Fault Feature Extraction And Diagnosis Of Rolling Element Bearing Under Variable Speed Condition

Rolling element bearing is a kind of general and precise components and parts of rotatingmachinery while it is one of the most important sources of troubles. Therefore, it hasimportant significance to research the technique of bearing fault diagnosis. In this paper, thestudy of fault feature extraction and diagnosis of rolling element bearing was carried outspecific to the variable speed condition. The main research contents are as follows:1. Vibration signals acquisition and noise reduction. Collect test data of rolling elementbearings (including normal condition, inner raceway fault, outer raceway fault and ball fault)under variable speed condition using QPZZ-II rotating machinery experimental system. Anew EMD de-noising method based on3criterion–correlation coefficient–kurtosiscriterion was proposed by the combination of EMD threshold de-noising and EMD filterde-noising. It can reduce noises in vibration signals effectively.2. Fault feature extraction and condition monitoring. Specific to the problem of conditionmonitoring and feature extraction of rolling element bearing under variable speed condition,principal component analysis was adopted. In this paper, a new concept named “full speedsample” was proposed. The PCA modeling samples were composed by time-domaincharacteristic parameters calculated from every1024points of vibration signals. Through T2and SPE statistics and control limits, the bearing fault can be detected successfully. Moreover,fault feature extraction was finished based on projection of bearing’s data on principalcomponents.3. Bearing fault mode identification and diagnosis. The improved K-nearest neighborclassifier was used to classify the bearing’s fault characteristics and Euclidean Distance wasreplaced by Mahalanobis Distance in KNN algorithm. Experiments show that this method canexclude correlation interference between variables and classify bearing’s faults accurately. Inaddition, the result is better than the Neural Networks and Support Vector Machine.The simulation and test results show, it has achieved the desired goal that the fault feature extraction and diagnosis of rolling bearing under variable speed condition. Moreover,the method proposed in this paper has broken the bottleneck of traditional fault diagnosismethod limits to stationary condition. So, it has practical significance.