Deep learning based porosity segmentation in X-ray CT measurements of polymer additive manufacturing parts
Simon Bellens, Patrick Vandewalle, Wim Dewulf
Abstract
X-ray computed tomography (XCT) is the only non-destructive technique able to perform a complete quality control of additive manufactured products in a single inspection. Yet, high-quality scans are associated with large acquisition times and limited to high-end AM parts. In this paper we investigate a deep learning U-net segmentation algorithm to improve the segmentation of low-quality XCT scans. A high-quality XCT scan is acquired and aligned with low-quality XCT scans to create the ground truth data. The accuracy of the segmentation is quantified with the Jaccard index and physical properties of the parts.
Topics & Concepts
Jaccard indexSegmentationGround truthPorosityQuality (philosophy)Artificial intelligenceMaterials scienceDeep learningComputer scienceComputer visionPattern recognition (psychology)Composite materialPhysicsQuantum mechanicsAdvanced X-ray and CT ImagingAdditive Manufacturing Materials and ProcessesMedical Imaging Techniques and Applications