Applied Thermal Engineering 218 (2023) 119232

Contents lists available at ScienceDirect

Applied Thermal Engineering

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

Research Paper

Numerical and experimental performance investigation of a heat exchanger

designed using topologically optimized fins
Mario Petrovic ∗, Kenichiro Fukui, Kenichi Kominami
Thermal control systems department - Aerospace, Sumitomo Precision Products Co.,LTD., 1-10 Fuso-cho, Amagasaki, 660-0891, Hyogo, Japan

ARTICLE INFO ABSTRACT

Keywords: Topology optimization for thermal-fluid problems has been successfully used to improve performance of small
Topology optimization structures used for heat transfer. Optimization of full heat exchangers, due to analysis method limitations
Heat exchanger design and computational costs, has not been able to produce manufacturable designs complex enough to satisfy
Numerical analysis
heat exchanger specifications. The goal of this study is determining if a practical heat exchanger design
Experiment
can be obtained by the use of topology optimization. By considering a plate-fin heat exchanger, topology
optimization is used to obtain a locally optimized fin pattern which is extended to form a heat exchanger
core. A modification of the Number of transfer units method is presented to deal with the unique topology
obtained by optimization and evaluate the performance of the exchanger. The performance, when compared
to traditional serrated fin designs, show that optimization improvements extend beyond the optimized region
onto the entire exchanger. The results show that the optimized design is an improvement outside optimization
flow parameters. However, the limitations in 3D printer technology result in slight deviation of experiment
results compared to the analysis, and prevent the manufacturing of designs with comparable total performance
values to heat exchangers manufactured through traditional methods.

1. Introduction function. While improvements in performance were achieved, obtained

designs were variations of existing designs. An approach to generating
Plate-fin heat exchangers allow for compact and lightweight heat new designs without topology optimization was the use of design mod-
exchanger designs, making their use widespread throughout various ifications, such as perforations [5] and vortex generators [6]. Again,
industries [1]. As various applications such as electronics, motors, improvements in total heat transfer were achieved with the trade-off in
jet engines, etc, increase in power output, the need to manage the higher pressure drop, and the question whether this trade-off is optimal
produced heat is also increasing. While this can be achieved with larger remained.
heat exchangers, reducing costs and improving thermal control system The natural evolution from parameter optimization of existing de-
efficiency means that heat exchangers need to be more compact and signs is generation of completely new designs through topology opti-
efficient. mization. Topology optimization is a mathematical method that seeks
One of the most common approaches to improving heat exchanger an optimal distribution of material within a given space, referred to
efficiency is improving the heat transfer that occurs within the core
as the design domain. Initially developed for structural problems, it
of a heat exchanger. A standard design plate-fin heat exchanger core
has since been applied many different physics problems including the
consists of different fluid passes separated by thin plates forming the
coupling of fluid flow with heat transfer [7]. For fluid and thermal-
primary heat transfer surface. These plates are connected by fins which
fluid problems, the method obtains an optimal shape of a flow path
form the secondary heat transfer surface, further increasing the total
by evaluating if a point in the design domain should be part of the
heat transfer area between the two fluids. In general, the majority of
fluid or part of the solid. The challenge for thermal-fluid problems is
heat transfer occurs on the secondary surface, therefore improving fin
the need to model both solid and fluid behavior in the same space.
design has a significant impact on exchanger performance.
Initial approaches to improving performance were focused on pa- The most widespread approach is the use of porous media, were a
rameter optimization for fin geometry parameters and exchanger per- design dependent penalty force is applied onto the fluid if the point
formance parameters [2–4]. The approach uses genetic algorithms to in space is determined to be inside the solid domain [8]. The penalty
identify the optimal combination of parameters for a given objective force reduces the velocity of the fluid to almost zero, simulating the

∗ Corresponding author.
E-mail address: [email protected] (M. Petrovic).

