Vous êtes sur la page 1sur 6

EVOLUTIONARY METHODS FOR DESIGN, OPTIMIZATION AND CONTROL

P. Neittaanmäki, J. Périaux and T. Tuovinen (Eds.)


© CIMNE, Barcelona, Spain 2007

MULTI-OBJECTIVE OPTIMISATION IN MODEFRONTIER FOR


AERONAUTIC APPLICATIONS

Alberto Clarich* Carlo Poloni


ESTECO srl Department of Energetics
Area Science Park University of Trieste
Padriciano 99, 34012 Trieste, Italy Via Valerio 10, 34100 Trieste, Italy
Email: clarich@esteco.it Email: poloni@units.it
web page:http://www.esteco.it web page: http://www.units.it

Abstract. This paper shows how the optimisation environment modeFRONTIER is


applied to solve multi objective problems in aeronautic field. These problems include
solution robustness on uncertain definition of parameters, and automatic and distributed
communication between parametric CAD and simulation software.
Two application cases are presented: first, a 3D wing defined and parameterised in
CATIA V5, is to be optimised for two different mission points, clean wing in cruise
transonic regime and opened high-lift devices for subsonic take-off, following
aerodynamic criteria and constraints. Second, the shape of a transonic airfoil is to be
optimised under uncertainties of flight conditions, in order to optimise not only the
design point, but to keep efficiency as well after operating condition fluctuations.

Key words: Multi objective optimisation, Robustness analysis, CAD/CAE distributed


environment, Game Theory, Metamodels, Evolutionary Algorithms, Aeronautic
applications

1 INTRODUCTION

Multi objective optimisation, including solution robustness on uncertain definition of


parameters, in an automatic and distributed environment that allows direct
communication between parametric CAD and simulation software, is becoming
continuously a key factor in aeronautic industries and not only.
This paper shows how the optimisation environment modeFRONTIER allows to
implement practically this methodology, whatever software is used in the simulation
phase; two different applications will be illustrated.
modeFRONTIER [3] is a multi-objective design environment software that allows
the integration of any commercial or built-in house computational code (CAD, FEM,
CFD, etc..) into a common environment, in order to run automatically a series of
designs, scheduled by the available optimisation algorithms, until the defined objectives
are satisfied.
A. CLARICH, C.POLONI/ Multi-objective aeronautic optimization with modeFRONTIER

In this modular environment, each component of the optimisation, including input


variables, input files, scripts to run the commercial software, output files, output
variables and objectives, is defined as a node to be connected to the other components.
In this way, the complete logic flow from CAD parameterisation to performances
evaluation is defined by the user, that can select among several available optimisation
algorithms, accordingly to the defined objectives; they include Genetic Algorithms,
Evolutionary Algorithms, Game Strategies [1], Gradient-based Methodologies, Robust
Design Optimisation as well as main DOE (Design Of Experiments) algorithms (Sobol,
Factorials, Latin Square, Montecarlo, D-Optimal, etc.).
These algorithms drive automatic series of simulations, allowing when available
distributed and parallel computations to fully exploit the computational resources, until
the objectives are met.
In addition, the influence of all the parameters in the process can be analysed in
detail by the use of statistical analysis (correlation matrix, t-Student, etc. ) and response
surface methods (Kriging, Neural Networks, Radial Basis Functions, SVD, Parametric,
Gaussian, etc.), that can be also used to reduce the number of computations required in
the optimisation, allowing an extrapolation of the results.
In particular, Game Strategies and Response Surface Methodologies are used to
reduce the global number of design evaluations required by a Robust Design Analysis
(design under uncertainties or fluctuations of input parameters), as it will be illustrated
in the second application of this paper.

