Abstract
X-ray computed tomography (CT) is a recognized nondestructive testing solution for additively manufactured (AM) components. Each time a new part is CT-scanned, questions remain regarding the expected probability of detection (POD) of AM-induced defects. To investigate the impact of CT parameters on the defect POD, this study introduces a digital twin (DT) workflow that replicates the complete CT analysis pipeline, from image acquisition to defect segmentation. This workflow combines a CAD model with randomly sampled and scaled defects from a custom database of over 67 k experimentally observed AM void and lack-of-fusion defects (30 µm to 3.1 mm in diameter). Realistic CT volumes with ground-truth annotations are generated using physics-based CT simulation of the defect-augmented CAD. These synthetic data enable training a customized U-Net deep learning (DL) segmentation model. Specimen- and CT parameter-specific POD estimations are carried out by testing the DL model on the DT-generated volumes. Experimental validation using four Ti64 AM specimens containing process-induced defects, scanned under two different sets of CT parameters, demonstrates accurate reproduction with image quality metrics showing a maximum relative error of 5% between experimental and simulated volumes. Segmentation results confirm the robustness of the proposed DT workflow, with the largest undetected pores falling within DT-predicted POD confidence intervals. This experimentally validated end-to-end integration of CT simulation, DL segmentation, and POD estimation algorithms within a single workflow enables virtual inspection and quantitative defect detectability assessment in complex AM parts under varying CT conditions, thus opening the way for implementation of the design for inspection paradigm.
| Original language | English |
|---|---|
| Journal | Journal of Intelligent Manufacturing |
| DOIs | |
| Publication status | In press - 2026 |
!!!Keywords
- Additive manufacturing
- Deep learning
- Defect segmentation
- Digital twin
- Simulation
- X-ray computed tomography
Fingerprint
Dive into the research topics of 'Reliability of CT digital twins for defect segmentation and probability of detection estimation in AM parts'. These topics are generated from the title and abstract of the publication. Together, they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver