Study On The Multi-fault Diagnosis Method For Rotating Machinery Rotor-rolling Bearing

To rotor-rolling bearing system of rotating machinery,the probability of multi-fault is greater than that of single fault.And it’s more important to identify the type of defects accurately in order to arrange the maintenance plan in advance and prevent the accidental shutdown.Multi-fault diagnosis method,in addition to the fault diagnosis of multi-fault situations,should be down compatible with the ability to diagnose a single fault.In this paper,the features of 6 kinds of multi-fault are analyzed,which are the rotor unbalance,misalignment and the rolling bearing multi-fault.It is found that the multi-fault vibration signal contains the feature information of each single fault,and the coupling feature information also exists.Moreover,the multi-fault has a great influence on the amplitude at fault frequency.The vibration signal pickup on the bearing seat is very complex,but it is generally summarized as rolling bearing fault signal,deterministic signal and noise.The weak fault signal may be masked by other signals,because of which it is difficult to extract the fault feature.In order to solve this problem,a fault feature extraction method based on LMD-MED is proposed in this paper.LMD is used as the preprocessing method of MED and the first four PF components are processed by MED to eliminate the noise and enhance the weak impulse signal.In this paper,through the test data of SKF6205 rolling bearing ball,it is proved that this method can effectively extract the fault feature.Besides the ratio of the peak value at fault frequency versus the mean value of the spectrum in 200 Hz band and the signal-to-noise ratio are respectively increased by 150% and 18.3%.Aiming at solve the problem that the LMD-MED method cannot directly determine the MED filtersize and the PF component,an improved LMD-MED method based on kurtosis-filtersize and SNR-component selection is proposed.This method searches for the kurtosis inflection point after LMD-MED to determine the fitersize.Then,SNR at fault frequency of rolling bearing are calculated in Hilbert envelope signal.Find the maximum SNR and determine the corresponding PF component and fault type.Finally,the fault features are extracted accurately in the envelope spectrum.The test results show that the method has a good effect on the single fault and multi-fault rolling bearing fault diagnosis.Based on these experimental data,EMD-MED method and LMD-MED method were compared.Results show that the energy of the components decomposed by LMD is concentrated and LMD has lighter mode mixing than EMD.Besides,LMD decomposes part high frequency noise to the target component,but EMD can decompose most noise of the target component with higher signal-noise-ratio.Aiming to establish muti-fault feature set,the time domain and frequency domain features are extracted in low frequency band and resonance frequency band of rolling bearing.In this paper,wavelet packet transform is adopted to decompose and reconstruct the collected vibration signal.2 frequency features and 8 time-domain features are extracted in low reconstructed frequency signal.In medium frequency band,the signal of rolling bearing resonance frequency band is reconstructed and then processed by EMD.6 frequency domain features and 8 time domain features are extracted in IMF1 component.In this way,a multi-fault feature set with 24 feature attributes can be established.Kernel principal component is extracted by kPCA,and the parameters of the penalty and RBF kernel function in SVM are optimized by PSO method in the train set.The research on 12 kinds of fault types of rotor unbalance,misalignment and N205 rolling bearing fault has confirmed the validity of this method.Moreover the 3-fold cross-validation classification accuracy of train set and fault recognition accuracy of test set can reach 99.44% and 99.58%,respectively.