2 3D WING MULTI-POINT OPTIMISATION

2.1 Problem definition


In fig.1 below we can see a representation of the two design points of the
optimisation problem.
The original wing is defined by three main sections derived by the RAE-16 airfoil,
and the geometry of these sections should be modified in order to minimise the drag of
the smooth configuration (a) for a fixed angle of attack (2°) and transonic Mach number
(0.7), and to maximise the efficiency (lift/drag ratio) of the three-element deployed
configuration (b) that corresponds to the take-off mission point for another fixed angle
of attack (17.12°) and Mach number(0.12; Re=1 106 for both configurations).

a) b)

Figure 1 Smooth transonic design point (a) and three-element deployed design point (b)

There is an aerodynamic constraint on the lift coefficient for the transonic point,
since it should be kept greater than the original one (0.18), another constraint on the lift
EVOLUTIONARY METHODS FOR DESIGN, OPTIMIZATION AND CONTROL (EUROGEN 2007)

coefficient for the subsonic point (0.78) and also a geometric constraint on the
maximum thickness that should be greater than the original one (16% of the chord
length).

2.2 CAD Parameterisation


The whole geometry model is defined inside Catia V5 CAD system; in particular, the
geometry of the two mission points is controlled by two set of variables: the first 30
ones control the three main section shapes, while the further 26 ones define the internal
shape and the position of the aerodynamic flaps deployed from the smooth wing.
About the first set of 30 variables, there are 10 parameters for each of the three main
sections (root, kink, tip). For each one of the three sections, the 10 parameters controls
the control points co-ordinates of four NURBS curves [1], that define the upper and
lower side of the profile.

Z1 Z2 X345 Z6
Z7

Z7d
Z1d Z2d Zmax Z6d

Figure 2 : Airfoil section: it is defined by 4 NURBS, the second one is represented here

About the second set of 26 variables, 13 ones of them parameterise the aerodynamic
elements located between the root and kink sections, while the other 13 ones are relative
to the elements located between the kink and tip sections.

X1s

Dx_slat
Xc
X1f alfa_slat

Dz_slat
X2s Dx_flap

X2f alfa_flap
Dz_flap
Figure 3 : Parameterisation of shape and position of the flaps deployed from the smooth section

For each subset of 13 variables, five of them define the shape of the flap and slat that
are detached during the subsonic mission point (fig.3, left), in order to increment the lift
coefficient, while the other 6 ones define respectively the position (translation and
rotation) of the same aerodynamic elements (fig.3, right).
A. CLARICH, C.POLONI/ Multi-objective aeronautic optimization with modeFRONTIER

2.3 modeFRONTIER optimization workflow

Direct CATIA node

Link each optimisation variable to a


parameter defined in the CATIA model

Batch script to run mesher

Figure 4 : modeFRONTIER optimisation workflow


Figure 4 above illustrates how modeFRONTIER can, in its modular workflow, drive
this optimization problem. Two applications chains are defined, from top to bottom, in
parallel: they are relative to the simulation of the two mission points (cruise at left and
take-off at right) for each design proposed during the optimization. The input variables
are defined in an opportune subsystem, and they are linked directly to the two CATIA
model nodes, that are updated automatically for each design.
The output file in IGES format is sent by a relative transfer node to the following
application, a UNIX batch command that run the mesher code, ICEM, that rebuild the
mesh over the IGES file, following instructions recorded in a macro file (during the
creation of the original model mesh). In a similar way, the output file, an .inp file, is
transferred to the following application, a script that run in batch StarCD for the CFD
evaluations. Output variables to evaluate the design fitness, drag coefficient in cruise
condition and efficiency in take-off conditions, are automatically extracted from the
StarCD output file accordingly to the specified pattern.
Once the modular workflow is built, the DOE and optimisation algorithm can be
chosen and the optimisation automatically run.

2.4 Optimisation results: MOGT + GA strategy


In this application, a combined strategy using the algorithms available in
modeFRONTIER has been used. First, MOGT (Multi-Objective Game Theory)
algorithm has been run [1]. The philosophy of this algorithm is that two players divide
the objectives and the search space. Each player tries to optimise his own objective
applying an efficient and fast mono-objective algorithm, Downhill Simplex [1], sharing
EVOLUTIONARY METHODS FOR DESIGN, OPTIMIZATION AND CONTROL (EUROGEN 2007)

in several steps the best variables found, and influencing in this way the search of the
other player. During the optimisation, the decomposition of the variable space is
updated accordingly to the statistical significance of the variable to the objective, in
such a way that if a variable is not significant for one objective, it is given to another
player. The solution is an equilibrium point, that represent the best compromise for the
two objectives.
After 300 designs, the (Nash) equilibrium pint has been found, but in order to extend
the results into a wider Pareto frontier, or set of not-dominated designs, the best results
from MOGT optimisation phase have been used to initialize a MOGA (Multi-Objective
Genetic Algorithm) algorithm of other 200 designs (4 generations of 50 designs). The
results are reported in table below. Since a cluster of 8 CPUs has used and 1 analysis
takes 2 hs, about 5 days have been spent to obtain the results.

