Journal of Molecular Liquids 313 (2020) 113492

Contents lists available at ScienceDirect

Journal of Molecular Liquids

journal homepage: www.elsevier.com/locate/molliq

Analysis of transport processes in a reacting flow of hybrid nanofluid

around a bluff-body embedded in porous media using artificial neural
network and particle swarm optimization
Javad Mohebbi Najm Abad a, Rasool Alizadeh b, Abolfazl Fattahi c, Mohammad Hossein Doranehgard d,
Ebrahim Alhajri e, Nader Karimi f,g,⁎
a
Department of Computer Engineering, Quchan Branch, Islamic Azad University, Quchan, Iran
b
Department of Mechanical Engineering, Quchan Branch, Islamic Azad University, Quchan, Iran
c
Department of Mechanical Engineering, University of Kashan, Kashan, Iran
d
Department of Civil and Environmental Engineering, School of Mining and Petroleum Engineering, University of Alberta, Edmonton, Alberta T6G 1H9, Canada
e
Department of Mechanical Engineering, Khalifah University, Abu Dahabi, United Arab Emirates
f
School of Engineering and Materials Science, Queen Mary University of London, London E1 4NS, United Kingdom
g
James Watt School of Engineering, University of Glasgow, Glasgow G12 8QQ, United Kingdom

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates heat and mass transfer in a hybrid nanofluid flow impinging upon a cylindrical bluff-body
Received 6 May 2020 embedded in porous media and featuring homogenous and heterogeneous chemical reactions. The analysis in-
Received in revised form 26 May 2020 cludes mixed convection and local thermal non-equilibrium in the porous medium as well as Soret and Dufour
Accepted 30 May 2020
effects. Assuming single-phase mixture, a laminar flow of Al2O3-Cu-water (Aluminium oxide-Copper-water) hy-
Available online 2 June 2020
brid nanofluid is considered and coupled transport processes are simulated computationally. Due to the signifi-
Keywords:
cant complexity of this problem, containing a large number of variables, conventional approaches to parametric
Hybrid nanofluid study struggle to provide meaningful outcomes. As a remedy, the simulation data are fed into an artificial neural
Artificial intelligence network to estimate the target responses. This shows that the volume fraction of nanoparticles, interfacial area of
Chemically reacting flow the porous medium and mixed convection parameter, are of primary importance. It is also observed that small
Mixed convection variation in the volume fraction of nanoparticles can considerably alter the response of thermal and solutal do-
Predictor algorithms mains. Further, it is shown that the parameters affecting the thermal process can modify the problem chemically.
Particle swarm optimization In particular, raising the volume fraction of nanoparticles enhances the production of chemical species. Further-
more, particle swarm optimization is applied to predict correlations for Nusselt and Sherwood numbers through
a systematic identification of the most influential parameters. The current study clearly demonstrates the advan-
tages of using the estimator algorithms to understand and predict the behaviours of complex thermo-chemical
and solutal systems.
© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).

1. Introduction years, nanofluids were manufactured of one type of nanoparticles. Ap-

plying the advantageous of different nanoparticles, recently, multiple
The rapidly growing concerns on environmental pollutions and kinds of nanoparticles have been simultaneously embedded in the
global warming [1,2], have motivated many attempts to improve the ef- base fluid. These nanofluids, regarded as hybrid nanofluid, feature
ficiency of transport processes. Utilizing various shapes of fins [3], ribs higher thermal conductivity than those prepared by monotype nano-
[4], turbulators and vortex generators [5] are some examples of these ef- particles [9]. Heat exchangers, heat pipes, chillers, solar heating and
forts. Further, blending nanoparticles with the base working fluid has electrical chips are the instances of potential use of hybrid nanofluids
been demonstrated to be an effective way of boosting thermal conduc- [9,10].
tivity and results in improved heat transfer [6]. Many articles have ap- Utilizing porous materials with high thermal conductivity is also a
peared in recent years on the metallic, non-metallic and quasi-metallic well demonstrated method of improving heat transfer. It has been al-
nanoparticles (e.g. [7,8]) and the work in this area is ongoing. For ready shown that combining porous media and nanofluids can enhance
heat transfer considerably [11–14]. Some recent studies targeted increas-
⁎ Corresponding author at: School of Engineering and Materials Science, Queen Mary
ing heat transfer by using porous media and hybrid nanofluids. Xiong el
University of London, London E1 4NS, United Kingdom. al. [15] numerically investigated free convection of hybrid nanofluid
E-mail address: N.Karimi@qmul.ac.uk (N. Karimi). (Fe3O4 + MWCNT (Iron Oxide + Multi-walled carbon nanotubes)) in a

https://doi.org/10.1016/j.molliq.2020.113492
0167-7322/© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

Nomenclature λ1 dimensionless mixed convection parameter

ε porosity
A1, A2, A3, A4, A5 constants μ dynamic viscosity
a cylinder radius υ kinematic viscosity
asf interfacial area per unit volume of porous media ρ fluid density
hsf asf :a ϕ1, ϕ2 solid volume fraction of nanoparticles of nanoparticles 1
Bi Biot number Bi ¼
4k f and 2.
c1 concentration of species 1
c2 concentration of species 2
Subscripts
Cp specific heat at constant pressure
w related to the external wall of the cylinder
Cs concentration
∞ far field
D ratio of the diffusion coefficient
hnf hybrid nanofluid
D1 solute diffusion coefficient of species 1
nf nanofluid
D2 solute diffusion coefficient of species 2
f fluid
D:k f C∞
Df Dufour number Df ¼ Þυ s solid
C s :C p ðT w −T ∞
f(η) radial velocity function
f ́ (η) axial velocity function
g gravitational acceleration porous enclosure under magnetic effects. It was shown that variation of
h heat transfer coefficient Reynolds and Darcy number as well as thermal radiation were in the
hsf interstitial heat transfer coefficient same direction of Nusselt number, while Hartmann number renders an
k thermal conductivity adverse effect. Natural convection of Al2O3-Cu water hybrid nanofluid in
k freestream strain rate a differentially heated porous cavity was studied by Mehryan et al. [16].
k1 permeability of the porous medium It was found that heat transfer enhancement by increasing nanofluid vol-
km mass transfer coefficient ume fraction might be completely suppressed by applying certain kinds of
kT thermal diffusion ratio porous materials.
Kc HOM parameter Sajjadi et al. [17] numerically simulated MHD (Magneto Hydro Dy-
Ks HET parameter namics) natural convection of MWCNT–Fe3O4/water hybrid nanofluid
m shape factor in a porous cavity. The existence of magnetic field could smear the influ-
N∗ ratio of concentration to thermal buoyancy forces N  ¼ ence of Rayleigh number on heat transfer, while adding the hybrid
g:βc :C ∞ nanofluid to the base fluid augments the role of Rayleigh number. Free
2
4ak convection of hybrid nanoparticles of MWCNT and Fe3O4 in water in a
Nu Nusselt number porous circle sector was investigated by Manh et al. [18]. They presented
Num average Nusselt number a correlation for mean Nusselt number as a function of Hartmann, Ray-
p fluid pressure leigh and Darcy number. Natural convection of hybrid nanofluid of
P non-dimensional fluid pressure MWCNT-Fe3O4-water in an enclosure filled with a two-layer porous
P0 the initial fluid pressure media was conducted by Mehryan et al. [19]. Heat transfer was aug-
Pr Prandtl number mented at lower values of solid-liquid interface convection parameter
qm mass flux on the wall and higher values of porosity ratio. This study considered the local ther-
qw heat flux on the wall mal non-equilibrium assumption for calculating the temperature field
r radial coordinate within the porous medium. Similarly, Alizadeh et al. [20] showed the
k:a2
Re freestream Reynolds number Re ¼ critical effect of the non-equilibrium assumption on the thermal mode
asf D1 1=2 2υ
S interfacial area S ¼ of the system through numerical simulation of an external hybrid
ðkυÞ nanofluid flow over a cylinder surrounded by porous materials.
υ
Sc Schmidt number Sc ¼ Jerry et al. [21] studied mixed convection of Ag-MgO (Silver
D -Magnesium oxide) hybrid nanoparticles in water through a horizontal
D:k f ðT w −T ∞ Þ
Sr Soret number Sr ¼ rectangular porous channel. The variations of thermal and hydrody-
T C ∞ :α
Sh Sherwood number ∞ namic parameters in the temporal domain were scrutinized. The heat
Shm average Sherwood number transfer and fluid flow were characterized by a pattern in spatial and
T temperature temporal domain. Darcy number determined the thermal response of
Tm mean fluid temperature the system; at low Darcy numbers, heat transfer showed a single-
u, w velocity components along (r − z)- axis maximum trend, while it had a monotonic increasing trend at high
z axial coordinate values of Darcy number. The forced convection of Al2O3–CuO–water hy-
brid nanofluid in a cylinder filled with porous medium was conducted
Greek symbols by Aminian et al. [22]. These authors reported that declining Darcy
α thermal diffusivity number intensifies the heat transfer rate, while it poses an adverse ef-
βC coefficient of thermal expansion with concentration fect on the pressure drop.
βt coefficient of thermal expansion The fluid flow near a stagnation zone is called stagnation-point
kf
γ modified conductivity ratio γ ¼ flow [23]. This kind of flow usually occurs on the solid surface of
k :a 1 ks
γ∗
R
Damköhler number γ ¼ the bluff bodies [24] and has been widely applied in aerospace tech-
2D
!2 C ∞
nologies and cooling of electronics [23]. Stagnation-point flows over
r a permeable stretching sheet in a nanofluid flow [25], porous me-
η similarity variable, η ¼
a dium [26–28] and by considering radiation and mixed convection
θ(η) non-dimensional temperature 2 heat transfer [24] have been already studied. Recently, a stagnation
a
λ permeability parameter, λ ¼ point flow over a stretching-shrinking surface merging in a hybrid
4k1
nanofluid has been conducted [29]. The Joule heating was assumed
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 3