https://doi.org/10.1016/j.applthermaleng.2022.119232
Received 27 May 2022; Received in revised form 8 August 2022; Accepted 25 August 2022
Available online 11 September 2022
1359-4311/© 2022 Elsevier Ltd. All rights reserved.
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Nomenclature

𝛾 characteristic function
𝜑 level-set function
𝐻 Heaviside function
𝐮 fluid velocity
𝑝 fluid pressure
𝑇 temperature
𝝈 stress tensor
𝑞 heat flux
𝑘 thermal conductivity
𝑓𝑏 body force
𝛼 inverse permeability
𝛽 heat generation coefficient
𝑄 inbound heat flux
Re Reynolds number
Pr Prandtl number
𝛷 energy loss Fig. 1. Design process flowchart.
𝑉𝑚𝑎𝑥 maximum fluid volume
𝑉𝑆 solid volume
𝑙𝛺 design domain center heat transfer when directly compared to traditional heat exchanger
𝜂 energy loss constraint multiplier designs. In order to achieve comparable heat transfer, the level of
topological detail needed would result in unfeasible computational
𝜏 regularization parameter
costs. To avoid this problem, studies focus on simple heat exchanger
𝐶 coefficient of proportionality
designs with small sections of two fluid domains [18,19] .
𝑆 topological sensitivity The goal of this study is to investigate and determine if a practical
𝜆 Lagrange multiplier plate-fin heat exchanger design can be obtained by using topology opti-
𝐮′ adjoint fluid velocity mization on a smaller section of the exchanger. The choice of a plate-fin
𝑝′ adjoint fluid pressure type exchanger is due to fewer simplification and assumptions needed
𝑇′ adjoint temperature when applying results from a small section to the whole exchanger.
Such an optimized heat exchanger could satisfy target specifications
and have a smaller size/mass compared to existing exchangers, while
also requiring less pumping power. This would make them ideal for
solid. Modeling of heat exchange required precise calculations of fluid compact applications such as the transportation or aerospace industries
flow around the topology. Initial studies focused on Stokes flow [8,9], by increasing the overall energy efficiency.
eventually progressing to Navier–Stokes flow [10]. With respect to heat A necessary step in achieving this goal is the performance cal-
transfer, many problems require 3D models to simulate the correct culation of the designed heat exchanger in order to confirm if the
boundary conditions. However, when trying to obtain complex topolo- improvements obtained by optimization of a small section are main-
gies within large design domains, 3D models are not practical due to tained. When calculating the performance of a heat exchanger, due to
computational costs. To that end, out-of-pane effects and boundary the complexity of the internal structure, different models use different
conditions are simulated using a pseudo 3D approach by coupling a assumptions and simplifications. One of the most common calculation
thermal conductive layer with a fluid design layer [11]. This approach methods is the Number of Transfer Units (NTU) method as it does not
require discretization of the exchanger, resulting in a fast calculation
is also capable of obtaining a 3D structure for heat sinks [12]. The
with highly accurate results. However, the assumptions the method
topology optimization method used in this study, builds on the previous
is based on are not valid in case of irregular fin shapes. This paper
work by adapting the methodology to the specific problem of cross-
introduces slight modifications to the NTU method in order to more
section fin optimization [13]. The results showed the potential to derive
accurately model topologically optimized fins. The results are verified
fin patterns from localized topology optimization.
by comparison to the CFD analysis of the whole heat exchanger, and by
The wast majority of studies on topology optimization for thermal-
comparison to experimental results. For the topology optimization, the
fluid problems are focused on the advancement of optimization meth-
selected approach was to focus on a small section, which is further sim-
ods needed to treat more complex flow and on the reduction of com-
plified by performing optimization in two dimensions. This approach
putational costs [14]. Results are simple models intended to test the
generates a topology with a high level of detail with relatively low
developed optimization method rather than generate a practical design computational costs.
for a heat exchanger. As a consequence, many are not manufactured The contents of this paper are divided as follows: Section 2 outlines
and tested, but rather evaluated through a Computational Fluid Dynam- the optimization method used to obtain fin topology. Section 3 intro-
ics (CFD) analysis [14]. In the context of this study, a practical design duces the design process from topology optimization to the final heat
is considered as a design with total heat transfer values comparable to exchanger core. In Section 4, the used evaluation methods are outlined.
commercially available heat exchangers while having similar of lower Also, the modifications to the NTU method are introduced. Section 5
pressure losses. The primary challenge in designing and manufacturing presents the results of each evaluation method, and Section 6 provides
practical topologically optimized designs is in the complexity of the ob- the conclusions of this study.
tained topology relative to the size of the entire part design. Heat sinks,
due to their scale a ease of manufacture, have been experimentally 2. Design methodology
analyzed in a number of studies [15–17]. For heat exchangers however,
performing the topology optimization where the design domain encom- The design process for the heat exchanger was divided into two
passes the entire exchanger, results in designs with significantly lower major segments indicated in Fig. 1: design and analysis of the local

