Research On Machine Vision Laser Welding Defect Detection Algorithm

Machine-vision-based laser welding has the advantages of high quality,high precision,high speed and so on.It realizes the automation of machinery and is widely used in industry.For example,the weight of the car and the production cost are reduced when laser welding is used to weld parts in the automotive industry.In the Ship industry,there is no deformation after welding when laser welding is used to weld boat deck.It is very important to detect the defects on the surface of welding parts by the machine vision technology with the development of machine-vision-based laser welding.Machine-vision-based defect detection system is generally fast and real-time,but it is difficult to improve the accuracy without sacrificing speed.Here Circular Hough Transform(CHT)is introduced into the defect detection.CHT is mainly to find the location of defects in defect detection.The accuracy and rapidity of defect detection have been greatly improved by introducing Hough Transform.This subject comes from a company.The project name is “Machine-vision-based Automatic Equipment in Laser Welding”.The main work of this paper is as follows:Firstly,CHT is applied in the field of machine-vision-based surface defect detection in laser welding.The process is to use the industrial Charge-coupled Device(CCD)for image acquisition,then use Canny operator for edge detection,use Circular Hough Transform to locate the weld position,and finally compare the number of bad pixels with the defect criteria to report detection results.And this algorithm is compared with template matching for defect detection.Experimental data show that the false positive rate and false negative rate are reduced by about 5% and 3% after the Hough Transform is introduced into the defect detection.Secondly,Random Hough Transform(RHT)is applied in the field of machine-vision-based surface defect detection in laser welding.RHT is introduced to detecting circle.The speed of RHT is improved compared with CHT,but this will introduce invalid accumulation.Therefore,the improvement of RHT is as follows.We should judge whether the three points of random sampling are collinear.And we also should limit the distance between the three points,so that a large number of invalid calculations are avoided.When judging whether the candidate circle is a real circle,we should determine whether the edge points are between the inner and outer squares of the candidate circle.And then we can calculate whether the edge points are on the candidate circle.This method can avoid a large number of calculations.The circle parameters obtained by solving the perpendicular bisector of chord.This improves the running speed.The algorithm is simulated on MATLAB and applied to the machine.The accuracy and feasibility are verified.The accuracy of the improved RHT is basically the same as that of RHT and CHT.The speed of the improved RHT is improved about 100 ms compared with RHT and about 600 ms compared with CHT.