Study On Methodology Of Product Surface Defects Online Detection And System Implementation

With the advancement of manufacture and techniques, the product quality becomes increasingly important, and the online quality inspection system based on machine vision has been an important quality control method.This paper studies the online defects inspection theory and algorithms for product surface based on distribution machine vision, and implements the online defects inspection sytem of model product and non-model product using the defects inspections of presswork and float glass as the instances.Firstly, to meet the requirement of large dimensionality and high precision in product surface inspection, an inspection system based on distribution machine vision is designed, and several inspection sub-systems are used to complete the inspection task cooperatively.For settling image registration of model image, a fast contour sub-space based image registration algorithm is presented to aim at the large computing complexity of traditional image registration. To detect image contour fast and correctly, a direction based multi-resolution morphology contour detection algorithm is presented, and a forecast based bi-thresholding value contour segmentation is used to thresholding the contour pixels. An evaluative method of contour sub-space is established, and the images are registrated through the contour sub-space for improving the efficiency of image registration.As to fulfilling the detection tasks, different defects segmentation methods are designed for model image and non-model image, and a fast defect merging method based on sequence and journey space is presented in order to realize defect clustering fastly. A defect segmentation method based on image subtraction is designed to realize defect segmentation for model image. A defect segmentation method based on thresholding surface is presented to segment defects for non-model image, and a fast thresholding surface construction method based on the statistic of grey level is presented to solve the great complexity and inveracity of traditional interpolation method. A contour subtraction based fake defect discrimination method is designed to eliminate contour fake defects for model image inspection, and a texture based fake defect discrimination method is presented to tide over the fake defects such as insects, smeariness and dust.Based on the analysis of several pattern recognition methods, all kinds of defects characters are picked up. According to the different classification requirements of different production defects, a rule based defects classification method is designed for presswork defects, and an improved neural network based defect classification method is realized for float glass defects.The synchronization and network congestion of distribution machine vision system are also studied here. For meliorating the network congestion in distribution system, an improved congestion controlling mechanism based on the predictable RTT is proposed. Futhermore, a self-diagnosis and self-recovery mechanism is realized for distribution machine vision system.At last, based on the aforementioned theory and algorithms, two patent production, such as an online presswork defect inspection system based on distribution machine vision and an online float glass defect inspection system based on distribution machine vision, are designed and implemented. They have been applied to manufacturing, and achieved good economic and social benefit.