to be active owing to electrical conductivity of the base fluid. It was methods to predict the behavior of a problem with many intercon-
shown that increases in Eckert number improved the thermal char- nected parameters. It is requires choosing the algorithms with reason-
acteristic of hybrid nanofluid. able errors in their predictions. Therefore, the numerical results are
The current problem includes stagnation-point hybrid nanofluid applied to train the ANN, leading to prediction of the targeted parame-
flow over a bluff body embedded in a porous media that further includes ters in a wide range. Further, particle swarm optimization (PSO)
surface chemical reactions. The large number of influencing parameters method is applied to precisely extract correlations for Nusselt and Sher-
render heat and mass transfer predictions too laborious. This is due to wood numbers [36,37].
many interconnections amongst the parameters and the complex phys-
ics of the problem. Therefore, a predictive tool having the capability of 2. Physical description of the problem and mathematical formulation
capturing the nonlinear and complex underlying physics with a high de-
gree of accuracy is highly desirable. Comparing to the other similar algo- Fig. 1 shows the geometry and physical domain of the current prob-
rithms, the artificial neural network (ANN) is the most capable of lem. The nanofluid passes over an infinitely long cylinder embedded in a
learning and estimating complex and non-linear problems, widely en- porous medium and develops a stagnation-point flow. The flow in-
countered in the engineering [30]. Given the powerful data-driven, cludes heterogeneous and homogenous chemical reactions. The follow-
self-adaptive and flexible features of a machine learning [31], ANN is ing assumptions are made throughout this work.
employed in this study. This method has been already utilized to under-
• The nanofluid flow is steady and laminar.
stand the responses of thermal systems with respect to pertinent vari-
• The nanofluid is assumed to be Newtonian and single phase [38–41].
ables. Abdollahi et al. [32] used ANN to find the optimum hydro-
• The cylinder is assumed to be infinitely long, and the porous medium
thermal working point of a grooved channel filled with a hybrid
is homogenous, isotropic and under local thermal non-equilibrium.
nanofluid and equipped with reflectors in the laminar regime. Similar
• The viscous dissipation of kinetic energy of the flow is ignored.
work was performed in a vortex-generator channel involving hybrid
Also, porosity, specific heat, density and thermal conductivity are
nanofluid using multi-objective genetic algorithm by Abdollahi and
assumed to be constant and thus the thermal dispersion effects
Shams [33]. The results of numerical simulation of Eulerian-Lagrangian
are ignored.
approach were used to train the model. Prediction of entropy generation
• A moderate range of pore-scale Reynolds number is considered in
of Ag-MgO-water hybrid nanofluid in a mini-channel was also carried
the porous medium and hence non-linear effects in momentum
out using ANN [34]. Bagheri et al. [35] predicted Nusselt number in a
transfer are deemed to be negligibly small.
C-shaped cavity occupied with a hybrid nanofluid by means of ANN.
The numerical simulation of Cu-Al2O3-water hybrid nanofluid fol-
lowing over a cylinder in a porous domain is conducted in this work. The continuity of mass, radial and axial direction of momentum and
Mixed convection as well as homogenous and heterogeneous chemical energy equations are written respectively as follows.
reactions are considered. To capture the heat transfer caused by mass
diffusion and vice versa, Dufour and Soret effects are further considered. ∂u u ∂w
þ þ ¼0 ð1Þ
The current study evaluates the ability of the machine-learning ∂r r ∂z

Fig. 1. Schematic configuration of bluff-body under radial impinging flow surrounded by porous media.
4 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

  2 2
!
ρhnf ∂u ∂u ∂p μ hnf ∂ u 1 ∂u u ∂ u μ hnf The HET reaction rate then can be related to the solute gradient as [44]
u þ w ¼ − þ þ − þ − u; ð2Þ
ε2 ∂r ∂z ∂r ε ∂r 2 r ∂r r 2 ∂z2 k1
∂c1 ∂c2
D1 ¼ −D2 ¼ ks c2 ; ð12Þ
  ∂n ∂n
ρhnf ∂w ∂w ∂p
u þw ¼− ð3Þ in which n is the normal direction towards the fluid.
ε2 ∂r ∂z ∂z
! The hydrodynamic boundary conditions can be written as
μ hnf ∂2 w 1 ∂w ∂2 w  
þ þ þ 2 ∓g:ðρ:βt Þhnf T f −T ∞
ε ∂r2 r ∂r ∂z r ¼ a : w ¼ 0; u ¼ 0; ð13Þ
μ hnf
þg:ðρ:βc Þhnf ½c2 −c∞ − w;  
k1 a2
r ¼ ∞ : w ¼ 2kz; u ¼ −k r− : ð14Þ
r
2 2
!
∂T f ∂T f ∂ T f 1 ∂T f ∂ T f
u þw ¼ α hnf þ þ Eq. (13) demonstrates the no-slip boundary conditions over the cyl-
∂r ∂z ∂r 2 r ∂r ∂z2
" 2 2
# inder walls, with radius a. Also, Eq. (19) shows that analogous to the
hsf :asf   D2 kT ∂ c2 1 ∂c2 ∂ c2 Hiemenz flow, the current viscous flow approaches the potential flow
þ  T s −T f þ þ þ ;
ρ:C p hnf C ps :C p hnf ∂r2 r ∂r ∂z2 in the limit of r → ∞[48–50].
ð4Þ The thermal boundary conditions give

r ¼ a : T f ¼ T w ¼ Constant;
where all parameters defined in the nomenclature.
The energy Eq. (4) contains the Dufour effect [42,43] by the last term T s ¼ T w ¼ Constant;
on the right hand side. Further, the heat transfer within the solid phase ð15Þ
of the porous medium can be expressed by r ¼ a : T f ¼ T ∞;
!
2
∂ T s 1 ∂T s ∂ T s
2
  T s ¼ T ∞;
ks þ þ 2 −hsf :asf T s −T f ¼ 0; ð5Þ
∂r 2 r ∂r ∂z
in which Tw and T∞ indicate the cylinder surface and the free-stream
temperature, respectively.
in which ks and Ts are thermal conductivity and temperature of the solid The solute transport boundary conditions are given by
porous medium. Considering homogenous-heterogeneous (HOM-HET)
reactions, mass transfer of chemical species is presented by an ∂C 1 k3 ∂C 2 k3
r¼a: ¼ þ C1 ; ¼ − C 2 ; r→∞ : C 1 →C 2 C 2 →0 ð16Þ
advective-diffusive model, as the following [44]. ∂r D1 ∂r D2

2 2
! c∞ is the concentration at the free-stream limit.
∂c1 ∂c1 ∂ c1 1 ∂c1 ∂ c1
u þw ¼ D1 þ þ 2 −kc c1 c22 −asf :ks :c1 ; ð6Þ
∂r ∂z ∂r 2 r ∂r ∂z
3. Self-similar solution of the governing equations

2 2
!
∂c2 ∂c2 ∂ c2 1 ∂c2 ∂ c2 Applying the following similarity transformation, Eqs. (1) to (7) are
u þw ¼ D2 þ þ 2 reduced to the forms that can be solved more conveniently.
∂r ∂z ∂r 2 r ∂r ∂z
2 2
!
D2 kT ∂ T f 1 ∂T f ∂ T f k:a h i
þ þ þ þ kc c1 c22 þ asf :ks :c1 : ð7Þ 2
u ¼ −pffiffiffi f ðηÞ; w ¼ 2kf ð́ ηÞ z; p ¼ ρ f k a2 P; ð17Þ
Tm ∂r 2 r ∂r ∂z2 η

!2
The mass transfer Eqs. (6) and (7) involve the classical Fickian diffu- r
sion as well as solute diffusion by means of thermal gradient, called where η ¼ is the non-dimensional radial variable. Some algebraic
a
Soret effect, which is the second bracketed term on the right hand manipulations reveals the following non-dimensional form.
side of Eq. (7) [45,46]. In Eqs. (3)–(7) the subscript “f” refers to the
h i
fluid properties. Further, D1 and D2 indicate molecular diffusion coeffi- ε:½ ηf ́ ́ ́þ f ́ ́ þ A1 :A2 : Re 1 þ ff ́−ð f Þ́
2

cient of species 1 and 2, having respectively the concentration of c1

́ 2 :A1 :A6 : Re:λ1 θ f þ ε2 :A1 : Re:N :½C 2 −1
þ ε2 :λ½1−f ∓ε
and c2. ks and kc are the HET and HOM reaction rate constant, corre-
¼ 0; ð18Þ
spondingly. The HOM-HET reactions used in the current study are re-
spectively defined as ! " ! Z #
2 ´f η
A2 f 1 λ f A2 λ z 2
P−P 0 ¼ − − þ dη −2 2 þ ; ð19Þ
2ε 2 η A1 :ε Re A1 : Re η ε A1 : Re a
S2 þ 2S1→3S1; ð8Þ 1

k:a2 a2
in which Re ¼ depicts the Reynolds number of freestream, λ ¼
S2→S1; ð9Þ 2υ f 4k1
Gr g:βt ðT w −T ∞ Þ
is the permeability parameter,λ1 ¼ 2
¼ 2
denotes the di-
Re 4ak
where S1 and S2 are the species 1 and 2. The HOM reaction rate for is
 g:βc :C ∞
given by [44] mensionless mixed convection parameter , N ¼ 2
ratio of con-
4ak
∂c1 ∂c2 centration to thermal buoyancy forces and prime symbol indicates the
¼ ¼ −kc c2 c1 2 : ð10Þ derivative against η. The constant A1 and A2 are defined as
∂t ∂t

