|About this Abstract
||2021 TMS Annual Meeting & Exhibition
||Additive Manufacturing for Energy Applications III
||Multi-scale Multi-fidelity Metamodeling for Advanced Materials
||Mohammad G. Abdo, Yu-Lin Shen, Cam Binh T. Pham, Isabella J. Von Rooyen
|On-Site Speaker (Planned)
||Mohammad G. Abdo
Machine learning has proven indispensable for predicting effective material properties, and other modeling aspects of heterogeneous materials such as composites, multi-layered structures, or functionally graded additively manufactured materials. This work proposes utilization of a multiscale technique that bridges the gap between different scales of modeling and experimental sparse measurements based on sensitivity studies. The goal is to construct a metamodel that can predict bulk properties by lumping the information learnt from the different scales as well as the experimental data into a single model. This model should no longer need to utilize either of these various scale simulators, yet still resolves a vast range of physical behaviors extracted from the training process and reveals the propagated uncertainties. This can guide nuclear fuel and reactor components design, help to understand the effect of porosity and geometric distribution of pores, interface modeling on the overall effective material properties such as thermal conductivity.