You are on page 1of 12

The propeller optimization with the use of

CFD
Cite as: AIP Conference Proceedings 2027, 030124 (2018); https://doi.org/10.1063/1.5065218
Published Online: 02 November 2018

A. V. Lysenkov

ARTICLES YOU MAY BE INTERESTED IN

Marine propeller modeling and performance analysis using CFD tools


AIP Conference Proceedings 2121, 040012 (2019); https://doi.org/10.1063/1.5115883

Aerodynamic characteristics of a quadcopter with propellers


AIP Conference Proceedings 2351, 040053 (2021); https://doi.org/10.1063/5.0051951

Efficiency study: Contra-rotating propeller system


AIP Conference Proceedings 1980, 060001 (2018); https://doi.org/10.1063/1.5044369

AIP Conference Proceedings 2027, 030124 (2018); https://doi.org/10.1063/1.5065218 2027, 030124

© 2018 Author(s).
The Propeller Optimization with the Use of CFD
A.V. Lysenkov

The Central Aerohydrodynamic Institute named after N.E. Zhukovsky


1 Zhukovsky Street, Zhukovsky, Moscow Region, 140180, Russian Federation

Corresponding author: lysenkov@tsagi.ru

Abstract. Description of the EWT-TsAGI solver modification for a flow simulation around rotating objects is given in the
paper. Explicit and implicit computational schemes, boundary conditions, object families rotating with different frequencies
are modified. New methods are implemented in a computer program EWT-RoS. The modified method verification and
validation for different propellers types are carried out. An example of developed program usage for an optimization of a
propeller aerodynamic characteristics of turboprop engine with acoustic characteristics taken into account is described.
Efficient global optimization algorithm (EGO) with constrains of first and second order is used for optimization.

INTRODUCTION
To increase the aircraft fuel efficiency, around the world, one returns to investigations propellers as engines.
Turboprop engines (TPE), in comparison with turbojet engines (TJE) have greater fuel efficiency. The greatest gains
from their use and their highest fuel efficiency are achieved at flight velocities that correspond to a value of
M ~ 0.5-0.6. To design effective propellers, it is necessary to develop methods for calculating their characteristics.
The main aerodynamic characteristics of the propeller are thrust coefficient, power coefficient and efficiency factor
[1]. Experimental investigations permit to obtain exact values of these characteristics. Numerical calculations allow
to consider a large number of variants of the propeller blade geometry and to evaluate the influence of various
parameters on the considered characteristics.
One of the most common methods for the rapid calculation of the propeller aerodynamic characteristics is a two-
dimensional linear vortex theory. Its founder is N.E. Zhukovsky [1]. On the basis of this theory, several methods that
are used to calculate and design the propellers have been developed. One example is the A.N. Kishalov’s method [2].
This method permits to determine the aerodynamic loads in the blade sections without taking into account the
deformations and oscillations of the blades. The main advantage of this method is the calculation speed and the
accuracy that is acceptable for preliminary design. Modern three-dimensional calculation methods based on the
Reynolds equation solution, permit to investigate both the characteristics of an isolated propeller and the mutual
influence of the propeller and plane elements [3].
TsAGI has developed EWT-TsAGI application package [4,5], which is successfully used to solve practical
problems for a long time. The scheme used in the solvers is based on the Godunov-Kolgan-Rodionov scheme [6–8].
Using EWT-TsAGI application package, the external flow around the aircraft and its elements, as well as internal
flows in the different engine ducts are investigated. However, the solvers in EWT-TsAGI permit to calculate the flow
around the immovable objects only.
The purpose of this paper is a modification of the solver to calculate the viscous gas flow around the rotating
objects. As it basis, Zeus solver [5,9,10] from the EWT-TsAGI application package has been used. An explicit and
implicit computational scheme, boundary conditions are modified and families are introduced to calculate the objects
rotating with different speeds. The developed modifications are realized as EWT-RoS computer program [11]. After
the development, the numerical method is validated and validated. Thus, the developed solver extends the capabilities
of EWT-TsAGI application package. It can be used to calculate the aerodynamic characteristics of propellers (both
single-row and coaxial), compressors, fans, and so on.

