To accurately extract the bearing fault-induced impulse features from the vibration signals corrupted by heavy noise and large-amplitude random impulses, an adaptive multi-band denoising model based on the Morlet wavelet filter and sparse representation is put forward. First, to locate the desired frequency band associated with fault components, the Morlet wavelet filter is employed to band-pass the signal from the perspective of the frequency-domain. Herein, an improved Protrugram-based index, termed as windowed envelope spectral kurtosis, is designed as the objective function to choose the optimal center frequency and the bandwidth of the Morlet wavelet filter. Furthermore, benefitting from the time-domain characteristics of the vibration signal, the in-band noise is eliminated by sparse representation. One of the critical parameters (resonance frequency) of the wavelet atom used in the sparse representation dictionary is directly taken as the center frequency of the Morlet wavelet filter, which makes full use of the information derived from the filter, and thus significantly improves the calculation efficiency. Finally, the recovery signal is demodulated by the Hilbert transform to extract the fault characteristic frequency. The effectiveness and superiority of the proposed method are demonstrated through a complete analysis of the simulated, experimental, and engineering signals, as well as a comparison with such prevalent methods as Kurtogram, individual sparse representation, and Morlet wavelet filter combined with the maximum correlation kurtosis deconvolution.

Given the complexity of the application scenarios of rolling bearing and the severe scarcity of fault samples, a solution to the issue of fault diagnosis under varying working conditions along with the absence of fault samples is required. A numerical model-driven cross-domain fault diagnosis method targeting variable working conditions is proposed based on the cross-Domain Nuisance Attribute Projection (cDNAP). Firstly, the simulation datasets consisting of multiple fault types under variable working conditions are constructed to solve the problem of incomplete fault samples. Secondly, the simulation datasets are expanded by means of generating adversarial network to ensure sufficient samples for subsequent model training. Finally, cDNAP is used to obtain the cross-domain simulation projection matrix, which eliminates the variance in the distribution of measured and simulated sample features under varying working conditions. The experimental results of cross-domain for variable working conditions show that the diagnostic accuracy reaches up to 99%. Compared with DANN, DSAN, and DAAN domain adversarial neural networks, the proposed method performs better in bearing fault diagnosis.

In the existing rolling bearing performance degradation assessment methods, the input signal is usually mixed with a large amount of noise and is easily disturbed by the transfer path. The time information is usually ignored when the model processes the input signal, which affects the effect of bearing performance degradation assessment. To solve the above problems, an end-to-end performance degradation assessment model of railway axle box bearing based on a deep residual shrinkage network and a deep long short-term memory network (DRSN-LSTM) is proposed. The proposed model uses DRSN to extract local abstract features from the signal and denoises the signal to obtain the denoised feature vector, then uses deep LSTM to extract the time-series information of the signal. The healthy time-series signal of the rolling bearing is input into the DRSN-LSTM reconstruction model for training. Time-domain, frequency-domain, and time–frequency-domain features are extracted from the signal both before and after reconstruction to form a multi-domain features vector. The mean square error of the two feature vectors is used as the degradation indicator to implement the performance degradation assessment. Artificially induced defects and rolling bearings life accelerated fatigue test data verify that the proposed model is more sensitive to early failures than mathematical models, shallow networks or other deep learning models. The result is similar to the development trend of bearing failures.