Research On The Typical Defects Classification And Recognition For Aircraft Rolling Bearings

At present, some common defects detection method is based on large sample data sets, but, in a practical application, the small data can be got due to environment restriction, in view of the poor accuracy of traditional classification algorithm for small data, building a model of data defects detecting based on small sample is necessary. And for this, A method of combing Bayes classifier and image processing technology is established to test, classify and identify for some typical defects of aircraft rolling bearing. The classification and recognition of Naive Bayesian classifier based on Bayes network is systematic researched by means of using the methods of theoretical analysis of Bayesian network, digital simulation and experimental research, to improve the classification accuracy of the typical defects of aircraft rolling bearings. A brief introduction to the project and its main achievements are as follows:Based on the analysis of the principle of Bayesian theory and Naive Bayesian classifier, due to the shortage of the traditional classification algorithm, the method is raised to improve the data parameter evaluation of Naive Bayesian classifier to avoid the extreme value of0or1for parameter estimation, the method is of great help to improve the classifier learning and classification accuracy for the bearing defect detection.The feature of aircraft rolling bearing and its defects have been extracted by using image processing technology, in this feature parameters are extracted, the feature of binary ratio, circularity, rectangularity and7invariant moments are added, the experiment data analysis results show that the binary ratio characteristic can make a distinction between good image and defect image of the bearings, the circularity and rectangularity can make a distinction between damage and scratch image of the bearings, because the7invariant moments have the invariance when the image is moved or rotated, the position requirement of collected image is greatly reduced.The Naive Bayesian classification model is established, typical defects classification and recognition for aircraft rolling bearing is realized. Based on typical defects feature parameters of aircraft rolling bearing are extracted, it was randomly divided into two thirds as the training sample and a third as the test sample. To cross training and learning for classifier, a better accuracy classification model is built. In order to prove the Naive Bayesian classifier has higher classification accuracy than other classifier, at the same time, the Neural Network classifier is built to comparison with it, the result shows that the classification accuracy of Naive Bayesian classifier was3.3%bigger than the Neural Network.Aircraft rolling bearing database is established, aircraft rolling bearing testing standards are formed. First, the specification of aircraft rolling bearing is made, to determine the bearing is good or not. And the causes of defect bearings and preventive measures are analyzed. This database is built using SQL Server based on the designation, structure parameter, defect type and so on of aircraft rolling bearing, to serve the defect detection of aircraft rolling bearing.