Rolling Bearing Fault Diagnosis Based On The Neighbor Function Criteria And Support Vector Machine

Roller bearing plays a key role in various types of machining equipment, and it is also one of the equipments which are apt to produce fault. The normal work of roller bearing is of great significance to ensure the safety, efficiency and high quality operation of the manufacturing processing. Once fault happens, the whole work system, economic benefit and personal safety will be influenced. So the fault diagnosis research of the rolling bearing is very important.During the operation process of the damaged bearing, a series of impulse force will produce when damage spot contact with the surface of its adjacent components, its vibration signal is non-stationary signal. The traditional Fourier transform can only get the whole spectrum of the signal, but it's difficult to obtain the local characteristics, obtaining the desired results becomes very difficult for non-stationary signal. Because of that, fault diagnosis of roller bearing confronts great difficulties.In order to overcome the weaknesses of Fourier analysis, wavelet analysis is used in the paper. Wavelet transform is a time scale analysis method of the signal. It has a high frequency resolution and a low time resolution for the low-frequency part of the signal, and has a high time resolution and a low frequency resolution for the high-frequency part, which can effectively accomplish filtering and feature extraction of roller bearing signal.Then the fault type identification is carried out using the extracted feature vectors. The advantage of fault type identification using SVM multi-classification algorithm is that it requires less samples. It can solve those problems many machine learning methods can't solve, such as: model selection and over-learning problem, nonlinear, and the curse of dimensionality problem, local minimum points, and so on. However, the algorithm itself is flawed, such as DAG-SVMS (Directed Acyclic Graph SVMs, which is abbreviated as DAG-SVMS) could produce error accumulation in the classification process. That is, if the classification error occurs in one node, the classification error will be extended to the lower one. The closer to the place where the classification error happens, the more serious error accumulation will be, the worse classification performance will be.To solve the problem, we carry out the initial classification via neighbor function criterion. Then we process more accurate classification via multi-class support vector machine algorithm. Thus the defects of SVM will be overcomed using the method, and the classification results will be better.In order to verify the validity of the method, experiment simulation is carried out using tested data of the engineering vehicle's roller bearing. Firstly signal de-noising and energy feature extraction were carried out using wavelet packet analysis, then we underwent type classification using neighbor function criterion algorithm combined with DAG-SVMS. We compared it with Fourier analysis, the artificial neural network method and DAG-SVMS. In the end, we found that its diagnosis result was significantly better than the other methods, so it can be used in the rolling bearing fault detection and diagnosis.