Research On Surface Defect Detection And Classification Of Arc Additive Components Based On Magneto-optical Imaging

Products manufactured by wire arc additive manufacturing(WAAM)technology have excellent mechanical properties and have a promising prospect in the filed of mechanical manufacturing.Nondestructive testing technology can effectively guarantee the usability of workpieces and the safety of industrial application.Nowadays,the detection technology for surface quality of WAAM products is immature.In particular,it is difficult to detect the small defects on the surface of low carbon steel WAAM formed parts.To cope with this issue,the magneto-optical imaging(MOI)technology,combined with neural network models,is proposed to realize the detection and classification of the surface defects of WAAM products.According to WAAM process and MOI technology,the process of using MOI to realize the surface quality inspection of low carbon steel WAAM products is designed,and the test system including the experimental system of WAAM products,pre-magnetization device and MOI detection system are established.A batch of workpieces with various surface qualities is produced by the experimental system of WAAM products.The pre-magnetization system is built and the yoke structure is designed,and the output of magnetic field is measured.The MOI detection system is built and the conversion relationship between light intensity and grayscale is introduced,then MO images are collected.The gray scale characteristics of MO images are summarized according to comparative analyze of the MO images with no defects and defects.Besides,the images effects are compared and analyzed.To further improve the effect of MO images,an improved imaging method based on light intensity thresholds is proposed.According to the effect of image enhancement and background noise by different light intensity thresholds,an algorithm to calculate the optimal thresholds based on the light intensity data of testpiece is designed.By acquired by improved imaging method based on the optimal threshold,the background noise is little and the defect information is significantly enhanced.Moreover,the improved imaging method is used to detect the specimens which the situation may occur in the actual inspection,which the effectiveness of this improved imaging method is verified.In addition,the enhancement effect of the optimal thresholds imaging method is verified by compared with the original MO images of four typical surface qualities and the images enhanced by image enhancement methods.Finally,the eigenvalues of MO images are extracted by gray level co-occurrence matrix.The BP neural network and support vector machine model is constructed to realize the classification of the surface quality of WAAM products.At the same time,the convolutional neural network model is established,which can realize the automatic recognition by directly inputting the MO images into the model.The prediction results of three models are counted by confusion matrix,and the corresponding classification indicators are calculated and analyzed.The prediction results prove that the proposed method can effectively realize the automatic defects detection and classification on the surface of WAAM products.