International Journal of Heat and Mass Transfer 182 (2022) 121971

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

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

Numerical optimization of the fin shape experiments of a heat

conjugate problem surface air/oil heat exchanger (SACOC)
Miguel Chávez-Modena a,b,∗, Leo Miguel González a,b, Eusebio Valero a,b
a
ETSIAE–UPM - School of Aeronautics, Universidad Politécnica de Madrid, Plaza Cardenal Cisneros 3, Madrid E-28040, Spain
b
Center for Computational Simulation, Universidad Politécnica de Madrid, Campus de Montegancedo, Boadilla del Monte, Madrid, 28660, Spain

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

Article history: Efficient aircraft engine designs imply the presence of a surface air/oil heat exchanger placed at the by-
Received 5 April 2021 pass flow separation of the engine. The heat exchanger consists of several parallel longitudinal fins that
Revised 10 June 2021
increase the contact area to obtain a higher heat dissipation rate. The design of these fins is an impor-
Accepted 10 September 2021
tant task as the pressure drop generated by the presence of the heat exchanger is the largest drawback
Available online 24 September 2021
to be minimized in the final setup. Consequently, an efficient design of the fins that optimizes the global
Keywords: performance of the heat exchanger is highly demanded. The result of this optimization process should
SACOC heat exchanger minimize the pressure drop caused by the exchanger without decreasing the heat transfer. The optimiza-
Aircraft engines tion methodology proposed in this work is split in two independent parts: in the first one, the fin shape
Optimization is optimized using the adjoint method and the derived sensitivity function that controls the position of
Adjoint method the fin shape design. In parallel, the range of fin thickness and distance between fins was investigated
Finite volumes
to have an optimized fin distribution for a given SACOC width. Once the geometry was optimized, the
Conjugate heat problem
coupled conjugate heat problem is numerically solved using realistic conditions showing good accuracy
in the two presented validations. The result for the optimized geometry presents a substantial pressure
drop reduction with little heat transfer variation, addressing the objectives that motivated the present
study.
© 2021 The Authors. Published by Elsevier Ltd.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)

1. Introduction changer assembled on the inner wall of the secondary duct of a

turbofan, could be an interesting quantitative study. The goal of
Advanced oil cooling strategies for more efficient engine archi- such design is to use the available surface as a heat exchanger be-
tectures are one of the technological challenges that demand more tween the air and the oil. To increase the thermal performance,
intense research. New engine designs based on inlet air stream the wet area is increased by adding longitudinal fins, reaching
and the fuel circuit are approaching their limits as typical cold the required heat dissipation rate. Such a design implies a strong
sources [1]. Complex coupled problems are found in this kind of compromise between the aerodynamic penalties, quantified by the
mechanical systems, where the thermal part must be treated ad- drag increase or pressure drop, and the improvement in the ther-
equately. It should be taken into account that the heat removal mal performance of the heat exchanger. The developed research
by the aircraft structure will be limited using composite materi- presents an optimization study of a heat exchanger in realistic flow
als with lower operational temperature and thermal conductivity conditions within the bypass of an engine.
properties. Furthermore, the limitation on the maximum fuel tem- This surface air/oil heat exchanger (SACOC) can be located on
perature decreases the viability of the fuel tank as a cold source the lip separating the engine bypass flow from the core flow, be-
[2]. tween the fan and the downstream outlet guide vane. The SACOC is
Therefore, a novel engine cooling strategy seems to be a new used to evacuate, in the bypass air stream, a large quantity of heat
industrial demand [3]. In this context, the evaluation of the ther- contained in the oil (contributing therefore to the thrust of the en-
mal performance of an Air Cooled Oil Cooler (ACOC) heat ex- gine) with minimal perturbation and therefore a minimal impact
on Specific Fuel Consumption (SFC). The coolant source is the by-
pass mass flow delivered by the fan. Indeed, the performance of
∗
Corresponding author at: ETSIAE–UPM - School of Aeronautics, Universidad the SACOC is measured in terms of maximum heat release capacity
Politécnica de Madrid, Plaza Cardenal Cisneros 3, Madrid E-28040, Spain.
E-mail address: m.chavez@upm.es (M. Chávez-Modena).

https://doi.org/10.1016/j.ijheatmasstransfer.2021.121971
0017-9310/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

We should remark that despite much literature can be found

Nomenclature regarding the heat exchangers in, for example, the electronic in-
dustry [5,6], here we are particularly interested in the flow and
X, Y, Z cartesian axis temperature conditions present in aircraft engine applications. This
x, y, z spatial coordinates SACOC heat exchanger concept was designed to reproduce the by-
ui ith component of the velocity pass flow of a high-bypass ratio turbofan at the most critical op-
field erational point for the oil thermal system, i.e., cruise velocity and
ui ith component of the turbulent take-off atmospheric conditions. The use of heat exchangers inter-
deviation of the fluid velocity acting with the engine bypass flow has been proposed in many
p pressure field patents [7–10], also the open literature provides numerous refer-
uexp experimental inflow profile ences [11–13].
T temperature field The geometry involved in the heat exchange is one of the most
θ conjugate boundary tempera- critical variables of the system performance. Starting from the sim-
ture plest case where forced convective heat transfer in a boundary
T∞ inflow temperature layer flow was studied over a flat plate by Pelerman [14] and Vyn-
n wall normal direction nycky et al. [15], other fin geometries such as plates, strips, and
U∞ maximum inflow velocity pins can be used as heat sinks giving different performance. The
ρ∞ inflow density performance of plate fin heat sinks has been studied extensively
ps outflow static pressure by Sparrow and co-workers [16–18], Lau and Mahajan [19], Wirtz
L, H, W domain dimensions and co-workers [20,21] by experimental tests, numerically and an-
LF IN, HF IN, W F , SF IN, LU, LD SACOC geometrical dimensions alytically by Kadle and Sparrow [22], Sata et al. [23] and Jonsson
a, b, LB, HB, W B base geometrical dimensions and Moshfegh [24,25] and both experimentally and numerically by
ρ fluid density Iwasaki et al. [26].
Q˙ heat transfer per time unit The heat transfer enhancement by introducing strip fins and its
m˙ mass flow comparison between plate and staggered strip fin heat sinks has
Re Reynolds number been studied by Sparrow [27,28] for laminar flow cases, and by
Axy lateral fin area other authors [29–31].
N number of control points An exhaustive review of the heat-exchange technology applied
Pr Prandtl number to cooling-air cooler applications was performed by Min et al. [32],
Nu Nusselt number and particularized for cooling oil systems in Kim et al. [33]. The
M Mach number surface air-oil heat exchanger (SACOC) is located inside the engine
μ fluid viscosity fan casing, and dissipates the heat from the oil into the air stream
ν fluid kinematic viscosity in the bypass duct (BPD), as shown in Fig. 1. The main difficulty
dh hydraulic diameter of SACOC design is the existence of the bypass stream, which af-
w average air velocity fects the pressure drop inside the bypass duct. There have been a
V volumetric flow rate number of studies aimed at enhancing aero-thermal performance
k thermal conductivity by varying the fin shape of the heat exchanger. It was found that a
Rth thermal resistance modified fin surface may have a high heat transfer coefficient, but
δ boundary layer thickness that its pressure drop is sometimes excessive for wide applications.
p pressure drop Several studies using computational fluid dynamics (CFD) have
Cp specific heat capacity been extensively applied to the study and comparison of different
Tc temperature at the bottom of SACOC designs. Kim et al. [34] compared numerically the perfor-
the fins mance of a plate and pin-fin shaped geometry as the surface of an
h local heat transfer coefficient air-oil heat exchanger. Basic heat transfer and pressure drop char-
hg global heat transfer coefficient acteristics were examined using a simplified channel model. Using
Af total fin area a parametric study, the optimal fin pitches of the pin-fin geome-
Hc characteristic mesh size close to tries in stream- and spanwise directions were determined. Finally,
the fin surface the high-speed bypass effect of the surface air-oil heat exchanger
H f in characteristic fin mesh size was calculated using the geometry of a real engine. He also ex-
amined [33] the sensitivity of finned coolers to perturbations on
their location and orientation using RANS simulations. Another nu-
with minimal perturbation of the flow, in terms of pressure losses merical comparison between the aerodynamic disturbances caused
and acoustic emission. by heated fins, where two different designs, continuous and inter-
Surface Air Cooled Oil Coolers are normally composed of par- rupted fins were tested in Sousa et al. [4].
allel fins integrated in the inner wall of the secondary duct of a Regarding the experimental approach to the problem in realis-
turbofan. The geometrical characteristics and location of the fins tic conditions, a particular air/oil heat exchanger design, see Fig. 1,
within the engine are designed to minimize aerodynamic effects composed by fins aligned with the flow direction and integrated
(drag and noise increments) while maximizing thermal exchange. in the surface of the core/bypass flow splitter, downstream of the
This heat exchanger configuration introduces less aerodynamic per- flow bifurcation was studied in Villafañe and Paniagua [35]. We
turbations than current plate/tube heat exchanger modules. These should also mention the experimental contribution based on the
losses were estimated at approximately 1% for the full annular sec- sensitivity of the complex transonic and three-dimensional tur-
tion of a turbofan with a bypass ratio equal to 6 [4]. The analysis bofan bypass-flow to arrays of fins embedded on the splitter is
of the interaction between the three-dimensional high velocity by- performed by Villafañe and Paniagua [35]. They assess the flow
pass flow and the heat exchangers is essential to evaluate and op- modifications introduced by two different fin heat exchanger de-
timize the aero-thermodynamic performance, and to provide data signs, with continuous and interrupted fins. This work demonstrates
for engine modelling.