International Conference on the Methods of Aerophysical Research (ICMAR 2018)


AIP Conf. Proc. 2027, 030124-1–030124-11; https://doi.org/10.1063/1.5065218
Published by AIP Publishing. 978-0-7354-1747-2/$30.00

030124-1
Now one of the main fields numerical methods using is to optimize the characteristics of aircraft and its elements
[12–14]. In the current paper, one describe an example of the RoS solver using for the propeller optimization with
taking into account the acoustic characteristics. Using an automatic rebuilding of computational grids, an optimization
cycle is developed. This cycle takes into account the constrains for integral aerodynamic and acoustic parameters. As
an optimization algorithm, the EGO algorithm [16], which has proved itself well for aerodynamics [15], is used. This
algorithm is realized in the free-distributed cross-platform software with open source Dakota code developed by
Sandia National Laboratories [17].

DEVELOPMENT OF THE SOLVER IN THE ROTATING COORDINATE SYSTEM

Numerical Method
The Reynolds equation system with the SST turbulence model in a rotating coordinate system [18] can be written
as follows:
& &
wU r wF &
dt
 ¦ wxi
Wr (1)

ª º
ªU º «
Q
»
« Uu » « Qu r  psx  I xn »
« r» « »
« Uv r » « Qv r  ps y  I yn »
& « » & « »
Ur « Uwr » F «
Qwr  psz  I zn
»
« UE » «Q( Er  p / U )  ( I xnur  I ynvr  I zn wr )  4 n  Tnk »
« r» « »
« Uk » « Qk  Tnk »
« UZ » « »
¬ ¼ «¬ Qu r  Tn w
»¼
ª 0 º
« & & & »
«  U (2(: y wr  : z vr )  (: u (: u R)) x ) »
& &
« U (2(: w  : u )  (: u (: u R)) )» &
& « x r z r
& & &
y
»
Wr «  U (2(: x vr  : y u r )  (: u (: u R)) y ) »
« »
« 0 »
« S (k ) »
« »
« S (Z ) »
¬ ¼

Q U uSx  vS y  wS z
&
where : – rotation frequency:

ª: x º ªu r º
& « » & & « »
: «: y » : a , ur «vr » – velocity in a rotating coordinate system,
«: » «¬ wr »¼
¬ z¼
& & & &
a – rotation axis, Er E  Uu& ˜ (: u R) – energy with rotation taking into account, R – radius vector of a point.
& &
In this system, the value F is analogous to the value F from the Reynolds equation system in a fixed coordinate
&
system. Additional members appear in the source Wr . These terms are responsible for the centrifugal force and the
Coriolis force. For the system (1), one can write the scheme and solve it.

030124-2
& & & &
In the current paper, the system is proposed to solve for absolute velocity [3]. For that, ur u  : u R is substituted
into the system (1). Then the system has the following form:

ªU º ª 0 º
« Uu » «  U (: w  : v ) »
« » « y z »
& & « Uv » «  U ( : x w  : z v ) »
wU wFr & & « » & & & & & & « »
dt
 ¦ wxi
Wrc ; U « Uw » ; Fr F  (: u R)U ; Wrc «  U (: x v  : y u ) » (2)
« UE » « »
« 0 »
« »
« Uk » « S (k ) »
« UZ» « »
¬ ¼ ¬« S (Z ) ¼»

In the system (2), additional terms are added into the calculation of the fluxes associated with the flow due to
&
coordinate system rotation. Coriolis force correction is introduced in the sources Wrc .
When rotating around the X axis:

