The critical issue for fault diagnosis of wheelset bearings in high-speed trains is to extract fault features from vibration signals. To handle high complexity, strong coupling and low signal-to-noise ratio of the vibration signals, this paper proposes a novel multi-branch and multi-scale convolutional neural network that can automatically learn and fuse abundant and complementary fault information from the multiple signal components and time scales of the vibration signals. The proposed method combines the conventional filtering methods and the idea of the multi-scale learning, which can extend the breadth and depth of the feature learning process. Consequently, the proposed network can perform better. The experimental results on the wheelset bearing dataset demonstrate that the proposed method has better anti-noise ability and load domain adaptability, and can diagnose 12 fault types more accurately compared with the five state-of-the-art networks.

For the data-driven remaining useful life (RUL) prediction for rolling bearings, the traditional machine learning-based methods generally provide insufficient feature representation and adaptive extraction. Although deep learning-based RUL prediction methods can solve these problems to some extent, they still do not yield satisfactory predictive results due to less degradation data and inconsistent data distribution among different bearings. To solve these problems, a new RUL prediction method based on deep feature representation and transfer learning is proposed in this paper. This method includes an offline stage and an online stage. In the offline stage, the Hilbert-Huang transform marginal spectra of the raw vibration signal of auxiliary bearings are first calculated as the input, and then contractive denoising autoencoder is introduced to extract deep features with good and stable fault representation. Second, by using the obtained deep features and Pearson’s correlation coefficient, a new health condition assessment method is proposed to divide the whole life of each bearing into a normal state and a fast-degradation state. Finally, using the extracted deep features and their RUL values, a RUL prediction model for the fast-degradation state is trained by means of a least-square support vector machine. In the online stage, a kind of transfer learning algorithm, i.e., transfer component analysis, is introduced to sequentially adjust the features of target bearing from auxiliary bearings, and then the corresponding RUL is predicted using the corrected features. Results using the PHM Challenging 2012 dataset show a significant performance improvement when using the proposed method in terms of predictive accuracy and numerical stability.

In this study, a novel time–frequency (TF) analysis method, referred to as the scaling-basis chirplet transform (SBCT), is developed by extending the conventional chirplet transform. This method includes a replacement kernel function that can vary the chirp rate with frequency and time by scaling the TF basis at and around the corresponding time center. This enables the corresponding chirplets to accurately match the targeted slopes for every trajectory of a multicomponent signal and within any window length. Therefore, the TF representation obtained via the SBCT can achieve significantly higher energy concentrations even for multicomponent signals with close-spaced frequencies and high levels of background noise. The effectiveness of the proposed SBCT approach was demonstrated by analyzing a numerical multicomponent signal and a vibration signal obtained from a gearbox test rig. Both numerical and experimental results showed that the SBCT can satisfactorily handle multicomponent signals with nonlinear frequency trajectories, close-spaced frequencies, and noisy backgrounds, demonstrating its superiority.

Effective signal processing methods are essential for machinery fault diagnosis. Most conventional signal processing methods lack adaptability, thus being unable to well extract the embedded meaningful information. Adaptive mode decomposition methods have excellent adaptability and high flexibility in describing arbitrary complicated signals, and are free from the limitations imposed by conventional basis expansion, thus being able to adapt to the signal characteristics, extract rich characteristic information, and therefore reveal the underlying physical nature. This paper presents a systematic and up-to-date review on adaptive mode decomposition in two major topics, i.e., mono-component decomposition algorithms (such as empirical mode composition, local mean decomposition, intrinsic time-scale decomposition, local characteristic scale decomposition, Hilbert vibration decomposition, empirical wavelet transform, variational mode decomposition, nonlinear mode decomposition, and adaptive local iterative filtering) and instantaneous frequency estimation approaches (including Hilbert-transform-based analytic signal, direct quadrature, and normalized Hilbert transform based on empirical AM-FM decomposition, as well as generalized zero-crossing and energy separation) reported in more than 80 representative articles published since 1998. Their fundamental principles, advantages and disadvantages, and applications to signal analysis in machinery fault diagnosis, are examined. Examples are provided to illustrate their performance.