Вы находитесь на странице: 1из 8

J. of T h e r m a l Science Vol.8, No.

Genetic Algorithms Development for Multiobjective


Design Optimization of Compressor Cascade

Jun LI
V e n t u r e L a b o r a t o r y , G r a d u a t e School, K y o t o I n s t i t u t e of Technology, M a t s u g a s a k i , S a k y o - k u , K y o t o
606-8585, J a p a n

Koji Morinishi Nobuyuki Satofuka


D e p a r t m e n t of M e c h a n i c a l a n d S y s t e m E n g i n e e r i n g , K y o t o I n s t i t u t e o f Technology, M a t s u g a s a k i ,
S a k y o - k u , K y o t o 606-8585, J a p a n

Aerodynamic optimization design of compressor blade shape is a design challenge at present because it
is inherently a multiobjective problem. Thus, multiobjective Genetic Algorithms based on the multi-
branch simulated annealing selection and collection of Pareto solutions strategy have been developed
and applied to the optimum design of compressor cascade. The present multiobjective design seeks high
pressure rise, high flow turning angle and low total pressure loss at a low inlet Mach number. Pareto
solutions obtain the better aerodynamic performance of the cascade than the existing Control Diffusion
Airfoil. From the Pareto solutions, the decision maker would be able to find a design that satisfies his
design goal best. The results indicate that the feasibility of multiobjective Genetic Algorithms as a
multiple objectives optimization tool in the engineering field.

Keywords: multiobjective optimization, genetic algorithms, Pareto optimal set, com-


pressor cascade, design.

INTRODUCTION ary layer separation and shock wave. Thus, trade-


off between high pressure rise and low total pressure
Turbomachinery, especially compressor cascade, loss must come to the best comprise. The design of
optimum design presents a grand challenge to numer- compressor is inherently a multiobjective optimization
ical optimization. The goal of compressor cascade de- problem. In addition, aerodynamic performance anal-
sign is to produce the highest pressure rise with the ysis of compressor cascade is computer intensive and
lowest total pressure loss at the constant flow condi- the resulting aerodynamic performance is very sensi-
tion. Parameter of pressure rise is an important per- tive to geometry shape of the cascade, therefore, a
formance parameter of compressor. The higher pres- robust multiobjective optimization algorithms is in-
sure rise at the given flow condition, the better perfor- dispensable to this research field.
mance of compressor. But, higher pressure rise may Development of aerodynamic design methods for
lead to large adverse pressure gradients and bound- compressor cascade has always been of strong interest.
Successful design methods have been reported [1,2] and
Received, 1999. most of them are based on an inverse design through
Jun LI et al. Genetic Algorithms Development for Multiobjective Design Optimization of Compressor Cascade 159

