Documentos de Académico
Documentos de Profesional
Documentos de Cultura
1 © 2011 ANSYS, Inc. September 22, 2011
Brief History of Aerodynamics CFD
1970s 1980s 1990s 2000s
2 © 2011 ANSYS, Inc. September 22, 2011
Brief History Aerodynamics CFD
1970s 1980s 1990s 2000s
2010s – How to leverage simulation extensively?
Decade of Optimization
3 © 2011 ANSYS, Inc. September 22, 2011
CFD in Aerodynamics Development
The main purpose of CFD is to enable optimization via thorough
The main purpose of CFD is to enable optimization via thorough
design exploration
Thorough Design Exploration
Design Alternatives Deep Insights Multiple Aspects
4 © 2011 ANSYS, Inc. September 22, 2011
CFD in Aero Development
Current state‐of‐the art
(commercial, non‐commercial codes)
5 © 2011 ANSYS, Inc. September 22, 2011
Technological Building Blocks
f A d
for Aerodynamics Optimization
i O i i i
Ability to simulate detailed aerodynamics for large DOE
matrix in a short time (e g simulating 50 design
matrix in a short time (e.g. simulating 50 design
alternatives automatically over a weekend)
Major technological advances needed in –
1. Model creation
2
2. S l
Solver speed
d
3. Process management
4. Data management
ANSYS is first in overcoming these technological
challenges
6 © 2011 ANSYS, Inc. September 22, 2011
Tech. Building Blocks for Aero Optimization
1 M d l C ti
1. Model Creation
Surface Wrapper Cut Cells
Model Creation
Model Creation
Reference:
SAE Paper 2009‐01‐0335
Cut Wrap
Cut Wrap Morpher
7 © 2011 ANSYS, Inc. September 22, 2011
Tech. Building Blocks for Aero Optimization
2 S l
2. Solver Speed
S d
Truck (111 million cells)
• Basic solver enhancements
Perfrormance
– P
Pressure based coupled solver, pseudo‐
b d l d l d 13.0.0
transient relaxation, higher‐order term 14.0.0
relaxation (HOTR), hybrid flow initialization, 0 768 1536 2304 3072 3840
non‐iterative transient, hierarchy based hybrid Number of Processor Cores
Number of Processor Cores
AMG, higher‐order numerics improvements
F1 140M AMD Magny‐Cours
• High performance computing (parallel
solver)
nce
Performan
– ANSYS Fluent is a clear leader in high‐
14.0.0
performance computing for external Ideal
aerodynamics
• Linear scalability shown on > 3000 processor
Li l bili h 3000 384 832 1280 1728 2176 2624 3072
cores (using wall clock time) Number of Processor Cores
• Parallel file I/O
• Smart and efficient partitioning and load
Smart and efficient partitioning and load
balancing algorithms
8 © 2011 ANSYS, Inc. September 22, 2011
ANSYS Aero Capabilities Leadership
ANSYS Fluent 15.0
Fluent–Tgrid integration (parallel)
HPC: linear scalability to >10,000 cores
Feaatures
ANSYS Fluent 14.x
GPU for radiation, Fluent–Tgrid
g integration (serial)
g ( )
HPC: scalability and I/O
ANSYS Fluent 14.0
Adjoints, higher order numerics and wall treatment
improvements, GPU for radiation
HPC li
HPC: linear scalability to >3000 cores
l bilit t >3000
ANSYS Fluent 13.0
Pseudo transient solver, bad mesh robustness, mesh‐morpher‐
optimizer, cut‐cell remesh, wall‐film, viewfactor speedup
HPC: file I/O, load balancing
ANSYS Fluent 12.0
One‐billion (1E9) cell cases
Anisotropic boundary layer remeshing
HPC: Parallel File I/O, Linear scalability to >1000 cores
Fluent 6.3
Pressure based coupled solver, polyhedra
HPC: improved file I/O
Fluent 6.2
Linear scalability to 256 cores
Linear scalability to 256 cores
Ideal process
Ideal process – “One‐click”
One click response surface
response surface
• Supply baseline model
• Identify parameters and ranges
• Submit job –
S b it j b obtain response surface
bt i f
Ideal process automatically manages
• Model modification and remeshingg
• Distributed solve
• Data collation and reporting
10 © 2011 ANSYS, Inc. September 22, 2011
Tech. Building Blocks for Aero Optimization
4 D t M
4. Data Management
t
When extensive simulations are rapidly run, simulation data is
produced at a faster rate than can be meaningfully
d d t f t t th b i f ll
understood and used by an engineer
Solution – Simulation Process Data Management
Solution Simulation Process Data Management
• Data handling and storage
• Meta data extraction
• Post simulation database studies
ANSYS EKM
11 © 2011 ANSYS, Inc. September 22, 2011
Case Study
Objective:
A “One Click” process for running an aerodynamics DOE
(D i
(Design of Experiments) study, after initial setup
fE i t ) t d ft i iti l t
Setup baseline model
Setup parameters
Run and post process entire DOE matrix automatically without user intervention
Run and post‐process entire DOE matrix automatically without user intervention
12 © 2011 ANSYS, Inc. September 22, 2011
Shape Exploration & Optimization Work Flow
