| About this Abstract |
| Meeting |
2023 TMS Annual Meeting & Exhibition
|
| Symposium
|
Refractory Metals 2023
|
| Presentation Title |
Data-augmented Property Modeling for Accelerated Closed-loop Multi-Objective Design of Refractory High Entropy Alloys for ULTIMATE |
| Author(s) |
Brent G. Vela, Danial Khatamsaz, William Trehern, Cafer Acemi, Prashant Singh, Douglas Alliare, Raymundo Arroyave, Ibrahim Karaman, Duane Johnson |
| On-Site Speaker (Planned) |
Brent G. Vela |
| Abstract Scope |
Refractory multi-principal element alloys (MPEAs) are thought to be likely candidate materials to enable next-generation gas turbine engines. These refractory MPEAs must be ductile at room temperature such they are formable and simultaneously retain their yield strength (> 200 MPa) at 1300C. However, the ‘strength-ductility trade-off’ makes the design of such an alloy difficult. Bayesian optimization frameworks enable the efficient exploration of such high-dimensional design spaces and identification of Pareto-optimal alloys. Importantly, Bayesian optimization can incorporate prior knowledge into the search scheme. Despite this, often high-fidelity prior knowledge of these systems is sparse, yet reduced order models (ROMs) exist that can inject the optimization scheme with useful information concerning objectives. In this work we use ROMs as informative priors, update the priors with high-fidelity data, creating machine learning augmented ROMs, and use said augmented ROMs within a multiple objective batch Bayesian optimization scheme to design high strength ductile MPEAs. |
| Proceedings Inclusion? |
Undecided |
| Keywords |
ICME, Modeling and Simulation, Computational Materials Science & Engineering |