a prescribed pressure distribution. However, this ap- pled from the analysis Pareto optimal set uniformly.
proach leaves the problem of specifying an appropriate Then, the present MOGAs have been applied to the
pressure distribution to designers at large. An arbi- multiobjective optimum design of the compressor cas-
trarily target pressure distribution may correspond- cade. Pareto solutions obtain the better aerodynamic
ingly yield an unrealistic geometry. On the other performance of cascade than the existing Control Dif-
hand, it is very difficult to decide geometry con- fusion Airfoil (CDA)[9]. From the Paxeto solutions,
straints. So, direct problem design is applied in this the decision maker would be able to find a design that
paper. Direct problem design is formed by coupling satisfies his design goal best. The results indicate that
aerodynamic analysis method with numerical opti- the feasibility of the present MOGAs as a multiple ob-
mization method. They optimize a given aerodynamic jectives optimization tool in the engineering field.
objective function by iterating directly on the geome-
try.
M U L T I P L E O B J E C T I V E S G E N E T I C ALGO-
Genetic Algorithms(GAs) [3] axe known to be ro- RITHMS (MOGAs)
bust and have been enjoying increasing popularity
in the field of numerical optimization, multiobjective Solving a multiple objectives problem generally re-
optimization [4,5] in particular, in recent years. Multi- quires the identification of Pareto optimal set, a con-
objective problems seek to optimize components of a ception introduced by V. Pareto, a prominent Italian
vector valued objective function. Unlike the single ob- economist at the end of the past century (1896). A
jective optimization, the solution of the multiobjective solution is said Paxeto optimal, or non dominated, if,
problem is not a single point, but a family of points starting from that point in the design space, the value
known as Pareto optimal set. Each point in this set of any of the objective functions can not improved
is optimal in the sense that no improvement can be without deteriorating at least one of the others. All
obtained in one objective component that does not potential solutions to the multiple objective problem
lead to degradation in at least one of the remaining can thus be classified into dominated and non domi-
components. nated (Paxeto optimal) solutions. The set of non dom-
GAs can search for many Pareto optimal solutions inated, solutions of a multiple objectives problem is
in parallel by tile Paxeto criteria in the constant popu- nanmd as Paxeto front or Pareto optimal set [4].
lation of the solutions [a] . GAs can be very efficient, if Therefore, it is the most important for the solution
they can sample solutions uniformly from the Pareto of multiple objectives problem to find the Paxeto op-
optimal set. Since GAs are inherently robust, the com- timal set. To define the notion of domination, let us
bination of efficiency and robustness makes them very think about the following problem:
attractive for solving multiobjective problems. There Min: f ( ~ = {fl(~?), f2(:~), . . . , f~(£)}
have many applications in the aerodynamic optimum where ~ is a vector of m decision variables.
design using different multiobjective GAs. Such as Suppose xi and 37j are solutions of the f(~?). Zi
multidisciplinaxy optimum design of the wing using is said to be dominated by ~?j, there exist Vk C
niching and elitist models of GAs [6] multiobjective X , . . . , m , f ( ~ i ) ~ f ( £ j ) and 3if(e~) ~ f ( e j ) . We of-
optimization of airfoil and wing design through hy- ten say that f ( x i ) is partially less than f ( £ j ) . £ i is
brid GAs [7] and direct problem design of compressor said to be non dominated if there does not exist any
cascade using multiobjective GAs based on Fonseca- £j in all the potential solutions that dominates £~.
Fleming's Paxeto-based ranking and fitness sharing All non dominated solutions Z"i of multiple objec-
techniques Is] . tives problem can make up Paxeto optimal set. If a
The Multiple Objectives Genetic Algorithms (MO- solution belongs to the Paxeto optimal set, it is im-
GAs) based on the multi-branch simulated annealing possible to improve one of the objectives without de-
selection and collection of Paxeto solutions strategy teriorating some of the others.
have been developed in this paper. To demonstrate GAs are applied to solve multiple objectives prob-
the feasibility of the present algorithms, the two objec- lem using Paxeto criteria to drive the evolution of the
tives and three objectives of mathematical test func- population. The characterizing feature of multiple ob-
tions are used. The obtained Pareto solutions are sam- jectives GAs is thus the introduction of Pareto criteria
160 Journal of Thermal Science, Vol. 8, No.3, 1999

in the m e t h o d used for individuals selection. Then cess, even though individuals are randomly chosen
through selecting individuals in reproduction phase from the population. In each branch of selection, in-
according to Pareto criteria, the Pareto optimal set dividuals are selected without replacement, so t h a t all
of multiple objectives problem can be developed. individuals will have one chance to compete on each
In this paper, a new type of Multiple Objectives objective. T h a t is to say, all individuals of current
Genetic Algorithms (MOGAs) have been developed. generation compete multiple times according to the
Various techniques used in the M O G A s are to be in- numbers of objectives. This feature provide the same
troduced briefly in the following: pressure toward individuals on each objective. This
can maintain diversity t h a t is needed to represent a
1. C o d i n g , C r o s s o v e r a n d M u t a t i o n spectrum of design across the Pareto optimal set.
The binary representation is traditionally used in
GAs. The binary alphabet offers the m a x i m u m num- 3. P a r e n t s S e l e c t i o n S t r a t e g y
bers of the schemata per bit of information of any The extended elitism strategy is adopted to parent
coding. In addition, the binary coding facilitates the- selection. T h a t is, the assigned percentage of parents
oretical analysis and allows elegant genetic operators. come from the actual set of Pareto solutions of local
Thus, binary coding m e t h o d is adopted in the MO- generation. These Pareto solutions can directly turn
GAs. into parents without undergoing genetic operation.
In the MOGAs, the one point crossover operator [3] The other percentage of parents come from individ-
is used. Non-uniform m u t a t i o n [1°] is adopted and uals via genetic operation. After multi-branch sim-
work on the word level of the binary code. T h a t is to ulated annealing selection has been finished, the as-
say, mutation performed directly on the integer range signed percentage of parents' pool is full. Individuals
from 0 to 2 n - l , n being the number of bit adopted in the pool are then m a t e d in crossover and m u t a t i o n
for each variable. In this case, the binary string is operation to form offspring. These children eventually
converted into an integer list, m u t a t i o n operator is become the parents of next generation. The parent se-
applied and then the list is converted back into a bit lection strategy is shown in Fig.1.
string.
1.0 \
..........................................................................................................................................

