Beruflich Dokumente
Kultur Dokumente
of Powerplant
Sylvain MOUTON – Gérald CARRIER (ONERA)
Matthieu MEAUX – Julien LAURENCEAU (Airbus)
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 1 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Outline
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 2 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Case presentation
• W pylon width
2 disciplines involved
engine
• Structure
• Aerodynamics
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 3 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Strategy for exchange of information
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 4 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Optimisation problem formulation
Pylon design
• Combined objective: minimize drag + weight / k
• Parameters: engine position and pylon width (X,Z,W)
• Constraints: technological feasibility
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 5 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Sampling and Surrogate models
Low-level design
Best achievable
Individual 1 low-level
Optimisation objective
X1,Z1,Wloop
1 Data
interpolation:
Low-level design
Best achievable Kriging
Individual 2 low-level method*
X2,Z2Optimisation
,Wloop
2
objective
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 7 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Structure optimisation
Automated remeshing
Uses Finite Element Model and
linear static solution for stress
• each sample (X,Z,W) corresponds
to a new overall positioning of the
pylon spars and panels, with
automatic meshing tool
• 27 low-level design variables
(thicknesses of spars and panels)
Max. shear stress in pylon
• 45 load cases (fan blade off and
fatigue)
• constraints on maximum allowed
stresses to sustain fatigue and static
loads
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 8 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Structure response surface
Structure optimum
X = -200 mm
Z = +200 mm 0 kg
W = +200 mm +82.5 kg
Mass = -100 kg
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 9 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Aerodynamic
optimisations and
response surface
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 10 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Parameterisation
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 11 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Parameterization
Baseline shape
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 12 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Flow analysis
A1
Gradient method: BFGS* quasi-Newton algorithm
• using Airbus tools Optalia and DOT software
• objective function: pressure drag CDp
• constraint on lift CL introduced by penalty
• bounds on design variables are defined to satisfy geometry
constraints
* Broyden-Fletcher-Goldfarb-Shanno
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 14 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Gradient assessment
Innovative method for gradient calculation: adjoint
method*
• state of the art in the field of aerodynamics for sensitivities
computations (not yet used for industrial design)
• solve the exact derivative of RANS equations with elsA
(discrete adjoint equations) → aerodynamic gradient
(equivalent d.c.)
satisfactory
• gain of 0.60 d.c. on pressure
drag of full aircraft
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 16 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Aerodynamic optimisation: results
increased radius of
curvature for outboard
intersection
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 17 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Aerodynamic optimisation: a posteriori
analysis of optimal result
Far-field drag analysis*
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 19 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Aerodynamic response surface
Optimised response
surface
• 9 samples in the high-level
design space
aerodynamic optimum
X = +90 mm -0.5 d.c.
Z = +200 mm
W = +200 mm
Drag = -0.9 d.c. 0.0 d.c.
+0.5 d.c.
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 20 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Multi-disciplinary
response surface
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 21 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Combined response surface
The combined cost
function is obtained by
adding response drag + weight / k
surfaces for drag and
weight
• Trade-off coefficient k
can easily be varied to
observe its effect on
overall cost
Multi-disciplinary optimum
X = +90 mm
Z = +200 mm
Driven by aerodynamics
W = +200 m
Drag -0.9 d.c.
Weight -77 kg
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 22 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
High level optimisation using surrogate
models
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 23 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
High level optimisation using surrogate
models
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 24 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Conclusions
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 25 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Conclusions
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 26 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Thank you for your attention
Integrated Project
AIP3 CT-2003-502917
Priority 4 “Aeronautics and Space”
http://www.vivaceproject.com
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 27 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Adjoint method: convergence
Convergence history of L2
On 3D turbulent cases, the iterative residual on first adjoint variable
scheme used to solve adjoint 10
1
Residual-rho
improve the convergence rate 10-3 0_2
0_5
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 28 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved
Adjoint method: accuracy
Convergence history of 6
components of drag gradient
• 500 iterations are sufficient to
reach converged value of the
gradient Finite diff
Adjoint
• Comparison to 2nd order finite
difference shows fair agreement
on major components (5 – 30%)
but larger error on smaller
components
→ robust and accurate enough
to compute aerodynamic
gradient
VIVACE Fourm 2 © 2006 VIVACE Consortium Members. Page: 29 Multidisciplinary Optimisation of Powerplant
24-26 October 2006 All rights reserved