Research On Large Bearing Defect Detection Based On Image Processing

With the society progress, national economy development, the national economicconstruction urgent need,and large bearing rings market demand increasingly growth,thequality of large bearing rings has been paid more and more attention. The existing largebearing surface defect detection methods mostly rely on artificial visual or outdatedsimple instrumentation,which result in lower labor efficiency, lower recognition rate,bad defect detection and classification precision. Not only would inspection personnelbe suffered physical and psychological damage but also may be not work properly whensometimes working in high temperature,high dust,noise,vibration and other harshenvironmental conditions. So research on the large bearing surface defect detection ishot in resent years.Large bearing surface defect detection was studied based on digital imageprocessing technology in this paper,the main contents are as follows:1.Large bearing surface defect’s typical performance type and defect area analysis.Three kinds of classical filtering denoising methods were adopted in this paper for thecollected large bearing surface defect images,and the denoising effect is analyzed andevaluated. The median filtering algorithm was selected according to the experimentresults.2.Image edge detection algorithm analysis. Large bearing surface defect imageswere compared using a variety of classical edge detection operator. An improved Sobeledge detection operator was proposed. After analyzing the contrast experimental effectdiagram, it is founded that different defects types should be analyzed according to theclassical edge detection operator applicable scope and defect image type. Good imageedge sharpness and continuity and noise suppression capability can achieved using theimproved Sobel operator proposed in this paper.3.Defect feature extraction and selection. Hu moment invariant feature,morphological features and texture features were extracted from defect image.Systematic analysis and argumentation were made and Hu moment invariant featurerequired in the classification identification were determined.4.Classification identification algorithm based on the BP neural network. Thedefects are classified using identification method based on the traditional BP neural network in this paper. The experimental results show that,the BP neural network canidentify the large bearing surface defect types to make large bearing surface defectdetection without omissions. The classification accuracy is effectively improvedcompared with the manual inspection, which can improve enterprise economicefficiency and market competitiveness.