2
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 1. EPNdB configuration studied by UPV, VKI and Purdue University [4,35].

the importance of aerodynamically optimized designs to minimize

detrimental effects on propulsive efficiency.
The goal of this work is clear, given the typical SACOC design
and flow conditions, we will develop a strategy to optimize the
design of the fin geometry and setup to minimize the pressure
drop due to the exchanger while keeping the heat transfer. Previ-
ous works based on shape optimization problems can be found in
Morimoto et al. [36], Gkaragkounis et al. [37], Subramaniam et al.
[38] but most of these studies are not referred to the conjugate
heat transfer problem and just optimize the shape of a body in
terms of drag, or are far from the flow conditions expressed here.
This work is organised as follows: First, the numerical method-
ology is presented in Section 2. Then, a validation of the multi-
physics methodology for a 2D flat plate and a 3D finned geome- Fig. 2. Computational scheme of one half of 3D symmetric finned geometry tested.
try is performed and compared with the experimental results in
Section 3. Finally, in Section 4 the optimization methodology is ex-
plained and applied to a realistic SACOC geometry, the problem is
solved for the optimized geometry and compared to the original
one.

2. Numerical methodology.

2.1. Description of the problem.

In this section, we are going to simplify the intrinsic complexity

that is found in the technological problem described in Section 1.
The first step is to use a simplified geometrical model to exam-
ine the aero-thermal performance of a finned surface in a reduced
part of the global geometry. In this simplified geometry, the slight
curvature of the model normally present in industrial designs will Fig. 3. Computational scheme of the periodic 3D geometry tested.
be neglected in the computational domain. The modeled geometry
is formed by fins lying on a flat base which is heated through its
lower surface. The fin and base are both made up of aluminium.
The coordinate system is such that the X axis follows the stream- The problem domain can be decomposed in two parts: the fluid
wise direction, the Y axis is in the vertical direction and Z axis is part ruled by the compressible Reynolds-averaged Navier–Stokes
orthogonal to the XY plane. Due to the symmetry of the problem, (RANS) equations, and the solid part where a heat transfer prob-
only half of the complete geometry (only 8 of the 16 fins) will be lem is solved. Both problems are coupled by the appropriate com-
considered for the computations, see Fig. 2. The geometrical pa- patibility conditions through the cooling fin surface, where no-
rameters shown in Fig. 2 are defined in Table 2. slip velocity, temperature, and heat flux continuity are applied.
The configuration shown in Fig. 2 is used for the analysis of When possible, the distances will be non-dimensionalized with the
the fin geometrical parameters of Sections 4.3.1 and 4.3.2. How- length of the fins in the flow direction LFIN, velocity with the max-
ever, when tasks such as the shape fin optimization performed in imum inflow value U∞ , density with the inflow value ρ∞ pressure
Section 4.3.3, a periodic single fin geometry is used, see Fig. 3, with the outflow static pressure ps , and temperatures as the Conju-
−T∞
which implies fewer cells and consequently there is less compu- gate Boundary Temperature, θ = T  T
, where Tc is the temperature
tational demand. at the bottom of the base and T = Tc − T∞ is the temperature dif-
ference between the heated wall base and the inflow. Below, the
2.2. Governing equations and boundary conditions subindexes f and s are used to determine the fluid and solid tem-
perature regions, respectively.
The conjugate heat transfer problem couples a turbulent com- Considering the air as an ideal gas, the continuity, momentum,
pressible flow to the heat transfer coming from a finned surface. and energy non-dimensional equations for the steady, compress-

3
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 4. Boundary conditions imposed at the different surfaces of half fluid domain
(symmetric hypothesis).
Fig. 5. Boundary conditions imposed at the different surfaces of the solid domain.

ible, and turbulent flow are:

