Research On Fault Diagnosis Methods For Rotor Systems Based On Empirical Mode Decomposition And Support Vector Machine

The process of rotor systems fault diagnosis includes the acquisition of information and extracting feature and recognizing conditions of which feature extraction and condition identification are the priority. A novel method of time-frequency analysis, Empirical Mode Decomposition (EMD) and the comparatively recent development of pattern recognition techniques, Support Vector Machines (SVMs), are applied to the rotor system fault diagnosis.EMD is a new signal process method proposed recently, it has proved to outperform than other signal process methods. SVMs have better generalization that artificial neural network and guarantee the local optimal solution is exactly the global optimal solution. SVMs can solve the learning problem of a small number of samples. Some research on rotor systems diagnosis based on EMD and SVM has been done. The main work is done as follows:Singular value entropy based on EMD is defined to identify the condition and fault patterns of a rotor system, further research on the effect of sampling rate to the method has been done. Analysis results of the experimental data shows that the proposed method can be applied to fault feature extraction efficiently.The concept of time-frequency entropy based on Hilbert-Huang transform is proposed, Hilbert spectrum offers a complete time-frequency distribution. time-frequency entropy based on Hilbert-Huang transform is used to describe the time-frequency distribution. The analysis results from rotor system vibration signals show that this method can extract fault features efficiently and classify working condition and fault patterns accurately.Statistical learning theory and support vector machine are briefly introduced. an investigation on support vector machine for application in the rotor system fault diagnosis is done. Analysis results of the experimental data shows that support vector machine outperform than BP neural network in train time, generation and robustness.A fault diagnosis approach for rotor systems based on support vector machines predictive model is proposed. This method can recognize and diagnosis multi-faults directly or has not need signal preprocessing to collect fault features. It has the advantages such as simply calculation, strong classify ability and so on. Practical examples demonstrate that the SVMs predictive model is efficient and generally outperform than BP neural networks model.