:x : , :y 0 , :z 0

The developed solver is a modification of the Zeus solver in the EWT-TsAGI package [5,9,10]. Therefore, the
algorithms realized in this solver have been taken as a basis. Approximation of convective flows for the developed
solver is similar to the case without rotation (i.e. to the basic solver).
The explicit and implicit schemes realized in the EWT-TsAGI application package [9,19] have been chosen as a
basis for implementing the new method. For the calculation in a rotating coordinate system, the solution of the
Riemann problem about the arbitrary discontinuity decay is modified. The modification of the original method [6]
consists in comparison of wave velocities with the velocity of the cell side to choose the solution type.
The boundary conditions are modified in the solver:
x in the case of the non-slipping heat-insulated wall boundary condition «solid_insulated», the rotation velocity
is given at the wall
& & &
x in the case of solid wall (slip wall) u u flow  u side
x in the case of soft boundary condition of free flow «riemann», rotation velocity is taken into account to
determine the flow direction.
For the explicit scheme, a correction is made for the time step, and for the implicit scheme, the smoother matrices
are modified.
The global time stepping provides a more accurate wave propagation velocity, but requires more time for
calculation. It was shown in [9] that it is possible to use a local time stepping to obtain a stationary solution. In the
case of a calculation in a rotating coordinate system, it is impossible to use the local stepping in a whole computational
domain, because, in this case, different cells on the blade surface will rotate with different velocities.
In the current paper, as in [9], it is proposed to use a combined approach. It is necessary to use a global time
stepping near the blades and a local step in far zones. The criterion to choose the time stepping type is the time step
restriction value for the explicit scheme. The cells near the blades have a small size and, therefore, they have a small
time step. If the time step is larger by a certain number of times than the minimum time step, then the calculation in
this cell goes with a local time step. Thus, the scheme is a combined explicit-implicit scheme with local and global
time stepping.
The next modification is connected with the necessity to obtain a solution for objects rotating at different velocities.
A special case of such problems is the calculation of the propeller in combination with a plane. The task is divided
into several zones. Some of them are calculated in a fixed coordinate system and the others are calculated in a rotating
coordinate system. To realize this opportunity, each block belongs to a certain family, which is attributed at the stage
of a computational grid generation. In the configuration file, the rotational velocity is given for each family (it can be
zero and can have different signs depending on the rotation direction). In cells where the rotation frequency is zero, a
local time stepping is always used.

030124-3
Method Testing
As a verification, convergence by Richardson is considered. The calculation is performed on nested calculation
grids (detailed, middle and coarse), which differ twice by the cell number in each direction. The total number of cells
differs in 8 times. The dependence of the thrust coefficient α upon the function 1 / N 2 / 3 , where N is the number of
cells, is shown in Fig. 1. In these axes, the graph should be a linear function in the case of second approximation order
in space

T 2SM D V
D ;E ;K O; O (3)
Un 2 D 4 Un 2 D 5 E nD

The resulting dependence (Fig. 1) shows that one formally obtains the convergence order greater than the second.
In addition, it can be seen that the results for the middle and detailed grids differ by less than 2%. Thus, one can use
the middle calculation grid for practical work.

0,0086
α
0,0084

0,0082

0,008

0,0078

0,0076

0,0074

0,0072

0,007
1/ܰ2/3
0 0,00005 0,0001 0,00015

FIGURE 1. Extrapolation by Richardson

FIGURE 2. The propeller geometry and the problem FIGURE 3. Block structure of the calculation grid
formulation

030124-4
For validation, following test case is selected: the flow around a rotating classical six-blade propeller with a small
saber, as shown in Fig. 2. The flow direction coincides with the rotation axis. The propeller chosen for testing is
typical, it is designed for aircraft cruise flight with a velocity of ~ 450 m / s. As a result of calculations, it is necessary
to obtain the characteristics of thrust α, moment β and propeller efficiency η (3).
The mathematical model has the main elements of the propeller: the blade and the shaft. All six blades are
considered. The size of the computational domain is ~ 8 diameters of propeller. The domain is divided into two
subregions: rotating and stationary. To exclude the shaft influence on the propeller, the semi-infinite shaft is given.
The boundary conditions are presented in the table Table 1.
TABLE 1. Boundary conditions
Name Description Surfaces of a geometric model
Solid Solid wall with integration of forces Propeller blades
Wall Solid wall Shaft
Curve Degenerated side On the rotation axis
Riemann Weakly reflecting boundary condition Outer boundary

