In Situ Three-Dimension Monitoring of Laser Powder Bed Fusion Melt Pool and Keyhole by Binocular Imaging

Abstract

In laser powder bed fusion (LPBF) additive manufacturing, unstable melt pool and keyhole can result in defects such as pores, lack of fusion, and cracks. In three-dimension (3D) monitoring of melt pool and keyhole is essential for preventing process deviations and optimizing part quality. This study proposed a novel binocular imaging system for in situ 3D monitoring of melt pool and keyhole. A coaxial binocular imaging optical path is designed to capture dual-view melt pools and an unsupervised adaptive weighted-loss residual U-net (Res-Unet) is adopted to achieve accurate disparity extraction. The performance of the network is validated, demonstrating subpixel accuracy using the HCI light field dataset. The binocular imaging system’s spatial resolution is validated at $6.2~\mu $ m using a standard resolution board, while its surface 3D reconstruction accuracy is confirmed to be $10.6~\mu $ m through a standard gauge block. The effectiveness of the binocular imaging system for in situ monitoring of melt pool keyhole depth is validated through both experiments and simulations, which reveals dynamic variation in keyhole depth. This work represents the first integration of optical imaging and artificial intelligence (AI) for coaxial in situ monitoring of 3D morphology of both LPBF melt pool and keyhole. It provides valuable tool for monitoring the evolution of keyhole depth, serving as a critical reference for enhancing the reliability and consistency of additive manufacturing processes.