| About this Abstract |
| Meeting |
2023 TMS Annual Meeting & Exhibition
|
| Symposium
|
Simulations/Experiments Integration for Next Generation Hypersonic Materials
|
| Presentation Title |
How Do You Integrate Both Simulations and Experiments into a Materials Discovery Optimization Campaign? A Case Study in Multi-fidelity Optimization |
| Author(s) |
Ramsey Issa, Sterling G. Baird, Taylor D. Sparks |
| On-Site Speaker (Planned) |
Sterling G. Baird |
| Abstract Scope |
Fidelity is the "degree to which the simulator replicates reality" (Alessi 2000). In materials science, simulations and experiments can have multiple fidelities: for example, increasing the mesh resolution of a finite element analysis (FEA) simulation (continuous fidelity), using low-temperature tests as a proxy for high-temperature ones (continuous fidelity), or using hardness measurements as a proxy for expensive ASTM tensile tests (discrete fidelity). Here, we focus on the case of continuous multi-fidelity optimization. We compare adaptive design cost of a ten-variable physics-based particle packing simulation for low-fidelity, high-fidelity, and multi-fidelity (knowledge gradient acquisition) optimization with the number of dropped particles as the fidelity parameter. When designing composites, ceramics, or multi-principal-element alloys (MPEAs), the design spaces are often high-dimensional and subject to similar constraints. State-of-the-art multi-fidelity optimization algorithms that synergistically integrate simulations with experiments in an adaptive design setup have the potential to dramatically accelerate the design of next-generation extreme-condition materials. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
High-Temperature Materials, Modeling and Simulation, Machine Learning |