|
CSME 2019/08
Volume 40 No.4
:
413-422
|
|
Research on Fault Diagnosis of Aeronautic Gear Based on Permutation Entropy and SVM Method
Jia-Yin Liua, Xiao-Guang Yub and Qing-Kai Hana
aSchool of Mechanical Engineering, Dalian University of Technology, Dalian, 116024, PR China. Collaborative Innovation Center of Major Machine Manufacturing in Liaoning, Dalian, 116024, PR China.
bSchool of Mechanical Engineering and Automation , Liaoning University of Science and Technology ,Anshan, 114051, PR China.
|
Abstract:
The permutation entropy method is a new method for the spatial characteristics of time series. It can not only represent the complexity of the signal, but also can measure the uncertainty of a system or a piece of information. It can detect the dynamic mutations of time series of complex systems well. Small changes in time series data can be well demonstrated by it, which is beneficial to deal with nonlinear problems. It can be considered for fault diagnosis. Due to non-stationary operating conditions, driving force of the equipment, damping force, elastic force and its own non-linearity, the mechanical system usually exhibits strong complexity, nonlinearity and non-stationarity. Therefore, it is necessary to perform permutation and entropy analysis on the vibration signal. Support Vector Machines (SVM) is a new machine learning method. It can transform the non-linear non-separable model into a linear separable model by mapping the sample space to the high-dimensional feature space through nonlinear mapping, and construct an optimal classification hyperplane in the high-dimensional feature space, thereby achieve pattern recognition. Based on this, a gear fault diagnosis method based on permutation entropy and SVM is proposed.First of all, starting from the phase space reconstruction theory, the permutation entropy method is used to calculate the entropy of the gear fault vibration signal, and the permutation entropy value and the normalized permutation entropy value can be obtained. Then, the permutation entropy value and the normalized entropy value are taken as the fault feature vector and input into the SVM and neural network classifier respectively for training. Finally, SVM and neural network classifiers are used to identify and classify the faults, and the classification results are compared. The method is applied to the gear experimental data. The analysis results show that this method still can effectively achieve gear fault diagnosis even in the case of small samples.
|
|
Keywords: gear, feature extraction, fault diagnosis, permutation entropy,SVM.
|
Download PDF
|
|
*Corresponding author; e-mail:
|
©
2019
CSME , ISSN 0257-9731
|