Study On Weak Fault Feature Extraction Of Rotating Machinery In Heavy Noise

For the fault diagnosis of rotating machinery, due to the field environment of mechanical equipment is very complex, collected vibration signal contains a lot of noise. Especially in the vibration signal contains weak fault, fault characteristics of low energy, has also been covered by a lot of noise, seriously affecting the feature extraction and fault recognition. This paper takes the rotating machinery as the research object, studies the feature extraction of weak signal under strong noise background. In recent years, the Empirical Mode Decomposition(EMD) time-frequency analysis method is widely used in the analysis of rotating machinery vibration signal processing, however, in the vibration signal with strong noise, the noise can interfere with the extraction of weak signal by EMD,The decomposition results will have serious end effect, have great influence to the quality and effect of decomposition. This paper, based on the multi-resolution singular value decomposition and the empirical mode decomposition of weak signal extraction method by correlation of signal in noise and useful signal, through the layers of singular value decomposition can improve the signal-to-noise ratio, the removal of strong noise is realized and the weak signal is kept, through the EMD and Hilbert envelope spectrum of the signal frequency components was extracted after denoising. Through simulation and real fault data, it is proved that this method has a good effect on extracting weak signal in strong noise.Singular value decomposition, as a kind of nonlinear noise reduction method, is widely used in the detection of signal denoising, however, for weak signal detection in strong background noise, the singular value decomposition effect is not good, about the problem, the multi resolution singular value decomposition is proposed. Through the decomposition of the signal, achieved the level of noise removal and improve the SNR of signal. The effective singular value is selected by the method of singular value decomposition difference spectrum theory, after obtained the denoising signal, the EMD decomposition to pick up the different component signals, the extraction of weak signal is realized. However when the signal contains two or more bands are too close to or amplitude ratio is very small, The EMD decomposition cannot guarantee single frequency component is separated correct, maybe will get false components, so that the impact of the failure of the results of the judge.According to the similar frequency signal is difficult to be separated, put forward the Variational Mode Decomposition(VMD) and the method of combining the singular value Decomposition. VMD decomposition is based on the Alternate(ADMM) Direction Method of Multipliers optimization algorithm through the iterative search for the optimal solution of the variational model to get the center frequency and bandwidth of each component, able to adaptively implement signal in frequency domain decomposition and effective separation of each component. VMD decomposition has a solid theoretical foundation, also has stronger noise robustness, has higher frequency resolution in modal separation. When used in VMD to decompose the signal but need to be set to the layer number of modal decomposition. This paper adopts VMD combined with singular value decomposition, using the method of VMD strong antinoise ability and high resolution characteristics of similar frequency signal, and the singular value decomposition method can effectively remove the false frequency characteristics. First of all, the signal is decomposed by VMD to get a series of component. Then detect the false composition by singular value decomposition to select the appropriate VMD decomposition layers, to achieve effective separation of similar frequency components. Aiming at the problem of VMD decomposition of the end effect, using the Calman filter to predict continuation of endpoint data, to ensure that the endpoint of signal energy is not lost. Finally the characteristic frequency of fault signal is obtained through the Hilbert transform and its envelope spectrum.