Research On Bearing Shield Surface Defect Detection Based On Image Processing

As an important method of non-contact detection, defect detection based on image processing has been widely used in industry. In this dissertation, static images of bearing are the main research objects. The suitable defect detection algorithms for the defects on the bearing shield surface are in-depth studied. The main research of this dissertation can be summarized into the following aspects:1. Considering the shape features and the industry standards of bearing, a fast algorithm to extract the shield region of the bearing based on least squares method is proposed. After rough thresholding, contour tracking and outline extraction, the contour of the bearing's outer circle is obtained. Least squares method is used to locate the center of this circle and its radius is calculated. Finally, the shield region of the bearing, which is a ring, can be segregated from the original image based on the priori knowledge about the bearing's parameters.2. Suitable pre-processing algorithm to the original images is designed, then an adaptive thresholding algorithm is used to extract the defects, and a post-processing algorithm to amend the output images is proposed. After using the pre-processing algorithm, which combines the conventional image processing algorithms like gray-scale transformation, median filter, Gaussian smoothing and Laplacian sharpening, the image is enhanced greatly. The post-processing algorithm to reject noise based on region growing is used after the OTSU thresholding algorithm.3. To detect the shallow defects of the bearing's shield surface, which are difficult to detect by thresholding, a novel edge detection algorithm is proposed, based on wavelet transform, texture feature and Canny operator. One-layer fast wavelet decomposition is used on the pre-processed image to weaken the random texture on the shield region. To detect the edge, Canny operators with different parameters are separately used in the low frequency and the high frequency parts which are generated by the wavelet decomposition before. Then the output image is generated by wavelet reconstruction. The relationship between the chosen parameters and the texture features of the shield region is summarized, and accordingly, an adaptive strategy is designed, which is based on the histogram's entropy of the shield region.4. After comparing and analysing the algorithms to extract the defects in this dissertation, a final algorithm to detect the defects and make judgments is proposed. The output images generated are fused to combine the advantages introduced in Chapter 4 and 5. Then a judgment algorithm based on calculation of connected domains is used to realize intelligent detection of the surface defect on bearing shield.