Research On Rolling Bearing Intelligent Diagnosis Based On Information Fusion And VPMCD

Roller bearings are pivotal supporting parts in mechanical systems,whose performances in operation have a close link with the whole device,so it has a practical significance to carry out fault diagnosis of roller bearings.In order to make the collected condition information complete and reliable,multi-sensors are usually arranged in the operational condition monitoring of the roller bearing.Traditional time-frequency methods,however,can barely analyze multi-channel signals synchronously,and multivariate empirical mode decomposition(MEMD)overcomes the problem effectively,which has advantages in the fusion of multi-channel information.That combing MEMD with multi-scale multivariate sample entropy(MMSE)and full vector spectrum respectively in the paper is applied to extract fault features of the roller bearing,and the selected features are recognized by variable predictive mode based class discrimination(VPMCD).The main research work of the paper is as follows:1.A method of extracting degradation features is proposed based on multivariate empirical mode decomposition(MEMD)and multi-scale multivariate sample entropy(MMSE).Firstly,multichannel signals corresponding to various degradation condition of roller bearing are decomposed adaptively by using MEMD,the reconstructed signals by multi-scale IMFs are then analyzed with MMSE.At last,by doing example analysis,it shows that the proposed method can reflect the degradation trend of the roller bearing.2.A method of fault diagnosis for roller bearing is proposed,which is called FV-NA-MEMD,combining noise-assisted multivariate empirical mode decomposition(NA-MEMD)and full vector spectrum.At first,a series of IMFs can get after the NA-MEMD adaptively decomposing the multiple sources information compounded of homologous double channel signal and a noise assisted signal,the IMFs components containing main fault information are then selected from homologous double channel signal according to correlation coefficients to conduct signal reconstruction.Finally,the full vector information fusion is used to merge the reconstruction signal and extra fault feature.In order to verify the effectiveness of the proposed method,a simulation signal and actual signal is applied with this method.3.Selected features are applied to Variable predictive mode based class discrimination(VPMCD)to identify the fault degree and classify faults of roller bearing,which obtain from those two fault features extracting method based on information.Firstly,multivariate sample entropy considered as a feature is used to set up a feature vector,imputing it to VPMCD to identify the fault degree of the roller bearing.Then,amplitude selected from characteristic frequency of all sorts of faults in FV-NA-MEMD envelope spectrum,imputing it to VPMCD to classify faults.At last,it realizes the goal that roller bearing's faults can be intelligently diagnosed qualitatively and quantitatively by the two methods.