Study On Detection Technology For Steel Strip Surface Defects

Steel strip surface defects on-line detecting is an important mean for improvingthe quality of steel strip, increasing production levels, reducing energy consumption,and strengthening the steel mills competitive ability in marketplace. But with produc-tion technology constant improving, and the user's requirements on steel strip surfacequality ever-increasing, the conventional inspecting methods can not meet the require-ments of steel strip production. Now with improvement on electron, computer, andartificial intelligence, the machine vision technology has become a research focus onsteel strip surface inspection system. In the 1990s, Germany, American, and Japanhad separately developed on-line steel strip surface defect inspection system (SIS)based on machine vision, that made important contribution on improving their steelstrip surface quality. However, until today, our research on this field still only in itsinfancy, to a certain degree, this condition in?uences steel strip quality improving.In SIS based on machine vision, the key technologies are grabbing surface im-age rapidly, and on-time identifying defects. Now, grabbing image is almost all im-plemented by standard camera, which has two styles: linear and area. Linear camerasgrab image through relative motion; area cameras grab image through time expo-sures.But standard camera output all data produced, a mass of data need to be com-municating and prossing on-time, this is a task for system, by this reason, the existingSIS be complex, and expensive. It is the key course that the SIS can not be popu-larized. On-time finding defects, describing defects and classifying defects from allimage data, and meeting the requires of product inspection, are a waiting resolvingtechnology question.After analyzing typical SIS system,on the principle of modularization, low price,and high performance, a new scheme of SIS was presented and accomplished. Allmain part's structure and function in the system are optimized. Synchronization ofinspectors with object, and inspector with inspector, are researched.Using high speed linear CCD and high level FPGA, a fully hardware smart linearcamera, which has special pre-processing function, is developed. It is implementedat the same time, grabbing image and identifying defects. This strategy reduces the requirement on communication, data storing and processing.System optical calibration methods and geometrical imaging model calibrationmethods are researched. System elector-optical response model and geometricalimaging model are established. After analyzing the courses for elector-optical re-sponse non-uniform, and estimating system noises in?uence, a calibration method forelector-optical non-uniform is implemented, and the non-uniform error is corrected.For imaging model, from statical style and dynamical style, image coordinates andobject coordinates are calibrated.Defects are segmented by using 2D maximum between-cluster variance algo-rithm and particle swarm optimization algorithm. The strategy of inertia improving inPSO is improved, reducing defect segmentation time-consuming. Defects first orderfeatures, second order features and moment features are extracted. Using principlecomponent analysis (PCA) and independent component analysis(ICA), the featuresdimension are reduced.After comparing classifiers, using the principle of support vector machine(SVM)space sphere for multi-value classification and ordered weighted averaging(OWA), adefect classifier is implemented and experimented. Under limited samples, classingaccuracy is up to 84%.