The structured calculation grid, which topology is shown in Fig. 3, (O-topology around the blade), consists of
240 blocks and 1.387 million cells.
Comparison of the results obtained with the use of the developed solver with the results of the commercial software
and with the experiment is shown in Fig. 4 - 5. The calculation accuracy at the regimes without flow separation
(λ = 0.2 ÷ 0.8) is 3-4% for all solvers, as it can be seen in Fig. 4, a. Differences in solutions appear at large and small
values of the propeller step (Fig. 5). The accuracy of determining the power coefficient functions (Fig. 4, b) is similar.

(a) (b)
FIGURE 4. The dependence of (a) the thrust coefficient α and (b) the power coefficient β upon the propeller step

FIGURE 5. The thrust coefficient at small values of the propeller step

030124-5
OPTIMIZATION OF THE PROPELLER BLADE

Method for estimation of the acoustic characteristics of propeller


Empirical methods were widely used in engineering calculations of propeller noise. They are based on the
generalization of the noise measurement results for full-scale and model propellers. In calculations, total
characteristics of the propeller are used: thrust, power consumption, propeller rotary speed, diameter and number of
propeller blades.
In the calculation of the noise on the ground, the total noise level at the observation point located on the ground is
determined by the formula:

L C N lg N B  CM M u  CD lg kDB  CS lg S  C1 (4)
¦

where N B – power to the propeller, k – number of blades, DB – propeller diameter, C1 – constant, depending on the
YDв
aircraft class and the propeller type, S – distance from the sound source to the observation point. M u –
2a
circumferential end Mach number, where Z – rotation frequency, a – sound speed, CN, CM, CD, CS – empirical
constants.
In the framework of the current paper (see section 2.3), the propeller geometric parameters in formula (4), which
influence on the power N B to the propeller, will change. But the optimization algorithm (see section 2.2) is developed
for a general formulation. I.e. when other geometric parameters (propeller diameter, propeller rotary speed, etc.)
change, noise will be affected by other terms.

Description of the optimization cycle

To find the geometry with optimal characteristics, a parametric model that includes 25 parameters is developed:
the propeller diameter, the parameters of the 4 profiles (installation angle, vertical offset, 4 parameters of shape: 2
parameters per top surface and 2 parameters per bottom surface). The profiles change in 4 sections of the propeller
r
blade: 0.25, 0.5, 0.75, 1 (Fig. 6), where r is the distance from the propeller rotation axis. The parameters for
D
changing the shape of the profiles are the coefficients of the Bézier surface.

FIGURE 6. Sections for geometry parametrization

Following designations have been introduced for the parameters:


ai – installation angles of the ith profile (i=1..4)
l_lbase – the relative diameter of the propeller
cuij – downwind surface of the blade, the jth parameter of the Bézier surface in the ith section (j=1..2, i=1..4)
The relative propeller diameter (l_lbase) is equal to one for the base propeller. For the other parameters, the zero
values correspond to the basic geometry of the propeller blade.

030124-6
Let's consider the variation ranges of the selected parameters:
The relative change in the propeller diameter (relative to the base one): l_lbase = 0.95 ÷ 1.05.
The installation angles of the blade profiles: a1, a2, a3, a4. The variation range of each angle is from -2q to +2q.
Such a small range of angles is explained by the fact that the basic blade geometry was previously designed using the
numerical method for calculating the aerodynamic characteristics of propellers according to the vortex theory [2].
The coefficients of geometry variation: cu21 – the upper surface of the 2nd section, cb21 – the lower surface of the
2 section, cu31, cb31 – similarly for the 3rd section. The range of each coefficient varies from -0.3 to 0.3. A positive
nd

