Fault Prediction Method Research Based On Full Vector Principal Component Analysis

Rolling bearing is an important part of rotating machinery,and its running status is related to the operating status of the whole unit.The industrial level is constantly improving,fault diagnosis of mechanical equipment has increasingly attracted attention.As the important content of the fault diagnosis,fault prediction can effectively track and predict the development trend of fault.Most of the traditional prediction models are based on single channel analysis;therefore,there is a shortcoming that data information is one-sided and inaccurate.However,the use of the full vector spectrum technique for information fusion of homologous dual-channel signals can fully characterize the operating state of the rotor.In this paper,firstly,we use the full vector VMD method to fuse multiple feature principal vectors to extract uncorrelated feature principal components,and establish a KPCA model to monitor the process data in real time,using the AR prediction model to predict the trend of monitoring indicators to predict whether the equipment fails,and further track the development trend of failure,finally,full vector Hilbert envelope demodulation is carried out to extract the feature frequency of fault data for the fault diagnosis..The main work is as follows:(1)A feature extraction method based on Full Vector Variational Mode Decomposition(FV-VMD)is proposed.Firstly,the homology dual-channel signals is adaptively decomposed by using VMD to obtain several IMF components,and then the full vector spectrum is used to extract the feature vector of the reconstructed signal.Experiment shows that this method can not only extract the characteristic frequency of vibration signal,but also overcome the mode aliasing and have good applicability in feature extraction.(2)A method combining the full vector variational mode decomposition(FV-VMD)algorithm and Kernel Principal Component Analysis(KPCA)is proposed to detect.Firstly,the above method is used to extract the feature vector of the signal in the normal state and the KPCA model is established.Then,the sample data is monitored online using the KPCA model,when the model'sT~2 and SPE statistics exceed the set limit,Initially,the bearing may start to malfunction..The experimental results show that the method can detect the running state of rolling bearing.(3)A prediction model of rolling bearing failure combined with the full vector KPCA and AR is proposed.First,the test data is input into the above method to establish a full vector KPCA monitoring model,andT~2 and SPE statistics are outputted,and then the value is used as input of AR prediction model,finally,according to whether the predicted value exceeds the control limit of the KPCA monitoring model,it is judged whether the equipment is faulty,and the development trend of the fault is tracked and an fault type is diagnosed.The result shows that the full vector KPCA-AR model gives the fault development trend when judging whether the rolling bearing is faulty,and the fault is further diagnosed..