A1 ¼ ð1−ϕ1 Þ2:5 ð1−ϕ2 Þ2:5 ;

The HET reaction rate performing on the fluid-solid interface of the
cylinder surface is represented by [47] " !# !
ρs1 ρs2
A2 ¼ ð1−ϕ2 Þ ð1−ϕ1 Þ þ ϕ1 þ ϕ2 :
r p ¼ −ks asf c2 : ð11Þ ρf ρf
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 5

Considering Eqs. (13), (14) and (17), the boundary conditions for asf D1 1=2
Eqs. (18) and (19) are given by is dynamic viscosity of the nanofluid. S ¼ is the non-
ðkυÞ
η ¼ 1 : f ð́ 1Þ ¼ 0; f ð1Þ ¼ 0; ð20aÞ dimensional interfacial area of the porous medium. The mass transfer
equations are closed by the following boundary conditions.
η→∞ : f ð́ ∞Þ ¼ 1: ð20bÞ
η¼1: C 01 ð1Þ ¼ þγ  C 1 ð1Þ
Taking the following transformation, γ
C 02 ð1Þ ¼ − C 2 ð1Þ;
T f ðηÞ−T ∞ D
θ f ðηÞ ¼ ; ð21Þ
T w −T ∞ η→∞ : C 1 ð∞Þ ¼ 1
ð29bÞ
the energy Eq. (4) is turned into non-dimensional form, which yields C 2 ð∞Þ ¼ 0;
      k3 :a
A4 :A5 ηθ ́ f́ þ θ f́ þ Re: Pr:A3 :A5 f :θ f́ þ Bi:A5 θs −θ f where γ  ¼ indicates Damköhler number.
þ Df : Pr:A3 :D½ ηC 2́ ́ þ C 2́  2D1
¼ 0; ð22Þ The numerical procedure aims to solve Eqs. (18), (22), (25), (28a)
and (28b) along with the boundary conditions of Eqs. (20a), (20b),
hsf asf :a D1 :kT (24a), (24b), (26), (29a) and (29b) using an implicit, iterative finite-
in which Bi ¼ represents the Biot number and Df ¼
c∞ 4k f C s :C p f difference method [28]. One of the important non-dimensional param-
is the Dufour number. The constants A3 to A5 in Eq. (22)
ðT w −T ∞ Þυ f eters to understand the thermal performance of the current problem is
are defined by the Nusselt number, Nu, which should be calculated using the definition
"   #   of the convection heat transfer coefficient, h:
ρ:C p s ρ:C p s  
A3 ¼ ð1−ϕ2 Þ ð1−ϕ1 Þ þ ϕ1   1 þ ϕ2   2; ∂T f
ρ:C p f ρ:C p f −khnf
qw ∂r r¼a 2khnf ∂θ f ð1Þ
    h¼ ¼ ¼− ; ð30Þ
ks1 þ ðs−1Þk f −ðs−1Þϕ1 k f −ks1 ks2 þ ðs−1Þkbf −ðs−1Þϕ2 kbf −ks2
T w −T ∞ T w −T ∞ a ∂η
A4 ¼   :   ;
ks1 þ ðs−1Þk f þ ϕ1 k f −ks1 ks2 þ ðs−1Þkbf þ ϕ2 kbf −ks2
and
" #
C p s1 C p s2 2khnf ∂θ f ð1Þ
A5 ¼ ð1−ϕ2 Þ ð1−ϕ1 Þ þ ϕ1 þ ϕ2 : ð23Þ qw ¼ − T w −T ∞ ; ð31Þ
Cp f Cp f a ∂η

The thermal boundary conditions subsequently reduce to h:a khnf

Nu ¼ ¼− θ ́ ð1Þ ¼ −A4 :θ f́ ð1Þ: ð32Þ
2k f kf f
η ¼ 1 : θ f ð1Þ ¼ 1 ð24aÞ

η→∞ : θ f ð∞Þ ¼ 0: ð24bÞ Also, the coefficients of mass transfer and mass transfer rate are
given by
Substitution of Eq. (21) into Eq. (5), the energy equation for the solid  
∂c1
porous media takes the form of −D1
qm ∂r r¼a 2D1 ∂C 1 ð1Þ
  km ¼ ¼ ¼− ; ð33Þ
ηθ ́ś þ θ ś −Bi:γ θs −θ f ¼ 0; ð25Þ C∞ c∞ a ∂η

with the boundary conditions as follows. where qm is the mass flow rate at the cylinder wall, which is

η¼1: θs ð1Þ ¼ 1 2D1 ∂C 1 ð1Þ

qm ¼ − C w −C ∞ : ð34Þ
ð26Þ a ∂η
η→∞ : θs ð∞Þ ¼ 0:
Hence, Sherwood number is given by
kf
In Eq. (25), γ ¼ is the modified conductivity ratio. Aiding Eq. (21) km :a
ks Sh ¼ ¼ −C 1́ ðη ¼ 1Þ: ð35Þ
and the definition of 2D1

cðηÞ
C ðηÞ ¼ ; ð27Þ
c∞ 4. Hybrid nanofluid characterization

the mass transport Eqs. (6) and (7) are transformed into dimensionless The hybrid nanofluid is made of Al2O3 and Cu nanoparticles
forms of suspended in water as the base fluid. The volume fractions of the
h i Al2O3 and Cu nanoparticles are denoted by ϕ1 and ϕ2, respectively.
Pr½ ηC 1́ ́ þ C 1́  þ Re: Pr:Sc f :C 1́ −K c :C 1 :C 22 −K s :S:C 1 ¼ 0; ð28aÞ
The hybrid nanofluid is assumed to be a mixture of various volume frac-
  tions of Cu nanoparticles into the Al2O3-water nanofluid with fixed
Pr:D:½ ηC ́2́ þ C
h 2́  þ Sr:Sc:D: ηθ ́ f́ þ θ f́ i value of ϕ1 = 0.1%. Therefore, in the current study, only the volume frac-
2
þ Re: Pr:Sc f :C 2́ þ K c :C 1 :C 2 þ K s :S:C 1 tion of the Cu nanoparticles, ϕ2, is varied [51–53]. The determining
¼ 0; ð28bÞ equations for thermo-physical properties of the hybrid nanofluid of
υf D1 :kT ðT w −T ∞ Þ Cu-Al2O3-water are depicted in Table 1. Owing to the precision of its
in which, Sc ¼ and Sr ¼ describes Schmidt and Soret thermo-physical correlations [54], this type of hybrid nanofluid is cho-
D1 Tm c∞ :α2
k c c∞ ks υ1=2 sen in this study.
number, respectively. Further, K c ¼ and K s ¼ 1=2
denote corre-
k Dk Thermal conductivity of the hybrid nanofluid in Table 1 includes a
spondingly the non-dimensional HOM and HET parameter, in which υ shape factor parameter, m, which is set to 3 for the spherical nanoparticles
6 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

Table 1 Table 4
Equations for thermo-physical properties of the hybrid nanofluid [54]. A comparison of the current simulation results and those of Ref. [56] for unity porosity and
very large permeability for Re = 10 and 1.0.
Properties Hybrid nanofluid
" !# η Re = 1.0 Re = 10
Density ρs1
ρhnf ¼ ρ f ð1−ϕ2 Þ ð1−ϕ1 Þ þ ϕ1 þ ϕ 2 ρ s2 Ref. [56] Current results Ref. [56] Current results
ρf
" #
Heat capacity ðρ:C p Þs1 f ´f f ´f f f́ f f́
ðρ:C p Þhnf ¼ ðρ:C p Þ f ð1−ϕ2 Þ ð1−ϕ1 Þ þ ϕ1 þ ϕ2 ðρ:C p Þs2
ðρ:C p Þ f
1.2 0.02667 0.25302 0.02693 0.25993 0.06638 0.58982 0.06631 0.06610
Viscosity μf
μ hnf ¼ 1.4 0.09665 0.43724 0.09652 0.43710 0.21400 0.84821 0.21393 0.21379
2:5
ð1−ϕ1 Þ2:5 ð1−ϕ2 Þ 1.6 0.19836 0.57315 0.19828 0.57329 0.39532 0.94852 0.39541 0.39535
Thermal khnf
¼ ks2 þ ðm−1Þkbf −ðm−1Þϕ2 ðkbf −ks2 Þ 1.8 0.32361 0.67444 0.32365 0.67438 0.58919 0.98380 0.58914 0.58926
kbf
conductivity 2.0 0.46674 0.75054 0.46683 0.75046 0.78731 0.99522 0.78735 0.78729
ks2 þ ðm−1Þkbf þ ϕ2 ðkbf −ks2 Þ
kbf
kf
¼ ks1 þ ðm−1Þk f −ðm−1Þϕ1 ðk f −ks1 Þ
ks1 þ ðm−1Þk f þ ϕ1 ðk f −ks1 Þ