coefficient means an increase in the profile thickness, a negative coefficient means a decrease in its thickness.
After choosing the parameters, it is necessary to rebuild the mathematical model. For this, the surface calculation
grid is rebuilt and then the space grid is rebuilt. To reconstruct the surface grid, an algorithm for automatic regeneration
of the mesh has been developed. Regeneration is performed according to the following algorithm:
x Using the parameters, the Bézier surfaces are generated for independent variation of the windward and
downwind blade surfaces. The generation is performed using the OpenCASCADE library [21];
x Using the generated surfaces, for each node of the mesh, the change in its coordinates is determined;
x Using known angles of sections, the nodes of the surface calculation grid are rotated by the corresponding
angle;
x The calculation grid blocks that belong to the model surface (they are highlighted in red, Fig. 7), deform
together with the surface;
x In the calculation grid blocks of the next layer (they are highlighted in green, Fig. 7), the outer side remains
constant, and the inner face is deformed. The lateral sides are deformed linearly.
As a result of this procedure, minimal variations of the calculated grid are performed to minimize the influence of
the calculated grid variation on the calculation accuracy. It is should be noted that to use such an algorithm it is
necessary to build a calculation grid with taking into account the maximum possible variations of the geometry.
For optimization, a rare calculation grid is used: the cell number of the space grid is 700 thousands, 338 blocks.
The calculation grid is developed and deformed for single blade. Then the grid is copied using its own automatic
means.

FIGURE 7. The block structure near the blade (the grid is thinned for visualization)

To realize a fully automatic approach, an optimizer program is necessary. In this paper, the free cross-platform
open-source Dakota software developed by Sandia National Laboratories [17] is used as an optimizer program. The
Dakota application package (Design Analysis Kit for Optimization and Terascale Applications) is a number of libraries
that allow to perform factor analysis, experiment planning, to solve optimization problems, to perform calibration,
and to evaluate the sensitivity and reliability of systems.
Effective global optimization (EGO) is a method of global optimization [16], which uses simulation models. At
each iteration of the method, an approximation of the objective function is created using the Gaussian process (GP).
For this, the points obtained directly from the calculation are used. GP permits to define a new vector of input
parameters together with the uncertainty of the objective function prediction at this point. The key idea of the EGO is
to maximize the Expected Improvement Function (EIF), which is defined as the expectation that any point in the

030124-7
parameter variation area will provide a better solution than the current best solution based on the expected values and
options predicted by the GP model.
The optimization can be performed with constrains. The constrains are of the first and second order. Constrains of
the first order are constrains that directly depend on optimization parameters. In the current problem, these are the
geometric parameters of the blades. Constrains of the second order are constrains that depend on the calculated
parameters. In aerodynamic problems, these are constrains on aerodynamic characteristics. For example, in this work,
during the optimization at cruise regime, the takeoff thrust is used as a constrain. The EGO algorithm implemented in
Dakota software allows to perform the optimization with constrains.
The final step in development of an optimization methodology is to link all the elements in a whole optimization
cycle. For this, programs written in scripting languages (BASH, Python) are used. The diagram of the optimization
cycle is shown in Fig. 8. The feature of the optimization cycle for propellers is that the installation angles of blades
differ at different regimes. Accordingly, the calculation grids will be different for each calculation regime.

