About this Abstract |
Meeting |
MS&T21: Materials Science & Technology
|
Symposium
|
Additive Manufacturing: Advanced Characterization for Industrial Applications
|
Presentation Title |
Domain Adaption for Enhanced X-ray CT Reconstruction of Metal Additively Manufactured Parts |
Author(s) |
Amir K. Ziabari, Abhishek Dubey, Singanallur Venkatakrishnan, Michael Sprayberry, Curtis Frederick, Paul Brackman, Philip Bingham, Ryan Dehoff, Vincent Paquit |
On-Site Speaker (Planned) |
Amir K. Ziabari |
Abstract Scope |
Nondestructive evaluation (NDE) of additively manufactured (AM) parts is important for understanding the impacts of various process parameters and qualifying the built part. X-ray computed tomography (XCT) has played a critical role in rapid NDE of AM parts.
However, standard reconstruction algorithms may face challenges when processing XCT of metal AM parts due to phenomenon such as noise, streaks/metal artifacts and beam hardening.
In this work, we present our deep learning-based CT reconstruction method, which we call SIMURGH, that leverages computer-aided design (CAD) models of the AM parts along with accurate XCT simulations, CycleGAN domain adaptation and a residual CNN to remove metal artifacts and beam hardening from the reconstructed 3D volumes. Promising results both on synthetic and real data sets are shown, demonstrating significant improvement in defect detection capability in XCT of AM parts, which is confirmed by multi-scale high resolution XCT reconstructions as the ground truth. |