Key Technologies Of Surface Defect Detection Of The Tiny Precision Bearings

Precision bearings are widely used and have high requirement for their accuracy. The defect of bearings surface has a great impact on their sale and use. So, it is necessary to detect the bearings defect. At present, the defect detection of the bearings mainly depend on people, however, when the defect is less than 0.075mm, it is difficult for people to identify. In this paper, the charge coupled device (CCD) camera and image processing techniques were adopted to design a detection method on line, which can greatly improve the detection efficiency and accuracy. Through the arrangement of the research work, it can be divided into three parts.1. The whole research program was designed according to the detection objects and requirements. The system hardware equipment were selected and image collection platform was established. The pixels of CCD camera, field of view (FOV) of lens, resolution, depth of field and the characteristics of light source were studied and calculated detailedly, selected hardware equipment which can meet the requirements, established experiment platform and collected bearings images.2. A series of image processing work were done and post-processing defect characteristics were extracted. In this paper, the source, characteristics and removal of image noise were analyzed and the image unfolding and stitching were studied. Furthermore, the ratio comparison algorithm image matching and direct image fusion, linear weighted image fusion were highlighted. The factor that different dots on the side surface of the bearings have different distance from light source and will be affected by it was also taken into account and nonlinear weighted image fusion were put forward. The experimental results show that the gray-scale value mean square errors of images collected directly and the images obtained by the three kinds of fusion algorithm are 4.5,1.3 and 1.1 respectively. This means compared with the other two fusion algorithms, nonlinear weighted fusion has the best results. After the image subtraction, threshold segmentation and binary morphological processing, the defect geometric characteristics were extracted and thus, the detection system camera distortion calibration and size calibration were carried out. The precision bearings with nominal diameter being respective 2mm,5mm and 9mm were carried out defect detection and obtained data. The results show that 9um defects can be detected reaching the design target.3. The bearings surface defect classification system was designed. Three layers BP neural network classifier with three-input and three-output was designed to achieve the classification of the rust spots, scratches and cracks. The input variables of the classifier are the defect’s aspect ratio, squareness and roundness and output results are the rust spots, scratches and cracks. The precision bearings with nominal diameter being respective 2mm,5mm and 9mm were carried out defect classification.The result shows that correct recognition rate of the rust spots, scratches and cracks was respectively 94%,90% and 94% and average defect recognition rate is 92.7%. Besides, GUI interface was designed. Using this interface can deal with discrete operations concentratedly, which will greatly improve the detection efficiency. What’s more, it can show the test results and data on the interface concisely, clearly and conveniently.