Base geometry
calculation

Optimization Geometry Copy of mesh


Base mesh
algorithm parameters and fields

Rebuilding of
New surface
3D mesh rebuilding surface
grid
computational grid

Running
EWT on
regime 1
Running of EWT
New 3D mesh
prepocessing
Running
Convergence EWT on
checking regime 2
Calculation of
objective meaning
Optimal and 2-nd order Propeller
parameters constrains characteristic

FIGURE 8. The scheme of the optimization cycle

For approbation of the developed technology, optimization is performed by varying 9 parameters:


x The relative propeller diameter (l_lbase);
x Installation angles of 4 sections (ai);
x For the 2nd and 3rd sections of the blade, one parameter for the windward side and one parameter for downwind
side (cu21, cb21, cu31,cb31).
Since the problem of a minimum determination is solved, the function f 1 K is used, where K is the
normalized value of the propeller efficiency at cruise regime. The stage of the experiment planning for 9 parameters
is 55 iterations. The process of solving the optimization problem (Fig. 9) took 153 iterations. The optimum is found
at the 124th iteration.
Analysis of the optimization cycle (Fig. 9) shows that the algorithm converges. Immediately after the end of the
experiment planning stage (the 56th iteration), the improvement is obtained and the objective function scatter by
iterations is decreased. The constrains (Fig. 10) after almost 55 iterations are almost all fulfilled (the value of the
constrain is below 0).
To analyze the influence of parameters on the objective function, one-dimensional sections passing through the
optimal point are generated (Fig. 11). On the same graphs, the cross shows the optimal parameter value obtained. The
optimal value isn’t at the minimum for all parameters (for example, a4, l_lbase). This is because the optimization is
performed with constrains.
Using these sections, one can recognize the nature of the influence of the chosen parameters on the propeller
efficiency. If the objective function changes slightly, then the parameter does not affect and it can be excluded from

030124-8
further optimization. The degree of influence can also be seen on the graph of the efficiency versus the dimensionless
parameters (Fig. 11). Dimensionless parameters are obtained according to the formula:
p  pmin
p , where p – parameters, and pmax, pmin - its maximum and minimum values, respectively.
pmax  pmin

FIGURE 9. The process of optimization cycle convergence

(a) (b)
FIGURE 10. Dependence of (a) thrust and (b) noise constrains on iterations

FIGURE 11. Dependence of the efficiency upon dimensionless parameters

030124-9
The parameters cu21 and cu31 (parameters of shape of downwind side of variable sections) have the most influence
(see Fig. 11). Moreover, the influence of the parameters of the second and third sections has a different gradient sign.
In the case of the 2nd section, the efficiency increases with the thickness increase (the parameter is greater than 0, and
the dimensionless parameter is greater than 0.5). In the case of the 3rd section, on the contrary, the efficiency increases
with decreasing the thickness, the efficiency increases. The weakest effect is provided by the parameter a4 (the
installation angle of the section far from the axis).
In addition, the chart (Fig. 11) shows that all the dependencies have all global extremum, for some parameters it
is reached at the boundary. For such problems, the optimization methods work reliably.
To analyze the interrelation of various parameters, we can consider two-dimensional efficiency cross-sections that
pass through the optimum point (Fig. 12).

(a) (b)

(c) (d)
FIGURE 12. Two-dimensional section of the efficiency function, the marker corresponds to an optimum
a) parameters a1 and a2; b) parameters a3 and a4; c) parameters cu21 and cb21; d) parameters cu31 and cb31

CONCLUSION
A modification of the method for solving the Navier-Stokes equations, which is realized in EWT-TsAGI
application package and based on Godunov-Kolgan-Rodionov scheme, is proposed. The method permits obtain a
solution of the problem about the flow around the bodies in a rotating coordinate system. A special feature of the
method is the solution of Navier-Stokes equation system in a rotating coordinate system for flow velocities in an

030124-10
absolute coordinate system. A combined local-global time stepping is used, families are introduced for the calculation
with different rotational speeds. The solver has been tested. Testing has shown:
The test case of calculating the aerodynamic characteristics of a classic six-blade non-scimitar propeller has
demonstrated the accuracy of 3-4% for regimes without flow separation.
A computational technique for optimizing the aerodynamic characteristics of the propeller has been developed
taking into account the constrain of the second order: acoustic and thrust characteristics of the propeller. The full-scale
propeller optimization has been performed during the cruise flight regime with the constrains of the takeoff
performance: the propeller thrust is not less than base one thrust, the noise on the ground is not more than the base
one. During the optimization, the propeller efficiency at the cruise regime is increased by 0.35%. At that, the thrust at
the cruise regime and at the takeoff regime is increased, and the noise at the takeoff regime is the same practically.