providing the continuity of thermal fields and heat flux between
∂ρ ui
= 0, (1) the body and flow at the interface.
∂ xi Similarly to the fluid case, a symmetry plane is used for the
  thermal problem, being the rest of the boundaries adiabatic, with
∂ρ ui u j ∂p ∂ 1 ∂ ui ∂ u j the only exception of the bottom wall which is isothermal with
= + ( + ) − ρ ui uj , (2)
temperature θs = 1, see Fig. 5.
∂xj ∂ xi ∂ x j Re ∂ x j ∂ xi
In summary, the problem depends on the geometry, the inflow
  profile, and the parameters Re, P r and k.
∂ρ ui θ f ∂ 1 ∂θ f
=− − + ρ ui θ f , (3) To solve correctly the velocity boundary layer, we will ensure
∂ xi ∂ xi ReP r ∂ xi that y+ < 1. As the fluid is a gas, the P r ≈ 1, the velocity and the
where ρ is the fluid density, ui is the fluid velocity in the i direc- thermal boundary layers thickness will be both equivalent δv ≈ δT .
tion, p is the fluid pressure, ui is the turbulent deviation of the
fluid velocity, θ f is the fluid temperature. 2.3. Computational tool
The SST k − ω turbulence model is implemented to compute
the eddy viscosity [39]. The equation of state for perfect gases is As compressible and turbulent effects are taken into account,
also added to the system (1)–(3). The dominant non-dimensional the energy equation is coupled with the momentum equation and
numbers are: the Reynolds number defined as Re = ρ∞U∞ LFIN/μ, the speed of sound is not fixed. A finite volume code is used to
the Prandtl number P r = C p μ/k f and the non-dimensional conduc- discretize both the fluid mechanics and the heat transfer coupled
tivity k = ks /k f . Where μ the fluid viscosity, k f = kair and ks = problem.
kAl the fluid (air) and solid (aluminium) conductivities respec- The coupled system of Eqs. (1)–(3) and (4) are solved using
tively and C p specific heat. Eqs. (1)–(3) are complemented with the an implicit second-order discretization method to compute the
boundary conditions shown in Fig. 4. Navier–Stokes equations and to calculate the density, temperature,
As the problem presents a symmetry plane z = 0, a symmetry pressure and velocity fields at the steady state. Despite the equa-
boundary is included in the simulation, see Fig. 2. tions are written in non-dimensional form to list the number of
For the inflow boundary conditions, a non-dimensional total non-dimensional parameters that influence the problem, the inter-
pressure profile pt (y ), provided by the aircraft engine company nal computation of the code uses the dimensional version of the
SAFRAN and considered representative of this flow configuration, equations, and consequently when non-dimensional results such as
a constant inflow temperature θ f = 0, and the turbulence intensity the reduced temperature are presented they require to be postpro-
and viscosity ratio values are imposed. cessed. For these steady simulations, an iterative process is used to
For the outflow, a non-dimensional static pressure, ps = 1, out- ensure the steady state condition to be satisfied with low residuals.
let boundary condition is used and the temperature is fixed to the To accelerate the solver convergence, a coupled algebraic multigrid
inlet value, θ f = 0. method [40] based on V fixed cycles is used.
Adiabatic no-slip boundary conditions are used for the lateral For the conjugate heat transfer analysis, the energy equation
right and top walls of the tunnel. is solved throughout the fluid and solid domains with an im-
For the bottom wall, several zones are considered: the fins, plicit thermal coupling at the fluid/solid interface. The coupling be-
where no slip boundary conditions are used for the velocity and tween Eqs. (1)–(3) and (4) is performed at each iteration on the
heat transfer is allowed, and the region with no fins, where adi- body/fluid interface, forcing the continuity of the temperature field
abatic and no-slip boundary conditions are assumed for tempera- and energy conservation between the solid and the flow at the
ture and velocity respectively. interface. In the absence of source terms at this interface, energy
The non-dimensional equation that models the heat conduction conservation implies heat flux continuity. These compatibility con-
effects of the conjugate heat transfer problem in the solid region ditions are additional equations that must be satisfied in combina-
inside the fins is: tion with the fluid and thermal subsystems to solve the coupled
problem.
∇ 2 θs = 0 , (4)
where θs is the solid temperature. The boundary conditions ∂θ
∂n = 0
s
2.4. SACOC optimization
and θs = 1 are used for the fin surface and the fins horizontal base,
respectively. At this point, we should explain what are the optimization
Additionally, the coupling between both the compressible fluid criteria that are followed for this kind of problem. According to
problem and the heat transfer problem is given by the CHT bound- the industrial criteria agreed with outstanding companies such as
∂θ f
ary conditions θs = θ f and ∂θ
∂ n = k ∂ n on the body/fluid interface
s
SAFRAN, the minimization of the pressure drop, see Eq. (5), is the

4
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

main objective of the optimization process. 1. V1 : This verification computes a conjugate heat transfer prob-
    lem for 2D flat plate (without fins) in turbulent regime. The
|m˙ | pt  |m˙ | pt 
p =  −  . (5) Nusselt number and the temperature isolines will be compared
|m˙ |  inlet
|m˙ |  outlet to previously published solutions of the problem [15].
2. V2 : A 3D validation will be performed including the presence of
Of course, the largest heat exchange achieved better, but this
fins in the geometry. The case selected for comparison is pre-
is a secondary objective, being the pressure drop as the real cost
sented in Kim et al. [34], where computational and experimen-
function. In this section, an optimization process will be described
tal results are shown.
to modify the shape of the fins according to the following objec-
tive:
• Minimize the pressure drop,  p, between two planes before 3.1. V1 : flat plate
and after the fin location.
Once the pressure drop is minimized during the optimization For this verification without fins, a previous study by Vynnycky
process, the heat exchange is monitored, Q˙ , between the fluid et al. [15], has been used as reference. In this work, the heat trans-
and the fin surfaces, in order to confirm that no strong varia- fer associated with the forced convection flow over a rectangular
tions were observed. To optimize this magnitude, the fin shape conducting slab sited in an aligned uniform turbulent stream is
and the setup of the different fins will be studied. Apart from investigated in 2D analytically and numerically. Both internal and
the fin shape on the XY plane, the additional geometrical de- external thermal conductivities are taken into consideration using
grees of freedom of the system analyzed are: a conjugate heat transfer model based on the full compressible
• The gap between fins, Navier–Stokes equations and the heat equation for the slab.
• the fin thickness.

We will split the global optimization process in two indepen- 3.1.1. Setup
dent parts: first, we will perform a parametric study of the fin A schematic description of the modeled problem is given in
thicknesses and the distance between fins of the symmetric con- Fig. 7. The fluid domain is rectangular with a height, H/b = 1, and
figuration (see Fig. 2), both directly related to the number of fins. length, L/b = 3. The conducting slab is defined by a height, a/b, and
Second, computing the periodic model (see Fig. 3), we will opti- length, b. And the distance from the slab leading edge to the inlet
mize the geometric design of the fin shape on the XY plane based boundary condition is x/b = 0.5.
on the adjoint method [41,42], using the pressure drop per lateral We assume that the lower side of the slab y = −a/b is held at
fin surface area  p/Axy as the objective function, which means that a uniform temperature θ = 1, whilst the vertical boundaries of the
we will improve the aerodynamic design of the fin while the heat slab are both insulated.
exchange is kept as constant as possible. The ambient forced flow is characterized by a uniform veloc-
To perform the second part, the adjoint method is used. The ity, U∞ , and temperature, T∞ . It is also assumed that in the fluid
adjoint method is an efficient process to predict the influence of boundary upstream and downstream the slab the heat flux, the
input design geometry, D, on some engineering cost function of normal outflow and the viscous shear are zero. All lengths were
interest, L, which as we already said, in our case, is based on non-dimensionalised with the slab length b, velocities with U∞ and
the pressure drop generated by the SACOC divided the lateral fin temperatures as explained in Section 2.2.
area,  p/Axy . The process is performed in two major steps. First, Regarding the domain spatial discretization, a hybrid mesh
an steady state solution, also known as primal solution, Q, is com- formed by structured elements for the fluid boundary layer and
puted on an initial grid with control points X0 . In the second step, the slab, and unstructured elements for the upper part, is used.
the adjoint problem based on the previously numerical solution, The boundary layer is fully resolved using y+ values below one.
is evaluated. This evaluation computes the sensivities of the cost This verification test was computed for Re = 104 , k = 20, a/b =
function with respect to the different design parameters: 0.25 and two different Prandtl numbers, P r = 0.01, 100. The re-
  sults obtained were compared with numerical and analytical re-
dL ∂ L ∂ L ∂ Q dX sults from the reference work [15].
= + (6)
dD ∂ X ∂ Q ∂ X dD The efficiency of this problem can be quantified in terms of the
local Nusselt number Nu(x ), defined on the top of the slab as:
dX
where dD is the Jacobian matrix. The linear system created is
solved by means of a GMRES algorithm. The cost of solving the ∂θ f
Nu(x ) = − | 1/2 ≤ x ≤ 3/2. (7)
linear system of equations is similar to solving the primal flow ∂ y y=0
solution in terms of iterations and computational time. As result,
The average Nusselt number is defined as:
the most sensitive zones of the geometry are detected on the grid
points. The gradient of this sensitivity map indicates the move-  3/2
ment of the grid points through the control points to obtain an Nu = Nu(x )dx. (8)
1/2
optimized cost function. An example of the sensitivity map for
the initial periodic configuration described will be illustrated in
Section 4.3.3. The control points are moved according to the previ- 3.1.2. Grid convergence
ous step of the optimization process, as a consequence the geome- First, a grid convergence test was performed for P r = 0.01 using
try and mesh are modified and a new iteration is performed until three different grid resolutions. The number of elements for each
an optimized geometry is obtained (Fig. 6). grid are 7.44 × 103 (Coarse), 2.26 × 104 (Medium) and 9.55 × 104
This iterative process is finished when the variation of the cost (Fine). Fig. 8 shows the mesh convergence process, consequently,
function value is below a given tolerance. the local Nusselt number, Nu, given by the Eq. (7), and the Con-
jugate Boundary Temperature, θ , approximate the reference results
3. Numerical verification and validation [15] as the mesh quality is increased. A convergence analysis based
on the infinity norm, not included here, gives an error ≈ 1% be-
In this section, two test cases are defined to verify and validate tween the medium and fine grids. Once the grid convergence test
the computational tool. The two cases, named as V1 and V2 , are: is satisfied, the following results are computed with the finest grid.

