OVERFLOW and the OVERSET Tools for CFD Analysis

1
Engineering Advanced CFD Simulations
OVERFLOW and the OVERSET Tools for CFD
Analysis The OVERSET Tools for CFD Analysis is a
set of codes developed at NASA Ames for
performing CFD flow simulations. OVERSET
customers include NASA research centers, major
aerospace companies, various DoD research
centers, and NASA focused programs for Advanced
Subsonic Technology (AST) and High Speed Research
(HSR). NASA recognized OVERSET Tools and its flow
solver, OVERFLOW, with an honorable mention in
the 1998 Software of the Year competition.
OVERSET tools have been optimized to run
efficiently on several platforms and development
is ongoing. However, this project will be the
first integrated restructuring of the application
code, system hardware, and system software to
optimize performance. The flow solver OVERFLOW
consists of 120,000 lines of Fortran, 1,500 lines
of C, 1,200 subroutines, and 1,100 files.
OVERFLOW is designed to solve simulations on
Chimera-style overlapping grids, permitting the
creation of complex and extensive geometries. It
was written in vector-oriented style and can
achieve over 400 MFLOPS on a C90 vector
processor. However, on other processors like the
IBM SP2, Cray T3E, and SGI Origin 2000, OVERFLOW
only achieves 20 to 90 MFLOPS per processor.
Preliminary results using a single Athlon
processor have achieved twice the MFLOP rate of
these supercomputers, and the inclusion of 3DNow!
macros may allow performance comparable to the
C90. Combined with the highly scalable FNN design
and AFN technology, a cluster of PCs should scale
up the single-processor performance to achieve
optimal parallel-computer speeds.
The OVERSET Tools for CFD Analysis have been used
many NASA projects, including Reusable Launch
Vehicles. Shown here is a simulation of the
shuttle and booster rockets, with color
representing Mach number.
LESTool LESTool is a parallel CFD code under
development at the University of Kentucky for
simulating challenging, three-dimensional,
time-dependent flows such a turbine blade/flow
interaction, high-speed compressible flows, and
other complex engineering simulations. LESTool
combines the most advanced numerical and
computational fluid dynamics techniques in a
single, flexible package that solves a wide
variety of aerospace and industrial design
problems. Towards this end, LESTool employs
multiple, high-order spatial and temporal
numerical schemes and turbulence models ranging
from Reynolds-Averaged Navier-Stokes (RANS) to
direct numerical simulation (DNS). LESTool is
highly portable, based on standard FORTRAN90, C,
OpenMP, and MPI, and exhibits good scaling
characteristics with increased number of
processors. LESTool simulations can use Chimera
overset grids to solve simulations of complex
geometries. LESTool has been optimized for SGI
multi-processor platforms, and it has been run on
NASA supercomputer facilities to solve
turbine-related flows in support of the
Low-Pressure Turbine Physics program. A precursor
to LESTool, DNSTool, has been ported onto KLAT2
with excellent results and the knowledge gained
in this process will be applied to the
development of the LESTool PeTS.
Chimera overset grids allow for the simulation of
geometrically complex flows through the use of
multiple, overlapping grids. Both OVERFLOW and
LESTool can use this style of grid, as
illustrated by the Space Shuttle Launch Vehicle
grid (left) and the turbine cascade grid (right).
Advanced Turbulence Models (DNS, LES, DES) With
researchers and designers employing computational
fluid dynamics to solve increasingly difficult
flows, the traditional engineering turbulence
model, RANS (Reynolds-Averaged Navier-Stokes) is
proving increasingly inadequate. In combustion,
acoustic-structural interaction,
laminar-turbulent transition, and other complex,
compressible, and unsteady flows, other
turbulence models are necessary to achieve
accurate results. Three possible alternatives are
direct numerical simulation (DNS), large eddy
simulation (LES), and hybrid models like detached
eddy simulation (DES). In DNS, there effectively
is no turbulent modelall important scales of
motion are assumed resolved and solved by the
Navier-Stokes equations. LES takes a step back
and assumes that the largest turbulent eddy
scales are resolved, the effects of smaller-scale
motions captured by a sub-grid-scale (SGS) model.
Hybrid methods like DES blend RANS models with
LES/DNS models, using RANS in areas where it is
sufficiently accurate (or where LES/DNS
techniques are impracticable) and switching
smoothly to LES or DNS elsewhere in the
flow. The barriers to widespread implementation
of these advanced turbulence models are their
need for higher grid densities (and
correspondingly higher computational costs) than
RANS for accurate results and the lack of an
experiential record to inspire the necessary
confidence for use in real design work. The first
problem is gradually being eliminated by the
steady growth of computer powerreal engineering
DNS work is probably several decades away, but in
the coming decade both LES and DES should become
viable options for engineering simulations.
Addressing the second problem is a major focus of
UK CFDusing the flexibility of LESTool, we have
begun to systematically evaluate the strengths
and limitations of DES techniques through
simulation comparisons with LES, DNS, and
experimental results. However, this process is
computationally expensive, so the development of
CFD cluster systems will greatly aid this
research while simultaneously testing the
potential of these clusters.
The LESTool scaling characteristics are good for
a complex CFD code for large grids, the speedup
is close to linear.
Above A LES simulation of homogeneous
turbulence, revealing the complex eddy
structures. Left Flow over a cylinder is a
good test case for DES validation, given its
unsteady and transitional characteristics. The
image shown here (left) reveal the complex,
unsteady structures that form in the downstream
wake.
For more information on research related to this
project, visit the following websites CPL
http//www.louisville.edu/research/cpl/ KAOS
http//www.aggregate.org/ UK CFD
http//www.engr.uky.edu/cfd/
Given current computer power, DNS can only
realistically be applied to simple flows. The
above flow visualization is the result of a DNS
simulation for doubly-periodic channel flow. To
the left are comparisons between the LESTool DNS
results, other DNS results, and theoretical
models, with the LESTool outcomes showing good
agreement.