Fault Diagnosis Method Based On Vector-HOS And Its Application

HOS (Higher order spectral) analysis method is a powerful tool to non-linear, non-Gaussian signal processing. However, HOS method is based on single channel signal, which means that it can not embody the vibration characters of the rotor integrally. The vector-spectrum based on data fusion is one of the full information techniques. It fuses the double channels information of one rotor's cross section, and reflects the vibration characters of this section factually. In this paper, the vector-spectrum technique was introduced to the HOS analysis, and the theory of the HOS was developed. Based on this, some new approaches of mechanical fault diagnosis were put forward, and applied to the recognition of typical faults. The main content is as follows:1.As one style of the HOS,bispectrum, it posses all the properties of HOS,and has the lowest order. The paper presented the HOS theory briefly, and studied the definition, character, algorithm and physical meaning of bispectrum weightily. Then the deficiency of bispectrum was analyzed, and the start and the necessity of further study were pointed out.2.To the the deficiency of bispectrum analysis method, the vector-spectrum technique was introduced to the bispectrum, and a new vector-bispectrum analysis method was put forward and studied by simulation and experimentation. The results indicated that, the vector-bispectrum could fuse the double channels information, and reflect the vibration characters of the rotor integrally, which were important in next fault diagnosis.3.The BP neural network (BPNN) has excellent classified and extensible ability. The paper introduced its structure and learning arithmetic. Combining vector-bispectrum and BP neural network, a new fault diagnosis approach was proposed. In the new approach, vector-bispectrum was used as signal preprocessing to extract the feature vector. Then the feature vector could be served as the input parameters of BP neural network for classification. The experiment results showed that this proposed approach was effective.4. The support vector machine (SVM) is a new machine learning method based on statistic learning theory, it can solve small-sample learning problems better. A new fault diagnosis approach based on vector-bispectrum and SVM was proposed. Applying this approach to the rolling bearing and gearbox fault diagnosis, the results showed that the proposed approach was very efficient to extract the signal feature and improve the veracity of SVM in fault diagnosis significantly.5.Support vector data description (SVDD) is a one class classification method. It can solve the problem of insufficient fault samples in fault diagnosis. Firstly, Combining bispectrum and SVDD, a new fault diagnosis approach of bispectrum-SVDD was presented, and the results of experiment indicated this approach was useful.Then, the vector-spectrum technique was introduced to the bispectrum-SVDD to form another new diagnosis approach of vector-bispectrum-SVDD. And at last, the two approaches were all applied to the research of gearbox experiment. Compared with the bispectrum-SVDD, the approach of vector-bispectrum-SVDD could achieve classification of fault diagnosis better.