Research On Automatic Detection And Identification For Surface Defects Of Silicon Steel Strip Based On Machine Vision

In recent years,our country's economy starts the fast development,which makes the demand of electricity increase.At the same time,the power transmission loss,energy regional structure,transmission and other issues are also getting more serious.In order to solve these problems,state grid starts to research and develop the extra-high voltage power transmission grid.It makes the industry of transformer and generator rapid growth.Silicon steel,the main raw material of transformer and generator internal core,also gets great development.Usually,steel factories make bare steels,which cannot be directly used in electric power equipment,and it needs to make insulation coating on the surface of steel strip.The quality of Insulation coating determines directly the quality of the downstream products,so defect detection of surface coating of steel belt is one of the important the production process on steel strip.However,now there is not automation equipment to test the quality of surface coating of steel strip and the traditional manual detection is randomness,not accurate and inefficient.It is based on this background that we propose to research on automatic detection and identification for surface coating of silicon steel strip.According to the need of a manufacturer in Fujian,we study this system.It completes the detection for surface defects of silicon steel strip through image processing and machine vision theory.The main work includes the following parts:Firstly,the design of image acquisition system.According to the light scattering properties of space,a reasonable light source is designed to combine with the BRDF illumination model.It makes the collected images of silicon strip present the light uniform state.At the same time,it needs to complete the task of image acquisition to use a industrial camera during the rapid transmission of silicon strip.Then we need to preprocess the collected images.It designs a kind of compound filter to complete the de-noising processing of images,and makes images have less noise.It uses projection transformation and bilinear interpolation methods to solve image distortion problem.For detecting the defects of the edge on images,the sequence similarity detection algorithm and weighted average method are used to complete image splicing.The judgment of defect-image is obtained by counting the subtraction of no-defect image's gray value and average gray value,then taking absolute value and cumulative.Last,it uses Laplace sharpening processing to get clear the images of the defects.Then the segmentation and feature extraction of the defects.The extraction of the edge on defects is realized by adaptive Canny edge detection algorithm.Then image morphology algorithm is used to repair and perfect the edges of defects.Then using area filling algorithm realizes the fill of area on defects and get a complete binary image on defects.Then using contour tracking algorithm gets the outline of defects.It is used to complete the segmentation and positioning of defects.After completing the segmentation,it uses the relevant algorithms of image-feature to extract shape features,gray features,texture features of defect images,clearly to obtain all kinds of feature values.Finally,it needs to research the knowledge of machine vision and establish BP neural network classifier model about surface defects of silicon steel belt and realize the classification of surface defects on silicon steel strip.