Bienvenue à l`ENSICA

Transcription

Direction Générale Technique
Shape optimization for aerodynamic
design: Dassault Aviation challenges
and new trends
Forum TERATEC 2014
Atelier « Conception numérique optimale des systèmes complexes : état de l'art et
verrous technologiques »
École Polytechnique - Palaiseau- France - July 2nd 2014
VA OÙ TON RÊVE TE PORTE
CFD Optimization Team






Frédéric Chalot
Laurent Daumas
Quang Dinh
Steven Kleinveld
Ximun Loyatho
Gilbert Rogé
page : 3
Civil > 50 %
Rafale
Falcon 7X
Spatial
UAV
NEURON
Outline
F8X
 CFD Aerodynamic Shape Optimization in
the framework of Aircraft Design
 Control Theory, Adjoint, AD
 SBJ, Sonic Boom, Engine Integration
 High Lift Configuration
 Air Duct, Topology + Sizing Opt, Unsteady
 New Trends
Abstract: The aim of this talk is to discuss various cutting edge techniques
developed for the aerodynamic shape optimization of Dassault Aviation future products.
Level 1
Global Optimization,
Comparative assessments
Trade off studies
Open framework
Surrogate models
level 3
ARCHITECTURE
MOTEUR
FORMES
(et AMENAGEMENT)
AMENAGEMENT)
MASSES
(et CENTRAGE)
CENTRAGE)
Controleur
AERO
GV
PERFOS
GV
Local Optimization
Critical point investigation
Quantitative
assessments
Aerodynamics &
Sonic Boom
Engine cycle
5
Fine analysis
Airframe
Layout
Controleur
Level 2 & 3
(icl. weight estimation,
aeroelasticity optimisation)
(icl. certfication issues)
level 1
MDO: The art of efficiently
managing the design parameters
between disciplines and levels
level 2
MDO: Multidisciplinarity and Multi-level
Airplane design cycle
Mission (market requirements)
(size, range, weight, ...)
Preliminary configuration
Aerodynamic shape optimization
Structure defined in
complete details,
control systems, ...
Detailed
design phase
Preliminary
design phase
Improvements of
structural and
aerodynamic behavior
(wind tunnel testing)
6
Conceptual
design phase
Optimization Process
Volume mesh displacement
Baseline
volume
mesh
Volume mesh
deformation
Adjoint
Volume mesh
deformation
Adjoint
CFD solver
Modified
volume mesh
CFD solver
Aerodynamic
observations
Aerodynamic
observations
gradient
Modified surface
mesh & gradient
Aerodynamic
variables
CAD Modeler
Cost - Gradient
Baseline
geometry
Objective &
constraints
Optimizer
design variables
Cost, constraints &
gradients
Starting point
8
Geometric Modeller
Large deformation
Intersection.
Untrimmed surfaces.
Geometric Modeller
9
Shape (e.g. fuselage)
Osculator parameters:
point, tangent,
curvature
10
Unstructured Meshes
y+=1
~ 10 Millions vertices
~ 60 Millions tetrahedral elements
11
Navier-Stokes solver
RANS, GLS, entropic variables, implicit scheme, GMRES
Turbulence modeling: k-epsilon, k-omega, DRSM, …
10 Millions vertices: 20mn on Purflex 512 procs
Tera 1012
Peta 1015
Performance Development
10 Years: Power*100
SUM
1 Eflop/s
100 Pflop/s
10 Pflop/s
10.5 PFlop/s
N=1
N=500
1 Pflop/s
100 Tflop/s
10 Tflop/s
1 Tflop/s
100 Gflop/s
10 Gflop/s
1 Gflop/s
supercomputer
CURIE
2 PFlop/s
100 Mflop/s
Dassault Aviation follows Moore law
MPI, OPENMP, GPU acceleration
Implementation strategy
Formulation for derivatives 1 et 2
Notations:
E
W
X
x
PDE control theory
J-L Lions
Dunod, 1968
Non linear system for fluid (Euler or Navier-Stokes)
L
State, solution of
Linear system for mesh deformation
J
E


Volume coordinates
Surface coordinates


T
Fluid Adjoint, solution of 
T
Mesh Adjoint, solution of
Second derivative operator
2
G1 ,G2
D
Observation
Aerodynamic parameters
Geometric parameters
E
J

W W
L
J
E

 T
X X
X
2
2M
2M
T
T  M
M .[V1 ,V2 ]  V
V1  2V1
V2  V2
V2
2
2
G1
G1G2
G2
T
1
Implementation strategy
Formulation for derivatives 1 et 2
dJ J
E

 T
d 

d 2J
W
W
2
T
2
 DW , J .[
,1]  DW , E.[
,1]
2
d


Implicit CFD
Explicit
dJ
T L x
 
d
x 
2
d 2J

W

X

W

X

L

x
2
T
2
T

D
J
.[
,
]


D
E
.[
,
]


