Laser Dissimilar Material Quality Assessment by Deep Learning

T. Kim, C. Han, H. Choi
Keimyung University,

Keywords: AI, laser welding, weld quality monitoring, deep learning


In the era of the 4th Industrial Revolution, artificial intelligence(AI) based deep learning(DL) has been applied to many industrial applications. Recently, smart factories have applied AI or DL technology to increase manufacturing productivity and quality. Deep learning is applied to a variety of areas, such as self-driving cars, image recognition, video diagnostics, translation, statistics, and gaming, and is expected to bring dramatic changes to overall production technology. This paper presents a method for evaluating weld quality using deep learning algorithms from bead shape images in MIG welding and laser welding. A series of images from a CCD camera are collected and trained by developed deep learning and parallel computing technology. MIG welding quality monitoring and laser heterogeneous polymer joining were used to demonstrate the robustness of the DL-based quality assessment developed. For learning, we acquired image data of the bead shape of MIG welding with a camera, and divided it into a continuous weldment and a welding start point using MATLAB's deep learning and parallel calculation toolbox. In addition to MIG welding, laser beam was used for polymer joining of dissimilar materials. The laser beam penetrated the transparent polymer(Poly-carbonate) and was absorbed on top of the opaque polymer(acetate). The absorbed laser beam is converted as a heat source and diffused into both materials for bonding. The in-line monitoring system installed in the CNC system is used as an image acquisition device for deep learning training.