Defects Detection On Surface Of Thread Based On Digital Image Directed By FangYi

According to the situation that defect detection and quality evaluation of screw thread are operated by workers in the realm of industry nowadays, this paper puts the defect detection based on digital image on surface of thread as research object and tries to do them use the technology of computer vision and digital image processing, then enable it to perform defect recognition on surface of screw thread and assertion on the degree of defect graveness. It is the detection of digitizing, no-contacting, most efficient, highly automatic, strongly reliable, so it of not only great practical significance but economic.We start from the collection of images. Photo sources should be laid out appropriately so as to let the object photographed and the backgrounds separate clear. The screw concerned can be photographed by digital camera on the principle of optical projection. Rotating a round platform that sustains screw for each particular invariable angle and photographing the object at the same time, we obtain images of thread axis profile section accordingly. In order to detect and recognize defect of thread, we yet need to process the images advanced with technology of digital image processing like removing noises , edge detection, to binary transformation and so on. With the method of binary thinning of mathematical morphology, we compute the relative distance between two contiguous thread and feature points. Then we search each region of thread inside the edge line of screw. Because the surface structure and parameters are different among each thread of varied type, we have to work out a particular solution to determine each single thread region for each type.Because we can't assess defects with only one image and the information of all images of one screw is are related relatively, we need to synthesize all the screw axis profile sections and analyze them all. The result is a three-dimensional structure indicated by other multiple images, which depict the information of thread from outside to inside by a sequence from top to bottom. Then we compute and analyze these three-dimensional structures of standard non-defects thread and varied defective thread separately to obtain a series of statistics from which we can extract suitable features to form rules. With the guidance these rules, we can assess whether there are some defects and if yes, the sort, degree of each defect, and then get the final result. How to form these rules is important. This paper attempts to classify defect features using pattern recognition, and proceed with quantitative analysis with regional features based on qualitative analysis. Additionally, this paper applies mathematical morphology theory and arithmetic to the process (use dilation and erosion to remove the noises caused by light and thinning to obtain the relative distance between two contiguous thread in image) .This paper aims at the objective of expansiveness of screw type, effectiveness of result and efficiency of processing, and tries to make it a practical system which can be used for general types of thread.