Structural and Multidisciplinary Optimization (2022) 65:306

https://doi.org/10.1007/s00158-022-03415-6

RESEARCH PAPER

A novel approach for obtaining optimum shape design

with the minimum drag coefficient
Bahador Abolpour1 · Ramtin Hekmatkhah2 · Rahim Shamsoddini3

Received: 5 July 2022 / Revised: 20 September 2022 / Accepted: 26 September 2022

© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022

Abstract
In the traditional methods, optimization and mesh generation in computational fluid dynamics are separate procedures,
which increases the time and calculation effort. This issue has been addressed by a shape optimization approach based on the
binary genetic algorithm coupled with the image-processing method for drag reduction purposes. A system-specified code
in MATLAB software is developed to find the optimized configuration. Accurate filters in the image-processing method are
used to detect the edge and corners of the imported obstacles in the flow field to generate suitable grids. A turbulent flow
field over obstacles was modeled by Chien’s low-Reynolds k − ε model and finite volume methods with stagger grids to solve
governing equations. Combining the binary genetic algorithm with the image-processing method is an effective method to
optimize shape configuration and mesh generation. The drag coefficients of the generated shapes are determined and the opti-
mum one is obtained. A good agreement is obtained by comparing the obtained results with wind tunnel experimental data.

Keywords Shape optimization · Binary genetic algorithm · Image processing · Turbulent flow · Finite volume method

List of symbols kwall Kinetic energy near the wall (Lyn et al.
C𝜇 , C𝜀1 , C𝜀2 Constants (Lyn et al. 1995): 1995): kwall = 0
C𝜇 = 0.09, C𝜀1 = 1.35, C𝜀2 = 1.8 L𝜀 The dissipation scale
Dpc Dot per centimeter P Mean pressure (Pa)
D, E Additional source terms (Lyn et al. 1995): P Average pressure (Pa)
D = 2v yk2 , E = −2v y𝜀2 exp − 0.5y+ Pk The rate of production of turbulence kinetic
f𝜇 , f1 , f2 The damping functions (Lyn et al. 1995): energy (Younis and Przulj 2006):
𝜕u
f𝜇 = 1 − exp − 0.0115y+ , f1 = 1, Pk = −u�i u�j 𝜕xi
j
( )2
Re Ret Reynolds number of turbulence flow (Lyn
f2 = 1 − 0.22exp − 6 t k2
et al. 1995): Ret = 𝜐𝜀
gi Acceleration caused by an external physical
S Reference area ­(m2)
force (m/s2)
Sui Source term
k The turbulent kinetic energy (J/kg)
Tu Relative turbulence intensity (Younis and
k0 The initial kinetic energy term
Przulj 2006): Tu = u t
u
0
u∗ Friction
√ velocity (Younis and Przulj 2006):
𝜏w
u∗ =
Responsible Editor: Emilio Carlos Nelli Silva 𝜌
ui , uj Components of the average velocity vector
* Bahador Abolpour (m/s): u, v
bahadorabolpor1364@yahoo.com u′i , u′j Components of the turbulent velocity
1
Department of Chemical Engineering, Sirjan University vector (m/s) u′ , v′
of Technology, Sirjan, Iran ui uj Reynolds stress components ­(m2/s2)
′ ′

2
Faculty of New Sciences and Technologies, University (Abolpour et al. 2017): u′ v′ , u′2 , v′2
of Tehran, North Kargar, Tehran, Iran y + Normal distance in wall coordinates (Lyn
∗
3
Department of Mechanical Engineering, Sirjan University et al. 1995):y+ = u𝜐y
of Technology, Sirjan, Iran

