Optical imaging framework for melt pool monitoring and bead geometry prediction in blown powder directed energy deposition

Abstract

Abstract Blown powder laser directed energy deposition (DED-LP) is an emerging additive manufacturing process with applications in both new part manufacture and repair of high-value components. However, the deposition of consistent bead geometry requires precise control of melt pool dynamics. While in-situ monitoring methods have been developed, linking melt pool optical signals to solidified geometry remains an open challenge. Monocular imaging offers a low-cost, minimally invasive sensing approach, yet is inherently indirect due to projection effects and process-induced optical variability. This work presents a monitoring framework to interpret melt pool images by analytical and data-driven means. Contours detected from intensity-based image segmentation are translated to world coordinates using affine projection to provide calibration-based geometric proxies for comparison with solidified bead geometry. At nominal printing conditions, the analytical approach yielded mean absolute percentage errors of 2.5 % for bead width and 16.2 % for bead height. The affine projection method captured trends in solidified bead geometry across the traverse speed and laser power ranges investigated within the material-specific operating envelope. To refine estimates, a Gaussian process regression model (GPR) was trained on image-derived features and process conditions. GPR achieved a mean absolute percentage error of 1.4 % for width and 3.7 % for height across the tested traverse speed conditions. This study demonstrates the applicability of analytical interpretations of monocular melt pool images for analysis within the standard operating window of 316L stainless steel feedstock and demonstrates how GPR may be leveraged to learn underlying relationships not captured by analytical intensity-based image segmentation.

The authors' abstract, as published at the source. The International Journal of Advanced Manufacturing Technology, 2026 · DOI ↗