Study On The Method Of Weld Defects Detection Based On Deep Learning

Welding is the most efficient and economical method of permanent metal joining.Almost all metal products,from simple daily necessities to complex equipment,are mostly made by welding.However,in the process of manufacturing metal members,due to the influence of the welding environment and the process,the welding often causes defects such as cracks to affect the quality of the welded product,which greatly affects the performance of the structure.Therefore,it is necessary to perform defect detection on the formed weld of the metal member.In the traditional weld defect detection,it mainly relies on manual visual inspection.It is necessary for the inspectors to accumulate years of experience to judge the presence or absence of weld defects,the position,and the accuracy of the inspection cannot be guaranteed,and the product quality requirements cannot be met.Machine vision is a non-contact detection method with non-destructive,high sensitivity,high detection accuracy,and has been widely used in the field of defect detection.Based on machine vision and deep learning technology,this paper studies the weld defect detection of metal components.The main contents are as follows:(1)Using the 3D laser line scanning is used to obtain the 3D point cloud model of the welded part,and then perpendicularly to the plane of both sides of the weld and the angle of 45°,respectively,the weld is better presented on the depth map from 3D to 2D.(2)Image needs to preprocess,including image filling,image denoising,and image enhancement.For the problem that the sample data is insufficient,the loop consistency is used to generate the anti-network to expand the sample data.(3)In the part of weld defect detection,the deep learning fusion defect detection method based on feature and MNet based on Deeplab-V3 model is designed,and the detection effect is demonstrated and verified.The two methods of weld defect detection are evaluated by AHP and FCE system evaluation methods.It is concluded that the weld defect detection method based on deep learning is better in detection effect.(4)The weld defect detection system is developed,and the typical enterprise case data is used for testing and analysis.The feasibility and effectiveness of the weld defect detection model based on deep learning are proved.The innovation of this subject is to use the 3D laser line scanning to obtain the3 D point cloud model of the welding site,and to better show the weld defect through3 D to 2D,and then to use the cycle consistency to generate the anti-network weld sample.After the expansion,the weld defect detection algorithm based on feature and deep learning are designed and implemented respectively.The two schemes are systematically evaluated to obtain a better detection effect based on the deep learning fusion defect detection algorithm.The weld defect detection system is developed and applied to typical enterprises to prove the effectiveness and feasibility of the test method.