Technology Research On Surface Defects Inspection System Of Cold Hot-rolled Strip

With the further development of industrial modernization in china,the demands of steel strip in area such as mechanical equipment,automobile manufacturing,aerospace and the building construction and so on is increasing continuously. However, the surface quality has been the important indicator in estimating the overall quality of steel strip. Therefore, it is deadly important to control the steel surface quality in steel production process.In this paper, considering the steel surface inspection demerits of artificial inspection under the background of cold hot-rolled steel surface inspection in certain large domestic iron and steel group,a suite of cold hot-rolled steel surface inspection system was developed.However it adopted the machine vision technology to implement the image recognition that overcoming the shortcoming of artificial inspection in aspects of the lower speed restricted, strong subjectivity and not comprehensive inspection results.Therefore, this system had the advantage in detecting efficiency and lowering the burden of staff in defects detecting process.This paper’s main researching contents are as follows:①According to the basic performance requirements of cold hot-rolled steel surface inspection system in cold status,the overall structure plan was proposed.And the plan contains the hardware structure designing and software structure designing.② Under the researching of traditional steel edge extraction methods, a fast boundary-searching algorithm was proposed.Firstly,the method filtering the noise by the adaptive median filter in order to get the better image quality.Scondly,finding the first suspect boundary point through scaning the first line or column.Finally,detecting out the boundary curve in steel image through the way of boundary tracing.However,the results showed that the fast boundary searching algorithm has the characteristics of fast and accurate boundary searching ability.③Under the researching of common grayscale correction methods in filtering the uneven illumination noise,the top-hat transformation method was adopted to approach the defect image with the noise of uneven illumination.However,the results showed that it is able to filter the uneven illumination noise and get the better image quality.④ Considering the interference of dust noise, the traditional image filtering methods was researched.As the shortcoming of traditional image filtering methods,theimproved median filtering method was proposed.However,it was invalidate that the method has a good advantage in filtering the dust noise and has lower MSE value.⑤Considering the shortcoming of common methods in positioning target defects,a target detecting method based on the theory of projection principle was proposed.The experiment results showed that the method has a better ability in poisoning target region and has an excellent anti-interference ability.⑥According to the researching of typical feature extraction methods,a group of 19 dimensional feature vector,based on the foreknowledge,was determined to be the input value of classifier.However,the experiment results showed that the feature vector can effectively describe the characteristics of defects.⑦Considering the RBF neural network has the advantages of good classification ability and fast convergence rate and so on,the classifier based on RBF neural network was designed.However,the experiment results showed that the classifier has a good recognition rate and has a considerable application value in this system.