Search

• The evolving spall sizes were measured during run-to-failure experiments for REBs. • Four measurements (acceleration, AE, IAS, and displacement) were compared. • Displacement was the best for spall size estimation, and acceleration comes next. • IAS correlates well with spall size, albeit not a direct quantitative size. • AE was found poor in both quantification and tracking of natural spall size. Bearing prognostics is an important aspect in condition-based maintenance (CBM), and a key step to successful prognostic methods is the ability to quantify the fault severity of the bearing. Previous studies have resulted in some severity assessment methods based on certain types of signals, such as vibration, acoustic emission (AE) and instantaneous angular speed (IAS), however, their performances were not compared, especially in terms of their ability to track the severity of naturally growing spalls. In this paper, four measurement approaches were tested on the same rig for bearing run-to-failure experiments, and their signals were analysed individually and compared. It was found that IAS and radial load (a proxy for displacement) required less processing to provide a reliable assessment of bearing fault severity, acceleration required sophisticated techniques to extract spall-size estimates, whereas AE could not track fault evolution accurately. [ABSTRACT FROM AUTHOR]

• The duration of natural frequency perturbations is found informative of spall size. • The WVS is utilized to identify the cyclic behaviour of natural frequencies. • This approach overcomes the limitation of entry/exit-based methods for natural spalls. • The new approach is validated and compared with existing methods in three tests. Spalling caused by fatigue is the most common reason for rolling element bearing failure, and spall size can be a good indicator to predict the remaining useful life of the bearing. Previous studies reported that, in the acceleration signal, a low-frequency step response is caused by de-stressing or re-stressing of the rolling element during its entry into and exit from the spall zone, and a high-frequency impact response is often induced when the roller hits the trailing edge of the spall. The conventional spall size estimation methods thus attempt to capture these events (entry, impact, and exit) and use their time difference to estimate the spall size. But the limitation of these methods is that they were mostly built on artificial spalls and have proved scarcely effective in dealing with natural ones. A recent study investigated the effect of these events on bearing stiffness and reported that the stiffness of the bearing structure decreases when the rolling element traverses the spall zone. Inspired by the study, a novel approach is proposed in this paper, which utilises the duration of natural frequency perturbations (induced by stiffness variation) of the bearing structure to perform spall-size estimation. Compared to the existing methods, this approach investigates the instantaneous resonance frequency variation during the spall span instead of the signal events at the entry and exit points. This novel approach along with three other representative existing methods are tested on three experimental datasets: one obtained by an artificially induced spall and the other two by naturally extended spalls. The performance analysis results reveal that all the methods are effective for artificial spalls, but only the proposed approach is effective for naturally extended spalls. [ABSTRACT FROM AUTHOR]

In recent years, the rapid development of convolutional neural networks (CNNs) has significantly advanced the progress of intelligent fault diagnosis. Most currently-available CNN-based diagnostic models are built on the premise that the monitored machine operates under stable conditions. However, in real-world scenarios, rotary machinery usually operates at varying speeds, making the fault-related pulse features susceptible to noise oversaturation. To extract discriminative features from mechanical signals under non-stationary conditions, a global contextual feature aggregation network (GCFAN) is developed in this paper. To begin with, a global contextual module (GCM) is embedded in the CNN architecture to explore multimodal features. Then, a multiscale attention module (MSAM) is introduced to guide the model to focus on global and local discriminative information. Further, a multiscale feature enhancement module (MFEM) is established to enlarge the receptive field and eliminate useless features. Finally, the GCFAN architecture is constructed based on these improvements. To achieve favourable diagnostic results under fluctuating variable speed conditions, we apply the label smoothing algorithm and the AMSGrad algorithm to assist the training of the model. Two case studies using the benchmark variable speed bearing dataset and the HF-MS variable speed gearbox dataset were carried out to test the practicality of the developed approach. Experimental results demonstrated that the developed GCFAN performs better than seven state-of-the-art approaches. [ABSTRACT FROM AUTHOR]