13
Vol.:(0123456789)
306 Page 2 of 16 B. Abolpour et al.

to select the first sample shape as an initial attempt. They

{ }
1 i=j
0 i≠j
𝛿ij Kronecker delta 𝛿ij = investigated developed models on certain airfoils and wings
in transonic flows. He et al. (2019) applied the Euler time
ε Turbulent kinetic energy dissipation rate (J/
spectral based on the computational fluid dynamics method
kg)
to investigate flutter limitations. They used the adjoint
ε0 Initial kinetic energy dissipation rate term
method to evaluate this limitation concerning the design
𝜀wall Kinetic energy dissipation rate (Lyn et al.
variables. The Krylov solver was utilized in their works to
1995): 𝜀wall = 0
solve coupled-adjoint equations. He et al. (2019) developed
μ The viscosity of the fluid (Pa s)
a new method to dominate the robustness problem of shape
μt Turbulent viscosity of the flow based on the
2 parametrization and trial and error in the flow solver require-
k-ε model (Pa s) 𝜇t = C𝜇 f𝜇 𝜌 k𝜀
ments of classic methods. His approach was based on the
μeff The effective viscosity of fluid (Bosch and
design optimization discussion group as a benchmark.
Rodi 1996): 𝜇eff = 𝜇t + 𝜇
Massively multipoint design issues were solved based on
υ Kinematic viscosity of the fluid ­(m2 s)
the surrogate-assisted gradient-based optimization method
ρ The density of the fluid (kg/m3)
by Li and Cai (2020). They found satisfactory shape design
𝜎k , 𝜎s Turbulent Prandtl numbers (Lyn et al.
with the aerodynamic functions, computed values for gra-
1995): 𝜎k = 1, 𝜎𝜀 = 1.3
dients, and a large number of shape design variables by
𝜏w Wall shear stress (Pa)
this model. A gradient-based algorithm is used in one loop
∇. u Divergence of u vector
and in the other loop gradient-enhanced kriging and proper
orthogonal decomposition kriging models were used to esti-
mate aerodynamic functions and gradients. Li and Zhang
1 Introduction (2020) investigated machine-learning techniques to enhance
the efficiency of surrogate-based optimization. They pre-
Shape optimization of any object exposed to fluid flows has
sented a productive network that was designed to self-learn-
been the focus of aerodynamic and hydrodynamic research-
ing all existing airfoils and wing characteristics to generate
ers for many years. Reducing the drag force decreases fuel
a new optimized shape configuration. Mangano and Martins
consumption and increases velocity for a means of better
(2019) developed a model based on the Reynolds averaged
transport. This is a major and vital purpose of these areas of
Navier–Stokes method for aerodynamic optimization in
research. Reduction of fuel consumption due to the reduc-
supersonic, transonic, and subsonic regimes. Drag minimi-
tion of drag force on objects, because of the presence of
zation of wing sections and wings was done by the MACH
surface shear force and pressure distribution from the fluid
framework approach with single- and multipoint optimiza-
flow, leads to cost savings and reduction of fuel consumption
tion in different flight velocity regimes. Ukken and Sivapra-
emissions. Furthermore, producing noise of different shapes
gasam (2019) presented an adjoint-based aerodynamic shape
and aerodynamics shock generation on the surface are extra
approach to estimate modified aerodynamic characteristics
issues that scientists and researchers deal with.
in the extremely low-Reynolds number regime. Three key
Many different methods and algorithms have been devel-
factors (drag coefficient, lift coefficient, and lift-to-drag
oped for the optimization of shapes in order to improve aer-
ratio) were investigated for optimization.
odynamics efficiencies by researchers, experimentally and
Zhang et al. (2020) developed an effective new optimi-
numerically (Chen et al. 2020; Han et al. 2020; He, et al.
zation design system according to the Reynolds averaged
2019; He, et al. 2019; Li and Cai 2020; Li et al. 2020; Man-
Navier–Stokes method. They selected an initial rotor air-
gano and Martins 2019; Ukken and Sivapragasam 2019; Yan
foil configuration and then optimized the rotor shape. Their
et al. 2019; Zhang et al. 2020). Chen et al. (2020) developed
experimental data demonstrated the reliable condition
a new method in shape design, called the unsteady adjoint
for optimized shape design of rotor airfoil. Among many
method, to improve the transonic fluid–structure interaction
approaches, one of the effective algorithms that have been
stability. The relation between the system stability condition
utilized by researchers for aerodynamics shape design is
and aerodynamic damping derivatives was studied with three
Genetic Algorithms (Chan et al. 2018; Daróczy et al. 2018;
different conditions (i.e., stable, critical, and unstable point).
Darwish, et al. 2018; Ebrahimi and Jahangirian 2017; Gao
The stability enhancement of aeroelastic characteristics and
et al. 2017; Ghalandari et al. 2019; Li 2020; Liu et al. 2017;
flow stability, and improving the control efficiency were
Qin et al. 2018; Saleem and Kim 2020; Şumnu et al. 2020;
investigated in their optimization results.
Wang and Zhao 2019; Yi et al. 2019). Chan et al. (2018)
Han et al. (2020) presented a new approach for aerody-
developed an optimization model based on the genetic
namic design, based on the multi-level hierarchical krig-
algorithm to enhance the power coefficient of the Savonius
ing theory. They used the design of experiment techniques
wind turbine. They combined computational fluid dynamics

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 3 of 16 306

simulations and genetic algorithms in iterative steps to merged the PARSEC method of airfoil and Xfoil to esti-
couple blade geometry and adopted mesh configuration mate the aerodynamic characteristics of airfoils. Wang and
to investigate the optimized shape of wind turbine blades. Zhao (2019) established a high-efficient optimization of
Daróczy et al. (2018) represent the optimization model of blade shape by coupling the Kriging model with a genetic
the H-Darrieus wind turbine using the genetic algorithm. An algorithm. The final shape of the blade was obtained by the
initial airfoil configuration was reshaped to achieve an opti- modified sequential quadratic programming method.
mal airfoil geometry. Consequently, the maximum energy The analysis of the aerodynamic, acoustic, and stealth
output and performance coefficient were obtained. characteristics of helicopter rotor blades in hover flight
Darwish et al. (2018) used the adjoint method to estimate based on the genetic algorithm was represented by Yi et al.
the optimized shape of the airfoil which was used in the (2019). The Farassat theory and the blade element momen-
blade according to the multi-objective genetic algorithms tum theory were used to simulate the acoustics and the
optimization method. Ebrahimi and Jahangirian (2017) used aerodynamic characteristics. Additionally, the airfoil coor-
a parallel genetic algorithm method to optimize the shape of dinated and stealth performance was studied to obtain the
airfoils. To enhance convergence rate and efficiency, differ- minimum radar cross section. Bhattarai et al. (2020) pre-
ent computational methods and a population dispersion tech- sented an adapted Delaunay triangulation method for free-
nique were applied. Their results showed that the method form surfaces for stochastic optimization applications. They
used was able to estimate optimum shape with considerably attempted to search the answer domain for optimal results
lower computational cost. Gao et al. (2017) coupled com- with minimum computational costs and provide a novel
putational fluid dynamics and genetic algorithm in a three methodology for lightweight computational fluid dynamic
dimension of diffusion section shape. Their studies provided geometry for optimization.
more uniform discharge and velocity distribution in both The present study aims to investigate a new approach
inflow and outflow conditions. for square obstacle shape optimization based on the binary
The usage of a hybrid three-dimension model, containing genetic algorithm in the turbulent flow field. This approach
the genetic algorithm, artificial neural networks, and experi- provides the facility of drag force reduction for any con-
mental setup for optimization of the first blade in the com- struction aerodynamically and hydrodynamically. Mesh
pressor was investigated by Ghalandari et al. (2019). Their generation has been done by the Image-processing method.
parametric studies included mass flow, efficiency, stress, The finite volume method over a staggered grid has been
eigen frequencies, taper, and sweep angle of the blades. A utilized to discretize governing hydrodynamic equations and
semi-analytical MATLAB code that contained Whitehead overcome pressure–velocity coupling issues. After detecting
and forces response theory to estimate flutter conditions obstacle edges and related corners, the mesh is generated in
was used in their study. Zhou Li et al. (2020) presented a the flow field. Chien’s Low-Reynolds k − ε model (Chien
multi-round shape optimization method according to the 1982) is used to simulate the turbulence flow field, and the
aerodynamics reduced model. Certain airfoils were selected drag coefficient is estimated for each newly generated shape
as initial geometries and optimization based on a genetic configuration. The optimization algorithm and image-pro-
algorithm was implemented in both subsonic and transonic cessing method have been developed by MATLAB software.
flow regimes. Furthermore, the effects of optimized shape on Comparing the present model with previous studies (see
shockwave strength and smooth aerodynamics were demon- Table 1) indicates special advantages of the present model,
strated in their study. Liu et al. (2017) incorporated the com- which are the wide range of shape modifications during the
putational fluid dynamics simulation and multi-objective optimization procedure and the applicability of this method
optimization to enhance the performance of the exchanger for different body shapes. The main reason for mentioned
based on the non-dominated sorting genetic algorithm. abilities in this method is utilizing image processing for
Their results have shown that heat transfer and flow resist- importing the modified shapes by the optimizer algorithm
ance increased, considerably. Huan et al. (2018) used an to the computational fluid dynamic code which is the major
advanced genetic algorithm to optimize auto-body. The mass novelty of the present work. In addition, using the genetic
minimization, static stiffness, dynamic eigen frequency, and algorithm (or other similar optimizer methods) gives the
manufacturing constraints were implemented as key con- artificial intelligence optimization ability to shape opti-
trol factors for the optimization process and design vari- mizer models based on the Pareto approaches (Abolpour
ables. Saleem and Kim (2020) optimized the aerodynamic et al. 2021, 2022).
characteristics of airfoil-based shells to increase the buoy- Attending to the main aim of this study, which is obtain-
ant airborne turbine system performance based on a genetic ing the optimum shape design using a generated MAT-
algorithm using MATLAB software. Their developed code LAB code for a coupled numerical solution based on the