5
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 6. Location of control points at iteration design zero.

x/b = 0 to x/b = 0.5. Consequently, the boundary layer is not de-

veloped and no viscous and heat transfer effects are considered for
x/b > 0.5. This consideration explains the differences between the
numerical results obtained by the preliminary computation per-
formed by our results and the results found in Vynnycky et al. [15].
Additionally, it is important to mention that the thermal and
viscous scales are proportional to the Prandtl number, being the
thermal resolution required in the thermal boundary layer, T + =
y+ P r. Therefore, in the simulation P r = 100, the thermal resolution
(T + ) is more demanding than the viscous resolution y+ , it will re-
Fig. 7. Computational setup for the first 2D validation without fins V 1 (based on
quired a y+ = 0.01 to obtain a T + = 1 and to resolve correctly with
Fig. 1 [15]). more accuracy the thermal boundary layer. However, in an indus-
trial context where the Prandtl number required is P r ≈ 0.71 at
standard conditions, the thermal resolution is less demanding than
the viscous resolution. Therefore, ensuring y+ < 1 both the thermal
3.1.3. Results
and the viscous boundary layers will be numerically well resolved.
Fig. 9 compares the analytical and numerical values of the lo-
cal Nusselt number, Nu, and the conjugate boundary temperature,
θ , along the convective wall. A good agreement between the ref- 3.2. V2 : 3D flat plate with fins
erence results and those presented in this work is observed, pro-
viding a good approximation of the heat transfer dissipation with This second test case is inspired on the experiments performed
an accurate temperature distribution over the slab. In the reference by Jonsson and Moshfegh [43]. A straight channel model is used to
work [15], there are no Nu analytical results available for the case examine the aero-thermal performance of a plate fin surface hav-
P r = 100. ing a bypass passage. Several fin geometries were set inside the
Fig. 10 compares the isosurfaces of the conjugate boundary wind tunnel in Jonsson and Moshfegh [43], being the one with
temperature, θ , with θ = 0.1 for both configurations (P r = 0.01 parallel plate fins the one selected for this validation. As result, the
and 100). The simulation presents a very good agreement when fundamental heat exchanger performance with a 3D turbulent flow
compared to the results from the reference work [15], being able under ideal conditions was examined.
to reproduce the thermal conductivity through both subdomains
(fluid and solid). Notice how, for the lower P r number, the thermal 3.2.1. Setup
boundary layer is more developed in the fluid due to the high ther- As this second test case is inspired from an experimental paper
mal conductivity. On the other hand, the simulation with P r = 100 [43], where the geometry is expressed in dimensional units, see
shows a larger temperature gradient inside the slab, while in the Table 1.
fluid the temperature variation is confined to a thin boundary layer These dimensions correspond to the fin-channel configuration
close to the slab surface. with aspect ratios HFIN/H = 0.33 and WF/W = 0.84 for the width
To justify the small differences between the computed results and height respectively, see [43]. The number of fins is equal to
and those presented in Vynnycky et al. [15], it is important to men- nine and the tested heat sinks were milled in aluminum of thermal
tion that in the referenced work the inflow is kept uniform from conductivity ks = 200 W/(m K).

Fig. 8. Grid convergence study using a coarse, a medium and a fine grid compared with the reference work [15] for the local Nusselt number Nu (left) and Conjugate
Boundary Temperature θ (left) along the convective wall with Re = 104 , Pr = 0.01, κ = 20 and λ = 0.25.

6
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 9. Local Nusselt number Nu (left) and Conjugate Boundary Temperature θ (right),compared with the reference work [15], at the convective wall with Re= 104 , Pr =
10−2 , 100, κ = 20 and λ = 0.25.

Fig. 10. Isosurfaces of conjugate boundary temperature (θ =0.1) of analytical [15] (top) and numerical (bottom) results with Pr = 0.01 (left), 100 (right), Re = 104 , κ = 20
and λ = 0.25.

Table 1
Geometrical dimensions of the wind channel and fins ex-
pressed in meters for the second test case, see [43].

Dimension Notation Length [m]

Channel length L 0.2

Channel height H 0.03
Channel width W 0.0637
Fin base length LFIN 0.0528
Fin height HFIN 0.01
Separation between fins SFIN 5 × 10−3
Fin thickness WFIN 1.5 × 10−3
Fins width WF 52.8
Distance leading edge Fin-Inlet LU 0.025
Fig. 11. Computational setup for the 3D validation case V 2 with one fin and peri-
Table 2 odic boundary conditions.
Geometrical dimensions of the wind channel, SACOC and fins ex-
pressed in meters.

Dimension Notation Length [m]

The computation has been performed in the periodic configu-
ration described in Section 2.1 (see Fig. 3), consequently only one
−1
Channel length L 3 × 10
fin will be considered in the computation. In Fig. 11 the domain
Channel height H 1 × 10−1
Channel width W 1 × 10−1 is shown and the following boundary conditions are indicated: no-
Distance leading edge SACOC-Inlet LU 7.6 × 10−2 slip and conjugate heat transfer boundary conditions are applied
Distance trailing edge SACOC-Outlet LD 3 × 10−1 on the fin walls; the velocity, temperature, turbulence kinetic en-
Fin height HFIN 2 × 10−2 ergy and the dissipation rate of the turbulent kinetic energy is pre-
Fin base length LFIN1 1.25 × 10−1
Fin top length LFIN2 1.05 × 10−1
scribed with a uniform distribution at the inlet, and static pressure
Separation between fins SFIN 2.175 × 10−3 is imposed at the outlet. The values of the boundary conditions for
Fin thickness WFIN 1 × 10−3 the turbulence variables k and ω are given by Agonafer and Moffat
Fins width WF 4.9 × 10−2 [44]. Duct walls (top and bottom) are modeled as adiabatic walls
Fin base lenght LB 1.33 × 10−1
with no-slip boundary conditions.
Fin base height HB 1.5 × 10−3
Fin base width WB 5.8 × 10−2 Regarding the grid structure, a fine hybrid mesh formed by
structured elements for the fluid boundary layer and the slab, and

7
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 12. Pressure drop,  p, (left) and thermal resistance, Rth , (right) for the second validation case, V2 , at different Reynolds numbers. Experimental values are obtained from
[43].

