Research On Defect Detection And Quantitative Method For Magnetic Particle Inspection Of Bearing Rings

As an imperative part of the industry,the quality of the bearing ring specifically influences the security of products.Due to the convenient operation and high sensitivity of fluorescent magnetic particle inspection（MPI）,it is widely used in the surface crack detection of bearing rings.At present,the detection of bearing rings is mainly completed manually,which is inefficient and unreliable.Utilizing machine vision technology for defect detection is helpful to realize the automation for MPI of bearing rings.Therefore,the contents of this paper are as follows:Firstly,we analyze the structural characteristics of bearing rings and the formation principle of crack indication.Several indications in MPI images are obtained,such as crack indications,irrelevant indications,and false indications.Then,the MPI images of the bearing ring are collected by the MPI platform with cameras,and the image datasets are obtained.Based on the MPI images,we analyze the geometric features and grayscale distribution of indications in the images of the bearing ring.Secondly,a two-stage detection method is proposed to identify the crack indication.Firstly,in order to eliminate the reflection of the UV lamp,the MPI images are preprocessed.According to the linear structure characteristics of crack indications in MPI images,the Steger algorithm is applied to extract the centerline of crack indication.According to the grayscale distribution characteristics,Gaussian similarity measurement is applied to reduce irrelevant indications and obtain crack candidate regions.In order to further distinguish candidate regions,the improved Mobile Net V3 is proposed to extract the high-level features of MPI images and identify the crack indications.The experimental results show that the recall rate and precision rate of the identification algorithm are 96.5% and 91.7% respectively,and the average time of identification for a single image is 1.7s.Furthermore,based on the identification results,we propose an image quantitative method for different crack types.Firstly,the correction algorithm for the MPI image is applied to eliminate the image distortion and the A1 standard test piece is applied to image measurement.Then,the crack defects are divided into linear cracks,tree cracks,and network cracks.Based on the classification results,the geometric characteristic parameters of crack defects are extracted by the image processing algorithm to achieve crack quantification.Finally,the results of the quantitative method are compared with those of manual measurement.The test results have an average error of crack length of9% and an average error of crack width of 13%,which can meet the actual quantitative requirements.Finally,to test the recognition and quantitative algorithm,the MPI system for the bearing ring is developed.The MPI images are used to verify the system.The test results show that the system meets the industrial requirements for accuracy and efficiency,which promotes industrial application.