Search

• The GARCH family models from time series analysis is introduced for the first time into machine condition monitoring. • Health indicators constructed based on GARCH and EGARCH models demonstrate the capability to detect incipient faults in bearings, verified through two experimental cases. • In accelerated degradation tests, it is established that acoustic emission signals are closely associated with damage evolution. • Acoustic emission technology can be employed for quantitative assessment of bearing damage. 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. [ABSTRACT FROM AUTHOR]

The acoustic emission (AE) technique is known for its sensitivity to early damage and is well-suited for online condition monitoring of rolling element bearings (REBs) in various industrial application scenarios. Nonetheless, identifying weak faults under varying speed and strong background noise conditions still remains challenging. In addition, the comprehensive modelling of AE signal from faulty REB in electromechanical systems is still a pending issue. In light of this, a well-considered model is firstly developed for the AE signal of faulty REBs in this work. After that, a novel bearing fault diagnosis framework based on semantic segmentation networks is devised. Precisely, the proposed framework consists of three main components: a preprocessing step depending on the signal segmentation algorithm, a diagnosis step using fault fingerprint mapping, and a postprocessing evaluation step supplemented by a density peak clustering (DPC) approach. We evaluate the presented procedures through simulation analysis and an experimental case under varying speed conditions. Meanwhile, the comparison with the original threshold-based fault fingerprint recognition algorithm is also conducted. The comprehensive results demonstrate the efficacy of identifying fault-associated fingerprint feature (FPF), indicating that the proposed framework holds promise for condition monitoring. [ABSTRACT FROM AUTHOR]