Pareto front original Nash/Simplex


best
Nash point Transonic drag 0.0371 0.0357 (-4%)

Subsonic lift/drag 3.93 4.65 (+18%)


Original point

Figure 5 : optimisation results (transonic drag in abscissa, subsonic efficiency in ordinate)

3 ROBUST DESIGN OF TRANSONIC AIRFOIL

3.1 Problem definition


The objective of this test case is the robust design optimisation [2] of the RAE2822
airfoil under uncertainties of free stream Mach number and angle of attack, regarding
drag reduction, with constraints on lift and momentum coefficients. The design point is
defined by Mach=0.73 and angle of attack α=2° while, due to the not deterministic
events (like gusts of wind, atmospheric turbulence, instable conditions of flight,
manoeuvre inaccuracy, etc...), the range of operating conditions is fixed to
α=2±0.5° and Mach=0.73±0.05 (Re=6 106)
In order to evaluate the robustness of each candidate solution proposed by the
optimisation algorithm, we need to compute a set of sampling point around the nominal
design, varying the values of Mach and angle of attack accordingly to stochastic
definition of uncertainty (Montecarlo sampling): from the samples, mean and standard
deviation of the output variables will be computed.

3.2 modeFRONTIER workflow


Also in this case modeFRONTIER allows the definition of a modular workflow, of
the Robust Design sampling policy and the definition of optimisation algorithms.
Using the Kriging RSM, modeFrontier create a metamodel that can approximate the
response of the sampling points for each design: only 10 points for each design are
necessary for the training.
A. CLARICH, C.POLONI/ Multi-objective aeronautic optimization with modeFRONTIER

Figure 6 : modeFRONTIER optimisation workflow


Figure 6 above shows the modular workflow of modeFRONTIER: the input
variables are relative to Bezier control points ordinates (9 for each side) that modify the
airfoil shape; a script reads these points, modify the airfoil shape, update the mesh and
run the Navier Stokes solver based on Johnson-Coakley turbulence model. From output
file, aerodynamic performances are read for each sample, whereas the objectives and
constraints are defined on mean and standard deviation values of the sampling
distribution computed for each configuration design.

3.3 Optimisation results


The algorithm used is also in this case MOGT, and 250 designs were needed to
obtain the equilibrium point (these correpsonds, thanks to the application of Kriging
RSM, to a total of 2500 CFD computations). Table in fig.7 reports the comparison of
the best configuration with the original one, regarding mean and standard deviation of
aerodynamic coefficients (objectives and constraints are distincted).

MeanCd σCd MeanCl σCl MeanCm σCm


(obj) (obj) (constr) (constr) (constr) (constr)
>original <original <original <original

RAE2822 1.33e-2 7.6e-3 0.686 5.91e-2 -0.105 9.50e-3


original
best BEST 1.18e-2 5.27e-3 0.691 5.83e-3 -0.102 6.45e-3

Figure 7 : optimisation results

11 REFERENCES
[1] A.Clarich, J.Periaux, C.Poloni, “Combining Game Strategies And Evolutionary
Algorithms for CAD Parameterisation and Multi-Point Optimisation of Complex
Aeronautic Systems”, EUROGEN 2003, Barcelona, September 2003
[2] Clarich A., Padovan L., Pediroda V., Periaux J., Poloni C., “Application of game
strategy in multi-objective robust design optimisation implementing self-adaptive
search space decomposition by statistical analysis”, ECCOMAS2004, Jyvaskila
[3] www.esteco.it

Vous aimerez peut-être aussi