b
2. M u l t i - B r a n c h Simulated Annealing Selec- Percentage picked from Pareto front I
m
tion e-
0.8 (\ .....................!......................................................................................................................i
In MOGAs, selection operator plays an i m p o r t a n t t'~

role for obtaining the optimization results of the mul- Percentage selected for the first objetive ]
tiobjective problem. The m u l t i - b r a n c h simulated an- 0.6 using the first branch selection
nealing selection [11] has been presented and used for 0
I
0.4
the individuals selection. Multi branch selection ap-
proach is organized so that designs compete on one .o Percentage selected for the second objetive
of multiobjective. As the name implies, one branch 0.2
using the second branch selection
of the selection measure designs on the first objective,
the second branch evaluates individuals on the second 0.0 >
generation
objective and so on. To do this, the individuals are
selected randomly from the population of current gen- Fig.1 Parents selection strategy for two
eration on the first fitness using simulated annealing objectives optimization
selection. This continues until the numbers of selected
individuals come up to the fixed number. T h e n the
process is repeated with individuals competing on the 4. C o l l e c t i o n s o f t h e P a r e t o S o l u t i o n s
second objective. Selection process continues until in- Many solutions of multiobjective problem are mea-
dividuals selected on the last objective. sured during the run of the MOGAs. The goal of
In the multi-branch simulated annealing selection, multiobjective design is to find a range of Pareto solu-
there are certain guarantees about the selection pro- tions t h a t approximate the Pareto optimal set of prob-
Jun LI et al. Genetic Algorithms Development for Multiobjective Design Optimization of Compressor Cascade 161

lem. Therefore, any feasible, Pareto solutions should A 16 bits string is adopted as a binary coding for
be stored during the run of the MOGAs, not just the the variables representation. Probability of crossover
Pareto solutions from the last generation. In this way, is selected as 1.00 and probability of mutation works
a large number of solutions can be found that repre- at 0.02. The population size equals 100 and evolve 50
sent the Pareto optimal set [12]. generations. The results are shown in Fig.2. Fig.2(a)
The collection of these individuals begins with the shows the results which are obtained in the last gen-
starting generation. The Pareto solutions which are eration. The obtained Pareto solutions are all dis-
selected using Pareto criteria in the generation are tributed along the Pareto optimal set, but they can
stored in special list. In subsequent generation, the not sampled uniformly from the Pareto optimal set.
Pareto solutions are identified. These are then com- The results are obtained with collection Pareto solu-
pared with the stored list. If individuals currently in tions strategy and shown in Fig.2(b) and the Pareto
the list do not dominate a fresh identified individual, optimal set of the test function is almost uniformly
this fresh individual is added to the list. Similarly, covered by the set of solutions found.
if the fresh identified individual dominate those cur-
rently in the list, the dominated individuals are re- 10 - c ~ ' ~ ~ ~ _
moved from the list. In this manner, all Pareto solu-
tions ever encountered during the run of the MOGAs 0.8
are stored in this special list.
0.6

MULTIOBJECTIVE OPTIMIZATION
0.4

The multiobjective problems were selected to


0.2
demonstrate the capabilities of the present MOGAs.
The mathematical test functions have been tested,
0 , 0 . . . . I . . . . I , , , l I o , , [ I . . . .
then compressor cascade optimization design using 0.0 0.2 0.4 0.6 0.8 1.0
MOGAs has been presented in this section. L
(a)
1. M a t h e m a t i c a l T e s t F u n c t i o n s
1.O
Multiobjective optimization method with better
performance should have the following two charac-
0.8
teristics. The first is that the solutions obtained are
Pareto solutions. The second is that they are uni-
formly sampled from the Pareto optimal set [13]. 0.6

