Research On Vibration Signal Based Machinery Fault Feature Extraction And Diagnosis

Machinery fault diagnosis is importance to the safe and stability of machinery. With features of on-line, real time, non-destructive detection, convenient, fast and accurate, machinery fault diagnosis based on vibration signal processing is widely used in practice. This dissertation focus on bearing and gear fault diagnosis and take an intensive study on the machinery fault extraction and diagnosis. The work includes study of the Empirical Mode Decomposition (EMD) based signal denoising method; study of the wavelet transform, wavelet packet, Hilbert-Huang transform, Independent Component Analysis based machinery fault feature extraction; and study of the SVM and Nearest Neighbour based fault recogntion. Some new machinery fault feature extraction and diagnosis methods are proposed in this dissertation. The main researches include:(1) Due to the limitations of current EMD based de-noising method can't meet the deamand of both high frequency Intrinsic Mode Functions (IMFs) and low frequency IMFs. An improved EMD based de-noising method is proposed which hybrids the merits of EMD based threshold de-noising method and Savitzky-Golay filtering method. In the proposed method, the high frequency IMFs and the low frequency IMFs are using different de-noising methods. The method is tested on simulated data and real vibration signal and the performance of the improved method is better than the EMD based threshold method and the Savitzky-Golay filtering method used alone.(2) Study on the applicaion of relative wavelet energy and Support Vector Machine in machinery fault diagnosis. The original fault vilbration signal is decomposed by discrete wavelet transform. Then the relative wavelet energy is served as the feature vector. In the classification, the support vector machine method is used to identify the different machinery faults. Experiments were conducted on roller bearing fault diagnosis and the experimental results indicate that the proposed approach could reliably identify the different fault categories and levels of fault severity.(3) Due to the complexity and non-linearity of the machinery vibration signal refelct the occurrence of the fault. Recently the Sample Entropy (SampEn) is proposed and can quantify the complexity of a signal and has the advantage of being less dependent on time series length than Fractal dimension, Kolmogorov entropy and Lyapunov exponent. A machinery fault diagnosis method based on Wavelet Packet Transform (WPT) and SampEn is proposed in this dissertation. The original machinery vibration signal is decomposed by wavelet packet transform. The SampEn of the resultant wavelet packet coefficients are calculated and served as feature vector. In the classification, the support vector machine method is used to identify the different faults. By combined with the wavlet packet transform, the feature information of different frequency band can get. This is more comprehensive and accurate characterised the machinery fault. Machinery fault diagnosis experiments indicate that the method can get better recognition result.(4) Due to the problem of time-frequency feature extraction from the machinery fault vibration signal, the feature extraction from Hilbert spectrum method is studied. The Hilbert spectrum offers a time-frequency distribution of machinery fault vibraion signal. A new fault feature extraction method based on Hilbert spectrum and singular value decomposition is proposed and applied to the bearing falut diagnosis. In order to feature extraction from the Hilbert spectrum, the Sigular Value Decompostion based method which used the singular value of the Hilbert spectrum as the feature parameter is introduced. The proposed method has the merits of the good stability with the Singular Value Decompostion and can better describe the time-frequency matrix feature. For real vibration signal fault diagnosis experiment indicate that the satisfied results can be acquired. The proposed mehod proved to be valuable for engineering application.(5) Study the application of Independent Component Analysis (ICA) on machinery fault feature extraction. A new feature extration method based on ICA and correlation coefficient is proposed. The ICA is used for vibration signal of each fault category and the extracted independent components include the information of the fault. Then the sum of the absolute correlation coefficients of the sample and the extracted indepent components of each category are used as the feature vetor. Finally the support vector machine is used as the classification method for fault diagnosis. The proposed fault feature extraction method is applied to two tasks:gear feault diagnosis and roller bearing fault diagnosis tasks. The proposed mehod could extract the feature of the machinery fault and acquire satisfied recognition results.