Research On Bearing Fault Detection Based On Anomaly Detection Algorithm

Rolling bearings are the common and important parts in mechanical equipment.B-earing failure often cause serious problems in life and property safety.Therefore,the research of rolling bearing fault diagnosis has been the focus in domestic and international engineering field.Because of the actual engineering data acquisition environment,it will cause the problem of a sample imbalance,the limited number of samples.In order to solve these two problems,this paper proposes an innovative diagnosis method based on the combining of anomaly detection algorithm and support vector machine.Firstly,the vibration characteristics of bearing are analyzed.Study on the vibration characteristics and dynamic characteristics of bearing,determining the way of the fault diagnosis of vibration signals.Then the wavelet packet transform is used to process the vibration signal and extract the wavelet packet energy spectrum feature vector.The wavelet packet transform can not only remove noise,but also decompose the fault’vibration signal of the high frequency segment.The feature vector is the input of the following model.After the data is processed by wavelet packet,the problem of sample identification is analyzed.Firstly,a single step fault diagnosis method based on multi class SVM is studied.This paper analyzes the diagnosis effect of multi classification support vector machine with different kernel functions,and determines the best kernel function.In the case of unbalanced samples,the performance of the optimized SVM classifier in this case is still poor.After studying the single step fault diagnosis method,the fault diagnosis based on the combination of anomaly detection algorithm is studied.Based on the characteristi-cs of anomaly detection algorithm,the feature information of wavelet packet energy spectrum is further optimized.The first step is to use the anomaly detection algorithm for fault detection and then the SVM classifier is used to classify the faults.This method can detect a large number of normal bearings in the first step,greatly reducing the burden of subsequent of SVM classification.Finally,the evaluation criterion of the single step and two step fault diagnosis model is established.According to the loss caused by different misjudgment in the actual industry is not the same,to create a combination of the actual loss of the industry and the probability of false positives of the fault diagnosis.The guidelines are not only more comprehensive and more in line with the actual situation.Aiming at the two problems,such as the imbalance of samples and the limited number of samples,verify the effectiveness and superiority of the two step fault detection model based on anomaly detection algorithm combined with support vector machine.