In situ measurement of part geometries in layer images from laser beam melting processes

Laser beam melting (LBM) enables production of three-dimensional parts from metallic powder with very high geometrical complexity and very good mechanical properties. In LBM, a thin layer of metallic powder is deposited onto the build platform and melted by a laser according to the desired part geometry. Until today, the potential of LBM for critical applications such as medical devices and aerospace has not been exploited due to the lack of build stability and quality management. We present an image analysis method, which segments part contours in high-resolution images of LBM-produced layers. Based on the reference contour from 2D slices of the 3D part model and edge-detection results, a graph model is built and segmented using Graph Cuts (min-cut max-flow algorithm). Our method is evaluated on 124 part contours from 5 build jobs with different part geometries. Iterative GrabCut segmentation on nonlinearly smoothed images achieves the best results with a median Jaccard distance of 0.035 (32 % improvement over the reference geometry masks) and a mean contour distance below 2.4 px (36.4 % improvement).