Feature Extraction And Feature Reduction Of Rolling Bearing Based On Compressive Sensing

State data reflect the real-time equipment operation status,which is important for obtaining the state trend of equipment and analyzing the cause of the equipment failure.The state data needs to be processed and analyzed in real time for getting the equipment operation status in time and accurately.The increasing data from condition monitoring leads to many problems,such as data storage and transmission analysis and etc.These problems cannot be ignored in the field of condition monitoring and fault diagnosis.Compressive Sensing theory is a new data acquisition and processing framework based on information collection.Its core idea is to recover signal from a few observed data with high probability based on the sparsity characteristics of the signal.Compressive sensing provides a new solution to process the massive data collected from condition monitoring.The present study is supported by the National Natural Science Foundation of China(No.51405211),which combines the theoretical investigation and experimental investigation.Based on the theory of sparse representation and compressive sensing,this dissertation focuses on the feature extraction and feature reduction of the rolling bearing.Vibration and acoustic emission signal are processed and a framework of compressive sensing is improved for the fault diagnosis of the rolling bearing.The main content is given as below:(1)From the viewpoint of engineering application,the background and significance of the present study are elucidated.The state of art review on fault diagnosis based on compressive sensing are thoroughly summarized.The key theory and several core problems of compressive sensing are overviewed,respectively.This paper provides a reference for the application of compressive sensing in the field of fault diagnosis of mechanical equipment.(2)This dissertation studies the compression method and the feature extraction method of vibration signal based on compressive sensing.Signal processing and analysis are transformed from one classical time space to one low dimensional compression space.The reduced dimension projection is utilized to compress vibration data by a Gaussian random matrix.A method of compressing data is studied based on the approximate isometric projection.A fault feature extraction method,based on the approximate isometrc projection,is proposed in the compressed domain.At last,a fault diagnosis model is proposed for diagnosis of rolling bearing,which combines the future extraction and a support vector machine.(3)A compression method is proposed to solve the storage problem of acoustic emission signals.It is studied how to decompose an acoustic emission signal and extract features of bond energy in the compressed domain,which combines Acoustic Emission Technology and Compressive Sensing.A method is proposed to extract features of band energy form compressed data.This method is used to assess a state of rolling bearing life cycle.(4)This dissertation studies the sensitivity of the characteristics of the rolling bearing.The quantification of the characteristics sensitivity is transformed into the sparse representation of an eigenvector.The sparseness of the eigenvector of rolling bearing is studied based on Sparse Representation.The sparse dictionary is obtained by a dictionary learning method.The sparseness of high dimensional sparse features is verified in the sparse dictionary.(5)the feature reduction of rolling bearing is researched based on imporved compressive sensing framework.A high-dimensional feature reduction is considered as the problem of signal compression.Based on the approximate isometric projection.property,the Gaussian random matrix is used to reduce the dimension of an eigenvector.A fault diagnosis model is proposed based on the compression learning model.Experimental results demonstrate that the classification can be correctly obtained by using the reduced feature,and the calculation effect of this method is higher.