Research On Fault Diagnosis Of Gearing Box Based On Analysis Of Wavelet Packet And Support Vector Machine

Mechanical devices are becoming more and more high-performance, high efficiency, high automation, and high reliability as science and technology is progressing step by step. Gearing box is a key component of mechanical device and also a part which is broken down easily, which is used in transmission parts for changing speed and transmitting power because of such merits like fixed transmission ratio, huge driving torque and compact structure. So its running state poses great influences on working performance of the whole machine. Therefore it is of great significance for diagnosing faults with gearing box, not only time and cost of repairing could be cut down but also accuracy and repairing quality is increased to some extent to create considerable economic benefits.Vibrating signals of gearing box are so complex that a great amount of messages of moving parts and structures of machine are comprised despite of messages that reflects working situation of gearing and bearing themselves relevant. Therefore if vibrating signals of gearing box are analysed only in time kinds of faults of gearing box in accuracy. A new kind of technology of diagnosis of fault of gearing box-Method of diagnosis of fault of gearing box based on analysis of wavelet packet and support vector machine is proposed and researched in this paper. This method could be classified as two strategies to carry on as below for diagnosis of fault of gearing box.(1) Extraction of vectors of characteristic of fault. Firstly, the feature vectors are extracted in method of good performance in denoising based on wavelet packet transform. The energy eigenvalues are regarded as gear box feature vector that are extracted by the method of wavelet packet decomposition and reconstruction for the non-stationary signal of different condition of gearing box.(2) Research of data identification method. Eigenvectors of the gearing box extracted are taken into a multi-classifier Support Vector Machine built by number of two classifiers SVM to train to model. Then the left sample is taken in to test, drawing in a classification result of seven faults, combining with two fork number decision tree and the theory of voting.The experimental results have shown that this method behaves well in time and precision of classification, of which recognition ability is strong and accuracy is up to (95.2±0.88)%by using the mathematical statistic knowledge, and could be applied to practice.