Aerodynamic profile design using genetic algorithms and CFD.
Vilag, Valeriu ; Popescu, Jeni ; Petcu, Romulus 等
1. INTRODUCTION
Classical methods for aerodynamic profile design rely on previous
experimental results and simplified mathematical models describing the
phenomena. Since the development of Computational Fluid Dynamics (CFD)
codes the obtained geometries are verified once more before new
experiments and corrections to the original geometry are often
identified. The current development of computer hardware made this first
validation process shorter and shorter arising the idea to use this
great resource in the actual design process. The problem consisting in
the fact that classical methods described in the iterature pretend to
obtain the best geometry and the hand adjustments made using experiments
and CFD simulations shows that improvements can be obtained, is solved
using an optimization process for the whole design phase. This
optimization is regarded in the present paper like the evolution of the
profile geometry towards the best fulfilling of its aerodynamic purpose.
2. GENETIC ALGORITHMS OVERVIEW
Genetic algorithms are optimization algorithms following the
natural evolution model (Kalyanmoy, 1999). From an original population,
through certain mechanisms inspired by the nature, there are obtained
inheritors fighting for survival. After relatively small number of
populations the solution converges to the best generation which cannot
be further improved.
In order to use genetic algorithms the designer must define the
architecture of an individual, similar to DNA, representing the sets of
parameters, chromosomes, defining its unique features (Koehler, 1997).
The great number of combinations of possible values of these parameters
shows the necessity to use an efficient way to decide which one is the
best.
3. PROFILE PARAMETRIZATION
A relative simple way for obtaining theaerodynamic profile is
defined. Two parabola portions are used: one for the skeleton and on for
the thickness distribution of the profile. The parabola portions are
further computed in order to obtain usefull data for geometry
construction under Computer Aided Design software (CAD) (Yang, 2006).
[FIGURE 1 OMITTED]
Six parameters define the aerodynamic profile. [A.sub.1,2] for the
peaks of the parabolas, [B.sub.1,2] for the starting point of each
parabola portion and [C.sub.1,2] for the end point of the parabola
portion. [A.sub.1,] [B.sub.1] and [C.sub.1] are used for the skleton and
the others for the thickness distribution. Figure 1 shows the
positioning of the parameters defining the parabola chosen such as to
cross (0,0) and (1,0) point.
The "x" coordinate of the END point is given in equation
(1), meaning that the END is always on a greater "x"
coordinate with respect to the START point. The parabola is defined
using equation (2), which is used for computing all the points on it.
[X.sub.END] = B + C (1-B) (1)
y = 4 A(x - [x.sup.2]) (2)
Possible values for the parameters are:
[A.sub.1,2], [C.sub.1,2] [member of] {0.1; 0.2; 0.3; 0.4; 0.5; 0.6;
0.7; 0.8; 0.9; 1}
[B.sub.1,2] [member of] C {0; 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.7;
0.8; 0.9}
Each parabola portion is split into 25 parts of equal length
between the START and the END points, the parabola portion defining the
skeleton is scaled to a chord equal to 1m and then rotated to zero
incidence and the parabola portion defining the thikness distribution is
scaled such as to obtain a maximum given thickness and 24 pairs of
points are obtained to be arranged symmetrical on the already obtained
skeleton. In this stage, a CAD software is used to import the points for
the suction and pressure side through which a spline curve is
constructed from the trailing edge to the leading edge and back, as
shown in Figure 2.
[FIGURE 2 OMITTED]
4. EVOLUTION
From an original population, which is arbitrarily chosen, there are
two main techniques to obtain the inheritors fighting for survival
(Temesgen, 2007). We chose the following two individuals as primordial
profiles and we will show the crossover and the mutation.
#1 = {0.5; 0.5; 0.5; 0.5; 0.5; 0.5}
#2 = {0.6; 0.6; 0.6; 0.6; 0.6; 0.6}
The crossover can be done using one or multiple crossover point but
for this example we chose crossover at the middle of the DNA sequence.
We obtain the following inheritors:
#3 = {0.5; 0.5; 0.5; 0.6; 0.6; 0.6}
#4 = {0.6; 0.6; 0.6; 0.5; 0.5; 0.5}
If we compute the aerodynamic performances for the four individuals
obtained so far we see the numbers presented in Table 1, where CL and CD
are the computed lift and drag coefficients.
We can see that #1 is the best followed by #3. #2 and #4
didn't survive to give birth to new inheritors.
In order to increase the diversity of the population we can use the
mutation evolution. This technique is also arbitrarily, and we chose to
apply it for profile #1 by decreasing its second parameter. We obtain
profile #5:
#5 = {0.5; 0.4; 0.5; 0.6; 0.6; 0.6}
The results for the new individual shows a [C.sub.L] / [C.sub.D] =
30.9784, meaning that it will take the place of #3 and will form the new
generation along with #1.
5. CFD CONTOUR PLOTS
For profile #5, which is the best one so far, we show the relative
pressure distribution and velocity distribution.
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
Figures 3 and 4 are ony qualitative ones showing that the profile
can fulfill a certain aerodynamic purpose(Gelsey, 1998). The CFD code
predicts the pressure and velocity distribution on the profile but in
this paper we do not discuss its precission. We only add that the
convergence is obtain in less then 1000 iterations taking about 4
minutes to compute.
6. TECHNIQUE UTILIZATION
The technique can be exploited for aerodynamic design problems
where a 2D approach is sufficient by applying multiple times one of the
evolution methods described in an iterative process. A large spectrum of
profiles can be obtained and the optimization process will lead to the
best profile for the imposed working conditions. Several profile types
are displayed in Figure 5.
7. CONCLUSION
The presented technique, which combines CFD with genetic
algorithms, can be used for optimization processes when designing 2D
aerodynamic profiles. The parametrization allows to obtain many types of
profiles meaning that many aerodynamical problems can be solved using
this method. Several steps shall be undertaken in order to use the
method for 3D designing problems where the number of parameters will
increase and wings or blades can be generated and optimized.
8. REFERENCES
Gelsey, A.; Schwabacher, M. & Smith, D. (1998). Using modelling
knowledge to guide design space search, In: Artificial Intelligence 101,
pp 35-62, Elsevier
Kalyanmoy, D. (1999). An introduction to genetic algorithms Vol.
24, Sadhana, India
Koehler, G. (1997). New directions in genetic algorithm theory.
Annals of Operations Research 75, pp 49-68, Springer
Temesgen, M. & Wahid, G. (2007). Aerodynamic optimization of
turbomachinery blades using evolutionary methods and ANN-based surrogate
models, Springer Science+Business Media, DOI 10.1007/s11081-007-9031-1
Yang, Y.; Wu, Y. & Liu, S. (2006). Graphical model construction
based on evolutionary algorithms, Journal of Control Theory and
Applications 4, pp 349-354, Springer
Tab. 1. Computed results for the first four individuals
Profile [C.sub.L] / [C.sub.D]
#1 30.4685
#2 20.4332
#3 30.2376
#4 19.8012