Research On Bearing Life Prediction Based On Full Information Support Vector Regression

The research on the reliability and life of the bearing is an important part in the safety management of mechanical equipment.In order to accurately monitor and predict the running state of the bearing,it is necessary to make sure that the bearing vibration signal collected includes the entire vibration information of bearing.According to the vorticity law of the rotor,the spectral structure and vibration intensity of the bearing vibration signals collected from different directions in the same cross section are not necessarily the same.In order to solve this problem,this paper uses the homologous vibration signal fusion technique,Full Vector Spectrum(FVS)technology,through the dual channel signal acquisition,fusion,the more complete and accurate bearing vibration signal is obtained.Furthermore,after the bearing vibration information of each moment is obtained,the description of the bearing operation status is often multi-index and multidimensional.In order to be able to more directly reflect the bearing current health,this paper introduces Multivariate State Estimation Technique(MSET),to estimate the bearing current operation state according to the bearing history information,and the multi-state information of bearing is processed and transformed into a single index: health degree.Compared with other single index,health index based on MSET bearing running condition of multivariate information is a comprehensive estimation result,it is more helpful to detect the development process of bearing failure,which can be sensitive to the subtle and imperceptible incipient faults.In addition,this paper is based on mutual information theory and clustering algorithm makes the establishment of the health status set of bearing more accord with the actual situation of bearing operation,estimation model is more convenient to calculate,to enhance the rationality of health estimate.The failure type of bearing is not easy to be diagnosed in the early stage of failure.Therefore,in order to predict not only when the bearing fails,but also to diagnose what the bearing failure is,an improved Deep Neural Network(DNN)with multiple kernel structure is proposed,which can diagnose the fault type in the initial stage of bearing health degree curve decline,the reasons for the failure of the bearing are explained.In the end of this subject,according to the degradation law reflected by the bearing health degree curve,this project uses Support Vector Regression(SVR)to achieve prediction of bearing life,and considering space reconstruction theory,discussed the input parameters of forecasting model,makes this model more reasonable and accurate to predict the bearing health and realize the bearing life management.