Sandia Labs FY21 LDRD Annual Report

FY21 ANNUAL REPORT

Project Highlights – Mission Agility Sandia’s LDRD program is organized around three themes: mission agility, technical vitality, and workforce development. Mission agility and technical vitality are closely related but differentiated by the technical readiness levels (TRL) of the research outcomes. The research outcomes in the accomplishments below have a higher TRL and could impact Sandia’s mission work more quickly.

Unless otherwise noted, these highlights are for projects that ended in FY21.

Reducing the costs associated with high- fidelity aerothermal simulations of hypersonic vehicles. Thermal protection system designers rely heavily on high-fidelity computational simulation tools for the design and analysis of high-speed aerospace engineering applications due to the expense and difficulty of flight tests and experiments. Because necessary fine- grid resolution makes the computational fluid dynamics models expensive, analysts primarily use low-fidelity or surrogate models when simulations at many different flight conditions or designs are required. In this Autonomy for Hypersonics LDRD Mission Campaign project, researchers explored an alternative approach where projection-based reduced-order models

A grid-tailored ROM with three input parameters for 3D flight vehicle geometries accurately computed the flow field and also improved axial force and wall heat flux accuracy over a data- driven surrogate model.

(ROM) were used to approximate the computationally infeasible high-fidelity model. By applying cutting- edge ROM techniques, specifically the Petrov-Galerkin ROM equipped with hyper-reduction, the team demonstrated the ability to significantly reduce simulation costs while retaining high levels of accuracy in computed quantities of interest on a range of aerothermal problems related to national security. (PI: Pat Blonigan) Improving accuracy of neural network algorithms. Decreasing the risk for tailored hardware solutions in hypersonic flight system applications is essential. By using conventional and low-size/weight/power (SWaP) analog neural accelerator hardware, this LDRD project research team significantly improved accuracy of neural network algorithms. By comparing the energy and delay of analog-to-digital conversion (ADC) schemes, the team proposed a more energy- efficient ADC, enabling them to develop a detailed accelerator architecture model. By modeling the feasibility of implementing modern neural algorithms in current and future onboard systems where SWaP is constrained, system architects will be able to evaluate more hardware-conscious decisions. (PI: Matt Marinella)

22

LABORATORY DIRECTED RESEARCH & DEVELOPMENT