A high-precision in-situ monitoring system for laser directed energy deposition melt pool 3D morphology based on deep learning

Laser Directed Energy Deposition (L-DED) technology has been widely used in the production of parts with complex geometries and the repair of high-value parts. And the 3D dimensions of the L-DED melt pool have an important impact on the quality of the formed parts. However, no researcher has yet achieved 3D morphology monitoring of the L-DED melt pool. Therefore, this paper proposes a high-precision in-situ monitoring system for L-DED melt pool 3D morphology, which can achieve high-precision 3D imaging of the melt pool. Firstly, the system employed a visible light camera with a filter and a biprism for in-situ monitoring of the L-DED melt pool. Then, the proposed 3D imaging algorithm based on Superpoint + LightGlue was utilized to achieve 3D imaging of the L-DED melt pool. Superpoint and LightGlue were responsible for feature point extraction and matching, respectively. Finally, the experimental results showed that the mean absolute percentage error (MAPE) of the proposed system for imaging the length, width and height of the static L-DED track was 1.17%, 1.87% and 1.11%, respectively. The MAPE of imaging on the width and height of the dynamic melt pool was 2.26% and 2.48%, respectively, and the 3D imaging results on the length of the melt pool were consistent with the actual physical laws. This indicates that the proposed 3D morphology in-situ monitoring system for L-DED melt pool is not only of low complexity but also of high imaging accuracy, and it is expected to promote the development of LDED in-situ monitoring technology.