About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Atomization & Powder Metallurgy for Advanced Applications: An MPMD Symposium Honoring Iver Anderson
|
| Presentation Title |
Cost-Aware Bayesian Optimization Across a CALPHAD–Atomization–LPBF Function Network for Critical Mineral-Lean Aluminum Alloy Discovery |
| Author(s) |
Sterling G. Baird, Sam Charles, Gage Erickson, Ronnie Guymon, Adam Hope, Philip Lampkin, Seth Leavitt, Paul Mason, Will Mulberry, Devora Najjar, Alik Nielsen, Carl Robison, Ashley Spear, Luke Winters, Xavier Zaitzeff |
| On-Site Speaker (Planned) |
Sterling G. Baird |
| Abstract Scope |
Autonomous alloy discovery for additive manufacturing is a multi-stage decision problem with per-query costs spanning more than an order of magnitude: conservatively, ~$1500 per ultrasonic atomization run, ~$1–5k per LPBF build, ~$200 per ASTM E8 tensile bar, and ~$25k per gas atomization batch. Tiered screening (e.g., experimentally validating top-performing simulations) wastes budget on uninformative expensive queries and forfeits the algorithm's ability to strategically sample and learn cross-measurement biases early. We apply the partial Knowledge Gradient for Function Networks (pKGFN, Buathong et al. 2024) to a directed acyclic graph of CALPHAD (TC-Python/TCAl10/AM Module), atomization, LPBF, and microstructure–hardness–tensile nodes that feed information downstream. Ultimate tensile strength, yield strength, elongation, and a printability score are jointly maximized via expected hypervolume improvement while penalizing a USGS-derived critical-mineral burden. We present both synthetic-network benchmarks showing high sample efficiency and an optimization campaign spanning several dozen experimental atomization/print/characterize cycles. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Additive Manufacturing, Aluminum, Powder Materials |