through a finite difference method [58]. The temporal term was also
discretized using backward second-order scheme [28]. The second-
[54]. The thermo-physical properties of the basic components of the hy-
order discretization provides higher accuracy in transforming differen-
brid nanofluid are presented in Table 2 for the standard temperature of
tial to algebraic equations involved in this study.
25 ° C.
6. Estimator and optimizer algorithms
5. Validation and grid independency
6.1. Feedforward artificial neural network: Multi-Layer Perceptron (MLP)
To demonstrate independency of the results from the grid size, as
well as precision of the numerical procedure, various cell numbers
Artificial neural network is an optimizer and estimator tool, inspired
were compared. To do so, the surface averaged values of Nusselt and
by the biological neural networks that constitute animal brains [59].
Sherwood numbers were calculated for mesh sizes of 51 × 18,
ANN requires learning to do expected tasks by providing the examples
102 × 36, 204 × 72, 408 × 144 and 816 × 288. Table 3 shows there is
of the system. ANN does not need to include programming of task-
no considerable difference between the parameters computed on the
specific rules. It is structured based on the connected units, called artifi-
grid size of 408 × 144 and those of finer grids. Therefore, the grid size
cial neurons, playing similar to biological neural units [60]. In this paper,
of 408 × 144 was chosen for meshing the η − φ plane. It should be
MLP neural network is used to estimate the output parameters. The
pointed out that the grid features a non-uniform structure in the normal
structure of this network typically includes three layers, consisting of
direction to the cylinder, η, to capture the sharp gradient of the param-
an input layer, a hidden layer, and an output layer. An input signal trans-
eters, while it is uniformly sized in the other direction.
mits to the neurons, showed by balls in Fig. 2, through connections,
Tables 4 and 5 show that if the porosity approaches unity indicating
which are segment lines in this figure, and can be adjusted by the
no porous material, the current numerical results reduce to those re-
weight implemented on each connection.
ported by Wang [56] and Gorla [57] on impinging flow over a cylinder.
If the model has n input(s) then the input layer contains n + 1 neu-
Also, it was verified that for large values of Biot number, the current re-
rons. The first neuron, with a value of 1, is biased and the other neurons
sults under the assumption of local thermal non-equilibrium reproduce
receive input data. The number of neurons in the second layer, called
to the local thermal equilibrium results reported in Ref. [27].
the hidden layer, is m + 1. The appropriate value for m can be calculated
The numerical procedure was terminated when the difference be-
at the time of network creation by performing multiple tests. In this
tween the two subsequent residuals of each discretised equation be-
layer, similar to the first layer, one neuron plays the bias role. Other neu-
come less than 10−7. Due to its superior stability, Crank-Nicolson
rons apply an activation function to the sum of the weighted output of
second-order method was applied to discretize the governing equations
the previous layer that is [60]
0 1
Xn
f i ¼ g @w0;i þ w j;i x j A; ð36Þ
Table 2 j¼1
Thermo-physical properties of the components of the hybrid nanofluid [55].

Properties Water Al2O3 Cu in which, w0, i is the weight between the bias and the i-th neuron of the
! hidden layer. Also, wj, i is the weight between the j-th neuron of the in-
kg 997.0 3970 8933
ρ put layer and i-th neuron of the hidden layer. Furthermore, xj and fi
m3
! shows the j-th input of the neural network and the output of i-th neuron
J 4180 765 385
Cp of the hidden layer, respectively. For g as an activation function, a variety
kg:K
! of functions such as logistic, hyperbolic and exponential can be used.
W 0.6071 40 400
k
m:K

Table 5
A comparison of the present numerical results and those of Ref. [57] for unity porosity and
Table 3
very large permeability for various Reynolds numbers.
Grid independence test at Df = 1.0, Bi = 0.1, Sr = 0.5, Re = 5.0, λ = 10, Sc = 0.1.
Re θ
Mesh size Num Shm
Ref. [57] Current results
51 × 18 1.540873 0.588014
102 × 36 1.499256 0.555648 0.01 0.84549 0.84557
204 × 72 1.479155 0.527381 0.1 0.73715 0.73701
408 × 144 1.465920 0.467275 1.0 0.46070 0.46045
816 × 288 1.465773 0.468014 10 0.02970 0.02983
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 7

Input Layer Hidden Layer(s) Output Layer

1 0,1
0,1 1
1
1 1

1
2

, ,

Fig. 2. The schematic view of a typical artificial neural network with three layers.

Similar to the hidden layer, each neuron of the output layer also ap- an algorithm as Minimum redundancy maximum relevance (MRMR)
plies the activation function to the sum of the weighted output of its is applied to select the set of features in order to obtain the maximum
predecessor layer as follows [60]. correlation with the model output and the minimum correlation be-
0 1 tween the input features. In fact, the max-relevance set of features
Xm
may contain features with high mutual correlations. If two features
@
Out i ¼ h v0;i þ v j;i f j A; ð37Þ
have the same aspects of the target class in the same way, there will
j¼1
be no benefit in having both of them as the members of the selected fea-
tures set. The time complexity of the MRMR algorithm is exponential. It
where v0, i and vj, i are the weights between the bias and the j-th neuron
means that it is required to calculate exponentially increasing mathe-
of the hidden layer and the latter with i-th neuron of the output layer,
matical operations for retrieving a long list of parameters. For this rea-
respectively.
son, some methods, such as F-test correlation difference (FCD) and
Usually, the neural network training is performed using the error
Mutual Information Difference (MID) have been proposed to approxi-
propagation algorithm. This process learns the weights between the
mate it. The recent experiments, however, show the interests of MID
network layers through supervised learning. The learning error rate is
in non-linear problem [61]. The MID is computed as the following [62].
calculated from the difference between the neural network outputs
and the expected ones in the training dataset. It should be noted that 2 3
the configuration of the MLP network has a significant impact on its per-
4 1 X
MID ¼ maxi∈φðSÞ Iði; hÞ− Iði; jÞ5; ð39Þ
formance. Typically, to reduce the error rate, the appropriate network S f j∈S
configuration is determined by repeated testing of the various choices.
For this purpose, the number of neurons of hidden layer, and the activa-
tion function of the neurons in the hidden and output layer are deter- where Sf is the set of selected features. The advantage of using MID is
mined by various experiments. In these experiments, different values that although it has an acceptable estimation of MRMR, its computa-
are considered for these parameters and the error value of the model tional complexity is low. In other words, low overhead, high speed
is calculated. and reliability have led to the use of MID to select suitable features.

6.1.1. Feature selection based on mutual information 6.2. Particle Swarm Optimization (PSO)
Selecting useful features has been a challenge in system identifica-
tion problems. There are various methods to select the appropriate in- The PSO algorithm is one of the most important intelligent optimiza-
puts based on the correlation between the data and the model output. tion algorithms that fall into the field of swarm intelligence. This algo-
This choice can be made by maximizing the statistical dependency crite- rithm is inspired by the social behavior of the swarm of birds and fish
rion on the model output and minimizing the Mutual Information (MI) [63]. It starts by collecting random solutions and then generating up-
amongst the set of selected features. This is an applicable way to relate dates to search for the optimal solution. In each iteration, each particle
feature subset to output vector. MI is calculated between two features is updated with two “best” values. The first is the best solution the par-
of x and y with the following equation. ticle has ever obtained in its movement. This is called the pbest value.
Another “best” is the value so far earned by every single particle in the
pðxÞ population. This is the best global value and is called gbest.
Iðx; yÞ ¼ ∬pðx; yÞ log dxdy; ð38Þ
pðxÞpðyÞ The particles move around the search space to find the best solution.
For this purpose, in each iteration, the particle's velocity is separately
in which p(x), p(y) and p(x, y), respectively, are the probability density calculated for all of them using the following equation [64].
functions of the variables x, y, and their simultaneous occurrence.
In order to minimize the overall error of any models, maximal de-
pendence to the model output is essentially required. For this purpose, vi ðt þ 1Þ ¼ wvi ðt Þ þ c1 r 1 ½pbest ðt Þ−xi ðt Þ þ c2 r 2 ½gbest ðt Þ−xi ðt Þ: ð40Þ
8 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

Table 6 shown in Fig. 3. The mean absolute error (MAE) is used to clearly dem-
Ordering of feature prioritization for Nusselt number. onstrate the difference between the calculated and reference values.
Order 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Considering Fig. 3, MAE shows a decreasing trend if the number of fea-
Feature Re λ1 ϕ2 Pr γ∗ D N∗ Df S λ Kc Bi Sr γ Sc Ks
tures grows for all the considered models. However, MLP displays a de-
creasing error and offers the most precise prediction in comparison with
the other models, showing why this model is applied in this study.
Therefore, a feedforward neural network with one hidden layer is
1
MLP LMS SVR RBF used for MLP.
0.9
In addition, to obtain the number of inputs suitable for the model,
the effect of increasing the number of model input features on the accu-
0.8 racy of estimating target parameters is calculated. For example, first
only the feature Re, then two features Re and λ1 are fed. This is then con-
0.7 tinued until all features are inputs of the model. According to Fig. 3, four
features are sufficient to provide an appropriate MLP model, as the error
0.6
falls in the minimum value.
MAE