W ,X
W ,X
d 2
 
 
x  2
Explicit
Implicit CFD
Ludovic Martin PhD, 2010,
CANUM Award
Implicit deformation
3D CFD Optimization using Newton
ONERA m6 wing
Mach = 0,84 , aoa= 3,06°
Cz > 0.01
BFGS: cost = 20 (19 NF + 10 Ng)
Newton: cost = 16 (15 NF + 8 Ng + 8 Nh )
BFGS
Newton
BFGS
Newton
Wing twist and camber optimization
Supersonic cruise at Mach=1.6
Polar curve
Symmetric airfoil
0.01
25
20
0.005
Prescribed CL
Optimized airfoil
15
Optimized wing
Z/C
0
Lift
Lift (pt)
10
-0.005
improvement
5
0
-0.01
0
20
40
60
80
100
120
140
160
180
200
-5
-0.015
0
0.025
0.05
0.075
0.1
X/C
Symmetric wing
Drag
Drag (count)
16
-10
9 control sections
4*9=36 design variables
36+1=37 optimization variables
Sonic Boom minimisation process
Shape before / after minimisation
Wing dihedral
Nose camber
Extraction /
propagation
17
67 optimization variables:
• Angle of attack (aoa)
• fuselage: scale,
thickness, camber angle
for 18 sections, 1
dihedral angle (nose)
• wing: twist angle,
camber parameters for
3 sections, 2 dihedral
angles (inner and outer
wing)
Same Sonic Boom optimization drag increase allowed
HISAC project (IP - 6th FP)
Pareto front (dBA vs drag)
88,5
88
Ref (1ms)
Optim (1ms)
87,5
87
Optim (5ms)
Ref (5ms)
18
Ref
ASEL (dBA)
Optim
• The drag constraint has been relaxed at different levels in
order to see if more noise reduction (dBA and dBC) could
be achieved.
• 2 meteo conditions are considered => Rise time of the
ground overpressure is either 1ms or 5 ms.
86,5
86
85,5
85
84,5
84
83,5
-2,0%
0,0%
2,0%
4,0%
6,0%
8,0%
drag coefficient increase
V A O Ù(%)
T O N- RRef=99.16
Ê V E T E P O Rcount
TE
10,0%
Engine integration optimization
HISAC project (IP - 6th FP)
Gain: 60 %
on zero-lift drag
Z
Y
X
Y
19
X
Z
Z
KP-G001
0.5
0.458824
0.417647
0.376471
0.335294
0.294118
0.252941
0.211765
0.170588
0.129412
0.0882353
0.0470588
0.00588235
-0.0352941
-0.0764706
-0.117647
-0.158824
-0.2
X
Z
KP-G001
0.5
0.458824
0.417647
0.376471
0.335294
0.294118
0.252941
0.211765
0.170588
0.129412
0.0882353
0.0470588
0.00588235
-0.0352941
-0.0764706
-0.117647
-0.158824
-0.2
Y
X
Nose
Rear fuselage
Cp distribution
Y
(Design, Simulation and Flight Reynolds Number testing for
advanced High Lift Solutions)
Presentation with kind permission of activities funded within Seventh Framework
Programme
EC – Grant Agreement N° ACP8-GA-2009-233607
Steven Kleinveld
DESIREH- DASSAULT-HIGH LIFT OPTIMIZATION
Optimization of High Lift configuration within
European project DESIREH
2.5D High Lift optimization
Use of various techniques to search for improved performance at take-off and landing
SOM
FD
MOGA
Example of performance
improvement through
optimization at Take-Off
conditions for classical high
lift configuration
using setting variables
(gap,overlap,deflection angle)
AoA
2.5D High Lift optimization
3D High Lift optimization
Example of flap
position optimization at
Take-Off conditions of
3D configuration taking
into account variations
of the intersection with
fuselage
Control and optimization of separated
flows
24
• PhD thesis J. Chetboun : Dassault Aviation / Polytechnique /
DGA.
• Development of automated methods for the control and the
optimization of separated flows.
• Application to curved air ducts for UCAV.
• Use of mechanical vortex generators (VG).
Topological derivative and
parametric optimization
• Vortex generators modelled by source term added to NS
equations.
• Creation of a new VG : topological derivative.
• Sizing : parametric optimization.
• State equation and cost function :
E (V , f (V , l ))
J (  , l )  j (V (  , l ))
 E E f 
dj
(V , f (V , l )) 
P 

(V )
dV
 V f V 
T
• Topological and parametric derivatives :

g  P  f (V , l )
T
J
T  E f 
(V , f (V , l ))
  P 
l
 f l 
25
• Adjoint state equation :
Applications : S-duct, U-duct,
unsteady computations
Swirl
DES
Complete
optimization
26
Topological derivative
Robust Design (1)
Cl < 0.3
Minimal
drag
Cl > 0.3
Final
Population
Minimal
probability
ONERA M6 wing, 2 design parameters: twist and TE camber angles
Euler, RSM RBF like but with 1rst and 2nd derivatives (original approach, Duchon
extension)
MOGA, Robust design. Objectives: to control Drag and P(CL<0.3)
Ludovic Martin PhD
27
Robust Design (2)
P(Cl<0.3)
Minimal
drag
Pareto Front
Initial population
Final population
E(CD)
Minimal
probability
lift
28
Robust Design (3)
Decision: we accept a probability lift of p = x% with minimal drag
Determination of drag mean CD (Pareto front)
P(Cl<0.3)
Determination of nominal values of geomerical
parameters a1 and a2 (camber and twist angles)
Minimal
drag
Minimal
drag
Minimal
probability
p
Cd
E(CD)
Minimal
probability
a1
a2
29
Conclusion
 Aerodynamic Shape Design Optimization based on
CAD (geometric constraints, …) and Adjoint
 Euler, RANS, cruise, TO, Unsteady, turbulence and
transition modelling
 Bidisciplinary Optimization (Mass, Flutter, RCS, …)
 Robust Design, Uncertainties
Falcon ++
Aircraft Fighter ++
UCAV ++
31
Q&A

Bienvenue à l`ENSICA

Transcription

Similar documents

2013 NARCh Finals Results