The use of mathematical test functions for perfor-


mance evaluation of the algorithms, in the case of mul- 0.4
tiobjective optimization, is not as clearly established
as for single objective optimization. As an example, 0.2
let us consider the problem of the maximization of
mathematical test function which is taken from the 0.0
Ref.[14].
0.0 0.2 0.4 0.6 0.8 ]5
fl
Maximize: f l = x, f2 = Y
(b)
Subject to: x 2 + y2 < 1 and 0 < x < 1,
0_<y_<l Fig.2 Results of the test function
Multiple optimization seeks to optimize the com- with two objectives
ponents of a vector valued objective function. The
Pareto optimal set of this test case is a quarter arc of In order to increase the difficulty of the problem,
the circle X 2 -~- y2 = 1 at 0 < x < 1, 0 < y < 1. let us extend above test function into three objective
162 Journal of Thermal Science, Vol. 8, No.3, 1999

problem, T h a t is, a third objective f3 has been added the larger flow turning angle can cause flow separation
in the test function. In this way, the dimensions of the and corresponding larger total pressure loss.
search space are increased by one order of magnitude. The present multiobjective problem can be de-
Maximize: ft =x, f2=Y, fa=z scribed as follows:
Subject to: x 2 q_ y2 q_ z 2 < 1 and 0 < x < 1, Function 1: Maximize pressure rise (outlet pressure
0_<y_<l, 0<z<l by inlet pressure)
T h e Pareto optimal set of.this test case is a 1/8 Function 2: Minimize total pressure loss
spherical surface of the sphere x 2 + y2 + z 2 = 1 at Function 3: Maximize flow turning angle
O<x<l,O<_y<_l,0<z<l. Subject to: De hailer number to be greater t h a n
T h e same parameters of the M O G A s for above test 0.72 (outlet flow velocity by inlet flow velocity) Maxi-
function are used for this three objective function. m u m airfoil thickness to be the same as t h a t of CDA [91
But, compare to the previous case, population size To obtain the values of the objective function of
was increased to 200 individuals and evolve 80 gen- compressor cascade, the Euler equations solver has
erations, due to the larger search space of the prob- been applied to simulate the cascade flow field. The
lem. Fig.3 illustrates the results concerning the set of C - t y p e grid is used, where 151 points are used in the
Pareto optimal obtained. It is apparent t h a t the large streamwise direction and 25 points are used in the di-
range of the exact Pareto optimal set is covered by rection normal to the airfoil surface.
the obtained solutions. The results of the above test The given flow condition is set to inlet Mach number
functions indicate the feasibility and excellent search of 0.25, inlet flow angle of 40.0 deg., blade stagger
performance for the Pareto solutions of the present angle of 14.4 deg., and the pitch of cascade of 0.5988,
MOGAs. similar to Ref.[9]. The present numerical m e t h o d can

1.0

0.8

0.6

1.0 ) OA
0.2
0.8
0.0
0.6 1.0
,q 0.8
O.z i!.6
O. +0.4 v_ % , .+L(':".~ ~ 11)
r ~ ; % . ..... ~ 08
0 ,~0.2
0.0 0.2 0.4 0.6 0.8 1.0 u.~ \ ~ - ~ 0.4 _
l

Fig.3 Results of the test function with three objectives from different direction

2. Compressor Cascade Optimization Design accurately obtain the pressure coefficients of the CDA
Goal of the compressor cascade design is to produce airfoil and agree with the experiment [9]. The B-spline
the highest pressure rise at the lowest total pressure function has been used for representation of compres-
loss and maximized the flow turning angle in a given sor cascade shape. In this work, 13 control points are
flow condition. Flow turning angle is an important pa- adopted to change the pressure surface and the other
r a m e t e r in the design process. In general, the pressure 13 points for the suction surface. The control points
rise increase as the flow turning angle increase. But, for leading edge and trailing edge do not change dur-
there is a limitation of the flow turning angle because ing the optimization process. We can only change y
Jun LI et al. Genetic Algorithms Development for Multiobjective Design Optimization of Compressor Cascade 163