0.5 The following results were extracted by the estimator algorithm of

MLP. It should be pointed out that in the proceedings figures, the default
0.4 values of the parameters are presented in Table 7, unless the different
values are expressed.
0.3 It should be pointed out that amongst many variables involved in the
current physics, only the most effective parameters were chosen and
0.2
the results are drawn for them. Fig. 4 shows the variation of Nusselt
0.1 number versus different values of the effective parameters. It is clear
that for Re = 0 or stationary flow, by increasing the mixed convection
0 parameter, the Nusselt number raises as a result of intensifying natural
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
convection in comparison with heat conduction. This confirms the pow-
number of features erful impact of natural convection through heat transfer process in the
current problem. Similar effect is also found with increases in Reynolds
Fig. 3. Mean absolute error (MAE) to estimate Nusselt number for various potential number, which is intensifies the forced convection. Increasing the vol-
models. ume fraction of the nanoparticles can pull up the heat transfer through
enhancing conduction in nanofluid. This is more intensive at higher
In Eq. (40), i and t specify the particle's index and the iteration num- values of mixed convection parameter, which is due to stronger contri-
ber, respectively. x and v determine the position and velocity of the par- bution of conduction mechanism to heat transfer at lower Reynolds
ticle, correspondingly. r1 and r2 are the random values that are numbers. Comparing the contribution of these three parameters in
generated each time that the equation is used. c1 and c2 are acceleration heat transfer augmentation throughout the investigated range, volume
coefficients, which are often greater than zero and smaller than two fraction of nanoparticles is the most influential parameter. In keeping
[65]. The inertia coefficient of velocity, which is called w, is usually be- with the literature [27], Fig. 4b shows that reduction in Prandtl number
tween 0.8 and 1.2 [65]. aids the thermal diffusivity and thus cause heat transfer augmentation.
After determining the particle's velocity, the position of the particle As the value of Biot number grows, the temperatures throughout the
is calculated using [64] porous solid and nanofluid approach each other. This is in keeping
with the physical expectations, as significant heat exchanges between
xi ðt þ 1Þ ¼ xi ðt Þ þ vi ðt þ 1Þ: ð41Þ the solid and fluid phases tend to equalise their temperatures. Under
this condition, the local thermal equilibrium assumption is valid and
The PSO algorithm uses an iterative loop whose repetition can be the Nusselt number has slightly improved, rendering increasing heat
equal to a determined number or based on the finding solution with transfer through solid body. However, it is noted that this simple mono-
particular conditions. The detailed algorithm can be found in Ref. [64]. tonic trend only exists for relatively low Prandtl numbers (Pr ≤ 4). The
physical reason for this convoluted behavior is not immediately obvious
7. Results and discussion but it is certainly related to the competition between the solid and fluid
phase in extracting heat from the surface of the cylinder. The thermal
The numerical simulation results are next fed to the predictor and op- conduction of the fluid decreases by increasing Prandtl number and
timizer algorithm. Choosing suitable model for estimating target parame- therefore, the solid takes over the fluid by receiving larger heat from
ters requires taking certain steps. Utilizing MID, stated earlier at Section 5, the cylinder. Fig. 4b further re-confirms the well-known effect of the
needs that the features are firstly prioritized and then discretized. For in- permeability of the porous medium upon Nusselt number [20,27,28],
stance, five effective elementary features after prioritization for Nusselt in which decreases in permeability parameter results in higher values
number are found to be Re, λ1, ϕ2, Pr and γ ∗. The complete ordering of fea- of Nusselt number. It is to be expected that the fluid can be more pene-
ture prioritization for Nusselt number is shown in Table 6. trating in the solid body and more heat is to be received. Fig. 4c illus-
Then, to investigate the effect of this prioritization, some models in- trates that enhancing the interfacial area provides more heat transfer
cluding MLP, support vector regression (SVR) [66], Least Mean Squares surface area and therefore intensifies the heat transfer rate.
algorithm (LMS) [67] and radial basis function (RBF) [68] are tested to In accordance with the classical theory of heat and mass transfer,
estimate Nusselt number, precisely. The result of this experiment is Fig. 5a shows that Sherwood number enhances with increases in

Table 7
Default value of the parameters chosen to draw the figures.

Parameter Re λ1 ϕ2 Pr γ∗ D N∗ Df S λ Kc Bi Sr γ Sc Ks

Values 10.0 10.0 0.1 1.0 1.0 1.0 0.1 1.0 3.7 10 1.0 0.1 0.5 1.5 0.1 0.0
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 9

Fig. 4. Average Nusselt number for various values of Reynolds number (Re), mixed convection parameter (λ1), Biot number (Bi) and volume fraction of copper nanoparticles (ϕ2) at two
specified values of (a) copper nanoparticles volume fraction (ϕ2), (b) permeability parameter (λ) and (c) interfacial area (S).

Fig. 5. Average Sherwood number for various values of Reynolds number (Re), mixed convection parameter (λ1), HOM parameter (Kc), HET parameter (Ks) and volume fraction of copper
nanoparticles (ϕ2) at two specified values of (a), (b) volume fraction of copper nanoparticles (ϕ2) and (c) interfacial area (S).
10 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

Reynolds number by intensifying convective mass transfer. A similar for existence of more uniform fluid temperature throughout the me-
trend is observed by reducing mixed convection parameter which im- dium. Nonetheless, a weak dependency of fluid temperature on the
plies strengthening of forced convection of mass. Increasing the volume permeability is found.
fraction of nanoparticles makes the solutal boundary layer thicker Fig. 6b denotes that by extending the heat exchanging surfaces via
through increases in viscosity and thermal conductivity of nanofluid increasing S, the fluid temperature approaches the values closer to the
[69]. This results in lower mass transfer at higher volume fraction and temperature at infinity as the heat transfer becomes augmented. Varia-
low values of Reynolds number where the viscous flow is dominant. tion of dimensionless temperature with respect to increases in N ∗ indi-
However, this is not the case at higher Reynolds number flows in cates a growing trend. This is emanated from involving the same term
which the flow inertia is being powerful and viscous boundary layer be- Tw − T∞ in the denominator of the non-dimensional parameters of
comes weaker. Fig. 5b depicts that increasing HOM parameter, indicat- N ∗and θ. It is expected that the thermal energy received by the flow in-
ing higher species production through the chemical reaction, renders creases as Dufour number rises. Thus, although Dufour number has no
higher values of Sherwood number. However, increasing HET parame- considerable effect on the temperature in this problem, it can poten-
ter forms a sharp gradient of species on the cylinder surface (see tially enlarge the temperature through the way of developing hot fluid
Eq. (12)). This indicates that mass transfer is dominated by the weaker zones. The mixed convection parameter does not have a considerable
diffusion mechanism rather than convention transfer. This is the reason effect on the temperature, as shown by Fig. 6c. Yet, increasing this pa-
for lower mass transfer by the convection process as HET parameter rameter may decrease the non-dimensional temperature through
grows. Reducing the interfacial area, allows strengthening of the flow shrinkage of the hot fluid zones and spreading them throughout the do-
inertia in comparison with viscous layer. This expectedly results in main [20,71]. Increasing the volume fraction of nanoparticles from 0.1
higher mass transfer as Fig. 5c illustrates and has been also confirmed to 0.3 results in magnification of non-dimensional temperature. This is
in other studies [70]. All the Sherwood number values in Fig. 5 fall inferred from increasing thermal diffusivity, which in turn thickens
lower than unity, showing that mass diffusion is of greater importance the thermal boundary layer. Fig. 6d illustrates that further away from
than forced convection of species. the cylinder surface, the fluid non-dimensional temperature approaches
Magnification of Prandtl number intensifies heat transfer from zero, matching the boundary condition at the infinity. The temperature
the bluff-body, as stated earlier. This causes a smaller difference be- rapidly decreases at a small distance from of the cylinder surface, dem-
tween the temperature of the fluid and infinity. That is why the non- onstrating a sharp gradient.
dimensional temperature is diminished by increasing Prandtl num- Fig. 7a depicts that Biot number variation shows a single-minimum
ber in Fig. 6a. This figure also shows that the dimensionless fluid behavior for the non-dimensional temperature of the solid phase of
temperature is not sensitive to Biot number. Further, lower non- the porous medium. This is followed from the fact that the Biot number
dimensional permeability parameter, indicating higher permeability is a multi-variable function and its influence on the solid temperature
of the porous media, facilitates flow convection and this is the reason can be complex. At about Bi = 30, the solid dimensionless temperature

Fig. 6. The non-dimensional fluid temperature for various values of Prandtl number (Pr), Biot number (Bi), Dufour number (Df), ratio of concentration to thermal buoyancy forces (N ∗),
mixed convection parameter (λ1), copper nanoparticles volume fraction (ϕ2), non-dimensional radial distance (η) at two specified values of (a) permeability parameter (λ), (b),
(c) interfacial area (S) and (d) copper nanoparticles volume fraction (ϕ2).
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 11