13
306 Page 4 of 16 B. Abolpour et al.

Table 1  Comparing the present model abilities with similar previous studies
Turbulent Shape Fluid–struc- Three- Wide range of Applicable for Smart Multi-objective
flow CFD optimiza- ture interac- dimen- shape modifica- different shapes optimiza- optimization
tion tion sional tion tion ability

Chen et al. (2020) × × ×

Han et al. (2020) × ×
He et al. (2019) × ×
Li and Cai (2020) × × × ×
Li et al. (2020) × × ×
Mangano and Martins × × × ×
(2019)
Ukken and Sivapragasam × ×
(2019)
Yan et al. (2019) × × × ×
Zhang et al. (2020) × ×
Chan et al. (2018) × × × × ×
Daroczy et al. (2018) × × × × ×
Darwish et al. (2018) × × × ×
Ebrahimi and Jahangirian × × × ×
(2017)
Gao et al. (2017) × × × × ×
Ghalandari et al. (2019) × × × × ×
Zhou Li et al. (2020) × ×
Liu et al. (2017) × × × × ×
Qin et al. (2018) × × ×
Saleem and Kim (2020) × × × × ×
Sumnu et al. (2020) × × × × ×
Wang and Zhao (2019) × × × × ×
Yi et al. (2019) × × ×
Bhattarai et al. (2020) × × × × × ×
The present study × × × × × ×

computational fluid dynamics, image processing, and genetic A water flow has been considered over a 2 × 2 cm
optimizer algorithm, the computational cost of this method unshaped obstacle as an initial sample in the open-boundary
is notable, especially for three-dimensional engineering domain. The turbulent water flow enters the domain with
problems. In addition, utilizing the image-processing meth- constant uniform velocity. Also at the outlet, a constant pres-
ods for three-dimensional structures is different, difficult, sure gradient has been considered.
and needs special software that increases the complexity of
this method. 2.1 Governing equations

2 Methodology The two-dimensional governing conservation equations

based on Einstein’s notation have been illustrated as Eqs.
The combination of genetic algorithm and image-processing (1–5) in Cartesian Coordinates and assumed that the flow
method provides strong and beneficial tools to develop an is steady, turbulent, incompressible, and Newtonian fluid
optimization approach for industrial and construction usage. flow. The gravity direction has been considered in the z-axial
Additionally, shape optimization for drag force reduction is a direction, and in the present study, gravitational body force
crucial target for scientists and engineers to reduce structural has been omitted because of neutrality in the solution (Abol-
vibrations and fatigues in fluid–solid interactions also reduc- pour et al. 2017):
ing energy consumption. Figure 1 shows the configuration
𝜕ui
of the initial square obstacle and boundary condition of the = 0, (1)
computational domain. 𝜕xi

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 5 of 16 306

Fig. 1  Domain and boundary condition definition for square obstacle as an initial sample

