missed alarms.