unstructured elements for the upper part, with 5.8 × 105 elements Table 3
Location of mea-
is used. The boundary layer is fully resolved using y+ values below
surements sections.
one. A convergence analysis, not included here, shows a good com-
promise between accuracy and computational cost for this mesh. Section X [m]
wd
The Reynolds number is defined as Re = ν h , where the ref- UMS1 0
erence length is the hydraulic diameter of the wind tunnel, dh = DMS1 0.25
2·H·W V DMS2 0.4
H+W . And the average air velocity is w = A−Ayz , where V is the DMS3 0.5
volumetric air flow rate, A = H · W is the wind tunnel cross sec-
tional area and Ayz = 1.35 × 10−4 m2 is the fin frontal area. In our
Table 4
case, the Reynolds number range is 40 0 0 < Redh < 14, 0 0 0 and the Air and Aluminium physical properties.
Prandtl is P r = 0.764. In this case, due to the reduced velocities,
the compressibility effects in the fluid are negligible. Property Air Aluminium

The problem can be analyzed by the contour of the thermal re- 3

Density, [kg/m ] 1.225 2702
sistance, Rth , defined as: Specific heat [J/kg K] 1004.9 896.0
Thermal conductivity [W/m K] 0.024 150
T
Rth = , (9)
Q˙
where Q˙ is the heat exchange and T = Tc − T∞ is the temperature
difference between the heated wall base and the inflow.
Finally, in terms of aerodynamic penalties, the pressure drop,
 p, is measured using two pressure taps positioned over the flat
plate in the center line with a distance, d, upstream and down-
stream of the fin respectively.

3.2.2. Results
The experimental measurements available were performed on
the central fin where the flow and heat transfer process are con-
sidered insensitive to the presence of the lateral walls of the wind
tunnel. The pressure drop over the heat sink was measured us-
Fig. 13. Experimental non-dimensionalized total pressure profile at the inflow
ing two pressure taps positioned over the flat plate in the cen- boundary. Pressure data provided by the aircraft engine company SAFRAN.
ter line (Z = 0 m) a distance d = 0.05 m upstream (X = 0.0038 m)
and downstream (X = 0.1628 m) of the heat sink respectively. The
pressure drop  p and the thermal resistance Rth are both com- 4.1. Setup
pared to the experimental measurements in Fig. 12. As can be ob-
served, a good agreement is obtained between the computed val- As the geometry of this final case is inspired by future exper-
ues and the experimental measurements. imental campaigns, similarly to the previous case in Section 3.2,
this case will also be expressed in dimensional units except for
4. SACOC optimization the inflow pressure profile and the classical non-dimensional num-
bers such as the skin friction, the conjugate temperature and the
In this section, we will apply the optimization process de- Reynolds and Mach numbers.
scribed in Section 2.4 to the SACOC problem. As explained in The setup was introduced in Section 2.1 and illustrated in
Section 2.4, the process is divided in two independent tasks. First, Figs. 2 and 4, and the geometric dimensions are listed in Table 2.
the symmetry model is used (see Fig. 2) to investigate the fin Additionally, Table 3 contains the four streamwise coordinates x of
thickness and separation effects. Second, the periodic model (see the measurement sections used in this work during the numerical
Fig. 3) is used with the adjoint method to optimize the fin shape analysis (see Fig. 2), and Table 4 contains the values of the fluid
on the XY plane. Before the explanation of the optimization pro- and solid properties.
cess, the simulation setup and the grid convergence process are For the inflow boundary conditions, an experimentally mea-
presented. sured inflow total pressure profile is imposed (see Fig. 13), where

8
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Table 5 Table 6
Mesh convergence study. Six different meshes with different characteristic Pressure drop  p, heat transfer per fin and the total heat transfer Q˙ when the
mesh size Hc and grid size inside the fin. The pressure loss  p and the heat separation between fins and the number of fins is varied inside the fins width dis-
transfer Q˙ /NFIN are monitored. tance WF. The middle case NFINS = 16 is used as reference case to compute the
variations.
Grid Hc [mm] H f in [mm] Grid cells  p [Pa] Q˙ /NFIN [W]
SFIN/WFIN NFINS  p [Pa] Q˙ /NFIN [W] Q˙ [W]
1 12 0.5 5.01 × 105 909.51 85.12
2 8 0.5 1.06 × 106 844.92 88.04 1.0875 24 807.2 (+10.1%) 74.98 (-21.4%) 1799.5 (+17.8%)
3 4 0.5 2.80 × 106 796.01 95.07 2.175 16 732.7 (0%) 95.43 (0%) 1526.9 (0%)
4 2 0.5 1.14 × 107 731.40 95.47 4.292 10 667.2 (−8.9%) 104.4 (+9.4%) 1044.2 (−31.6%)
5 2 0.25 1.31 × 107 732.70 95.43
6 2 0.125 2.36 × 107 733.35 95.47
Table 7
Pressure drop  p, heat transfer per fin and the total heat transfer Q˙ when the fin
thickness and the number of fins is varied inside the fins width distance WF. The
middle case NFINS = 16 is used as reference case to compute the variations.

WFIN [mm] NFINS  p [Pa] Q˙ /NFINS [W ] Q˙ [W ]

0.647 18 737.2 (+0.6%) 73.09 (−23.4%) 1315.5 (−13.8%)

1 16 732.7 (0%) 95.43 (0%) 1526.9 (0%)
1.45 14 748.3 (+2.2%) 108.13 (+13.3%) 1513.8 (−0.8%)

