Research On The Defect Detection Method Of The Bead Ring In The Radial Tire X-ray Image

As the key component of automobile,tire affects the driving safety of automobile with its quality.In order to keep the quality problems of tire from affecting the driving safety of automobile,the quality inspection before the tire leaves the factory has become a key link in the tire production process.At present,most of the domestic tire manufacturers rely on quality inspectors to judge whether the tire has defects,but this method is inefficient and subjective,which easily affects the accuracy of tire defect identification.Therefore,there is an urgent need to develop a tire defect detection system which is suitable for domestic production conditions to meet the actual needs of domestic tire manufacturers.Against this background,ours research group and a company in Shandong Province jointly developed a radial tire defect detection system for tire manufacturers,combining image processing and machine learning theory to analyze tire X-ray images.Tire defects are mainly concentrated in the tire sidewall,shoulder,belt layer and bead ring,the research group has completed the defect detection of tire sidewall,shoulder and belt layer.This thesis mainly focuses on the detection algorithm of tire bead ring defects.The main work is as follows:(1)A similarity analysis method based on sampling is designed to segment the repeating part of the tire.We consider that the problem of repeating part of the tire will affect the detection efficiency,so we segment the repeating part of the tire before defect detection.Based on the same texture distribution of the repeating part and the first half of the image,we first downsample the image to reduce the calculation cost,and then use the mean-square error method to judge the similarity between the regions to achieve the repeating part of the tire segmentation.This method is efficient in calculation,fast in segmentation and suitable for large-scale tire image.(2)The detection algorithm of the bead ring defect of the chafer layer boundary height difference skew is designed.According to the characteristics of gray-scale changes in different regions of tire,the sidewall and the bead ring are segmented by vertical projection.Combined with the difference of the cord directions of the tire sidewall and the bead ring,the improved extreme value filter is used to eliminate the tire sidewall cords and retain the bead ring cords.Then we use adaptive binarization and morphological operations to solve the problem of background elimination in different gray-scale regions,and finally we get the boundary of the cover layer.The defect detection is realized by calculating the height difference of the chafer layer boundary on the same side.(3)The detection algorithm of the bead ring defect of the turn-up layer boundary height difference skew is designed.The horizontal projection curve is used to reflect the distribution law of the chafer layer and the influence of the turn-up layer pattern on the chafer layer cord.Combined with the change degree of the turn-up layer gray-scale in the projection curve,the period and variance of the gray-scale projection curve are calculated to generate the turn-up layer feature map.On the basis of binarization and morphological operation,the boundary of the turn-up layer is obtained and the defect detection is realized by calculating the height difference of the turn-up layer boundary on the same side.(4)The detection algorithm of the difference skew between the chafer layer and the turnup layer is designed.According to the coordinates of the turn-up layer boundary and the chafer layer boundary,the boundary difference of the turn-up layer and the chafer layer of the same side is calculated,and compared with the boundary difference of the other side which has the same horizontal coordinate.The defect detection is realized by judging the distinction between the two side boundary differences.(5)The detection algorithm of the bead ring cords bending,cords cracking and impurity defects is designed.We construct the bead ring defect dataset,and use the Faster R-CNN trained to realize the defect detection of the bead ring structure type.In order to solve the problem of insufficient feature extraction of small-size defects such as impurity,the feature maps of convolution layers at three different dimensions in VGG16 network,the convolution part of Faster R-CNN are fused to ensure the feature map include higher-level features and higher resolution features,which improves the accuracy of small-size defect detection.(6)The algorithms of tire bead ring defect detection are integrated into the tire X-ray defect detection system.We design the inputs and outputs that conform to the system process,and reserve parameter setting interfaces for the quality inspectors,which can set different defect detection standards for different defect types.On the basis of the radial tire defect detection system developed by the research group,we further add bead ring defect detection to make the system more versatile.At present,the system has been tested in many tire manufacturers,the speed and accuracy of the detection are good.However,for the bead ring images with poor X-ray imaging quality and extremely low average gray-scale,the detection results of the algorithms are still not ideal,and the robustness of the algorithms needs to be improved.