Sandia National Labs FY20 LDRD Annual Report


Physics-informed deep learning for fast and accurate multiscale simulations in computational solid mechanics. Sandia developed a deep learning framework to accurately predict the localized elastic microstructural responses at the microscale. This framework could replace crystal plasticity finite element models. In addition, the team developed an unsupervised deep learning framework for learning latent representations of microstructures, thereby reducing the dimensionality of the microstructure representations. Using these approaches may allow the solution of direct numerical simulation problems for macroscale objects at a small fraction of the computational cost of traditional simulations. Therefore, the methods

developed in this research can be used to lower the computational cost of high-fidelity simulation in

computational materials science, and to assess and improve the materials reliability in extreme and critical conditions. (PI: Anh Tran)

The machine learning framework can accurately predict strain with only a fraction of the computation burden of traditional methods.

Hafnium-boron-based precursors for vapor deposition of ultra-high temperature materials. High-temperature protection applications require refractory hafnium diboride ceramic coatings. This project employed density functional theory calculations to identify candidate hafnium- based precursors for chemical vapor deposition processing, synthesized the new precursors, and experimentally demonstrated the deposition of hafnium using the new precursors. Demonstrating the preparation of hafnium diboride precursors helps advance U.S. initiatives in the manufacture of thermal management and protection capabilities. (PI: Bernadette Hernandez-Sanchez)

(Left) Simulation of [Hf(BH 4 ) 2 L 2 ] precursors for metal-organic chemical vapor deposition (MOCVD) applications; (Middle) Precursors explored and synthesized, and (Right) MOVCD setup used to produce films.



Made with FlippingBook - professional solution for displaying marketing and sales documents online