direction coordinate value of the control points and Fig.5 shows variations of airfoil geometry shape of
keep x direction coordinates constant. Pareto solutions. To increase pressure rise, the airfoil
In this optimization process, the 60 individuals are tends to have a smaller curvature. However, in or-
evolved 60 generations. The optimization experiment der to increase flow turning angle, the airfoil should
work on the Tempest2 Work Station which CPU is has a larger curvature. In addition, during numeri-
the DEC Alpha AXP21164A. The convergence crite- cal analysis of cascade performance, generation of the
rion is that the fitness do not improve for 5 genera- C-type grid for cascade restrict the limitation of vari-
tions. Fig.4 shows that the Pareto solutions of the ables and also limit the search space. Above reasons
present MOGAs after 45 generations. Each axis is prove the difficulty in meeting both of these objectives.
linearly scaled according to the maximum and mini- Table 1 summarizes the performance of airfoil designs
mum fitness values. From Fig.4(b), with the pressure including the CDA performance. Fig.6 shows the cor-
rise increase, the total pressure loss also decrease in responding performance plane. The Pareto solution
some extent. However, when the pressure rise reach with the highest pressure rise is found to give better
the fixed high value, the total pressure loss will in- performance than CDA in all three objectives. Fig.7
crease with the pressure rise increase. We can see the illustrates the airfoil that has relatively high pressure
pressure rise versus flow turning angle in Fig.4(c), so, rise and relatively low total pressure loss. These re-
it is very difficult to increase both fitness values. sults confirm that the present MOGAs is able to
find improved designs.

1.0
//• '~ I 0.8
1.0
0.6
1.O-i ~ o~
,, .

0.81 / '~'° oo° ~0.4 "q . 0.6


~ 0O OQ
0.6 ~-o.2 Y< " ""
I ) .o--. ~o.o
0.4 4 / ~,oo,eDo g].o 0.4I • ..._eeDo
~_
O. 2t / -% 'oo ~ 6 4 < " - g 0.2 ' @0 ,,~ ,
0 . ~ 0-2 0.0 .............. I .... I -
0.0 0.2 0.4 0.6 0,8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
f~ f, (P2/P, )

t.0 1.O
°l
0.8 0.8

,-. 0.6
c~
v
a% 0.6
IP i
0.4 @ 0.4 e~

0.2 0.2
• ,,~,,~ t,o o-g-. ~ *~ Q*
0.0 i i i i i . . . . i . . . . i . . . . i , , , ~
w 0 . 0 - , , , ' . . . . , . . . . , , , , , i , , , ,

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
L (~/P,)

Fig.4 Pareto solutions in the objective space


164 Journal of Thermal Science, Vol. 8, No.3, 1999

T a b l e 1 Comparison of cascade airfoil performance


CASE P2/P1 w B(deg.)
CDA 1.0193 0.0331 38.0
Highest P2/P1 1.0199 0.0288 39.8
Lowest w 1.0196 0.0282 39.6
Highest B 1.0185 0.0396 46.3
Higher P~/P1 & lower w 1.0198 0.0284 41.2

High P2/Pl Point ( P2/Px )

I ] I :.". ......... Lower w


I : I '~h - ' - - - H i g h e s t B
1 ~ ~ - - - --- Higher Pz/PI
I ~ }~ ',',', ~, ~wer W
Low w

oint (B)

Point (W

-0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2


X/CL

F i g . 5 Variation of Pareto solutions F i g . 6 Performance plane of Pareto solutions


compared with t h a t of CDA

1.0

0.5

0.0

-0.5 CDA
o Higher t:'2/P1 & Lower w

--].0 , , L L I L L , , I , , , , I , , , , I , , , ;
0.0 0.2 0.4 0.6 0.8 1.0
X/CL

F i g . 7 Comparison of higher P2/P1 & lower w with CDA

B a s e d on t h e m u l t i - b r a n c h s i m u l a t e d a n n e a l i n g se-
CONCLUSIONS
l e c t i o n a n d collection of P a r e t o s o l u t i o n s s t r a t e g y , t h e
As a r e s u l t of t h i s effort, several conclusions have MOGAs have b e e n d e v e l o p e d in t h i s p a p e r . The
b e e n m a d e on t h e M O G A s for m u l t i p l e o b j e c t i v e p r o b - p r e s e n t M O G A s c a n b e d e m o n s t r a t e d as t h e useful
lem. t o o l s for t h e m u l t i p l e o b j e c t i v e o p t i m i z a t i o n p r o b -
Jun LI et al. Genetic Algorithms Development for Multiobjective Design Optimization of Compressor Cascade 165