Setup Baseline CFD model
Set Shape Parameters
Using Mesh Morpher
Set Output Parameter (Drag
Coefficient)
Generate and Run DOE Matrix
Generate Response Surface &
Perform Goal Driven Optimization
Perform Goal Driven Optimization
13 © 2011 ANSYS, Inc. September 22, 2011
Baseline CFD Simulation Setup
Mesh Details
• Tetrahedral + prism mesh
• 5 prism layers from all surface of vehicle
• Total cell count ~ 5.0 M
• Same mesh is used in all DOE points
p
Boundary Condition Setup
• Inlet : Velocity inlet (V = 80 mph)
• Outlet: Pressure outlet (0 Pa (gage))
• Tunnel Top & Sides : Symmetry
T l T & Sid S t
• Tunnel Road: No slip wall
Solver Setup
• Pressure based coupled solver
p
• RKE turbulence model
• 2nd order discretization schemes
15 © 2011 ANSYS, Inc. September 22, 2011
Shape Parameter Definition
Mesh Morpher Setup
3 Mesh morpher
3 M h h
parameters
corresponding to 3
p
shape factors defined
for vehicle
16 © 2011 ANSYS, Inc. September 22, 2011
Shape Parameter Definition
Correlating Shape Parameters with Control Point
17 © 2011 ANSYS, Inc. September 22, 2011
Movement
Design of Experiments
Leveraging ANSYS WorkBench Platform
18 © 2011 ANSYS, Inc. September 22, 2011
Design of Experiments
Fluent parameters are exposed to and driven by
DesignXplorer (DOE and Optimization application)
• Input Parameters
– Backlight angle
– Tumblehome
– Windshield angle
• Output Parameter
– Drag force on the vehicle
19 © 2011 ANSYS, Inc. September 22, 2011
Design of Experiment
Design Space is defined by the
following bounds
• Backlight angle(BA): ‐2 < BA <2
• Tumblehome (TH): ‐0.5<TH<0.5
• Windshield angel (WA): 0<WA<1
Windshield angel (WA): 0<WA<1
DOE Algorithm
• Central Composite Design (CCD)
• Design Type : Face Centered with
Enhanced Template
• 29 DOE Points Generated
29 DOE Points Generated
20 © 2011 ANSYS, Inc. September 22, 2011
Design of Experiments
After setup, running the DOE matrix of simulations is a
After setup, running the DOE matrix of simulations is a
“One‐Click” process
Workbench automatically runs simulations for all design
p
points one after the other
Post‐processing data is automatically gathered and
supplied to DesignXplorer
li d D i X l where response surfaces,
h f
etc are automatically generated
21 © 2011 ANSYS, Inc. September 22, 2011
Response Surfaces
Response Surface are generated based on Non‐Parametric
Regression Model
2D and 3D surfaces can be created to visualize the
2D d 3D f b t d t i li th
variation of output parameter within the design space.
Chart between predicted and
observed value of output parameter
Goodness of fit shows the accuracy of
the response surfaces
For bad goodness of fit one need to
refine the DOEs to get a better fitted
response surface
22 © 2011 ANSYS, Inc. September 22, 2011
Response Surfaces (3D)
23 © 2011 ANSYS, Inc. September 22, 2011
Response Surfaces (2D)
F Vs TH
F Vs BA
F Vs WA
24 © 2011 ANSYS, Inc. September 22, 2011
Parameter Sensitivity
Sensitivity Charts show that drag force is most sensitive to backlight angle
and least to windshield angle, for this case
25 © 2011 ANSYS, Inc. September 22, 2011
Pareto Plots
26 © 2011 ANSYS, Inc. September 22, 2011
Pareto Plots
27 © 2011 ANSYS, Inc. September 22, 2011
Goal Driven Optimization
• Goal driven optimization method works on the relative
importance of user specified conflicting goals to get an optimized
set of parameters
• Basic goal considered is that of minimizing drag force
Screening method for optimization
Goal: Minimize drag force
3 best designs suggested
3 best designs suggested
by DX
28 © 2011 ANSYS, Inc. September 22, 2011
Flow Field
Design Space Point:‐‐
Design Space Point: Baseline Design
Baseline Design
(Velocity Contours on x=0)
29 © 2011 ANSYS, Inc. September 22, 2011
Flow Field
Design Space Point: Worst Design
Design Space Point:‐‐ Worst Design
(Velocity Contours on x=0)
B li D i
Baseline Design
Optimized Design
Worst Design
32 © 2011 ANSYS, Inc. September 22, 2011
Shape Comparisons
33 © 2011 ANSYS, Inc. September 22, 2011
Summary
2010s is the Decade of Optimization
Four key technologies needed for aerodynamics optimization simulation
1. Model creation
2
2. S l
Solver speed
d
3. Process management
4. Data management
A “One‐Click” process is developed and demonstrated for running the
entire DOE matrix after initial setup
p
34 © 2011 ANSYS, Inc. September 22, 2011
Aerodynamics Optimization
G d Ch ll
Grand Challenges
A simulation tool that can simulate
A simulation tool that can simulate
50 design variants (extensive DOE matrix)
50:50:50 50 million cells (accurate model)
50 hours total elapsed time (weekend process)
35 © 2011 ANSYS, Inc. September 22, 2011