Research On Laser Welding Quality Monitoring Of Mesh-skin Structure Based On Visual Sensing

TC4 titanium alloy has the advantages of low density,high strength,high specific rigidity,and good corrosion resistance,and is widely used in aerospace,automotive manufacturing and other fields.Fiber laser welding has the advantages of fast welding speed,large penetration depth,small welding distortion,ability to weld at room temperature or under special conditions,and simple welding equipment,which has been widely used in many manufacturing fields.This paper takes the T-joint laser non-deep penetration and deep penetration welding as the background,and uses the image processing technology to extract the geometric characteristics of the weld pool,and uses the neural network modeling method to establish the geometric characteristics of the weld pool relationship with the penetration depth and the weld width of the joint surface further predicts the weld seam section shape based on the geometric features of the weld pool.First,in order to realize the quality monitoring of T-joint non-penetrating deep penetration welding,a fiber laser welding quality monitoring system was set up.The appropriate filter and auxiliary light source were selected,and the best the front image of the molten pool was obtained by adjusting the position of the shooting device.Aiming at the characteristics of the front image of the molten pool,an image processing algorithm was designed,including grayscale processing,piecewise linear stretching,morphological operations,and keeping the maximum communication area.It was used to extract the contour of the weld pool and keyhole,and further The geometry of the bath can be extracted.Secondly,the relationships between the laser power,welding speed,defocusing distance and the geometry of the molten pool are established,and then the relationship between the geometric characteristics of the weld pool and the penetration depth of the weld and the width of the joint weld is further established,using the traditional secondary Non-linear regression equations established the quadratic regression equations between weld penetration,the weld width of the joint surface and welding parameters,and regression analysis and variance analysis were performed on the equations to verify that the regression equation is highly significant.And the regression equation can be used to predict the size of the weld section and guide the choice of welding process parameters.Again,for the most common butt joints in the project,use the built coaxial monitoring platform to obtain the front image of the pool,and extract the pool length L and the pool width W,and use the value of L/W of the pool length to achieve welding.penetration monitoring.Under the premise of penetration,using the regression model to establish the relationship between the process parameters and the width of the weld at the 1/2 thickness and the width of the back of the weld,laid the foundation for precise control of the weld section size of the butt joint.Finally,taking the length of the molten pool,the width of the molten pool,the diameter of the keyhole and the angle of the tail of the molten pool as the input variables,the weld penetration and the weld width of joint surface width are the output variables.Taking into account the laser welding process.The high coupling between input variables and output variables,neural network prediction model was established using BP neural network.The global search feature of genetic algorithm is used to optimize the weights and thresholds of BP neural network.The experimental results show that the genetic neural network can effectively improve the prediction accuracy of penetration depth and weld width of joint surface.Then,the established genetic neural network was applied to the prediction of the width of the butt joint waist and the weld width of the weld seam.It was found that the neural network had good prediction accuracy and universality.