| In this paper,a smooth intrinsic time scale decomposition(SITD)method is used to extract the characteristics of rolling bearing vibration signals.By establishing a multi-index Support Vector Machines(SVM)classification model,the bearing health status is identified.On this basis,SVM regression model is used to predict the remaining life of the bearing.The main research contents of this article are as follows:(1)In order to improve the accuracy of extracting vibration information of rolling bearing vibration signals,and aiming at the problem of noise interference during the diagnosis of rolling bearing inner and outer ring faults,this paper proposes a smooth intrinsic time scale decomposition method SITD.The wavelet analysis method is embedded into the inherent time scale decomposition.In the process of method decomposition,an adaptive threshold function is used to select wavelet coefficients,which can obtain more detailed and useful information in the signal reconstruction process,thereby effectively eliminating background noise,while retaining the impact characteristics,prepare for bearing operation status identification and life prediction.(2)In the application of the state mapping method,there is a problem that the early running state variables of the parts cannot accurately predict the remaining life.This paper proposes the use of multiple sets of degradation indicators to establish a model for defining the health status of the parts to achieve accurate identification of the degraded running state of the bearings and improve the accuracy of the remaining life prediction.Based on the analysis of the correlation between Hilbert’s entropy and bearing state,the change of Hilbert’s entropy slope is used as the reference for state division,and feature parameters with high correlation are selected as input variables for state assessment The calibration value representing the state of the part is used as an output variable,and the SVM state definition model is established,which accurately realizes the accurate differentiation of different operating stages of the bearing,and provides the possibility for the establishment of a bearing life prediction model for different operating states.(3)Establishing an SVM regression model based on a single degradation index for the prediction of remaining life has the problems of high dependency on indicators and large fluctuations in prediction results.A multivariate SVM prediction model for remaining life is proposed,and the correlation is analyzed to screen the available degradation indicators.The indicators that characterize bearing degradation and the remaining life are used as input and output variables of the model,respectively.It is assumed that the input variables and residuals are independent of each other and the type of residual distribution is known.The maximum likelihood method is used to estimate the distribution parameters of the residuals.Based on this,the confidence interval of the SVM output variables at a given confidence level is derived.The mean square error is used as a measure of the prediction error of SVM,and the grid search method is used to determine the SVM parameters.Based on the bearing life test data and the predicted degradation signal of the bearing during the decline period,a remaining life prediction model is established.The example shows that the model can accurately estimate the interval of the remaining bearing life,which has strong engineering application value and versatility. |