[ ( ) ] [ ( ) ]
𝜕uj 𝜕ui 𝜕uj 𝜕 𝜕 𝜇 𝜕𝜀 )𝜀
𝜕 𝜌𝜀uj − 𝜇 + t
(
� � (𝜌𝜀) + = C𝜀1 f1 P − C𝜀2 f2 𝜌𝜀 + 𝜌E,
𝜌uj = −P𝛿ij + 𝜇 + − 𝜌ui uj + 𝜌gi , (2) 𝜕t 𝜕xj 𝜎𝜀 𝜕xj k
𝜕xj 𝜕xj 𝜕xj 𝜕xi
(7)
( ) which the steady-state forms of k and ε have been as fol-
𝜕 kui lows (Abolpour et al. 2022):
[( ) ]
𝜕 𝜇
𝜇+ t 𝜕k
(3)
� �
𝜌 = 𝜕xj
− 𝜌ui uj − 𝜌𝜀,
𝜕xi 𝜕xj 𝜎k [ ]
𝜕 ui k
[ ]
𝜕 vt 𝜕k
= v+ + Pk − 𝜀. (8)
( ) ( ) 𝜕xi 𝜕xi 𝜎k 𝜕xi
𝜕 𝜀ui 𝜕uj
[( ) ]
𝜕 𝜇 𝜀 𝜀2
𝜇+ t 𝜕𝜀 � �
𝜌 = 𝜕xj
+ C𝜀1 −𝜌ui uj − C𝜀2 𝜌 .
𝜕xi 𝜕xj 𝜎𝜀 k 𝜕xi k [ ]
𝜕 ui 𝜀
[ ]
𝜕 vt 𝜕𝜀 𝜀 𝜀2
(4) = v+ + C𝜀1 Pk − C𝜀2 . (9)
𝜕xi 𝜕xi 𝜎𝜀 𝜕xi k k
Additionally, the Boussinesq approximation has been
used as follows (Abolpour et al. 2017): The production term Pk in both of these √equations has
[ 𝜕u
]2
been calculated asPk = C𝜇 𝜀S2, where S = 𝜀k 12 𝜕xi + 𝜕xj .
( ) 𝜕u
� � 𝜕ui 𝜕uj 3
𝜌ui uj = 𝜇t + − 𝜌k𝛿ij . (5) j i

𝜕xj 𝜕xi 2 The modified form of the standard k-ε turbulent model has
been introduced by Kato and Launder (Kato 1993; Laun-
To solve the turbulent flow field, Chien’s low-Reynolds der and Sandham 2002). This model was developed by
k − ε model (Chien 1982) has been selected because of its replacing
√the strain stress
) term (S) with a vorticity term
high accuracy and acceptable numerical properties and the
(
[ 𝜕u
]2
in the kinetic energy equation
𝜕u
concept of the eddy viscosity of Boussinesq (Patel et al. Ω = 𝜀k 21 𝜕xi − 𝜕xj
j i

1985). The general form of this model is shown as follows:

Pk = C𝜇 𝜀S Ω . The velocity components involve two
( )

𝜕 𝜕
[ ( ) ]
𝜇t 𝜕k terms of stable and fluctuated parts (i.e., u = u + u� and
(𝜌k) + 𝜌ku j − 𝜇 + = Pk − 𝜌𝜀 − 𝜌D, (6) v = v + v� ). They substitute velocity components into
𝜕t 𝜕xj 𝜎k 𝜕xj
momentum equations, applying Reynolds averaging rules,

13
 306 Page 6 of 16 B. Abolpour et al.

and simplifying for a steady and incompressible flow calculation, automatically. This step increases the conver-
gives: gence rate and decreases the computation cost.

𝜕ui 𝜕uj
[ ( )]
𝜕 ( 𝜕 P̂ 𝜕
+ Sui , (10)
)
𝜌ui uj = − + 𝜇 +
𝜕xi 𝜕xj 𝜕xi eff 𝜕xj 𝜕xi
3 Numerical solution
𝜕𝜇eff
where 𝜇eff = 𝜇 + 𝜇t , P̂ = P − 13 𝜇eff ∇.u + 23 𝜌k and Sui = 𝜕xj
𝜕uj 𝜕𝜇eff 𝜕uj Figure 2 is a demonstration of a square obstacle as an initial
− .
𝜕xi 𝜕xi 𝜕xj shape. In the binary genetic algorithm that has been used
in this study, the first shape of the initiated obstacle was
2.2 Boundary conditions introduced as the first generated chromosome that contains
16 genes (numbered from 1 to 16).
The related boundary conditions of the computational field In the binary genetic algorithm approach, each gene can
are considered according to Fig. 1. At the inlet (x = 0), uni- select just values of 1 or 0. Number 1 is defined as a small
form velocity, kinetic, and dissipation rate distribution, have square that existed in the domain and 0 value is the definition
been concerned. In addition, a constant pressure gradient
along the x-axis has been taken into use.
𝜕P
U = U0 , V = 0 , k = k0 , 𝜀 = 𝜀0 , = 0 at x = 0. (11)
𝜕x
At the outlet (x = L), the gradient of V velocity along the
y-axis and the gradient of Ψ variables along the x-axis are
considered zero.

𝜕V 𝜕𝜓
= 0, = 0, 𝜓 = U, P, k, 𝜀 at x = L. (12)
𝜕y 𝜕x

On( the wall sides of the)obstacle, no slip, and no penetra-

tion Uwall = 0 , Vwall = 0 , and at the top and bottom of the
(computational field, )the open-boundary conditions
𝜕𝜓
𝜕y
= 0, 𝜓 = U, P, k, 𝜀 have been implemented (Grinstein
1994).
In the present study, the developed system-specified code
detects the downstream area behind the inserted obstacle and
sets all velocity values to zero as an initial value to run the

Fig. 2  The first generated chromosome as input for the binary genetic
algorithm contains 16 genes Fig. 3  The flow chart of the processing steps

