The Remaining Life Prediction Of Rolling Bearin Based On Data Fusion And Least Squares Support Vector Machine

Bearing is an important linking part of mechanical equipment.Among various kinds of bearings,rolling bearing has been a hot research object in the state of equipment failure.Thus,the study of the remaining life of rolling bearing is not only helpful to improve the service life of mechanical equipment,but also can help people to make reasonable maintenance measures in advance.And this will greatly reduce the economic losses and accidents caused by the bearing failure to the enterprise.In this paper,the least squares support vector machine(LS-SVM)model is used to predict the remaining life of the rolling bearings.And the feature which is obtained from the fusion of Mahalanobis distance(MD)and Kernel principal component analysis(KPCA)is chosen as a research object.Through the comparison of the prediction results,we conclude that the feature that is acquired from KPCA shows a better prediction effect.During the signal analysis and processing,it is necessary to pre-process the collected data.And an improved lifting wavelet transform(LWT)method was adopted to de-noise the vibration signal from rolling bearings.Firstly,the decomposed wavelet coefficients are obtained by lifting wavelet analysis against the full life data,and then the reconstructed signal can be obtained after lifting the inverse transform of wavelet against the wavelet coefficients.By calculating the correlation coefficient between the reconstructed signal and the original signal,the wavelet coefficients which are less than the set threshold can be set to zero.Then the processed wavelet coefficients are used to lift the wavelet reconstruction to realize the de-noising process.Finally,feature extraction is needed for the pre-treated data,and the parameters which chose the time domain feature,frequency domain feature and wavelet domain feature as the signal characteristics are studied in this dissertation.Before the establishment of the model,it is necessary to construct the input characteristic parameters of the model by using the extracted signal features.In the fourth chapter in this dissertation,the methods of Mahalanobis distance and Kernel principal component analysis(KPCA)were used to fuse the characteristic parameters,and then two different signal characteristics(multi-parameter and single-parameter characteristics,respectively)are obtained,while the fifth chapter is mainly about the establishment of the LS-SVM model and the prediction of the remaining life of the rolling bearing.The radial basis function is chosen as the kernel function of the model.The penalty factor and kernel function parameter which will have a better predicted result are acquired by parameter optimization,thus the LS-SVM model was obtained.In the final part in our dissertation,we use the LS-SVM model to predict the remaining life of rolling bearing with single parameter input and multi parameter input.Results indicate that the life prediction of the LS-SVM model with multi-parameter input which is acquired by using the KPCA method to fuse characteristics is relatively better and the precision is higher.It is of great significance in engineering application and scientific research.