Research Welding Penetration Characteristics Base On Different Conditions Of GMAW

With the development of modernization,various forms of welding process are developing gradually from manual welding to automatic welding,therefore,the requirement of welding quality is also gradually increased.Backing welding is the first step in the welding process.However,at present,it still depends on manual welding so that the welding quality can't be guaranteed.Therefore,it's great significant to study the automation of backing welding.It is a primary research direction to monitor and acquire the information of penetration characteristics in the welding process for the automation of backing welding.Visual sensing technology is one of the most widely used sensing technologies to obtain the penetration feature information.Because it can simulate the welder's observation method,get rich visual information of molten pool in real time,and will not affect the welding process.The backside melting width is the most direct information to characterize the penetration status,but it is difficult to monitor in real time,in some specific working conditions.In actual welding,a skilled welder can judge the penetration state of the current welding according to the shape characteristics of the front molten pool,and adjust the shape of the molten pool in real time by changing the welding parameters,so as to effectively control the penetration state of the welding.For that reason,on the basis of the original experimental platform,a vision sensor system based on Gas Metal Arc Welding GMAW for binocular molten pool is established in this paper.In the welding process,the images of the front and the back of the molten pool are collected synchronously in real time.In addition,the change rule between the shape characteristics of the front molten pool and the backside melting width are studied under differentwelding conditions of GMAW.A prediction model of penetration based on BP neural network is established,so as to determine the characteristics and provide basic data for the study of penetration control.The main research contents and results are as follows:(1)The original visual sensing experimental platform is improved.The binocular vision sensing platform of GMAW is established to realize the real-time synchronous acquisition of the binocular image on the front of the weld pool and the image of the back of the molten pool during the welding process.(2)The Halcon software is used to process the monocular image of the frontal molten pool.The edge contour of the front molten pool is extracted by means of median filtering,image enhancement,binary processing and morphological processing.The two-dimensional characteristic parameters of the molten pool are calculated.The backside images of the molten pool are processed by the same method and the backside melting width is extracted.(3)The three-dimensional shape of the molten pool is reconstructed by the parallax principle of binocular stereo vision,and the three-dimensional shape of the front and mid profile of the molten pool is reestablished,then three-dimensional characteristic parameters of the molten pool are calculated using the extracted 3D feature points from the front contour of the molten pool.(4)Two-dimensional and three-dimensional characteristics of the molten pool are taken as characteristic parameters of the frontal molten pool.The change rule between the characteristic parameters of front weld pool and the backside melting width under different welding conditions of GMAW are studied.The characteristic parameters of the frontal molten pool are taken as the input factors,the backside melting width is taken as the output factor.Then the prediction model of the backside melting width is established based on BP neural network,and the connection weight of the network are extracted.The influence weight between the each characteristic parameters of the frontal molten pool and the backside melting width are calculated by Garson algorithm.The characteristic parameters that strongly correlated to the back melting width are screenedout to determine the penetration characteristics.(5)Accroding the above research results,the welding tests are carried out under different welding parameters.The experimental values of the backside melting width are compared with the predicted values,thus the accuracy of prediction model of the BP neural network is verified.The results show that the relative error of the model is smaller,which can effectively predict the size of the backside welding width and predict the penetration status of the welding.