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 7 of 16 306

Fig. 4  The detected edge and

corners of the obstacle and also
the generated mesh by image-
processing method, the blue
nodes demonstrate the obstacle,
the green and orange nodes are
I, J and i, j calculation nodes of
the staggered grids in the fluid
domain, respectively

of an empty area in the computation domain. To generate have been saved as digital gray-scale images for export
acceptable shapes, numbers 1–4 genes remain constant and to image processing.
have the value of 1 to protect the aspect ratio of the gener- • In the second step, each image is initiated the image pro-
ated obstacles. The binary genetic algorithm as an optimi- cessing. Our system specified written MATLAB code-
zation algorithm starts to process on the generated initial detecting corners and edges of each digital image. In the
chromosome in the first population. This population includes gray-scale form, the values upper than 200 for each color
twenty randomly produced chromosomes each chromosome tone are estimated to be the white color that represents
represents a distinct shape. Figure 3 illustrates the flowchart the body edge, and lower than this value shows the black
of the process sequence of the optimization algorithm. color which defines the null space in the domain. In addi-
According to this flowchart, the executing steps of the tion, for recognition of the corners, the specified method
presented method have been listed below: has been utilized. By considering one of the obstacle
nodes, the left, right, upper, and bottom neighbor nodes
• In the first step, the first population with twenty randomly indicate the location of the mentioned node of the body.
chromosomes (shapes) containing 16 genes, which can After detecting the corners and edges of the generated
select just values of 1 or 0 (number 1 is defined as a small obstacles, the mesh generation step has been done over
square that existed in the domain and 0 value is the defi- the flow field that is shown in Fig. 4. The finite volume
nition of an empty area in the computation domain) have method has been used to discretize governing equation
been generated by the genetic algorithm. These shapes and solve the turbulent flow field over obstacles as a bluff
body. To overcome the checker-board problem in the

13
306 Page 8 of 16 B. Abolpour et al.

Fig. 5  A comparison between the model predictions with the experimental measurement from velocity along the horizontal line passing the cen-
terline of the obstacle (Lyn et al. 1995; Lyn and Rodi 1994; McKillop and Durst 1985; Durao et al. 1988) (Uinlet = 0.535 m/s)

pressure and velocity field, the staggered grid has been flow, the drag force coefficient is estimated by the follow-
generated by the image processing. ing relation (Anderson 2010):
• In the next step, the generated calculation mesh by the   The drag coefficient of each generated population
Image processing has been imported to the computa- by the binary genetic algorithm has been calculated
tional fluid dynamic for hydrodynamic calculations in the after the image processing and the computational fluid
fluid phase. In this work, the governing equations have dynamic steps. Each drag value has been saved for
been discretized based on the Power-law differencing comparison as a fitness function (Hamdia et al. 2021).
scheme of Patankar (1980). After solving governing

∫
equations in the calculation field, flow field variables
�
(14)
( )
D =𝜌 Uoutlet Uinlet − Uoutlet dy.
(U, V, P, k, 𝜀) are estimated. The inlet velocity has been •
chosen Uinlet = 0.535 m∕s , based on Lyn et al. study
�
( 1 9 9 5 ) a n d 𝜌w = 997 kg/m3 , 𝜇w = 8.9 × 10−4 Pa.s, D
CD� = . (15)
P = 1 atm . The inlet turbulent kinetic energy q∞ S
( out )2 )
k0 = 32 Tu u0 , was calculated from the equation of the
(
• In the final step, the binary genetic algorithm generated
isotropic turbulence and Kolmogorov
( expression
) has the next population based on the obtained fitness func-
tions in the previous step. The newly generated popula-
1.5
been used for the dissipation term 𝜀0 = C𝜇0 kL , which
was extracted from the eddy-viscosity relation (Younis
𝜀
tion, which contains 20 chromosomes composed of 16
and Przulj 2006). After convergence of the calculations genes, has been exported to the second step for image
of this step (computational fluid dynamic), the calculated processing and then computational fluid dynamic calcu-
values of the fluid hydrodynamic have been exported to lations for each one of the generated chromosomes. The
the next step for calculating the entered drag force to the iteration loop has been considered to repeat population
body. generation based on the flowchart in Fig. 3. Therefore,
• Drag force is calculated by velocity values at the outlet each iteration generates a new generation and each gen-
and inlet for each generated shape. For incompressible eration presents 20 chromosomes.

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 9 of 16 306

Fig. 6  The non-duplicated chromosomes for 100 iterations

4 Results and discussion study. The calculation times for generating mesh by image-
processing method and solving the turbulence flow over a
The results of the present study are validated by the experi- single square (using an Intel® Core™ i5 CPU at 2.2 GHz,
mental data. The velocity along the horizontal axis that 8 GB of RAM) were estimated at 15, 63, 233, and 837 s for
passes the centerline of the obstacle (Lyn et al. 1995; Lyn different mesh densities (from 0.5 to 5 dpc). The required
and Rodi 1994; McKillop and Durst 1985; Durao et al. 1988) memories are 1.39, 8.25, 18.19, and 31.48 MB, for 1.5, 2.5,
with the present model has been compared (see Fig. 5). A 3.5, and 4.5 dpc computational fluid dynamic calculations,
good agreement was achieved. For this comparison, differ- respectively.
ent calculative meshes (from 0.5 to 5 dot per centimeter The binary genetic algorithm procedure has been ended
(dpc)) were used, and similar model predictions have been after considering 100 iterations for finding the best shapes
observed for more than 3.5 dpc calculative meshes. Attend- with the target of the minimum drag coefficient. Figure 6
ing to the calculation cost, this mesh was used for other com- presents the non-duplicate chromosomes for 100 itera-
putational fluid dynamic calculations in this optimization tions with their drag coefficient. It should be noted that

13
306 Page 10 of 16 B. Abolpour et al.

Table 2  The drag coefficient Body shape Reference area Reynolds Measured drag coefficient Calculated drag
values for regular shapes based (S) Length (b) number (Re) (Aziz et al. 2008) coefficient in this
on the wind tunnel data (Aziz study
et al. 2008) in the comparison
with the calculated values in S = bD Re = ­105 R/D CD
this study, attending to the
presented limitations in Fig. 2

0 2.2 2.26
0.17 1.2 –
S = bD Re = ­105 L/D CD

< 0.1 1.9 –

0.25 – 2.03
0.5 2.5 –
S = bD Re = ­104 T/L L/D CD

1 0.25 1.65 –
0.5 0.5 – 1.58
S = bD Re = ­104 2.05 2.09

Fig. 7  The pressure contour and

streamlines of the turbulent flow
field around the square rod

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 11 of 16 306

Fig. 8  The pressure contour and

streamlines of the turbulent flow
field around the flat plate

Fig. 9  Turbulent flow field visu-

alization around the obtained
optimal shape configuration
with streamlines and pressure
contour plots

the presented geometries for each iteration in this Figure, the generated geometries (i.e., 45 × 20 = 900 cases) before
are the cases that were different from all of the evaluated that. It is clear that, by increasing the iteration number, the
cases before that iteration. For example, in the 46th itera- number of evaluated geometries was increased, and the gen-
tion, 3 new cases were found (the shown geometrics in this eration of new geometries (with different shapes) had less
Figure for this iteration), which were different from all of chance. Nevertheless, each population (for example 47th

13
 306 Page 12 of 16 B. Abolpour et al.

