This work presents fast, deep learning-based algorithms developed under the Simurgh framework, designed to optimize XCT for scalable characterization in metal AM. Unlike existing approaches, our algorithms uniquely leverage CAD models of the parts, physics-based information, and both supervised and unsupervised deep learning networks, including generative adversarial networks (GANs). This combination enhances the quality of the reconstruction while significantly reducing scan time. Our algorithms have shown considerable improvements in NDE for metal AM, enabling faster scans, lower costs, and labor, while providing higher-quality results without imposing additional computational burdens. We present results demonstrating how these approaches enhance defect detection capabilities and reduce acquisition, reconstruction, and analysis time and complexity for components made of both low- and high-density metallic materials. The application of these algorithms allows for high-throughput characterization of hundreds of coupons in a matter of weeks, offering a comprehensive solution for faster, more accurate, and cost-effective industrial CT imaging.
The speaker is:
Dr. Amir Koushyar Ziabari, R&D Staff Scientist, Oak Ridge National Lab
Tags
- Design und Produktentwicklung