基于emd近似熵和ls-svm的齿轮箱故障诊断研究-research on gearbox fault diagnosis based on emd approximate entropy and ls - svm.docxVIP

Research on Gearbox Fault Diagnosis Based on EMD approximate entropy and LS-SVMAbstractThe gearbox is important transmission parts in the mechanical equipment. Research on gearbox fault diagnosis has very real sense. In this paper, EMD (Empirical Mode Decomposition) approximate entropy and LSSVM (Least Square Support Vector Machine) are combined to achieve the gearbox fault diagnosis. EMD method has good localization for signal processing, moreover has very good decomposition effect for non-linear and non-stationary signal. Approximate entropy contains more information in the characterization of the signal dynamics, and has inherent advantages to extract the signal fault feature. The LSSVM make modifications and variations for the drawbacks of too long running time and excessive calculation of the SVM (Support Vector Machine) as a classification algorithm. Experiments show that LSSVM can achieve accurate and fast fault identification in the gearbox fault diagnosis.This paper firstly expounds significance, purpose and domestic and foreign research situation of gearbox fault diagnosis, simultaneously provides an overview of the current fault diagnosis technology. Secondly, the gearbox vibration mechanism and fault types are introduced. Then the endpoint effect drawbacks while EMD method is decomposing signal are focused on, in order to improve the EMD method, mirror extension, as well as combination of approaches of a windowing function in the signal sequence are proposed. The experiments show that the improved EMD method achieves very good results in the signal decomposition. Then gearbox fault feature extraction is completed through the application of EMD and approximate entropy method of combining, and LSSVM highlights the advantage of fault identification by comparing SVM and LSSVM from theory and specific experiments. Finally, the improved EMD method combined with the approximateentropy completes fault feature extraction. Then using LSSVM to identify the fault fe