Fig. 7. The non-dimensional solid temperature for various values of Prandtl number (Pr), Biot number (Bi), radial distance (η) and modified conductivity ratio (γ) at specified values of
(a) permeability parameter (λ) and (b) mixed convection parameter (λ1).

gains the closest value to that at infinity. However, at low or high Biot Fig. 9b). Keeping the diffusivity at a high values, the non-dimensional
number, the non-dimensional solid temperature is more intensified, in- ks υ1=2
dicating considerable difference between the temperature of body and HET parameter (K s ¼ 1=2
) is drastically reduced, such that there is
Dk
that of infinity. Both Fig. 7a and b show that the solid temperature is no reaction rate- or C2 consumption, until D = 15. Fig. 9b indicates
not a strong function of Prandtl number, permeability parameter and that Soret number, which might be positive or negative [72,73] can affect
modified conductivity ratio. The dimensionless temperature of the the recation through manipulating the variation of mass diffusion [20].
solid phase, however, falls down by taking distance from the cylinder One objective of the current study is to estimate the Nusselt and
surface, such that at η = 4 the solid temperature approaches that of Sherwood number by presenting precise correlations. To do so, similar
infinity. to what performed for the MLP model, the outputs of numerical simula-
Fig. 8a reveals the strong positive correlation between Reynolds tion were fed to PSO for training the algorithm. Further, according to the
number and concentration of species 1, such that increasing Reynolds prioritization of the features, different proposed equations are evaluated
number from zero to 100 causes an increment of about 7 times in con- for the Nusselt and Sherwood number estimation. Tables 8 and 9 show
centration of C1. Decreasing the mixed convection parameter, implying respectively the range of chosen parameters in the Nusselt number es-
magnification of Reynolds number, has a similar effect on the concen- timation and the final correlations extracted. The same type of informa-
tration of species 1. Intensifying Reynolds number, like rising volume tion is presented by Tables 10 and 11 for Sherwood number. In Tables 9
concentration of nanoparticles, allows the heat transfer to be acceler- and 11, the proposed correlations become progressively more accurate
ated. As the surface reaction is exothermic, any modification that aids as more parameters are added to them. This is inferred from mean abso-
the heat transfer makes the production of species more augmented lute error (MAE), rendering as a criterion of accuracy of the current pre-
(see Fig. 8b). This is also obvious from the decreasing concentration of dictor correlations against true values of numerical simulation. That is
the species 2 versus increasing ϕ2 in Fig. 9a. HOM parameter, it does
Xn
not present considerable impact on the concentration in this figure. In- jyi −xi j
creasing HET parameter of reaction rate diminishes the species produc- i¼1
MAE ¼ ; ð42Þ
tion by thickening the solutal boundary layer. Fig. 9a illustrates the n
opposite trend of Fig. 8 with respect to copper nanoparticles volume
fraction and HET parameter, as species 2 is a reactant. The concentration where yi and xi are respectively the predicted and calculated values and
demonstrates no substantial dependency on the Schmidt number (see n is the number of points evaluated. The developed correlations can be

Fig. 8. The non-dimensional concentration of species 1 for various values of Reynolds number (Re), mixed convection parameter (λ1), HOM parameter (Kc) and HET parameter (Ks) at
specified values of (a) and (b) volume fraction of copper nanoparticles (ϕ2).
12 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

Fig. 9. The non-dimensional concentration of species 2 for various values of HOM parameter (Kc), HET parameter (Ks), diffusion coefficient (D) and Schmit number (Sc) at specified values
of (a) volume fraction of copper nanoparticles (ϕ2) and (b) Soret number (Sr).

used in other problems that are physically similar to that investigated in parametric study of such cumbersome problem are computationally
this work [74,75]. quite demanding. More importantly, derivation of correlations for
Nusselt and Sherwood numbers would be extremely difficult, if not im-
8. Conclusions possible. This paper, therefore, took a novel approach and employed
techniques from artificial intelligence to resolve these issues. A predictor
A comprehensive analysis of heat and mass transfer in a chemically algorithm on the basis of artificial neural network, i.e. multi-layer per-
reacting flow of hybrid nanofluid impinging upon a cylinder embedded ception, was applied. This revealed the variations of temperature and
in porous media was put forward. The problem included a number of concentration fields as well as Nusselt and Sherwood numbers with re-
underlying physical and chemical processes including mixed convec- spect to pertinent parameters. Further, the correlations for Nusselt and
tion, Soret and Dufour effects, homogenous and heterogeneous chemical Sherwood numbers were presented using particle swarm optimization
reactions and local thermal non-equilibrium. Conventional methods for (PSO) algorithm. It was observed that increasing interfacial area could

Table 8
Range of variation for effective parameters involving the Nusselt number correlation.

Re λ1 ϕ2 Pr γ∗

Effective parameters 0.1 ≤ Re ≤ 150 0.1 ≤ λ1 ≤ 50 0 ≤ ϕ2 ≤ 0.4 0.1 ≤ Pr ≤ 7.0 0 ≤ γ ∗ ≤ 7.0

Table 9
Nusselt number correlations.

Nusselt correlation Effective parameters Mean absolute error

Nu = 1.755 + 0.23 × Re0.636 Re 0.2770

Nu ¼ 1:755 þ 0:23  Re0:636  λ1 0:146 Re, λ1 0.2560
Nu ¼ 1:832 þ 1:732  Re0:661  λ1 0:156  ϕ2 0:938 Re, λ1, ϕ2 0.2034
Nu ¼ 1:832 þ 1:732  Re0:661  λ1 0:156  ϕ2 0:938  Pr 0:549 Re, λ1, ϕ2, Pr 0.1568
Nu ¼ 1:832 þ 1:732  Re0:661  λ1 0:156  ϕ2 0:938  Pr 0:549  γ  8:642 Re, λ1, ϕ2, Pr , γ ∗ 0.1421

Table 10
Range of variation for effective parameters involving the Sherwood number correlation.

Sc γ∗ λ1 D Sr N∗

Effective parameters 0.1 ≤ Sc ≤ 3.0 0 ≤ γ ∗ ≤ 7.0 0.1 ≤ λ1 ≤ 50 0.1 ≤ D ≤ 30 −1.0 ≤ Sr ≤ 1.0 0 ≤ N ∗ ≤ 2.0

Table 11
Sherwood number correlations.

Sherwood correlation Effective parameters Mean absolute error

Sh = 0.027 + 0.604 × Sc0.385 Sc 0.1229

Sh = 0.099 + 0.512 × Sc0.462 × γ ∗1.031 Sc, γ ∗ 0.0932
Sh ¼ 0:099 þ 0:512  Sc0:462  γ  1:031  λ1 0:257 Sc, γ ∗, λ1, 0.0854
Sh ¼ 0:141 þ 0:454  Sc0:525  γ  1:159  λ1 0:305  D0:269 Sc, γ ∗, λ1, D 0.0789
Sh ¼ 0:221 þ 0:637  Sc 0:759
γ  1:617
 λ1 0:500
D 0:479
 Sr 1:011 Sc, γ ∗, λ1, D, Sr 0.0734
Sh ¼ 0:242 þ 0:547  Sc0:893  γ  1:876  λ1 0:621  D0:619  Sr 1:045  N
1:552 Sc, γ ∗, λ1, D, Sr, N ∗ 0.0677
 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492 13

