Research On Inspection Of Roller Surface Defects Based On Machine Vision

Roller is an important tool for rolling parts production and a key part of rolling mill equipment.It is easy to produce pitting,roll surface spalling,mark and other defects during the rolling process.If we do not timely find the defects on the roll and do not timely repair the defects,it may affect the structure of the rolled parts,so the factory attaches great importance to the surface inspection and maintenance of the roll.The roll will be replaced,and then the scene through the eyes of the workers to observe the surface of the roll and the experience of the workers on the surface defects to judge the extent of the damage and defect type of the surface of the roll,and then operate the spraying repair device to repair the surface defects of the roll.This detection method has a large workload and is not easy to operate.In order to prevent the roll defects from affecting the steel rolling production,we repaired the roll as soon as possible to meet the steel rolling production.Therefore,a method based on machine vision was proposed to detect the surface defects of the roll.The roll was disassembled for detection and repair,which was offline detection.The main research work and contents are as follows:(1)Acquisition of defects on the outer surface of the roll.In order to realize the detection and repair of the outer surface defects of the roller,the hardware design of the vision system is completed through the selection of the camera,the lens,the light source and the motor.In order to ensure a certain overlap between the stitched images and the stitched images,the roller needs to rotate at a certain Angle,but the rotation Angle needs to match the number of camera's acquisition frames to ensure that there is more than one-third overlap between the stitched images and the stitched images.(2)Image preprocessing.In the process of image acquisition,the image may produce noise because of various external factors.After comparing various smoothing filters and image enhancement algorithms,the appropriate algorithm is selected to meet the requirements of image preprocessing.(3)Image Mosaic.Due to the cylindrical shape of the roll surface and the small view of the camera,the image acquisition of the whole roll surface cannot be carried out,which may be due to the lack of surface defect features,resulting in the failure to realize the defect recognition of the roll surface.Due to the cylindrical property of the roller surface,the image taken will be distorted,so a cylindrical correction algorithm based on cylindrical back projection is proposed.After comparing the speed of all kinds of algorithms to extract the feature points,the selection of a suitable for roll surface image feature points extraction algorithm,approximate nearest neighbor and(FLANN)fast matching algorithm and use LMEDS algorithm refined match point,improve the accuracy of the transformation matrix,transform image is obtained by perspective transformation,the overlap together,Then the image fusion algorithm is used to eliminate the stitching lines and realize the image stitching.The stitching image does not appear ghost and double shadow,and the stitching effect is better.The stitched image can detect the surface defects of the roll very well,and analyze what types of defects are found.It is convenient for the industrial computer to control the spraying device to repair the surface of the roll,so that the repair effect is better.