Automatic defects detection and classification of low carbon steel WAAM products using improved remanence/magneto-optical imaging and cost-sensitive convolutional neural network

Abstract

Wire arc additive metal manufacturing (WAAM) is one of the most revolutionary and popular manufacturing processes. However, the poor quality is an important factor restricting the development of this technology. In particular, it is difficult to detect the small defects on the surface and subsurface of the manufactured products. To cope with this issue, we propose a new method for automatic defects detection and classification of low carbon steel WAAM products using improved remanence/magneto-optical imaging and cost-sensitive convolutional neural network. The improved remanence/magneto-optical imaging is used to obtain clear magneto-optical images. A convolutional neural network model is then deployed to detect the defects in magneto-optical images. The proposed method is effective in automatic detection of the surface defects of low-carbon steel WAAM products.