2
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 2. Plate-fin heat exchanger core cross-section.

fin section, and the design and analysis of the full heat exchanger. By where the inverse permeability parameter 𝛼(𝛾) changes depending of
focusing the topology optimization on a local fin section, higher com- the topology as:
plexity of the internal structure can be achieved as opposed to directly ( )
1 𝜏 (1 − 𝛾)
optimizing the full exchanger. The evaluation on the full exchanger 𝛼 = 𝛼(𝛾) = 𝜒 1 + (7)
Re 𝜏 +𝛾
investigates if the optimized results provide improvements outside the
domain and parameters of the optimization. As a result, the body force vanishes inside the fluid sub-domain
(𝛾 = 1) and is active inside the solid sub-domain (𝛾 = 0). The constants
𝜒 = 104 and 𝜏 = 0.1 ensure a sufficiently strong penalty to reduce the
2.1. Topology definition
flow in the solid sub-domain to near zero, without causing numerical
instability [12].
The design domain 𝛺 is defined as the space within which topology
For a plate-fin heat exchanger, the heat transfer occurs over the
is allowed to change. Topology inside the design domain represents the
primary and secondary surfaces. In 2D topology optimization of fins,
distribution of the fluid and solid regions, and is expressed using the
directly applying the heat flux or temperature boundary conditions
characteristic function:
{ onto a fluid flowing between the fins is not possible. Therefore, the
𝛾(𝐱) = 1 in 𝛺𝑓 heat that is conducted through the plates and fins is simulated by a
(1)
𝛾(𝐱) = 0 in 𝛺𝑠 heat generation term [10] that is, similarly to the body force, active in
only the solid sub-domain:
where 𝛺𝑠 represents the solid sub-domain and 𝛺𝑓 the fluid sub-domain
such that 𝛺 = 𝛺𝑠 ∪ 𝛺𝑓 . To relax the problem 𝛾 is calculated using 𝛽
𝑄= (1 − 𝑇 )(1 − 𝛾) (8)
the smoothed Heaviside projection as 𝛾(𝐱) = 𝐻(𝜑(𝐱)), where 𝜑(𝐱) is RePr
the Level-set function representing the topology. This results in a small where 𝛽 is the dimensionless heat generation coefficient, corresponding
gray-scale region around the solid fluid boundary, while allowing for to the average Nusselt number for the secondary surface. The different
differentiation on the boundary. thermal conductivity in the two sub-domains is calculated as:
𝑘𝑠𝑜𝑙𝑖𝑑
𝑘 = 𝑘(𝛾) = 1 + (𝐾 − 1)(1 − 𝛾), 𝐾≡ (9)
2.2. Thermal-fluid problem 𝑘𝑓 𝑙𝑢𝑖𝑑
Finally, the boundary conditions are defined as:
When considering topology optimization for a thermal-fluid prob-
lem, the challenge is to model both fluid and solid behavior at every 𝐮 = 𝐮𝑖𝑛 , 𝑇 = 𝑇𝑖𝑛 on 𝛤𝑖𝑛 ; 𝜎 ⋅ 𝐧 = 0 on 𝛤𝑜𝑢𝑡 ; 𝑢𝑦 = 0 on 𝛤𝑠𝑦𝑚