3.5% between grids #3 and #4. Finally, in the solid refinement,

very small differences were found between the grids #4, #5 and
#6 where the refinement inside the fins has little influence, show-
Fig. 14. Global perspective of grid number 5 (top), details of the fin zone (bottom). ing less than 0.2% variation. Therefore, balancing the computational
load and the accuracy demand, the final grid selected for the opti-
mization process is grid number #5, that contains 1.31 × 107 cells,
the pressure data were provided by the aircraft engine company where 1.19 × 107 are for the fluid part and 1.20 × 106 cells are for
SAFRAN. The complementary turbulence inflow values are 5% for the solid part, see Fig. 14.
the turbulent intensity and 10% for the turbulent viscosity ratio.
For the outflow, the pressure outlet value is fixed to ps =
4.3. Shape and setup optimization.
122, 364 Pa. Finally, the temperature difference between the bot-
tom heated aluminium fins and the inflow is T = 100 K, being
4.3.1. Separation between fins effect
T∞ = 300 K and Tc = 400 K. Regarding the fluid and solid proper-
The effect of the separation between fins is studied in this sec-
ties, Table 4 depicts these values. Besides, the dynamic viscosity is
tion. As part of the first task, we define the reference setup with
calculated through the Shuterland’s law[45].
16 fins and dimensions according to Table 2.
In contrast to the previous validation cases, the dominant non-
Keeping the fin base length LFIN1 and the fin thickness WFIN
dimensional numbers such as the Reynolds Re = 2 × 106 and Mach
constant and varying the distance between fins SFIN from 1.08 ×
M ≈ 0.5 numbers are high enough that the compressible and tur-
10−3 m to 4.29 × 10−3 m and consequently the number of fins
bulent effects should be taken into account.
NFIN. We computationally confirmed, using a medium grid that
In this section, the pressure drop is computed using the
reduces the size between fins and consequently includes a larger
Eq. (5) as the mass-flow averaged absolute total pressure differ-
number of fins in the WF length:
ence between the inlet and outlet sections.
The local heat transfer coefficient h and the skin friction coeffi- 1. The pressure drop,  p, and the total heat transfer, Q˙ , are both
cient C f are defined as: increased as expected (see Fig. 15).
  2. The heat transfer per fin, Q˙ /NFINS, decreases (see Table 6).
1 ∂ T  1 ∂ u 
h= k , Cf = μ , (10)
Tc − Tre f ∂ n  1
2
ρU∞
2 ∂ n A Reducing the distance between fins, which reduces the free
Af f flow area, the velocity increases within the fins. However, it does
where the reference temperature Tre f is the one that satisfies the mean a higher heat transfer. This effect is shown in Fig. 16 with the
equation of state for both the outflow static pressure Ps and inflow local heat transfer coefficient h and skin friction coefficient. The re-
density ρ∞ , and the total heat transfer coefficient hg is the integral sults have shown how reducing the distance between fins produces
of the local heat transfer coefficient on the fin area A f . a flow constrain within the fins that reduces the heat transfer effi-
ciency. Additionally, note how the flow is accelerated locally in the
4.2. Grid convergence lower part of the leading edge induced by the geometry angle. It is
directly related to the increasing of heat transfer and skin friction.
Using one half of the geometry (symmetric model), a grid con-
vergence process based on the pressure loss  p (see Eq. (5)) and 4.3.2. Fin thickness effect
the heat transfer per fin Q˙ /NFIN for seven different meshes is per- The second step of the analysis was performed keeping the hor-
formed. All meshes are designed using a minimum grid thickness izontal length occupied by the fins WF, the fin base length LFIN1
of 2.0 × 10−6 mm, which implies y+ ∼ 0.5 for the boundary layers. and the distance between fins SFIN constant and varying the fin
The mesh convergence process is summarized in Table 5, where thickness WFIN between 0.65 × 10−3 m and 1.45 × 10−3 m and
the first column is the characteristic mesh size Hc close to the wall consequently the number of fins NFIN. We computationally con-
fin, and the second is the grid size inside the fins, H f in , see Fig. 14. firmed, using a medium grid, that decreasing the fin thickness and
Notice that the grid refinement can be divided in two parts: consequently increasing the number of fins in the WF length, the
the fluid refinement (Grids #1, #2, #3 and #4) and the solid re- pressure drop  p remains almost constant and the heat trans-
finement (Grids #4, #5 and #6). Regarding the fluid refinement, fer per fins decreases (see Table 7). However, this increasing of
notice how the error between the different grids is reduced to fins number, could compensate the loose of heat transfer per fin.

9
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 15. Pressure drop,  p, (left) and the total heat transfer, h, (right) when the separation between fins and the number of fins is varied with a fixed fins width distance
WF.

Fig. 16. Heat transfer coefficient (left) and skin friction coefficient (right) with three different separation between fins: SFIN/WFIN = 1.0875 (top), 2.175 (middle) and 4.292
(bottom).

Fig. 17. Pressure drop  p, heat transfer per fin and the total heat transfer h when the fin thickness and the number of fins is varied inside the fins width distance W F .

Therefore, a local maximum is found for the total heat transfer Table 8
The lateral fin area, Axy , the pressure drop,  p, and
Q˙ when the W F IN = 1 × 10−3 m (see Fig. 17). As in the previous
the total heat tranfer, Q˙ , during the different itera-
test varying the distance between fins, the heat transfer per fin tions indicated by the Adjoint optimization process.
Q˙ /NFINS increases when the fin thickness is increased.
In contrast to the previous Section 4.3.1, Fig. 18 depicts how the Geometry Axy [ m2 ]  p [Pa] Q˙ [W ]

shear stress distribution barely changes in the three different con- Original 2.30 × 10−3 636.4 95.2
figurations when the separation between fins is constant. However, Mod 1 2.30 × 10−3 619.7 95.3
Mod 2 2.35 × 10−3 586.2 95.3
the heat transfer increases proportionally to the fin thickness as
Mod 3 2.38 × 10−3 575.1 95.2
the addition of material permits a higher heat transfer. Mod 4 2.43 × 10−3 564.6 94.7
Mod 5 2.50 × 10−3 562.4 94.7
4.3.3. Shape optimization of the periodic case
Finally, as part of the last task, Fig. 19 depicts the vector dis-
placements obtained from the sensitivity computation on the con- 74 Pa (11.6%). Despite the iterative lateral area growth, the total
trol points and used to modify the SACOC geometry at the iteration heat transfer h is only reduced in 0.5 W (0.5%). Besides, the evo-
design zero. Notice that we use the periodic model (see Fig. 3) ac- lution of the pressure drop and heat transfer coefficient during the
cording the values of Table 1. iterations of the optimization process is presented in Fig. 20.
Then, Table 8 shows the iterations of the process, where the The comparison between the final geometry of the fin obtained
control points adjust the optimal shape of the fin. The final geom- after the optimization process and the original one is shown in
etry Mod 5 presents a lateral area, Axy , increase of 8.6% compared Fig. 21. Notice how the shape of the optimized geometry elimi-
to the original design, but showing a pressure drop reduction of nates the sharp corners and smooths the perimeter of the fin. It

10
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 18. Local heat transfer coefficient (left) and skin friction coefficient (right) with three different separation between fins: WFIN = 0.647 mm (top), 1.0 mm (middle) and
1.45 mm (bottom).

Fig. 19. Vector displacement on control points at the iteration design zero.

Fig. 20. Evolution of the pressure drop (left) and the total heat transfer (right) for the different optimized designs.

shape optimization process. See Table 9 for comparison between

the different meshes.
As can be observed, the difference between the thermal and
aerodynamic results obtained for the two different meshes (5 and
Fig. 21. Comparison between the original fin shape and the fourth and fifth itera-
6) show that the optimization process has little mesh dependence.
tions Mod 4 and 5.
When the optimized geometry is compared to the original case,
the pressure drop variation is between 10% and 11% and the heat
transfer variation below 2%.
is also relevant that the bow shows a prominent bulb shape that
moves upwards and upstream the high pressure point, which opti- 4.4. Results for the optimized SACOC.
mizes the aerodynamic performance.
To answer the question of how much depends this method on As the main objective of the optimization process is to decrease
the mesh, the last two design iterations (Mod 4 and Mod 5) have the pressure drop caused by the SACOC presence losing minimum
been computed with an additional finer mesh. This finer grid is heat exchange. This statement comes from the experience achieved
based on the refinement values of the grid number 6 (see Table 5). by the aircraft engine industry. We obtained an optimized shape
The results are very similar when compared to the original mesh attending this requirement applying the adjoint method. Addition-
(Grid 5), concluding that the mesh has a small influence on this ally, the distance between fins and the fin thickness effect is an-

11
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Table 9
Comparison of the pressure drop  p and the total heat transferred power Q˙ during the
different iterations indicated by the Adjoint optimization process for two different meshes.
Grids 5(original) and 6 (finer) are detailed in Table 5.

 p [Pa] Q˙ [W]

Geometry Grid 5 Grid 6 Difference [%] Grid 5 Grid 6 Difference [%]

Original 636.4 633.3 −0.49 95.2 96.9 1.75

Mod 4 564.6 569.4 0.84 94.7 96.0 1.35
Mod 5 562.4 560.2 −0.39 94.7 95.0 0.31

Fig. 22. Comparison of the pressure profiles when the SACOC contains the original fins and when they are replaced by the optimized fins at the positions DMS1 and DMS2.

