Research On The Fault Feature Extraction Method Of Rolling Bearings Based On Graph Signal Processing

Rolling bearing is the key component of mechanical system. According to statistics, more than 30% of mechanical failures are associated with the rolling bearing failures, therefore, the safety and reliability of mechanical equipmen t was affected directly by the operation state of rolling bearing. Diagnosis technique based on vibration analysis is widely applied in condition monitoring of rolling bearing, which become one of the most important fault diagnosis analysis method. Because vibration signals contain a large number of fault components, so that, the purpose of fault diagnosis is detect the changes of vibration signal and extract the signal feature, achieving the fault mode identification.Fault diagnosis includes data collecti on, feature extraction and fault identification and diagnosis. The fault feature extraction is the key of fault diagnosis. The traditional feature extraction includes time domain analysis, frequency domain analysis and time-frequency domain analysis. With the increasingly complicat e of machinery equipment, the early fault feature often submerged in the noise, sometimes the traditional analysis method was difficult to extract the fault feature effectively. As a result, it is necessary to research new feature extraction methods, such as Graph Signal Processing(GSP).Supported by the project of Natural Science Foundation of China(Project Approval Number: 51275161)‖ and the independent research project of the state key laboratory of advanced design and manufac turing for vehicle body of Hunan University(Project Serial Number: 71375004)‖, this thesis investigated the feature extraction methods of rolling bearings in graph domain, and proposed three feature extraction methods of rolling bearings based on graph si gnal processing.The main researches and the acquired innovative achievements in the thesis are as follows:(1) To effectively extract fault features of vibration signals of rolling bearings, a feature extraction method of rolling bearings based on path gr aph Laplacian norm is proposed. Graph Laplacian norm is an indicator to measure graph signal smoothness. In the proposed method, the vibration signal of a rolling bearing is firstly transformed into the path graph signal. Then, the path graph Laplacian nor m is calculated for fault feature extraction and the standard feature space can be obtained. Finally, the Mahalanobis distance(MD) is used to identify the fault patterns of the roller bearing. The analysis results of the practical vibration signals of rol ling bearings demonstrate that the proposed method can be used to diagnose the roller bearing faults effectively. It provides a new way for feature extraction technology different from the traditional time domain, frequency domain and time-frequency domain analysis.(2) To fully reflect the complexity of the internal structure of rolling bearing vibration signals, Spectrum index is introduced into fault feature extraction of rolling bearings. In the proposed method, graph domain expression of the vibration signal of rolling bearings is firstly established, and original sample space is constituted; then, spectrum indicators are calculated for fault feature extraction and the feature space can be obtained. Finally, Laplacian eigenvector correlation spectrum of the known samples and testing samples is used to identify the fault patterns of the roller bearing. Practical measured signals demonstrate that this method can be used to diagnosis the fault of rolling bearings effectively with simple calculation process, fast calculation speed and high classification accuracy.(3) The Laplace eigenvalues are introduced into fault feature extraction of rolling bearings and a feature extraction method of rolling bearings based on Laplace eigenvalues is proposed. In the proposed method, the five largest eigenvalues reflecting the internal structure information for path graph is firstly selected, and the feature vectors could be obtained. Then, Support vector machine(SVM) is used to identify the fault patterns of the roller b earing. The analysis results of the practical vibration signals of rolling bearings demonstrate that the proposed method can be used to diagnose the roller bearing faults effectively.This thesis conducts a systematic research on GSP and its applications to feature extraction method of rolling bearings, such as the feature extraction, the feature selection and the pattern recognition. Besides, it proposes a complete and systematic rolling bearings diagnosis method based on GSP. The research ideas and propos ed approaches of this thesis have a good application prospect in the field of rotating machinery fault diagnosis.