sub header2

[View PDF]

 

Kenneth Jansen, U. Colorado
Michel Rasquin, Argonne
Berk Geveci, Kitware, Inc.

Objectives

  • Simulate and analyze synthetic jet flow control using the finite element method
  • Use mesh adaptivity to fully resolve all required physical scales
  • Analyze and visualize the resulting 1 billion element mesh per time step solution to study the effect of flow control
Geveci.Kitware.Vis.jet.flow-fig1

Impact

  • Synthetic jet flow control can dramatically alter aerodynamic flow with very small power input.
  • Advancing our understanding of the flow mechanisms involved is essential to developing novel flow control techniques

Accomplishments - FY13

  • Computations confirm experimental finding that flow control can produce 20%  improvement/increase in side force
  • Demonstrated that unsteady, separated flow with active flow control is not accurately simulated with time-averaged models
  • Using ParaView on ALCF Tukey enabled researchers to remotely analyze and visualize their data, an essential part of the discovery process at scale

Notes:

Researchers at Rensselaer Polytechnic Institute have observed that synthetic jets (1mmx20mm slits on top of acoustically resonating cavities) can raise the rudder side force of a vertical tail assembly by 20%. The physics behind this dramatic improvement are poorly understood.

A coordinated experimental and computational study has been carried out between UCB and RPI to understand the physical process on a real tail-rudder geometry.

At this high Reynolds number, the turbulent flow produces a broad range of length scales that must be resolved to properly predict the behavior as typical time averaged models (Reynolds Averaged Navier-Stokes (RANS)) have been shown to be inadequate.

In this project, the UCB team applied an adaptive, unstructured grid flow solver to resolve enough of the temporal and spatial scales to accurately match the experimentally observed flow control improvement. The left figure shows the real geometry, the CAD and an instantaneous isosurface of the vorticity Q criterion.

The key impact of this work was to adaptively improve the resolution to a level that this 20% improvement/increase in side force was captured by the computations.  The second figure illustrates that while it is possible to predict the baseline (no flow control) case with a carefully designed initial mesh (orange solid simulation curve whose average is the same color but dashed compared to dashed experiment), the same mesh captures very little of the expected improvement (dark blue solid and dashed) when the jets are turned on.
 
Uniformly refining this mesh makes a mesh that is 8 times bigger but the improvement to the predicted force (magenta)  is modest given the additional cost.

Local adaptivity, based on the simulation’s own measurement of the error, refines the mesh in only where needed and the force from the first adapted simulation is plotted in light purple (Adapt1) where the agreement with the experimentally observed force is apparent.  This mesh uses about 1/3 of the number of elements used in the uniformly refined case. Finally, the yellow curve shows a second adaptation (Adapt2) which resolves even more scales yet produces the same side force improvement giving more confidence to the grid independence (verification) of this validated prediction while still using less mesh elements than the uniform refinement case (1.2B vs 1.4B).  Note that the baseline is also verified to be grid independent with the adapted mesh (Adapt1 mesh is purple curve on Baseline).

These simulations generate over a terabyte of data per time step, which is impossible to transfer to a local computer at U. Colorado for analysis. The researchers used ParaView to remotely visualize their data on ALCF’s Tukey. This significantly reduced the data analysis barrier enabling the discovery process at scale.