Fig. 10  Turbulent flow field vis-

ualization around the obtained
optimal shape configuration
(using 10 × 10 chromosome
size) with streamlines and pres-
sure contour plots

Fig. 11  Turbulent flow field vis-

ualization around the obtained
optimal shape configuration
(using 20 × 20 chromosome
size) with streamlines and pres-
sure contour plots

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 13 of 16 306

Table 3  The drag coefficient values for regular shapes based on the wind tunnel data (Aziz et al. 2008) in the comparison with the calculated
values in this study with different chromosome sizes
Body shape Reference area Reynolds Measured The calculated drag coefficient for The calculated drag coefficient for
(S) Length (b) number drag coef- obtained optimal shape (10 × 10 chro- obtained optimal shape (20 × 20 chro-
(Re) ficient mosome size) mosome size)
(Aziz
et al.
2008)

S = bD Re = ­105 R/D CD 1.13 1.16

0 1.4
0.02 1.2
0.08 1.3
0.25 1.1
S = bD Re > ­104 1.15

population) was generated based on all of the 20 chromo- The black color zones in Figs. 7 and 8 demonstrate the
somes (3 different and 17 duplicate cases) of the previous obstacle in the flow field. The stagnation zone of both shapes
population (46th population). is the same but pressure contours behind the shapes have
At each iteration, 20 chromosomes as a new generation various values. The vortex zone of the flat plate is wider
have been generated, but in Fig. 6, just non-similar chromo- than the square rod.
somes as the new shapes have been illustrated, for brevity. According to Fig. 6, in the 43th iteration, the semi-
According to the graphical representation in Fig. 6, in the T-beam-shaped chromosome has been found that a drag
beginner iterations, the number of non-duplicated chromo- coefficient is 1.58. Figure 9 shows the turbulent flow field
somes is more than in the next steps. The drag coefficient around this optimal shape. The results from comparing
order is reduced by continuing the process of iteration. The Figs. 7, 8, and 9 indicate that the stagnation zone in front
drag force estimation for incompressible flow can be deter- of the optimized shape is smaller than others. It is con-
mined by the conservation of momentum equation according cluded that momentum dissipation is reduced for opti-
to Eq. (14). The drag coefficient for regular body shapes with mized shaped and drag force is decreased consequently.
known reference area (S) and length (b) in various Reynolds The near wake region due to vortex shedding and separa-
numbers has been presented in Table 2 according to the wind tion of the optimized shape is thinner than square rod and
tunnel data (Aziz, et al. 2008). The drag coefficient for the flat plat. The smaller pressure field contribution behind
square rod without any skewness at the corner (R∕D = 0) the obstacle is a demonstration of the smooth streamlines
is equal to 2.2 which has the highest value of other body and velocity field, which introduces the lower drag value.
shapes. To validate drag coefficient values as a result of our An important point about the presented method for
system-specified code, a comparison has been done between shape optimization is the fact that increasing the size of
the estimated values and wind tunnel data. Three different defined chromosomes (i.e., number of genes) increases
shapes (Square rod, flat plate, and T-beam) were selected as the calculation time (by increasing the necessary dpc)
case studies. and increases the model's ability for optimizing the body
The estimated drag coefficient for the first initiated chro- shape. The presented optimal body in Fig. 9 has been
mosome as a square rod has a value of 2.26 and the cal- obtained based on the defined problem limitations in Fig. 2
culated drag coefficient for the vertical flat plate is 2.03. (a 4 × 4 chromosome size). For approving this ability, two
Attending to the defined optimization problem in Fig. 2, the cases with the same boundary conditions and higher chro-
optimum drag coefficient for the selected chromosome has mosome sizes, i.e., 10 × 10 and 20 × 20, have been opti-
been signed in Fig. 6. The turbulent flow field and pressure mized and the obtained results are presented in Figs. 10
contours for the square rod and the flat plate have been illus- and 11 and also Table 3.
trated in Figs. 7 and 8. In addition, streamlines indicate the In this study, a simple body shape has been investi-
velocity field around the obstacles. gated for minimizing its drag coefficient. Nevertheless, the

13
306 Page 14 of 16 B. Abolpour et al.

Fig. 12  Validation of models predictions based on the presented (Uinlet = 0.03 m/s) (Mohammed et al. 2015), c Extended surface chan-
method in this study (i.e., combination of the image processing nel ­(Uinlet = 0.003 m/s) (Mohebbi et al. 2018), and d Tesla micro-
and the computational fluid dynamic) using the relevant experi- mixer (in this modeling, the concentration equation has been also
mental results: a Double-baffle heat exchanger (Uinlet = 7.8 m/s) solved) (Hong et al. 2004)
(Menni and Azzi 2018; Demartini et al. 2004), b Step geometry

presented method (i.e., the combination of image process- 5 Conclusion

ing and computational fluid dynamic) can be utilized for
more complex body structures. Therefore, the presented The shape optimization based on the binary genetic algo-
optimization method is applicable for each body shape and rithm was done as a new approach for drag force reduction
each solid–fluid problem. For approving this ability, the by system-specified programming code in MATLAB soft-
passing of fluids inside different solid structures has been ware and image-processing method. Combining the binary
modeled using the presented method in this study, and genetic algorithm with image processing in the written code
the results of these models have been validated using the has been used as a powerful and effective method to solve
relevant experimental results. Figure 12 shows the results turbulent incompressible flow fields around any obstacles. In
of these comparisons. the present study, shape optimization and mesh generation