enhance heat transfer and non-dimensional temperature, while it weak- [11] M. Akbarzadeh, S. Rashidi, N. Karimi, N. Omar, First and second laws of thermody-
namics analysis of nanofluid flow inside a heat exchanger duct with wavy walls
ened mass transfer. Decreasing mixed convection parameter, or increas- and a porous insert, J. Therm. Anal. Calorim. 135 (1) (2019) 177–194.
ing Reynolds number, promoted heat transfer and dimensionless fluid [12] D.G. Guthrie, M. Torabi, N. Karimi, Combined heat and mass transfer analyses in cat-
temperature. However, it had a negligible effect on the temperature of alytic microreactors partially filled with porous material-the influences of nanofluid
and different porous-fluid interface models, Int. J. Therm. Sci. 140 (2019) 96–113.
the solid phase. Although reducing permeability lead to intensification [13] D.G. Guthrie, M. Torabi, N. Karimi, Energetic and entropic analyses of double-
of heat and mass transfer, solid temperature had little dependency diffusive, forced convection heat and mass transfer in microreactors assisted with
upon that. Volume fraction of nanoparticles was shown to be an influen- nanofluid, J. Therm. Anal. Calorim. 137 (2) (2019) 637–658.
[14] M. Torabi, C. Dickson, N. Karimi, Theoretical investigation of entropy generation and
tial parameter on the heat transfer and species production. HOM param-
heat transfer by forced convection of copper–water nanofluid in a porous channel—
eter appeared to be ineffective on the species concentration. Yet, HET local thermal non-equilibrium and partial filling effects, Powder Technol. 301
parameter presented a strong contribution to species consumption (2016) 234–254.
and production. The importance of Soret effect on the species con- [15] Q. Xiong, I. Tlili, J.A. Ali, S.M. Hamad, A. Shafee, T. Nguyen-Thoi, H. Hamouda, Model-
ing of hybrid nanofluid behavior within a permeable media involving buoyancy ef-
centration was modified by magnification of diffusivity. As an im- fect, Physica A: Statistical Mechanics and its Applications (2019) 123940.
portant outcome, correlations for Nusselt and Sherwood numbers [16] S.A. Mehryan, F.M. Kashkooli, M. Ghalambaz, A.J. Chamkha, Free convection of hy-
were developed systematically through a progressive increase in brid Al2O3-Cu water nanofluid in a differentially heated porous cavity, Adv. Powder
Technol. 28 (9) (2017) 2295–2305.
the number of considered parameters. A high level of accuracy was [17] H. Sajjadi, A.A. Delouei, M. Izadi, R. Mohebbi, Investigation of MHD natural convec-
reported for correlations that involve more than five parameters. tion in a porous media by double MRT lattice Boltzmann method utilizing MWCNT–
This study clearly demonstrated the potentials of artificial intelli- Fe3O4/water hybrid nanofluid, Int. J. Heat Mass Transf. 132 (2019) 1087–1104.
[18] T.D. Manh, N.D. Nam, G.K. Abdulrahman, M.H. Khan, I. Tlili, A. Shafee, M. Shamlooei,
gence in tackling complicated problems in thermo-chemical and
T. Nguyen-Thoi, Investigation of hybrid nanofluid migration within a porous closed
solutal systems that are not amenable to conventional methods of domain, Physica A: Statistical Mechanics and its Applications (2019) 123960.
analysis. Future studies may concentrate on more complicated con- [19] S.A.M. Mehryan, M.A. Sheremet, M. Soltani, M. Izadi, Natural convection of magnetic
figurations or addition of further details to the chemistry and trans- hybrid nanofluid inside a double-porous medium using two-equation energy
model, J. Mol. Liq. 277 (2019) 959–970.
port phenomena. [20] R. Alizadeh, N. Karimi, A. Mehdizadeh, A. Nourbakhsh, Analysis of transport from cy-
lindrical surfaces subject to catalytic reactions and non-uniform impinging flows in
CRediT authorship contribution statement porous media, J. Therm. Anal. Calorim. 138 (1) (2019) 659–678.
[21] A. Jarray, Z. Mehrez, A. El Cafsi, Mixed convection Ag-MgO/water hybrid nanofluid
flow in a porous horizontal channel, The European Physical Journal Special Topics
Javad Mohebbi Najm Abad: Data curation, Formal analysis, Investi- 228 (12) (2019) 2677–2693.
gation. Rasool Alizadeh: Formal analysis, Validation. Abolfazl Fattahi: [22] E. Aminian, H. Moghadasi, H. Saffari, Magnetic field effects on forced convection
flow of a hybrid nanofluid in a cylinder filled with porous media: a numerical
Writing - review & editing. Mohammad Hossein Doranehgard: Soft-
study, J. Therm. Anal. Calorim. (2020) 1–13.
ware. Ebrahim Alhajri: Software. Nader Karimi: Conceptualization, Su- [23] N. Abbas, M.Y. Malik, S. Nadeem, Study of three dimensional stagnation point flow
pervision, Writing - review & editing. of hybrid nanofluid over an isotropic slip surface, Physica A: Statistical Mechanics
and its Applications (2020), 124020. https://doi.org/10.1016/j.physa.2019.124020.
[24] D. Pal, G. Mandal, Mixed convection–radiation on stagnation-point flow of
nanofluids over a stretching/shrinking sheet in a porous medium with heat gener-
Declaration of competing interest ation and viscous dissipation, J. Pet. Sci. Eng. 126 (2015) 16–25.
[25] F.M. Ali, R. Nazar, N.M. Arifin, I. Pop, MHD stagnation-point flow and heat transfer
towards stretching sheet with induced magnetic field, Appl. Math. Mech. 32 (4)
The authors declare that they have no known competing financial (2011) 409–418.
interests or personal relationships that could have appeared to influ- [26] F. Mabood, W.A. Khan, Approximate analytic solutions for influence of heat transfer
ence the work reported in this paper. on MHD stagnation point flow in porous medium, Comput. Fluids 100 (2014)
72–78.
[27] R. Alizadeh, A.B. Rahimi, N. Karimi, A. Alizadeh, On the hydrodynamics and heat con-
Acknowledgment vection of an impinging external flow upon a cylinder with transpiration and em-
bedded in a porous medium, Transp. Porous Media 120 (3) (2017) 579–604.
[28] R. Alizadeh, N. Karimi, R. Arjmandzadeh, A. Mehdizadeh, Mixed convection and
N. Karimi acknowledges the financial support of Engineering and thermodynamic irreversibilities in MHD nanofluid stagnation-point flows over a
Physical Science Research Council through grant EP/N020472/1. cylinder embedded in porous media, J. Therm. Anal. Calorim. 135 (1) (2019)
489–506.
[29] N.S. Khashi’ie, N.M. Arifin, R. Nazar, E.H. Hafidzuddin, N. Wahi, I. Pop, Magnetohy-
References
drodynamics (MHD) axisymmetric flow and heat transfer of a hybrid nanofluid
past a radially permeable stretching/shrinking sheet with joule heating, Chin. J.
[1] A.A. Nassani, A.M. Aldakhil, M.M.Q. Abro, K. Zaman, A. Kabbani, Resource manage-
Phys. 64 (2020) 251–263.
ment for green growth: ensure environment sustainability agenda for mutual exclu-
[30] S.B. Hacene, F. Ghomari, F. Schoefs, A. Khelidj, Probabilistic modelling of compres-
sive global gain, Environ. Prog. Sustain. Energy 38 (4) (2019), 13132.
sive strength of concrete using response surface methodology and neural networks,
[2] J.R. Vázquez-Canteli, S. Ulyanin, J. Kämpf, Z. Nagy, Fusing TensorFlow with building
Arab. J. Sci. Eng. 39 (6) (2014) 4451–4460.
energy simulation for intelligent energy management in smart cities, Sustain. Cities
[31] M. Mohanraj, S. Jayaraj, C. Muraleedharan, Applications of artificial neural networks
Soc. 45 (2019) 243–257.
for thermal analysis of heat exchangers–a review, Int. J. Therm. Sci. 90 (2015)
[3] A.M. Abdulateef, S. Mat, J. Abdulateef, K. Sopian, A.A. Al-Abidi, Geometric and design 150–172.
parameters of fins employed for enhancing thermal energy storage systems: a re- [32] A. Abdollahi, M. Shams, A. Abdollahi, Artificial neural network modeling of a deflec-
view, Renew. Sust. Energ. Rev. 82 (2018) 1620–1635. tor in a grooved channel as well as optimization of its effective parameters, Heat
[4] G.K. Chhaparwal, A. Srivastava, R. Dayal, Artificial repeated-rib roughness in a solar Mass Transf. 54 (1) (2018) 59–68.
air heater–a review, Sol. Energy 194 (2019) 329–359. [33] A. Abdollahi, M. Shams, Optimization of heat transfer enhancement of nanofluid in a
[5] T. Alam, R.P. Saini, J.S. Saini, Heat and flow characteristics of air heater ducts pro- channel with winglet vortex generator, Appl. Therm. Eng. 91 (2015) 1116–1126.
vided with turbulators—a review, Renew. Sust. Energ. Rev. 31 (2014) 289–304. [34] C. Uysal, M.E. Korkmaz, Estimation of entropy generation for Ag-MgO/water hybrid
[6] S.U. Choi, J.A. Eastman, Enhancing Thermal Conductivity of Fluids With Nanoparti- nanofluid flow through rectangular minichannel by using artificial neural network,
cles (No. ANL/MSD/CP-84938; CONF-951135-29), Argonne National Lab, IL Politeknik Dergisi 22 (1) (2019) 41–51.
(United States), 1995. [35] H. Bagheri, M. Behrang, E. Assareh, M. Izadi, M.A. Sheremet, Free convection of hy-
[7] M.H. Esfe, S. Esfandeh, M.K. Amiri, M. Afrand, A novel applicable experimental study brid nanofluids in a C-shaped chamber under variable heat flux and magnetic
on the thermal behavior of SWCNTs (60%)-MgO (40%)/EG hybrid nanofluid by fo- field: simulation, sensitivity analysis, and artificial neural networks, Energies 12
cusing on the thermal conductivity, Powder Technol. 342 (2019) 998–1007. (14) (2019) 2807.
[8] M. Afrand, Experimental study on thermal conductivity of ethylene glycol contain- [36] Q. Xiong, M. Ayani, A.A. Barzinjy, R.N. Dara, A. Shafee, T. Nguyen-Thoi, Modeling of
ing hybrid nano-additives and development of a new correlation, Appl. Therm. heat transfer augmentation due to complex-shaped turbulator using nanofluid,
Eng. 110 (2017) 1111–1119. Physica A: Statistical Mechanics and Its Applications 540 (2020), 122465.
[9] W. Yu, D.M. France, J.L. Routbort, S.U. Choi, Review and comparison of nanofluid [37] A. Izadi, M. Siavashi, H. Rasam, Q. Xiong, MHD enhanced nanofluid mediated heat
thermal conductivity and heat transfer enhancements, Heat transfer engineering transfer in porous metal for CPU cooling, Appl. Therm. Eng. 168 (2020), 114843.
29 (5) (2008) 432–460. [38] M. Sadeghzadeh, H. Maddah, M.H. Ahmadi, A. Khadang, M. Ghazvini, A. Mosavi, N.
[10] C. Kleinstreuer, Y. Feng, Experimental and theoretical studies of nanofluid thermal Nabipour, Prediction of thermo-physical properties of TiO2-Al2O3/water nanoparti-
conductivity enhancement: a review, Nanoscale Res. Lett. 6 (1) (2011) 229. cles by using artificial neural network, Nanomaterials 10 (4) (2020) 697.
14 J. Mohebbi Najm Abad et al. / Journal of Molecular Liquids 313 (2020) 113492

