Systematic Study Of The Locomotive Bearing Fault Diagnosis Based On Support Vector Machines

With the high-speed train coming forth and train speed increasing in railway, high precise fault diagnosis of locomotive bearing is very important. Accurately finding bearing faults and repairing or replacing blooey bearing can improve application level of the bearing and avoid the accidents. So, the bearing fault diagnosis is always a research focus.Bearing fault diagnosis includes the acquisition of information,extracting feature information and recognizing conditions of which feature extracting and condition identification are the priority. Intrinsic mode singular value decomposition and Support Vector Machine are combined to realize fault diagnosis of locomotive bearing. The empirical mode decomposition method is used to decompose the vibration signals of bearing into a number of intrinsic mode functions by which the initial feature vector matrixes are formed, then applying the singular value decomposition technique to the initial feature vector matrixes, the singular values are obtained, and then the locomotive bearing work condition and fault patterns could identified by the result output of the support vector machine classifier.This paper introduces origin cause of formation of locomotive bearing fault,basic form and vibration model, The widely used methods of bearing fault diagnosis are analyzed and discussed. Then the fundamental theory of EMD,SVD and SVM are studied, and an investigation on them for application in bearing fault diagnosis system is done. Recur to the MATLAB, the bearing fault signals analysis and disposal are done, and then obtaining the feature vector-singular values; the support vector machine classifier is established, and by which the fault patterns of locomotive bearing could be identified, the generation ability of the classifier is studied under smaller number of samples. The order,inner and outer race fault conditions of locomotive bearing are studied in this thesis, and their feature information extraction and status identification are done. Practical examples prove that the approach is validity and rationality and have good generation ability under small number of samples.