Applications

Electrodynamics:

Extreme-scale optimal control for electronic camouflage

In conjunction with the Mirage electromagnetics module of the extreme-scale simulator MrHyDE, ROL computes an optimal control strategy to mask a source of electromagnetic waves.

Left: Uncontrolled electric field emitted by the Sandia thunderbird. Middle: Electric field is controlled through four spheres. Right: Control current in the four spheres.

Fluid dynamics:

Optimization enables new additive manufacturing applications

Using the Darcy module of its PDE-constrained optimization test suite PDE-OPT, ROL designs porous media in columns used by chemical engineers, to enforce uniform transit time of the fluid between an inlet and an outlet.

Left: Double-cone column. Middle: Optimal permeability profile. Right: The resulting streamlines, corresponding to the uniform transit time of fluid particles.

H. Antil, D. P. Kouri, D. Ridzal, D. B. Robinson, M. Salloum (2023) Uniform flow in axisymmetric devices through permeability optimization, submitted.

Climate science:

Computing basal friction to accurately predict flow of land ice

Coupled with the ice-sheet model MALI (MPAS Albany Land Ice), ROL is used to estimate the basal friction coefficients of Greenland and Antarctic ice sheets, based on ice velocity observations.

Left: Greenland, modeled ice speed in m/yr. Middle: Greenland, reconstructed basal friction in kPa yr/m. Right: Antarctica, reconstructed basal friction in kPa yr/m.

M. Perego (2022) Large-scale PDE-constrained Optimization for Ice Sheet Model Initialization, SIAM News.

Structural dynamics:

Stochastic optimization reduces risk of structural failure

Using the elasticity module of its PDE-constrained optimization test suite PDE-OPT, ROL computes risk-averse designs of structures subject to random loads. Here, we use the conditional value-at-risk (CVaR) to generate structures that are less likely to fail.

Optimization under uncertainty. (a) Mean-value design, with vertical loads and no uncertainty. (b) Risk-neutral design, due to random uniformly distributed load angles and magnitudes. (c) Risk-averse design, with a CVaR confidence level of 99%.

Image processing:

New algorithms for enhanced super-resolution on GPUs

Coupled with the GPU-accelerated image processing routines from ArrayFire, ROL is used for super-resolution imaging, to recover a high-resolution image from a sequence of low-resolution images resulting from lateral jitter.

Super-resolution imaging. (a) Sample low-resolution image. (b) Reconstruction using the Drizzle algorithm. (c) Derivative-based optimization reconstruction using ROL.

D.P. Kouri, E.A. Shields (2014) Efficient multiframe super-resolution for imagery with lateral shifts, Applied Optics 53(24), F1-F9.

Geophysics:

Exploring Earth’s subsurface through full-waveform seismic inversion

Using Sandia’s discontinuous Galerkin finite element simulator DGM, ROL is used for full-waveform seismic inversion.

Given the seismic response from the synthetic SEG 2D Salt Model, ROL is used to recover the P-wave velocity field describing the subsurface.