Study On Magneto-optic Imaging Feature Classification And Three-dimensional Reconstruction Recognition For Welding Defects

As an efficient and low-cost processing technology,weld plays an important role in the manufacturing industry.However,because the welding process is relatively bad and there are many uncontrollable factors,it is easy to produce welding defects,which directly affect the normal use of welding products.In order to prevent these welding defects from leading to uncertain safety accidents,it is necessary to conduct non-destructive Testing(NDT)on the welding seam.The commonly used NDT technologies include X-ray testing,ultrasonic testing,magnetic particle testing,eddy current testing and penetration testing.All the above testing methods have their own advantages and limitations.This article improves a new type of welding defects magneto-optic(MO)imaging NDT technology based on Faraday effect,which can realize the detection of tiny defects on the surface and subsurface of welding samples.Its advantage is that the welding defects information can be converted into three-channel image information,and it has high sensitivity and accuracy.In this paper,a new MO imaging detection method for welding defects was studied based on the distribution law of magnetic leakage field and Faraday effect.First of all,the formation mechanism of common welding defects was analyzed,and a large number of test samples were obtained through three welding processes,including resistance welding,laser welding and tungsten inert gas(TIG)welding.Next,the platform for welding defects MO imaging detection was construct,including the MO imaging control platform,MO imaging sensors,excitation mechanism,welding parts,MO imaging acquisition system and the fixture,and the detail of three excitation device for constant magnetic field,alternating magnetic field and rotating magnetic field were communicated.At the same time,the paper researched the welding defects for different features of MO imaging under different excitation source,and summarized their respective advantages and disadvantages and range of application,including constant excitation magneto-optic imaging effect is more stable,alternating field that can detect deeper defect information,the direction of the rotating magnetic field can inspire a weld defects different leakage magnetic field,which can obtain more comprehensive defect information.Then,the welding defects MO imaging convolution neural network recognition method was adopted,and the influence of structure parameters on the training results of different model was studied.When the size of the convolution kernel of the first layer is 7×7 and the Relu activation function is adopted,the prediction model can achieve the best effect.The detection accuracy of the MO imaging welding defects is 98.61%,and the average prediction accuracy is 93.36%,including pits,cracks,no penetration,no fusion and no defects.Moreover,the finite element magnetic field simulation theory was analyzed,and Maxwell software was used to establish 3D finite element magnetic field model of welding defects,and the distribution law of leakage magnetic field of welding defects was studied,which included establishing 3D simulation model,loading excitation and setting boundary conditions,meshing and setting parameters of solving conditions.The simulation test results show that the extreme point of magnetic induction intensity of leakage magnetic field corresponds to the two side walls of the welding defect,and the point with zero magnetic induction intensity can be regarded as the defect center.The farther away from the center point,the greater the magnetic field intensity should be,and the greater the gradient of field intensity change,the closer it is to the center point in the direction of Y axis.Finally,combined with the distribution law of magnetic induction intensity of welding defects leakage field and MO imaging principle,a contour reconstruction method of welding defects was designed.The 2D plane contour of welding defects was extracted by threshold segmentation,Gaussian filtering,expansion etching,and so on.The relationship between gradient/deviation and the defect depth information was established,and the 3D contour reconstruction of welding defects from MO imaging was finally completed.It is consistent with the actual contour information obtained by the laser confocal scanning microscope.