13
A novel approach for obtaining optimum shape design with the minimum drag coefficient Page 15 of 16 306

were done automatically. This approach decreases the time a vertical tube containing uniform gas flow. Heat Mass Transf
and calculation effort of mesh generation in computational 53(12):3517–3528
Abolpour B, Hekmatkhah R, Shamsoddini R (2021) Multi-objective
fluid dynamics, especially in the finite volume method. The optimum design for double baffle heat exchangers. Thermal Sci
binary genetic algorithm as an optimization method has Eng Prog 26:101132
been utilized to optimize the drag coefficient. The square Abolpour B, Hekmatkhah R, Shamsoddini R (2022) Optimum design
rod as an initial chromosome that includes sixteen genes for the Tesla micromixer. Microfluid Nanofluid 26(6):1–8
Anderson JD Jr (2010) Fundamentals of aerodynamics. Tata McGraw-
was considered. Each gene is only able to contain 0 or 1 Hill Education, New York
value which demonstrates the digital pixel of the obstacle. Aziz E, Chassapis C, Esche S, Dai S, Xu S, Jia R (2008) Online Wind
The binary genetic algorithm generated randomly twenty Tunnel Laboratory Paper presented at 2008 Annual Confer-
chromosomes in the first iteration as a new population. Each ence & Exposition, Pittsburgh, Pennsylvania. https://​doi.​org/​10.​
18260/1-​2--​3402
chromosome is converted to a gray-scale digital image. The Bhattarai S, Dahal K, Vichare P, Chen W (2020) Adapted Delaunay
image-processing method detected the edge and corners of triangulation method for free-form surface generation from ran-
each imported gray-scale digital image to generate mesh dom point clouds for stochastic optimization applications. Struct
grids. Multidisc Optim 61(2):649–660
Bosch G, Rodi W (1996) Simulation of vortex shedding past a square
The pressure–velocity decoupling issue in the incom- cylinder near a wall. Int J Heat Fluid Flow 17(3):267–275
pressible flow field was solved by the staggered grid method. Chan CM, Bai HL, He DQ (2018) Blade shape optimization of the
The turbulent flow over generated new obstacles (new chro- Savonius wind turbine using a genetic algorithm. Appl Energy
mosome) was solved via the finite volume method and 213:148–157
Chen W, Gao C, Gong Y, Zhang W (2020) Shape optimization to
governing equations were discretized using the Power-Law improve the transonic fluid-structure interaction stability by
scheme based on the Patankar approach. In addition, the an aerodynamic unsteady adjoint method. Aerosp Sci Technol.
low-Reynolds k − 𝜀 model was used to model turbulent flow. https://​doi.​org/​10.​1016/j.​ast.​2020.​105871
The drag coefficient has been estimated by momentum dif- Chien K-Y (1982) Predictions of channel and boundary-layer flows with
a low-Reynolds-number turbulence model. AIAA J 20(1):33–38
ference for incompressible Newtonian fluids. The drag coef- Daróczy L, Janiga G, Thévenin D (2018) Computational fluid dynamics
ficient data were stored for comparison as a fitness function. based shape optimization of airfoil geometry for an H-rotor using
The new population at the next steps was generated a genetic algorithm. Eng Optim 50(9):1483–1499
according to the previous fitness function. After 100 itera- Darwish S, Abdelrahman M, Elmekawy AM, Elsayed K (2018) Aero-
dynamic shape optimization of helicopter rotor blades in hover
tions, the optimized shape in the 43rd iteration was found using genetic algorithm and adjoint method. In 2018 AIAA Aero-
and the drag coefficient is about 1.58. The method expressed space Sciences Meeting, p 0044
by binary genetic algorithm and image processing is a practi- Demartini LC, Vielmo HA, Möller S (2004) Numeric and experimental
cal concept in the industrial world to reduce the drag coef- analysis of the turbulent flow through a channel with baffle plates.
J Braz Soc Mech Sci Eng 26(2):153–159
ficient. Reducing the drag coefficient by optimizing the Durao D, Heitor M, Pereira J (1988) Measurements of turbulent and
obstacle shapes in the flow field leads to increasing the life periodic flows around a square cross-section cylinder. Exp Fluids
of the structure due to reducing fatigue due to vibrations of 6(5):298–304
the solid–fluid interaction. Ebrahimi M, Jahangirian A (2017) Accelerating global optimization
of aerodynamic shapes using a new surrogate-assisted parallel
genetic algorithm. Eng Optim 49(12):2079–2094
Gao X, Tian Y, Sun B (2017) Shape optimization of bi-directional flow
Funding No funding was received to assist with the preparation of passage components based on a genetic algorithm and computa-
this manuscript. tional fluid dynamics. Eng Optim 50(8):1287–1303
Ghalandari M, Ziamolki A, Mosavi A, Shamshirband S, Chau K-W,
Declarations Bornassi S (2019) Aeromechanical optimization of first row com-
pressor test stand blades using a hybrid machine learning model of
Conflict of interest The authors declare that they have no conflict of genetic algorithm, artificial neural networks and design of experi-
interest. ments. Eng Appl Comput Fluid Mechan 13(1):892–904
Grinstein FF (1994) Open boundary conditions in the simulation of
Ethical approval This article does not contain any studies with human subsonic turbulent shear flows. J Comput Phys 115(1):43–55
participants or animals performed by any of the authors. Hamdia KM, Zhuang X, Rabczuk T (2021) An efficient optimization
approach for designing machine learning models based on genetic
Replication of results Code and data for replication can be provided algorithm. Neural Comput Appl 33(6):1923–1933
up on request. Han Z, Xu C, Zhang L, Zhang Y, Zhang K, Song W (2020) Efficient
aerodynamic shape optimization using variable-fidelity surro-
gate models and multilevel computational grids. Chin J Aeronaut
33(1):31–47
References He X, Li J, Mader CA, Yildirim A, Martins JRRA (2019) Robust aero-
dynamic shape optimization—from a circle to an airfoil. Aerosp
Abolpour B, Mehdi Afsahi M, Soltani Goharrizi A, Azizkarimi M Sci Technol 87:48–61
(2017) Study of the motion and deposition of micro particles in

13
 306 Page 16 of 16 B. Abolpour et al.