lems. The Pareto optimal set approximated for the [3] Goldberg, D.E., "Genetic Algorithms in Search, Op-
test functions were fairly good in comparison to the timization, and Machine Learning," Addison-Wesley,
analytical results. In addition, compressor cascade (1989).
multiobjective design problem also proved their op- [4] Quagliarella, D., Vicini A., "Coupling Genetic Algo-
timization performance. rithms and Gradient Based Optimization Techniques, in
Quagliarella, D. et al., editors, Genetic Algorithms and
This work has illustrated that the multi-branch sim-
Evolution Strategies in Engineering and Computer Sci-
ulated annealing selection operator in the MOGAs can
ence, John Wiley & Sons Ltd., UK, pp.289-309, (1997).
provide a practical and successful means for multiob-
[5] Obayashi, S., "Aerodynamic Optimization with Evolu-
jective design. This method can provide the same se- tionary Algorithms," Inverse Design and Optimisation
lection pressure for each objectives of design problem Methods, Von Karman Institute Lecture Series, (1997).
and can keep diversity of the population. Coupling [6] Obayashi, S., Takahashi, S. and Takeguchi, Y., Nich-
parent selection strategy, this selection method can ing and Elitist Models for MOGAs, in A.E. Eiben, M.
efficiently improve optimization performance of the al- Schoenauer, and H.- P. Schuefel, editors, Parallel Prob-
gorithms for multiobjective designs. lem Solving From Nature-PPSNV, Amsterdam, Hol-
Collection of the Pareto solutions strategy play an land, Springer- Verlag, pp.260-269, (1998).
important role for the Pareto solutions of the multi- [7] Vicini A., Quagliarella, D., "Airfoil and Wing Design
Through Hybrid Optimization Strategies," AIAA paper
objective problems. The results shows that obtained
98-2729, (1998).
Pareto solutions can uniformly covered the analyti-
[8] Obayashi, S., Tsukahara, T. and Nakamura, T., "Cas-
cal Pareto optimal set for the mathematical test func-
cade Airfoil Design by Multiobjective Genetic Al-
tions. gorithms," Second International Conference on Ge-
The solutions of the compressor cascade design sug- netic Algorithms in Engineering Systems: Innovations
gests that it is difficult to design an airfoil satisfying and Applications, IEE Conference Publication No.446,
high pressure rise and high flow turning angle. From Glasgow, UK, (1997).
the Pareto solutions computed, the decision maker is [9] Elazar, Y. and Shreeve, R.P., Viscous Flow in a Con-
able to find a design that meet his design goal best. trolled Diffusion Compressor Cascade with Increasing
Incidence," Journal of Turbomachinery, 112, pp.256-
Acknowledgements 266, (1990).
The authors would like to thank Professor [10] Michalewicz, Z., Genetic Algorithms -t- Data Structures
OBAYASHI, Shigeru who provided many references = Evolution Programs, Springer-Verlag, Berlin, (1992).
[11] LI, J. and Satofuka, N., "Optimum Aerodynamic
on the multiobjective GAs applications of his research
Design Using Whole Annealing Genetic Algorithms,"
achievements.
Computational Fluid Dynamics Journal, 8, No.2, Spe-
The first author are grateful for Professor MAT-
cial Issue, (1999).
SUMOTO, Tsuguo for supporting him to work in the [12] William A. Crossley, Andrea M. Cook, David W. Fan-
Venture Laboratory of Kyoto Institute of Technology. joy, Vipperla B. Venkayya, "Using the Two-Branch
Tournament Genetic Algorithm for Multiobjective De-
REFERENCES sign," AIAA Journal, 37, No.2, pp.261-267, (1999).
[13] Obayashi, S., Aerodynamic Inverse Optimisation Prob-
[1] Hobbs, D.E. and Weingold, H.D., "Development of lems (chapter 9), in Zalzala, A.M.S. and Fleming, P.J.
Controlled Diffusion Airfoils for Multistage Compressor Eds., Genetic Algorithms in Engineering Systems, IEE,
Application," Journal of Engineering for Gas Turbines London, pp.203-228, (1997).
and Power, 106, pp.271-278, (1984). [14] Obayashi, S., Takahashi, S., Fejtek, Ian, "Transonic
[2] Dunker, R., et al., "Redesign and Performance Analy- Wing Design by Inverse Optimization Using MOGA,"
sis of a Transonic Axial Compressor Stator and Equiva- Proceedings of the Sixth Annual Conference of the
lent Plane Cascades with Subsonic Controlled Diffusion Computational Fluid Dynamics Society of Canada,
Blades," Journal of Engineering for Gas Turbines and Quebec, pp.II-41-46, (1998).
Power, 106, pp.279-287, (1984).

Вам также может понравиться