Visual Characteristic Regularity And Three-Dimensional Reconstruction Of Surface Morphology Of Weld Pool In Aluminum Alloy Tandem PMIG Welding

Intelligent sensing and control of welding process is an important research direction in recent years.Low melting point,large thermal conductivity and small viscosity of liquid molten pool lead to obvious thermal accumulation effect in the welding process of aluminum alloy.Thus welding collapse,burn-through,deformation,slag,porosity and unstable penetration are easily to occur in aluminum alloy welding process.Based on the research on welding process and weld forming experiment of aluminum alloy double wire PMIG welding,visual sensing of weld pool is implemented to make intelligent perception of welding process,and after image processing and visual feature extraction the two-dimensional visual characteristic rule of aluminum alloy weld pool in double wire pulsed PMIG welding were studied and the surface morphology of weld pool were reconstructed by binocular stereo vision method.Welding experiments of 5A06 aluminum alloy with thickness from 6mm to 60mm were carried out.After analysis of the influence of groove type,groove gap and the heat input on weld forming,the optimum welding parameters of aluminum alloy were obtained and the microstructure and mechanical property of the joints were tested.Two-directional synchronous visual sensing method with dual cameras was proposed in aluminum alloy double wire PMIG welding process and image acquisition software was developed based on Lab VIEW platform.The two cameras were simultaneously triggered by external trigger circuit,and the synchronous precision can reach 1?s.The special two-directional capturing fixture was designed with compact structure and convenient adjustment.Through arc spectrum analysis and exposure control,the infrared band filter with 980nm center wavelength was selected,using which the interference of the arc light can be effectively filtered.Clear weld pool images with deep field and abundant details in two direction were obtained when the camera exposure time was shorter than 2ms.Strong arc light,electromagnetic and mechanical interference,no obvious color distinction between liquid and solid state of aluminum alloy and other factors can cause the interference of the internal hole,edge gray discontinuity and horizontal grain in aluminum alloy weld pool images,which makes the edge extraction difficult.Two kinds of image contour extraction method were proposed for the images with internal hole and the edge gray discontinuity,one was based on spatial filtering,mathematical morphology,edge contour extraction of the gradient operator method,and the other one was edge detection method based on Snake active contour model.In view of the images existing horizontal stripes interference,frequency domain filtering method was proposed,and proper Butterworth low-pass filter was conducted,and the contour was extracted by using Canny edge gradient operator.The extraction results of weld pool profiles confirmed the validity of the three methods.The geometric feature parameters such as melt width,half length,area,perimeter,parabola coefficient and dispersion degree of weld pool are proposed,and the texture feature parameters of the mean of energy,entropy,contrast and inverse difference moment of weld pool were put forward,which formed the visual feature parameter system of aluminum alloy weld pool,and the visual feature parameter extraction algorithms of the rear and side aluminum alloy weld pool were designed.Based on the analysis of the visual characteristic parameters of the two-directional aluminum alloy weld pool images,the visual characteristic rule of the aluminum alloy weld pool in double wire PMIG welding was obtained,and the BP and RBF neural network penetration identification models were established,which realized the effective identification of penetration state in the single face welding with double face forming welding of aluminum alloy.Through network training and testing,the accuracy rate of BP and RBF neural network was 90%and 94%respectively.A three-dimensional reconstruction model for weld pool surface morphology was developed.The internal and external matrix of the two cameras were obtained through stereo calibration algorithm.The precision of the model is verified through the three-dimensional bevel plate of the known marker point.The error in the Yw direction is the maximum,the mean error is 0.16mm,and the standard deviation is 0.26mm.By using this model the three-dimensional static arc crater and weld bead surface were reconstructed,which verified the effectiveness of the model in aluminum alloy double wire PMIG welding system.And the disparity map of weld pool surface was obtained through the normalized cross-correlation matching algorithm,combined with internal and external matrix of the cameras the three dimensional weld pool surface morphology of aluminum alloy in double wire PMIG welding was finally reconstructed.