About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
Optimizing Bead Geometry during Laser Wire-Powder Directed Energy Deposition using a Multi-objective Machine Learning Approach |
| Author(s) |
Stanford White, Seongun Yang, Nahal Ghanadi, Somayeh Pasebani, Samrat Choudhury |
| On-Site Speaker (Planned) |
Samrat Choudhury |
| Abstract Scope |
This study presents a data-driven framework for modeling and optimizing directed energy deposition with simultaneous wire and powder feedstocks. Bead geometry is known to directly governs component integrity and dimensional accuracy. However, navigating the vast high-dimensional processing parameter space for a targeted bead geometry can be time consuming. We developed a multi-objective adaptive design approach using Gaussian process regression and Bayesian optimization to simultaneously target a 20 to 30 percent dilution ratio and an aspect ratio less than 3 of the bead. Benchmarked against traditional factorial design, our model demonstrated rapid convergence over 180 experiments, yielding a twofold increase in discovering optimal processing conditions. Feature importance analysis revealed that laser power strongly dictates penetration depth, while powder feed rate controls bead height. Ultimately, this integrated methodology enables enhanced process control and accelerates closed-loop feedback development, unlocking new efficiencies for aerospace repair, large-scale fabrication, and custom tooling. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Additive Manufacturing, |