He S, Jonsson E, Mader CA, Martins JR (2019) Aerodynamic shape Patel VC, Rodi W, Scheuerer G (1985) Turbulence models for
optimization with time spectral flutter adjoint. In AIAA Scitech near-wall and low Reynolds number flows-a review. AIAA J
2019 Forum, p 0697 23(9):1308–1319
Hong C-C, Choi J-W, Ahn CH (2004) A novel in-plane passive Qin H, Guo Y, Liu Z, Liu Y, Zhong H (2018) Shape optimization
microfluidic mixer with modified Tesla structures. Lab Chip of automotive body frame using an improved genetic algorithm
4(2):109–113 optimizer. Adv Eng Softw 121:235–249
Kato M (1993) The modelling of turbulent flow around stationary and Saleem A, Kim M-H (2020) Aerodynamic performance optimization
vibrating square cylinders. Turbulent Shear Flow 1: 10.4.1–10.4.6. of an airfoil-based airborne wind turbine using genetic algorithm.
Launder BE, Sandham ND (2002) Closure strategies for turbulent and Energy. https://​doi.​org/​10.​1016/j.​energy.​2020.​117841
transitional flows. Cambridge University Press, Cambridge Şumnu A, Güzelbey İH, Öğücü O (2020) Aerodynamic shape optimi-
Li J, Cai J (2020) Massively multipoint aerodynamic shape design zation of a missile using a multiobjective genetic algorithm. Int J
via surrogate-assisted gradient-based optimization. AIAA J Aerosp Eng 2020:1–17
58(5):1949–1963 Ukken MG, Sivapragasam M (2019) Aerodynamic shape optimization
Li J, Zhang M, Martins JRRA, Shu C (2020) Efficient aerodynamic of airfoils at ultra-low Reynolds numbers. Sādhanā. https://​doi.​
shape optimization with deep-learning-based geometric filtering. org/​10.​1007/​s12046-​019-​1115-z
AIAA J 58(10):4243–4259 Wang Q, Zhao Q (2019) Rotor blade aerodynamic shape optimization
Liu C, Bu W, Xu D (2017) Multi-objective shape optimization of a based on high-efficient optimization method. Proc Inst Mech Eng
plate-fin heat exchanger using CFD and multi-objective genetic G 234(2):375–387
algorithm. Int J Heat Mass Transf 111:65–82 Yan X, Zhu J, Kuang M, Wang X (2019) Aerodynamic shape optimi-
Lyn D, Rodi W (1994) The flapping shear layer formed by flow separa- zation using a novel optimizer based on machine learning tech-
tion from the forward corner of a square cylinder. J Fluid Mech niques. Aerosp Sci Technol 86:826–835
267:353–376 Yi M, Pan Y, Huang J, Wang L, Liu D (2019) A Comprehensive opti-
Lyn DA, Einav S, Rodi W, Park J-H (1995) A laser-Doppler veloci- mization design method of aerodynamic, acoustic, and stealth of
metry study of ensemble-averaged characteristics of the turbulent helicopter rotor blades based on genetic algorithm. Math Prob
near wake of a square cylinder. J Fluid Mechan 304:285–319 Eng 2019:1–12
Mangano M and Martins JRRA (2019) Multipoint aerodynamic shape Younis B, Przulj V (2006) Computation of turbulent vortex shedding.
optimization for subsonic and supersonic regimes. In: AIAA Comput Mech 37(5):408
Scitech 2019 Forum Zhang W, Sun J, Wang L, Wu J, He L (2020) Rotor airfoil aerodynamic
McKillop A and Durst F (1985) LDA experiments of separated flow design method and wind tunnel test verification. Chin J Aeronaut
behind a circular cylinder. In: 2nd International Symposium on 33(8):2123–2132
Applications of Laser Anemometry to Fluid Mechanics Zhou Li L, Jiu Li J, Zhang J, Lu K, Ni Yuan M (2020) Aerodynamic
Menni Y, Azzi A (2018) Numerical analysis of thermal and aerody- shape optimization by continually moving ROM. Aerosp Sci
namic fields in a channel with cascaded baffles. Periodica Polytech Technol. https://​doi.​org/​10.​1016/j.​ast.​2020.​105729
Mech Eng 62(1):16–25
Mohammed H, Alawi OA, Wahid M (2015) Mixed convective nano- Publisher's Note Springer Nature remains neutral with regard to
fluid flow in a channel having backward-facing step with a baffle. jurisdictional claims in published maps and institutional affiliations.
Powder Technol 275:329–343
Mohebbi R, Rashidi MM, Izadi M, Azwadi Che Sidik N, Wei Xian H Springer Nature or its licensor holds exclusive rights to this article under
(2018) Forced convection of nanofluids in an extended surfaces a publishing agreement with the author(s) or other rightsholder(s);
channel using lattice Boltzmann method. Int J Heat Mass Transf author self-archiving of the accepted manuscript version of this article
117:1291–1303 is solely governed by the terms of such publishing agreement and
Patankar SV (1980) Numerical heat transfer and fluid flow. Hemisphere applicable law.
Publ. Corp, New York, p 58

13