Auto-hub Defects Detection Technology Research Based On Dynamic Threshold Segmentation

X-ray nondestructive testing is widely used in car wheel quality testing method.Domestic car wheel manufacturers get the wheel hub X-ray image by the on-linedetection system, then access the quality of the products using artificial way mostly.This method greatly reduces the production of automation degree, and the results issusceptible to the staff’s subjective influence which reliability and stability is poorer.The researchers at home and abroad have made a lot of achievements in the car wheeldefects and label, but there are still many problems, such as inspection operation istoo complex、the wheel hub special geometric structure to the wrong result andleakage、detection speed too slow, etc. In view of the above situation, this thesis firstanalyzes the car wheel of the common defects, sets up a complete and practical defectextraction process. This process will deeply study the hub X-ray preprocessing, targetsegmentation and automatic tagging. The main work and innovation reflected in thefollowing respects:(1) According the common quality problem of X-ray image, the causes areanalysed, and the image preprocessing is divided into two steps, collected and afterthe collection. The best image can be got by denoising of the multi-frame acquisitionand the pull of grayscale dynamic range, at the same time, the choice of the bestframes is discussed. Noise reduction and local contrast enhancement will be madebefore image segmentation which set the complete image pretreatment process.(2) Considering the real time online of the nondestructive testing system, adynamic threshold segmentation method is used to the first division and the seconddivision of the image preprocessing. With the concept of the reconstruction ofmathematical morphology, the final segmentation reconstruction result is the firstreconstruct segmentation image based on the secondary division result. This methodresolves the problem of the leakage of convicted and misjudgment from the wheelhub geometric structure. (3) The segmentation result can be marked in the original image automaticallyand saved for technical staff. The wheel hub defects gray characteristics will beconsidered fully when marking, and the false operation is made with the standarddeviation and contrast of the local image which improve the accuracy of defectsegmentation. At the same time, the annotation can be set by the artificial settingparameters that facilitate to human-computer interaction.