ROM-Tools at Scale: Example Applications#

This feature page highlights two large-scale studies from the Pressio ecosystem showcasing ROM construction and deployment in advanced workflows.

SPARC: Turbulent Flow UQ and Sensitivity#

Pressio is coupled to SPARC for uncertainty quantification of a turbulent flow over a cone-slice-wedge geometry with a 3D RANS solver and SA turbulence model. The workflow targets high-fidelity UQ and sensitivity analysis while remaining computationally practical.

Cone-slice-wedge geometry and flow field

Cone-slice-wedge geometry with wall heat flux and a flow-field slice.#

Key results:

  • 48.27M finite volumes, 289.6M DOFs; FOM runs converge in roughly 2.0 hours on 12 nodes.

  • Massive sample-level parallelism via asynchronous workflows, scaling UQ studies across many parameter evaluations.

  • ROMs deliver sub-0.5% QoI errors, with the largest ROM below 0.1%.

  • Speedups range from about 30x (coarse ROM) to 15x (fine ROM).

  • Sobol sensitivity studies drop the estimated cost from ~110,000 to ~5,000 core-days.

FOM wall heat flux sample

FOM wall heat flux, sample case.#

ROM wall heat flux sample

ROM wall heat flux, same sample case.#

Aria: Thermal Protection System Inference#

Pressio powers a multi-fidelity ensemble Kalman inversion workflow to infer temperature-dependent material properties for a thermal protection system using SIERRA/Aria.

Paraboloid nose geometry rendering

Paraboloid nose geometry rendering.#

Paraboloid nose cross-section

Paraboloid nose cross-section.#

Key results:

  • 475,588 DOFs with FOM runs around 630 seconds; ROM runs around 60 seconds.

  • High-throughput multi-fidelity EKI leverages large, parallel batches of FOM and ROM evaluations per iteration.

  • Multi-fidelity EKI reaches ~0.03% parameter error versus ~7% for FOM-only.

  • MF EKI achieves comparable accuracy at roughly an order of magnitude less cost.

ROM sample mesh for TPS inference

ROM sample mesh focused near the nose.#

Cross-sectional temperature field

Cross-sectional temperature field.#

Figures are sourced from the Pressio SOM paper examples.