The Research Of Weak Fault Information Enhancement Method Of Rolling Bearing Based On Time-Frequency Analysis

Bearing, as a basis component in the modern industry, is used very popular. Bearing failure will lead the whole machinery and equipment to stop running, even the secondary damage of mechanical equipment. This will cause economic losses and safety incidents. If, we can detect the failure to warn in the early stage of bearing failure, then we can plan to take corresponding measures to reduce or avoid the bad effects. But during the early fault state of bearing, the fault vibration signals are often very weak. The traditional signal analysis technologies are not so perfect. Aiming at this problem, in this paper, we make some relevant studies about the bearing early faults based on the weak fault information enhancement of the construction and recognition of the time-frequency image. We discuss three kinds of methods including wavelet analysis and underdetermined blind source separation technology based on the time-frequency. We have achieved good results, through process the bearing vibration signals.In this paper, we study the model of shock response signals caused by bearing failure firstly. Besides commonly used cycle model, the relative sliding is also considered. The shock response signal is not strict periodic but with some slight random fluctuations. At the same time, we also consider the effect of modulation, which makes the model more accurate. Also we deduce the characteristic frequency under all kinds of bearing fault. This provides a theoretical basis for later analysis.A continuous wavelet transform-envelope spectral average analysis method is developed from the second FFT on the basis of short-time Fourier transform. Firstly, the bearing vibration signals are processed with continuous wavelet transform (CWT), then the envelope demodulation analysis is applied to the time dimension of the results of the analysis under each scale. At last we choose the medium and small scale envelope spectrum for average processing. Through processing the practical early fault bearing signals and simulation signals, we get the analysis results we want, which verifies the effectiveness of the proposed method.The early fault bearing feature is very weak and susceptible to noise interference. It is not easy to accurately identify the fault. The wavelet scale spectrum is seriously affected by the noise. The wavelet scale spectrum rearrangement can improve its time frequency aggregation. Therefore, we propose a feature extraction method based on the time-frequency ridge line, by associating the wavelet scales spectrum synchronous average with the advantages of the wavelet ridge line analysis. Firstly we apply wavelet continuous transform (CWT) to the multi periodic vibration signal. Then the wavelet scale spectrum is rearranged. Then, the wavelet ridge line of the scale is extracted after processing scale spectrum with synchronous average according to the periodicity of the signal. At last we calculate the envelope signal amplitude and give the frequency spectrum analysis and manage to extract the weak fault feature. The effectiveness of this method is verified by processing the simulation and practical example.In practice, observation signals we acquired contain a variety of source signals including the fault information source signal. Because the fault source is very weak, which increases the difficulty for its detection. Blind source separation is one of the ideas to solve this problem. In this paper, we process the signals with short-time Fourier transform to improve the sparse of the observation signal. Then, the mixing matrix is estimated through processing the observed signal time-frequency scatterplot image. Finally, by using the L1 norm which also called the shortest path method to restore the source signal time-frequency matrix, then we can get the estimated source signals in time domain. In this paper, by processing the simulation signals and the acquired practical vibration signals from the bearing seat with the underdetermined blind source separation based on the SCA (sparse component analysis) method, we can acquire the separated signals. Finally the fault characteristic frequency is extracted after envelope analysis. The result of the separation of each channel has its actual physical significance. It verifies the effectiveness of the proposed method, and is managed to detect bearing early weak faults. This can be used in the long-term real-time monitoring of bearing state.