Efficient image segmentation for in-situ lattice structure inspection in Additive Manufacturing: a comparison study

As additive manufacturing (AM) advances in high-value production, efficient methods for process and part qualification are increasingly critical. The geometric complexity enabled by AM necessitates costly inspection techniques that may be impractical for large-scale applications. However, the layerwise nature of AM offers the potential for in-line, in-process qualification using in-situ sensors, enabling early defect detection. This advantage is particularly significant for lattice structures. In laser powder bed fusion (LPBF) images of the solidified layer can be gathered in-situ, during the production of the structure. Image segmentation allows reconstructing the geometry of the printed part on a layer-by-layer basis, and estimate the deviation from the nominal shape, enabling an anticipated detection of distortions and anomalies compared with the current industrial practice. This study explores the potentials of artificial intelligence (AI)-based segmentation and compares it against competitor machine vision techniques. A real case study in metal LPBF of lattice structures is used as a benchmark. Methods are compared in terms of their accuracy and computational efficiency. Guidelines for the selection of most effective methods based on the natural characteristics of the input image data are also discussed.