Turbulence evolves through highly nonlinear interaction of a broad spectrum of spatial and temporal scales. Researchers have developed computational techniques to resolve all, some, or none of these…

This is a $576,004.84 Cooperative Agreement awarded by NASA's Shared Services Center to the Regents of the University of Colorado, a public research university, to study turbulence modeling across different computational approaches. The work aims to systematically assess the successes and failures of various turbulence models, including Reynolds-Averaged Navier-Stokes (RANS), Large Eddy Simulation (LES), and Direct Numerical Simulation (DNS), by applying them to a series of coordinated flow problems. This effort is intended to improve both scale-resolving simulation (SRS) models and RANS models, which is important for advancing computational fluid dynamics (CFD) capabilities used in aerospace engineering and other applications. The award has no set-aside designation and is structured as a Cooperative Agreement, allowing for close collaboration between the university researchers and NASA technical experts. This contract builds on the university's prior federal research engagements with agencies like NASA, the Department of Defense, and NOAA in areas such as space weather, AI, and environmental science.

Turbulence evolves through highly nonlinear interaction of a broad spectrum of spatial and temporal scales. Researchers have developed computational techniques to resolve all, some, or none of these scales and have developed turbulence models to capture the unresolved effects on the resolved scales or the mean flow field. Resolving all (direct numerical simulation or DNS), while most accurate, can require the largest supercomputer for months for relatively simple flows but is the gold standard for scale-resolving simulations (SRS). Modeling all (reynolds-averaged navier-stokes or rans) can give results on complicated flows in hours but the prediction has low confidence for important flows like separation from smooth surfaces. Intermediate models (e.g., large eddy simulation (LES) or detached eddy simulation (DES)) resolve larger scales and provide intermediate accuracy and cost. For too long, researchers have pitched a particular model in this spectrum as the key to improved engineering prediction. This proposal seeks to apply simulations across the entire modeling/resolving spectrum to systematically identify modeling successes and failures. Rather than promote a single class of models, each lower fidelity model will be assessed with the next higher fidelity model under matched flow conditions. While the idea of using statistics, budgets, and a priori tests from higher fidelity models to identify failures and improvements to lower fidelity models is not new, the effort proposed herein builds strongly on this prior work together with dramatic advances in massively parallel CFD to apply this approach across an unprecedented range of reynolds numbers, SRS models, coordinated flow problems, and boundary conditions. Specifically, this work will study an extrusion of a bump profile that matches the centerline coefficient of pressure from a recently started joint computational/experimental study at boeing. By resolving the wall layer at DNS resolution over momentum thickness reynolds numbers ranging from the tunnel speed (2130), the full tunnel speed (4260), and then up to 25,700 (using DNS near the wall and LES away from IT), wall models for LES can be tested in a 12-fold rise in reynolds numbers in a combination of zero, favorable, and adverse pressure gradients that occur in rapid succession and result in separation. These conditions replicate the most challenging separated flows on high-lift wing configurations. Using these three databases, models will be assessed using a priori and a posteriori closure modeling tests, improved, and then validated on a series of increasingly complex models, including, at the end, application of the improved SRS models to high-lift wing configurations. The proposed effort is significant to NASA due to its strong potential to improve both SRS models and ultimately rans models. This is important because without improving rans modeling, hybrid models that hold the greatest promise for SRS methods will not achieve their full potential.