Identification And Classification Of Magneto-Optical Imaging For Welding Defects

In order to ensure good welding quality,it is necessary to test the weld seam of the weldment.Non-destructive testing(NDT)technology completes the detection of surface or internal defects of the test piece without damaging the test piece.Aiming at the limitations of existing NDT methods in welding defect detection,a NDT for welding defects based on magneto-optical imaging(MOI)is proposed,the NDT system for MOI under rotating excitation and the identification and classification of welding defects are studied mainly.In this paper,the relationship between the applied magnetic field and the deflection direction of the polarized light and the deflection angle is explored by the Faraday magneto-optical effect,the principle of the leakage magnetic field and the working principle of the magneto-optical sensor firstly,and the angle between the rotation angle of the polarized light and the gray-scale distribution of the MO images is explained.Then the three-dimensional(3D)simulation model of welding defect is established by using the finite element method,the magnetic field distribution law above the welding defect is explored,and the magnetic field distribution under different lift heights of the welding defect is compared and analyzed to obtain the best lift-off to acquire MO images by using the MOI sensor.Afterwards the mechanism of alternating MOI is introduced in detail,and the alternating MOI experiment of medium carbon steel is carried out,the relationship between the alternating MOI and the excitation magnetic field distribution is studied.Finally,the principle of rotating magnetic field generation is introduced,and the mechanism of cross-yoke yoke to generate composite rotating magnetic field to excitation weldment is analyzed,by comparing the effect of rotating MOI and alternating MOI of low-carbon steel weldment,the applicability of the two MOI method in detecting welding defects is discussed,and the multi-directionality of the rotating MOI detection method is discussed.In order to realize the automatic identification and classification of MO images of welding defects,the classification model of welding defects was established by applying MO image sequences acquired under alternating excitation and rotating excitation.Aiming at the problem that the classification effect of existing welding defects classification methods is not very good,a better recognition method is proposed,the column pixel feature of alternating MO images after noise reduction and down-sampling is extracted by principal component analysis(PCA),is used as the input of the welding defect classification model,and the backpropagation neural network-AdaBoost(BP-AdaBoost)welding defects strong classification model is established by using the adaboost algorithm combined with the backpropagation(BP)neural network.At the same time,the welding defect classification model of BP neural network optimized by Levenberg-Marquardt(LM)algorithm is established,the results show that the BP-AdaBoost welding defects classification model can effectively improve the classification accuracy of welding defects.Finally,the MOI sensor is used to acquire the rotating MO images of the low-carbon steel weldment with defects,and a series of rotating MO image sequences are obtained to establish a classification model of the MO images of the welding defects.The gray-level co-occurrence matrix(GLCM)is used to extract the texture features of the reduced MO images,along with the gray-scale calculation feature of images is used as the input feature vector to construct a random forest(RF)classification model of the welding defects MO images.The classification results prove that the proposed method can effectively identify cracks,unmelted and sag in medium carbon steel weldments,the automatic identification and classification of welding defects MO images can be realized.