ACKNOWLEDGMENTS
The author is grateful to Dr.Sc. V.A. Titarev, E.V. Kazhan and A.A. Saveliev for their help.
The proposed optimization methodology was developed within the framework of the state contract of the Ministry
of Industry and Trade of the Russian Federation No. 17112.1770290019.18.016 for the performance of the research
work "Development of a methodology for calculating the aerodynamic characteristics of an inlet and a fan of a large
thrust turbojet engine with simulating their gas-dynamic interaction and its application for optimizing the power plant
inlet", "Inlet-2" under the state program of the Russian Federation.

REFERENCES
1. V. L. Alexandrov, Propellers (State Publishing House of the Defense Industry, Moscow, 1951), 475 p.
2. V. G. Dmitriev, G. A. Pavlovets, A. F. Chevagin, A. P. Byrkin, A. K. Ivanyushkin, S. K. Kim, and
A. N. Kishalov, Studies on the development of perspective methods for calculating the aerodynamic
characteristics and for designing the propellers with taking into account the requirements for aerodynamic
efficiency and resource (Zhukovsky, 2005).
3. R. T. Biedron and J. L. Thomas, AIAA Paper 2009-1360 (2009).
4. V. Neyland, S. Bosniakov, S. Glazkov, A. Ivanov, S. Matyah, S. Mikhailov, and V. Vlasenko, Prog. Aerosp. Sci.
25, 1–25 (2001).
5. S. M. Bosnyakov, Proceedings of TsAGI 2735, (2015).
6. S. K. Godunov, Mathematical collection 47(89), 271–306 (1959).
7. V. P. Colgan, Scientific notes of TsAGI 3(6), 68–77 (1972).
8. A. V. Rodionov, Journal of Computational Mathematics and Mathematical Physics 27(4), (1987).
9. E. V. Kazhan, Scientific notes of TsAGI 43(6), 1–38 (2012).
10. S. V. Mikhailov, Certificate of computer program state registration No. 2013610172 (November 12, 2012).
11. A. V. Lysenkov, Certificate of computer program state registration No. 2016619700 (2016).
12. J. A. Samareh, AIAA Paper 2004-4630 (2004).
13. M. Zhang, A. Rizzi, P. Meng, R. Nangia, R. Amiree, and O. Amoignon, AIAA Paper 2012-5402 (2012).
14. M. L. Ryerkerk, R. C. Averill, K. Deb, and E. D. Goodman, AIAA Paper 2012-5569 (2012).
15. N. A. Zlenko, S. V. Mikhaylov, A. A. Saveliev, and A. V. Shenkin, Scientific notes of TsAGI 46(6), 20–38
(2015).
16. N. Bartoli, I. Kurek, R. Lafage, T. Lefebvre, and R. Priem, “Improvement of efficient global optimization with
mixture of experts : methodology developments and preliminary results,” in 17th AIAA/ISSMO Multidisciplinary
Analysis and Optimization Conference, (2016).
17. B. M. Adams, M. S. Ebeida, M. S. Eldred, J. D. Jakeman, L. P. Swiler, J. A. Stephens, D. M. Vigil, T. M. Wildey,
W. J. Bohnhoff, K. R. Dalbey, et al. Sandia Tech. Rep. SAND2014-4633 (2014).
18. C. Hirsch, Numerical Computation of Internal and External Flows, 2nd ed. (Elsevier, 2007), 696 p.
19. S. M. Bosnyakov, Proceedings of TsAGI 2671, (2007).
20. S. P. Ostroukhov, Aerodynamics of propellers and propeller-driven engines (FIZMATLIT, Moscow, 2014).
21. http://dev.opencascade.org

030124-11

You might also like