Fig. 23. Local heat transfer coefficient (left) and skin friction coefficient (right) with reference (top) and optimized (bottom) geometry.

alyzed (see Figs. 15 and 17) and some comments are required. For the sake of completeness, in this section the fins of
Fig. 15, expresses that both the pressure drop and the heat transfer the particular setup described in Section 4.2 are replaced by
are proportional to the number of fins, then if no additional heat the ones obtained after the optimization process performed in
transfer is required, the number of fins should remain as 16. Re- Section 4.3 where the number of fins and the fin thickness is the
garding the fin thickness, we should mention that the optimum same in both configurations. After the shape optimization, a pres-
heat transfer was obtained for the thickness used in the design sure drop of 77.6 Pa (10.6%) is obtained, while the heat transfer
case where the pressure drop does not suffer relevant variations per fin and the total heat transfer both vary less than 0.3%. In
(see Fig. 17). As consequence, the optimum design in terms of Fig. 22 the pressure profile between the case with the original
number of fins and fin thickness is the one already presented as fins and the optimized ones is presented. As can be appreciated
the design case. However, these results give useful information for in the fin area at the bottom part of the figure, the pressure pro-
alternative designs of the geometry based on new requirements. files are clearly modified for the test positions DMS1 and DMS2
downstream the SACOC, where the flow is accelerated.

12
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Fig. 24. Comparison of the non-dimensional temperature field between the original design (top) and the one obtained after the fourth iteration Mod 4 for the fine mesh
(bottom).

Fig. 25. Comparison of the turbulent kinetic field between the original design and the one obtained after the fourth iteration Mod 4 for the fine mesh.

Fig. 26. Comparison of the vorticity profiles when the SACOC contains the original fins and when they are replaced by the optimized fins at different YZ sections along
x-direction.

13
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

Again, Fig. 23 compares both designs in terms of heat trans- Acknowledgements

fer and skin friction coefficients along the fin and shows the ar-
eas with highest heat transfer. In the optimized shape, this area This project has received funding from the Clean Sky 2 Joint
is located at the bulb of the leading edge. Besides, Fig. 24 shows Undertaking under the European Union’s Horizon 2020 research
the conjugate boundary temperature contour for both the fluid and and innovation programme under grant agreement no. 831977 Leo
solid. The thermal wake and the temperature distribution on the M. González acknowledges the financial support from the Span-
fin are similar in both cases, resulting a very similar heat transfer. ish Ministry for Science, Innovation and Universities (MCIU) under
Finally, regarding the aerodynamics, Figs. 25 and 26 depict a grant RTI2018-096791-B-C21. The authors acknowledge the com-
comparison of the flow pattern downstream and around the fins puter resources and technical assistance provided by the Centro de
respectively. Fig. 25 shows how the turbulent kinetic energy is re- Supercomputación y Visualización de Madrid (CeSViMa). The au-
duced with the optimized SACOC. Comparing both wakes, in the thors also wish to thank Safran Aircraft Engines for their kind per-
reference model the flow is detached by the sharp corner present mission to share the data presented in this publication.
at the leading edge, while in the optimized SACOC the detachment
occurs past the middle of the fin with less production of turbulent
kinetic energy.
References
In Fig. 26 a contour plot of the vorticity magnitude is repre-
sented for different sections along the x-direction for both designs. [1] D. Reay, C. Ramshaw, A. Harvey, Process Intersification: Engineering for Effi-
We can observe that the size and evolution of the primary vortex ciency, Sustainability and Flexibility, Butterworth-Heinemann, Oxford, 2013.
created at the top of the external fin presents significant differ- [2] J. Rishmany, C. Mabru, R. Chieragatti, F.R. Aria, Simplified modelling of the be-
haviour of 3D-periodic structures such as aircraft heat exchangers, Int. J. Mech.
ences in both cases. Note how the primary vortex is reduced in the Sci. 50(6) (2008) 1114e22.
optimized model producing a wake with less vorticity intensity. [3] J. Griffin, D. Singer, L. Summers, Lubrication cooling system for aircraft engine
accessory, U.S. Patent No. 4,151,710 (2008).
[4] J. Sousa, L. Villafane, G. Paniagua, Thermal analysis and modeling of surface
heat exchangers operating in the transonic regime, Energy 64 (2014) 961–969.
5. Conclusions [5] H. Iwasaki, T. Sasaki, M. Ishizuka, Cooling performance of plate fins for multi-
chip modules, IEEE Trans. Compon., Pack., Manuf. Technol. 18 (1995) 144–147.
[6] D. Gonzalez Cuadrado, A. Marconnet, G. Paniagua, Inverse conduction heat
In this work we have presented an efficient numerical method-
transfer and Kriging interpolation applied to temperature sensor location in
ology to optimize the global performance of a SACOC. This opti- microchips, J. Electron. Packag. 140 (2018) 1–8.
mization process requires the definition of an objective, which in [7] D. Bajusz, A. Cornet, J. Friedel, N. Raimarckers, Air-oil heat exchanger placed at
our case was given by the industry, according to which the new the location of the air separator nose of a turbojet, and a turbojet including
such an air-oil heat exchanger, TechSpace Aero S.A., Milmort (BE), US Patent
SACOC design should minimize the pressure drop between the in- 20 09/0165995A1 (20 09).
flow and outflow sections but keeping the heat transfer similar [8] B. Olver, A. Ciampa, R. Marrano, R. Trepanier, S. Lamarre, M. Bernard, P. Ger-
to the original case. The optimization methodology presented was main, Turbofan bypass duct air cooled fluid cooler installation, Pratt & Whit-
ney Canada Corp. (CA), US Patent 20 08/0 053059 A1 (2008).
split in two parts: first, the optimal setup in terms of fin thickness [9] E. Elovic, Cooling air cooler for a gas turbofan engine, General Electric Com-
and distance between fins for a given SACOC width is also studied pany, Cincinnati, Ohio (US), US Patent 004254618 (1981).
by parametric variation. Second, the fin shape was optimized us- [10] J. Elder, Gas turbine engine heat exchanger with tapered fins, Hamilton Sund-
strand Corporation, US Patent 8784047 B2 (2010).
ing a adjoint method, where several control points refined the fin [11] C. Zähringer, K. Stastny, S. Ardey, Towards the powerhouse for more electric
geometry according to the gradient of the sensitivity function com- aircraft-dedicated engine concepts, in: Proceedings in 19th International Sym-
puted. Once the optimal geometry is defined, a validated and effi- posium on Air Breathing Engines (ISABE), 2009.
[12] H. Streifinger, Fuel/oil system thermal management in aircraft turbine engines,
cient numerical method solves the conjugate heat transfer problem RTO-Applied Vehicle Technology Panel, Symposium on Design Principles and
coupling the Reynolds averaged compressible Navier–Stokes equa- Methods for Aircraft Gas Turbine Engines, Toulouse, France, 1998.
tions with the heat transfer equation for the solid fins. The final [13] M. Flouros, H. Streifinger, Lubrication systems for aircraft gas turbines engines,
new component developments for future applications, Lubrication Systems for
result, when the optimized geometry is used, presents a substan-
Aircraft Gas Turbine Engines, Biarritz, France, 2012.
tial pressure drop reduction 10.6% with a negligible heat transfer [14] T.L. Perelman, On conjugated problems of heat transfer, Int. J. Heat Mass
variation. The CFD methodology used permits a complete compari- Transf. 3 (1961) 293–303.
son of skin friction coefficient and the heat transfer coefficient than [15] M. Vynnycky, S. Kimura, K. Kanev, Forced convection heat transfer from a flat
plate: the conjugate problem, Int. J. Heat Mass Transf. 41 (I) (1998) 45–59.
explains both the pressure reduction and the heat transfer conser- [16] E.M. Sparrow, B.R. Baliga, S.V. Patankar, Forced convection heat transfer from
vation. a shrouded fin array with and without tip clearance, J. Heat Transf. 100 (1978)
572–579.
[17] E.M. Sparrow, T.J. Beckey, Pressure drop characteristics for a shrouded longitu-
dinal-fin array with tip clearance, J. Heat Transf. 103 (1981) 393–395.
Declaration of Competing Interest [18] E.M. Sparrow, D.S. Kadle, Effect of tip-to-shroud clearance on turbulent heat
transfer from a shrouded, longitudinal fin array, J. Heat Transf. 108 (1986)
519–524.
The authors declare that they have no known competing finan- [19] K.S. Lau, R.L. Mahajan, Effects of tip clearance and fin density on the perfor-
cial interests or personal relationships that could have appeared to mance of heat sinks for vlsi packages, IEEE Trans. Compon. Hybrids, Manuf.
influence the work reported in this paper. Technol. 12 (1989) 756–765.
[20] R.A. Wirtz, W. Chen, Longitudinal fin heat sink performance in arrays of
low-profile electronic packages, Adv. Electron. Packag. 58 (1993) 809–817.
[21] R.A. Wirtz, W. Chen, R. Zhou, Effect of flow bypass on the performance of lon-
CRediT authorship contribution statement gitudinal fin heat sinks, ASME J. Electron. Packag. 116 (1994) 206–211.
[22] D.S. Kadle, E.M. Sparrow, Numerical and experimental study of turbulent heat
transfer and fluid flow in longitudinal fin arrays, J. Heat Transf. 108 (1986)
Miguel Chávez-Modena: Conceptualization, Methodology, Soft- 16–23.
ware, Validation, Formal analysis, Investigation, Resources, Writing [23] Y. Sata, H. Iwasaki, M. Ishizuka, Development of prediction technique for cool-
ing performance of finned heat sink in uniform flow, in: Proc. 1996 InterSoc.
– review & editing, Visualization. Leo Miguel González: Concep-
Conf. Thermal Phenom. Electron. Syst., 1996, pp. 108–114.
tualization, Methodology, Formal analysis, Investigation, Resources, [24] H. Jonsson, B. Moshfegh, Influence of fin spacing, fin thickness, and inlet veloc-
Writing – original draft, Writing – review & editing, Supervision. ity on the performance of plate fin heat sinks under varying bypass conditions
Eusebio Valero: Conceptualization, Formal analysis, Writing – re- using CFD, Int. J. Heat Exch. 1 (2) (20 0 0) 177–196.
[25] H. Jonsson, B. Moshfegh, Modeling and characterization of plate fin heat sinks
view & editing, Supervision, Project administration, Funding acqui- under bypass flow conditions using computational fluid dynamics methods,
sition. IEEE Trans. Compon. Packag. Technol. 24 (2) (2001) 142–149.

