Order tracking has been widely used to diagnose failures of variable speed rotating machines. The performance of the TOT (Time-Frequency Domain Tacholess Order Tracking) methods is based on the correct separation of the target component strictly related to the shaft rotation frequency. Currently, most of the methods have focused on obtaining the instantaneous frequency with accuracy. In this paper, a new TOT method has been proposed that combines the inverse short-time Fourier transform (ISTFT) with singular value decomposition (SVD). The target component closely related to the shaft rotation frequency is selected and filtered approximately in the time-frequency domain. Hence, the ISTFT is adopted to reverse the target component into the time domain. Next, SVD is used to refine the roughly filtered target component. Finally, the phase of the refined signal is extracted to resample the original signal. The performance of the method was tested using real vibration signals collected from a large-scale test rig of a high-speed train traction system.

Deep learning-based models have been widely used in compound fault diagnosis of rotating machinery. A compound fault, distinct from a single fault, encompasses a variety of individual faults, suggesting a robust correlation among different compound faults. However, conventional data-driven methods often neglect this a priori knowledge, potentially impeding the performance of compound fault diagnosis. To counter this problem, our study endeavors to fully exploit the a priori knowledge of compound faults through label correlation modeling and knowledge transfer using a hierarchical neural network. Initially, a 2-D convolutional neural network (CNN) extracts features from short-time Fourier transform (STFT) time–frequency maps, which are subsequently processed using the multihead attention module to discern fault class features. Then label information is fed into the label correlation embedding (LCE) module constructed by static and dynamic graph CNNs (GCNNs), fully capturing the label relationships of different individual faults. Finally, the features that carry sufficient label correlation information are input into the classifier of sigmoid activation for decoupling compound fault diagnosis. Experimental validation on two separate datasets reveals that the proposed method surpasses existing methods in terms of exceptional classification accuracy, thus offering a novel paradigm for compound fault diagnosis in rotating machinery.

The development of diagnostics for rolling element bearings (REBs) in recent years makes it possible to detect faults in bearings in real-time. Squared envelope analysis (SEA), in which the statistical characteristics of the squared envelope (SE) or the squared envelope spectrum (SES) are analysed, is widely recognized as both an effective and the simplest method. The most critical step of SEA is to find an optimal frequency band that contains the maximum defect information. The most commonly used approaches for selecting the optimal frequency band are derived from measuring the kurtosis of the SE or the SES. However, most methods fail to cope with the interference of a single or a few impulses in the corresponding domain. A new method is proposed in this paper called “PMFSgram”, which just calculates the kurtosis of the SES in the range centred at the first two harmonics with a span of three times the modulation frequency rather than the entire SES of the band filtered signals. It is possible to avoid most of the interference from a small number of undesired impulses in the SES. PMFSgram uses several bandwidths from 1.5 times to 4.5 times the fault frequency and for each bandwidth has the same number of central frequencies. The frequency setting method is able to select an optimal frequency band containing most of the useful information with less noise. The performance of the new method is verified using synthesized signals and actual vibration data.

Transmission error (TE) and backlash are important parameters used to evaluate the transmission accuracy of cycloid-pin drives. Existing calculation methods are mostly based on two-dimensional tooth profile models, and these methods ignore the influence of some abnormal meshing phenomena caused by profile modifications (PMs), manufacturing errors (MEs), and assembly errors (AEs), such as the instantaneous mesh-apart of tooth pairs and the eccentric load on the tooth surface. To fill this gap, a novel approach to accurately calculating the TE and backlash of a cycloid-pin gear pair based on the error tooth surfaces is proposed, and its feasibility and effectiveness are validated by comparison with the theoretical analyses and the results from the literature. Based on this, the effects of the PMs, MEs, and AEs on the transmission accuracy are studied, which will be helpful in optimizing the tooth profile design of a cycloid gear and the tolerance allocation during the installation of a gear pair. The proposed method is also expected to provide accurate error excitation data for the dynamic analysis of cycloid-pin drives.

This paper addresses the challenge of designing efficient framework for real-time object detection and image compression. The proposed Compressive Convolutional Network (CCN) is basically a compressive-sensing-enabled convolutional neural network. Instead of designing different components for compressive sensing and object detection, the CCN optimizes and reuses the convolution operation for recoverable data embedding and image compression. Technically, the incoherence condition, which is the sufficient condition for recoverable data embedding, is incorporated in the first convolutional layer of the CCN model as regularization; Therefore, the CCN convolution kernels learned by training over the VOC and COCO image set can be used for data embedding and image compression. By reusing the convolution operation, no extra computational overhead is required for image compression. As a result, the CCN is 3.1 to 5.0 fold more efficient than the conventional approaches. In our experiments, the CCN achieved 78.1 mAP for object detection and 3.0 dB to 5.2 dB higher PSNR for image compression than the examined compressive sensing approaches.