[39] F. Nasirzadehroshenin, M. Sadeghzadeh, A. Khadang, H. Maddah, M.H. Ahmadi, H. [56] C.Y. Wang, Axisymmetric stagnation flow on a cylinder, Q. Appl. Math. 32 (2) (1974)
Sakhaeinia, L. Chen, Modeling of heat transfer performance of carbon nanotube 207–213.
nanofluid in a tube with fixed wall temperature by using ANN–GA, The European [57] R.S.R. Gorla, Heat transfer in an axisymmetric stagnation flow on a cylinder, Appl.
Physical Journal Plus 135 (2) (2020) 1–20. Sci. Res. 32 (5) (1976) 541–553.
[40] M.H. Ahmadi, B. Mohseni-Gharyehsafa, M. Farzaneh-Gord, R.D. Jilte, R. Kumar, K.W. [58] M. Hamid, T. Zubair, M. Usman, Z.H. Khan, W. Wang, Natural convection effects on
Chau, Applicability of connectionist methods to predict dynamic viscosity of silver/ heat and mass transfer of slip flow of time-dependent Prandtl fluid, Journal of Com-
water nanofluid by using ANN-MLP, MARS and MPR algorithms, Engineering Appli- putational Design and Engineering 6 (4) (2019) 584–592.
cations of Computational Fluid Mechanics 13 (1) (2019) 220–228. [59] M.H. Hassoun, Fundamentals of Artificial Neural Networks, MIT press, 1995.
[41] A. Baghban, A. Jalali, M. Shafiee, M.H. Ahmadi, K.W. Chau, Developing an ANFIS- [60] A.Y. Alanis, N. Arana-Daniel, C. Lopez-Franco, Artificial Neural Networks for Engi-
based swarm concept model for estimating the relative viscosity of nanofluids, En- neering Applications, Academic Press, 2019.
gineering Applications of Computational Fluid Mechanics 13 (1) (2019) 26–39. [61] N. AlNuaimi, M.M. Masud, M.A. Serhani, N. Zaki, Streaming feature selection algo-
[42] G. Hunt, N. Karimi, M. Torabi, Two-dimensional analytical investigation of coupled rithms for big data: a survey, Applied Computing and Informatics (2019)https://
heat and mass transfer and entropy generation in a porous, catalytic microreactor, doi.org/10.1016/j.aci.2019.01.001.
Int. J. Heat Mass Transf. 119 (2018) 372–391. [62] C.O. Sakar, O. Kursun, F. Gurgen, A feature selection method based on kernel canon-
[43] G.F. Al-Sumaily, H.M. Hussen, M.C. Thompson, Validation of thermal equilibrium as- ical correlation analysis and the minimum Redundancy–Maximum Relevance filter
sumption in free convection flow over a cylinder embedded in a packed bed, Inter- method, Expert Syst. Appl. 39 (3) (2012) 3432–3437.
national Communications in Heat and Mass Transfer 58 (2014) 184–192. [63] K.E. Parsopoulos, M.N. Vrahatis, Particle Swarm Optimization and Intelligence: Ad-
[44] C. Liu, M. Pan, L. Zheng, P. Lin, Effects of heterogeneous catalysis in porous media on vances and Applications, 2010.
nanofluid-based reactions, International Communications in Heat and Mass Transfer [64] T.V. Sibalija, Particle swarm optimisation in designing parameters of manufacturing
110 (2020), 104434. processes: a review (2008–2018), Appl. Soft Comput. 84 (2019), 105743.
[45] T. Hayat, M. Mustafa, S. Mesloub, Mixed convection boundary layer flow over a [65] F. Chan, M. Tiwari, Swarm Intelligence: Focus on Ant and Particle Swarm Optimiza-
stretching surface filled with a Maxwell fluid in presence of Soret and Dufour ef- tion, BoD–Books on Demand, 2007.
fects, Zeitschrift für Naturforschung A 65 (5) (2010) 401–410. [66] H. Drucker, C.J. Burges, L. Kaufman, A.J. Smola, V. Vapnik, Support vector regression
machines, Adv Neural Inf Process Syst (1997) 155–161.
[46] N. Freidoonimehr, M.M. Rashidi, S. Abelman, G. Lorenzini, Analytical modeling of
[67] T. Strutz, Data Fitting and Uncertainty. A Practical Introduction to Weighted Least
MHD flow over a permeable rotating disk in the presence of Soret and Dufour ef-
Squares and Beyond, Vieweg+ Teubner, 2010.
fects: entropy analysis, Entropy 18 (5) (2016) 131.
[68] D.S. Broomhead, D. Lowe, Radial Basis Functions, Multi-variable Functional Interpo-
[47] C.G. Hill, T.W. Root, An Introduction to Chemical Engineering Kinetics & Reactor De-
lation and Adaptive Networks (No. RSRE-MEMO-4148), Royal Signals and Radar Es-
sign, Wiley, New York, 1977 245–316.
tablishment Malvern, United Kingdom, 1988.
[48] J. Ahmed, M. Khan, L. Ahmad, Stagnation point flow of Maxwell nanofluid over a
[69] R. Chowdhury, S. Parvin, M.A.H. Khan, Finite element analysis of double-diffusive
permeable rotating disk with heat source/sink, J. Mol. Liq. 287 (2019), 110853.
natural convection in a porous triangular enclosure filled with Al2O3-water
[49] R. Alizadeh, A.B. Rahimi, R. Arjmandzadeh, M. Najafi, A. Alizadeh, Unaxisymmetric nanofluid in presence of heat generation, Heliyon 2 (8) (2016), e00140.
stagnation-point flow and heat transfer of a viscous fluid with variable viscosity [70] J. Cho, M.D. Annable, P.S.C. Rao, Measured mass transfer coefficients in porous media
on a cylinder in constant heat flux, Alexandria Engineering Journal 55 (2) (2016) using specific interfacial area, Environmental science & technology 39 (20) (2005)
1271–1283. 7883–7888.
[50] R. Alizadeh, A.B. Rahimi, M. Najafi, Unaxisymmetric stagnation-point flow and heat [71] A. Hamid, M. Alghamdi, M. Khan, A.S. Alshomrani, An investigation of thermal and
transfer of a viscous fluid on a moving cylinder with time-dependent axial velocity, solutal stratification effects on mixed convection flow and heat transfer of
J. Braz. Soc. Mech. Sci. Eng. 38 (1) (2016) 85–98. Williamson nanofluid, J. Mol. Liq. 284 (2019) 307–315.
[51] Q. Xiong, M. Vaseghi, J.A. Ali, S.M. Hamad, M. Jafaryar, M. Sheikholeslami, Z. Li, Nano- [72] G. Hunt, N. Karimi, B. Yadollahi, M. Torabi, The effects of exothermic catalytic reac-
particle application for heat transfer and irreversibility analysis in an air condition- tions upon combined transport of heat and mass in porous microreactors, Int. J. Heat
ing unit, J. Mol. Liq. 292 (2019), 111372. Mass Transf. 134 (2019) 1227–1249.
[52] Q. Xiong, E. Abohamzeh, J.A. Ali, S.M. Hamad, I. Tlili, A. Shafee, T.K. Nguyen, Influ- [73] C. Dickson, M. Torabi, N. Karimi, First and second law analysis of nanofluid convec-
ences of nanoparticles with various shapes on MHD flow inside wavy porous tion through a porous channel-the effects of partial filling and internal heat sources,
space in appearance of radiation, J. Mol. Liq. 292 (2019), 111386. J. Appl. Therm. Eng. 103 (2016) 459–480.
[53] Q. Xiong, M. Jafaryar, A. Divsalar, M. Sheikholeslami, A. Shafee, D.D. Vo, Z. Li, Macro- [74] R. Razavi, A. Sabaghmoghadam, A. Bemani, A. Baghban, K.W. Chau, E. Salwana, Ap-
scopic simulation of nanofluid turbulent flow due to compound turbulator in a pipe, plication of ANFIS and LSSVM strategies for estimating thermal conductivity en-
Chem. Phys. 527 (2019), 110475. hancement of metal and metal oxide based nanofluids, Engineering Applications
[54] S.S. Ghadikolaei, M. Yassari, H. Sadeghi, K. Hosseinzadeh, D.D. Ganji, Investigation on of Computational Fluid Mechanics 13 (1) (2019) 560–578.
thermophysical properties of Tio2–Cu/H2O hybrid nanofluid transport dependent on [75] M.H. Ahmadi, M. Sadeghzadeh, H. Maddah, A. Solouk, R. Kumar, K.W. Chau, Precise
shape factor in MHD stagnation point flow, Powder Technol. 322 (2017) 428–438. smart model for estimating dynamic viscosity of SiO2/ethylene glycol–water
[55] T.L. Bergman, F.P. Incropera, A.S. Lavine, D.P. DeWitt, Introduction to Heat Transfer, nanofluid, Engineering Applications of Computational Fluid Mechanics 13 (1)
John Wiley & Sons, 2011. (2019) 1095–1105.