GARCH family models oriented health indicators for bearing degradation monitoring

Recent progress in digital twin (DT) has significantly contributed to the advancement of predictive maintenance. The interaction of data between physical and virtual models is facilitated through carefully designed health indicators (HIs). Conventional condition monitoring HIs are inadequate for early-stage fault detection and lack the capacity to quantitatively assess defects. In light of this, the paper proposes HIs based on the generalised autoregressive conditional heteroskedasticity (GARCH) family time series model to characterise the evolution of bearings dynamic response, specifically the cyclostationarity of the repetitive transients. The verification of the proposed indicators is assessed using a publicly available vibration dataset and acoustic emission signals acquired from accelerated bearing degradation tests. The result shows that the GARCH oriented indicators can determine the incipient failure earlier than traditional statistical HIs and have the ability to quantify defects. Significantly, the improved Exponential GARCH model-based indicator demonstrates heightened stability throughout the monitoring process.