14
M. Chávez-Modena, L.M. González and E. Valero International Journal of Heat and Mass Transfer 182 (2022) 121971

[26] H. Iwasaki, T. Sasaki, M. Ishizuka, Cooling performance of plate fins for multi- [36] K. Morimoto, H. Kinoshita, Y. Suzuki, Adjoint analyses of enhanced solidi-
chip modules, in: Proc. 1994 InterSoc. Conf. Thermal Phenom. Electron. Syst., fication for shape optimization in conjugate heat transfer problem, in: APS
1994, pp. 144–147. Division of Fluid Dynamics Meeting Abstracts, APS Meeting Abstracts, 2016,
[27] E.M. Sparrow, B.R. Baliga, S.V. Patankar, Heat transfer and fluid flow analysis of p. H10.007.
interrupted-wall channels, with application to heat exchangers, J. Heat Transf. [37] K. Gkaragkounis, E. Papoutsis-Kiachagias, K. Giannakoglou, Adjoint-assisted
99 (1977) 4–11. Pareto front tracing in aerodynamic and conjugate heat transfer shape opti-
[28] E.M. Sparrow, C.H. Liu, Heat-transfer, pressure-drop and performance relation- mization, Comput. Fluids 214 (2021) 104753.
ships for in-line, staggered, and continuous plate heat exchangers, Int. J. Heat [38] V. Subramaniam, T. Dbouk, J.-L. Harion, Topology optimization of conjugate
Mass Transf. 22 (1979) 1613–1625. heat transfer systems: a competition between heat transfer enhancement and
[29] B. Boesmans, F. Christiaens, J. Berghmans, E. Beyne, Design of an optimal pressure drop reduction, Int. J. Heat Fluid Flow 75 (2019) 165–184.
heat-sink geometry for forced convection air cooling of multi-chip modules, [39] F. Menter, Zonal two equation k − ω turbulence models for aerodynamic flows,
in: Proc. Eurotherm Seminar 29 Thermal Manag. of Electron. Syst., 1993, in: 23rd AIAA Fluid Dynamics, Plasmadynamics, and Lasers Conference, Or-
pp. 267–276. lando, FL, US, 1993, p. 2906.
[30] C.L. Chapman, S. Lee, B.L. Schmidt, Thermal performance of an elliptical pin fin [40] Y. Notay, An aggregation-based algebraic multigrid method, Electron. Trans.
heat sink, in: Proc. 10th IEEE Semi-Therm Symp., 1994, pp. 24–31. Numer. Anal. 37 (2010) 123–146.
[31] P. Sathyamurthy, P.W. Runstadler, S. Lee, Numerical and experimental evalua- [41] F. Giannetti, P. Luchini, Structural sensitivity of the first instability of the cylin-
tion of planar and staggered heat sinks, in: Proc. InterSoc. Conf. Thermal Phe- der wake, J. Fluid Mech. 581 (1) (2007) 167–197.
nom. Electron. Syst., 1996, pp. 132–139. [42] P. Luchini, A. Bottaro, Adjoint equations in stability analysis, Annu. Rev. Fluid
[32] J. Min, J. Jeong, M. Ha, K. Kim, High temperature heat exchanger studies for Mech. 46 (2014) 493–517.
applications to gas turbines, Heat Mass Transf. 46 (2009) 175–186. [43] H. Jonsson, B. Moshfegh, Modeling of the thermal and hydraulic performance
[33] S. Kim, J. Min, M. Ha, C. Son, Investigation of high-speed bypass effect on the of plate fin, strip fin, and pin fin heat sinks influence of flow bypass, IEEE
performance of the surface air-oil heat exchanger for an aero engine, Int. J. Trans. Compon. Packag. Technol. 24 (2) (2001) 142–149.
Heat Mass Transf. 77 (2014) 321–334. [44] D. Agonafer, D. Moffat, Numerical modeling of forced convection heat transfer
[34] M. Kim, M.Y. Ha, J.K. Min, A numerical study on various pin.fin shaped surface for modules mounted on circuit boards, ASME J. Electron. Packag. 112 (1990)
air-oil heat exchangers for an aero gas-turbine engine, Int. J. Heat Mass Transf. 333–337.
93 (2016) 637–652. [45] Y. Fu, P. Hall, N. Blackaby, On the Görtler instability in hypersonic flows:
[35] L. Villafañe, G. Paniagua, Aerodynamic impact of finned heat exchangers on sutherland law fluids and real gas effects, Philos. Trans. R. Soc. Lond. Ser. A
transonic flows, Exp. Therm. Fluid Sci. 97 (2018) 223–236. 342 (1993) 325–377.

15