About this Abstract |
| Meeting |
MS&T26: Materials Science & Technology
|
| Symposium
|
Additive Manufacturing Modeling, Simulation, and Machine Learning: Microstructure, Mechanics, and Process
|
| Presentation Title |
Machine Learning-Based Defect Characterization from Fracture Surfaces in LPBF Ti-6Al-4V |
| Author(s) |
Johan Buecker, Brett E. Ley, Mingjian Lu, Justin Miner, Sneha Narra, Christian Gobert, Jack Beuth, Anthony Rollet, John J. Lewandowski |
| On-Site Speaker (Planned) |
Johan Buecker |
| Abstract Scope |
Ti-6Al-4V parts fabricated using laser powder bed fusion (LPBF) often contain process-induced defects that act as crack initiation sites and limit fatigue life. Defect type, size, and distribution depend strongly on print parameters. Common defect characterization methods, including micro-computed tomography (μCT) and metallography, are constrained by resolution and sampling limitations. In contrast, high-resolution SEM fractography enables direct identification of the initiating defect and can provide insight into defect populations, particularly in the fatigue overload region where cracks intersect defects at their maximum cross sections. In this study, four-point bend and axial fatigue specimens were fabricated using optimized and sub-optimal parameters and tested to failure. Full fracture surfaces were reconstructed using SEM image stitching, with complementary height data from profilometry. Machine learning-based segmentation was used to quantify defect populations, which were compared to μCT and metallography results. Additionally, fatigue crack-growth, overload, and shear lips were identified by this approach. |