Abstract:It is difficult to detect and identify small defects on the surface and subsurface of wire arc additive manufacturing (WAAM) formed parts. To solve this problem, the texture feature of images and neural networks are both utilized. A nondestructive detection method based on magnetooptical imaging is proposed to detect surface defects of low carbon steel WAAM formed parts detection and classification. Firstly, WAAM formed parts are magnetized after processing by surface finishing. Magnetooptical images of the surface of formed parts are obtained by the magnetooptical imager as test samples. Then, the texture feature of angular second moment, entropy, contrast and correlation of magnetooptical images are extracted by the graylevel cooccurrence matrix after preprocessing the images and texture feature data of four different surface qualities. To be specific, perfectness, poor fusion, depression and cracks are used to carry out comparison. Finally, the classification of formed parts is predicted by LevebergMarquard （LMBP） neural network. Experimental prediction results show that the surface defect detection rate of low carbon steel WAAM formed parts is 9733% and the classification accuracy rate of the surface quality can reach 9133%. These results verify that the proposed method can effectively detect and identify small surface defects on surface of low carbon steel WAAM formed parts.