Research On Weld Defect Detection And Classification System Based On Machine Vision

Welding is a significant basic technology,which plays an important role in today's manufacturing industry.Nowadays,nondestructive testing methods,such as radiographic sensing,have been widely used in the quality inspection of welding workpieces.However,due to the characteristics of the thin-walled metal cans,the commonly used nondestructive testing methods are not applicable,and manual detection is still the most commonly used method in the industrial production.This method,which has low efficiency and great randomicity,is dependent on the experience of the inspectors,greatly affecting the efficiency of production and the quality of products.Inspired by the artificial visual method,a lot of research of visual inspection method has been discussed in this paper,then the weld defect detection and classification system based on machine vision has been designed.The system collects sample images of welding seam by industrial camera,and deals with these samples through image processing operation.Improved background subtraction method and curve detecting method are used to detect and distinguish welding seam defects.Specifically,the work of this paper includes:Firstly,the whole framework of the system is introduced in this paper,including the image acquisition system and the weld image processing software.In addition,the causes and features of weld fusion,weld perforation and cold solder joint are also introduced in this part.Secondly,the process of weld image preprocessing is introduced.This process mainly includes the selection of sample images,the extraction of effective area,the rotation correction of sample images and the extraction of weld core area.Image preprocessing can eliminate most interference of the pseudo defects.It can also distinguish the body-area images and end-area images.Then,the procedure of the detection and classification of body-area weld defects is introduced.A modified background subtraction algorithm is presented to extract the defect feature of welding seam images by constructing the background model of weld image sequence.According to the difference of the area,the brightness and the cumulative curve of feature area,the weld defects are classified.Finally,the detection method of end-area weld defects is introduced.Mean threshold segmentation method is presented to eliminate the residual interference.The curve fitting method is used to determine weld defects.Horizontal accumulation method is proposed to detect the cold solder joint defect.The experimental results show that the system can meet the needs of actual production,and achieve the accuracy of more than 99%.