point of the design domain 𝛺. This is due to the fact that whether a
point belongs to a solid or to a fluid is not known prior to optimization. (10)
The approach used in this study is to model the design domain 𝛺 as
porous media, and express the porosity as a function of the topology. 2.3. Optimization problem
The dimensionless state variables needed to describe the behavior of
the system are velocity 𝐮(𝐱), pressure 𝑝(𝐱), and temperature 𝑇 (𝐱). The The objective of this study is obtain and evaluate a completely new
governing equations for steady, incompressible, laminar thermal-fluid fin design and compare its performance to a traditional serrated design.
problem are expressed as: To that end, the considered objective function is the maximization
of heat transfer defined in Eq. (11). The constraints imposed are the
−𝛁 ⋅ 𝐮 = 0 conservation of mass (2)
( 𝑇) maximum fluid volume constraint, maximum energy loss, and center
−𝛁 ⋅ 𝐮𝐮 + 𝛁 ⋅ 𝝈 + 𝐟𝐛 = 0 conservation of momentum (3) of mass constraint in Eqs. (12) to (14).
−𝛁 ⋅ (𝑇 𝐮) − 𝛁 ⋅ 𝐪 + 𝑄 = 0 conservation of energy (4)
inf 𝑓 = − (1 − 𝑇 )(1 − 𝛾)d𝛺 (11)
∫𝛺
where:
1 ( ) 𝑘 s.t. 𝑔𝛷 =
∫𝛺
[𝛷(𝛁𝐮) + 𝛼(𝛾)(𝐮 ⋅ 𝐮)] d𝛺 − 𝜂𝛷0 ≤ 0 (12)
𝝈 = −𝑝𝐈 + 𝛁𝐮 + (𝛁𝐮)𝑇 , 𝐪=− 𝛁𝑇 (5)
Re RePr
are the stress tensor and the heat flux vector, respectively. The porous 𝑔𝑣 = 𝛾d𝛺 − 𝑉𝑚𝑎𝑥 ≤ 0 (13)
∫𝛺
media is represented by the body force [8] 𝐟𝐛 and is used to reduce the 1 1
velocity of the fluid inside the solid sub-domain to near zero. The body 𝑔𝑐𝑚 = 𝐱(1 − 𝛾)d𝛺 − 𝑙𝛺 ≤ 0 (14)
𝑉𝑆 ∫𝛺 2
force is defined as:
The maximization of heat transfer is obtained by minimizing the
𝐟𝐛 = −𝛼(𝛾)𝐮 (6) temperature within the solid sub-domain. The energy loss constraint is

3
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 3. Topology optimization (a) Initial shape, (b) Optimized shape, (c) Single pass topology.

method [21]:
( )
( ) 1 ′ ( ) 𝛽
𝑆 = − 𝛼 ′ (𝛾) 𝐮 ⋅ 𝐮′ − 𝑘 (𝛾) 𝛁𝑇 ⋅ 𝛁𝑇 ′ + (1 − 𝑇 ) 1 − 𝑇 ′
RePr RePr
+ 𝛼 ′ 𝜆𝛷 (𝐮 ⋅ 𝐮) + 𝜆𝑣 − 𝐱𝜆𝑐𝑚
(17)

Fig. 4. Two topology segments which; configuration is repeated downstream with where 𝛼 ′ (𝛾) = 𝑑𝛼(𝛾)∕𝑑𝛾 and 𝑘′ (𝛾) = 𝑑𝑘(𝛾)∕𝑑𝛾. The Lagrange multipliers
variations in fin size.
𝜆𝛷 , 𝜆𝑣 and 𝜆𝑐𝑚 for the constraint functions control whether the con-
straint conditions are satisfied or not, while 𝑝′ , 𝐮′ and 𝑇 ′ are adjoint
variables for the governing equations. The adjoint governing equations
defined as a multiplier 𝜂 of the energy loss for the initial shape of the
topology, and is calculated as: are also obtained as:

1 [ ]2 𝛁 ⋅ 𝐮′ = 0 (18)
𝛷 = 𝛷 (𝛁𝐮) = 𝛁𝐮 + (𝛁𝐮)𝑇 (15)
2Re ( ) ( )𝑇 [ ( ( )𝑇 )]
1
−𝛁 ⋅ 𝐮′ 𝐮𝑇 − 𝛁𝐮′ ⋅ 𝐮 − 𝛁 ⋅ −𝑝′ 𝐈 + 𝛁𝐮′ + 𝛁𝐮′ +
This definition allows for simple comparison with the serrated fin Re
which is selected as the initial shape. The optimization method seeks ( )
𝛼(𝛾)𝐮′ − 𝑇 𝛁𝑇 ′ + 2𝜆𝛷 𝛁 ⋅ 𝐮𝐮𝑇 = 0 (19)
the optimal topology through an iterative process where the topology ( ) ( )
( ′ ) 𝑘 𝛽
is updated over a fictitious time 𝑡 by the following expression: 𝛁⋅ 𝑇 𝐮 −𝛁⋅ 𝛁𝑇 ′ − (1 − 𝛾) 1 − 𝑇 ′ =0 (20)
RePr RePr
𝜕𝜙
= −𝐶(𝑆 − 𝜏∇2 𝜙) (16) and are solved by applying the following boundary conditions:
𝜕𝑡
where speed of topology change is controlled by the coefficient of
𝐮′ = 0, 𝑇 ′ = 0 on 𝛤𝑖𝑛 ; 𝑢′𝑦 = 0 on 𝛤𝑠𝑦𝑚 ;
proportionality 𝐶, while the regularization parameter 𝜏 controls the [ (
1 ( )𝑇 )]
complexity of the obtained topology [20]. The variable 𝑆 is the sensitiv- −𝑝′ 𝐈 + 𝛁𝐮′ + 𝛁𝐮′ ⋅ 𝐧 = − (𝐮 ⋅ 𝐧) 𝐮′
Re
ity of the topology, and dictates where in the design domain and how ( )
the topology changes. The sensitivity is obtained through the adjoint − 𝐮 ⋅ 𝐮′ + 𝑇 𝑇 ′ 𝐧 on 𝛤𝑜𝑢𝑡 (21)

4
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 5. 3D printed heat exchanger core.

3. Heat exchanger design 10 ◦ C. The value of Re is selected as the target flow rate through the
design domain. Thermal conductivity is calculated as 𝐾 = 305.41, based
A plate-fin heat exchanger core consists of parting sheets that are on water flowing around aluminum fins.
connected by a large number of fins fabricated as corrugated sheets, The optimization method is implemented in the Finite Element
as shown in Fig. 2. The two fluids engaging in heat exchange, in the Method solver FreeFEM++. Square piece-wise quadratic continuous
most common configuration, alternate between a pair of parting sheets, finite elements are used for the vectors, while square piece-wise linear
forming a single pass. The type and specification of a fin used in a continuous finite elements are used for scalars. The mesh is adapted
particular pass is generally determined by the target performance and after every update of the topology, with high refinement on the solid–
type of fluid used. Because if this, most plate-fin heat exchanger designs fluid boundary and at points of high velocity gradients.
will use two different fin designs, one for each fluid. The optimized topology is obtained as in Fig. 3(b). For clarity, the
figure applies symmetric boundary conditions once. By periodically
3.1. Topology optimization result repeating the obtained topology, the entire fluid pass is obtained as
in Fig. 3(c). The obtained topology can be segmented into similar parts
In topology optimization, higher levels of complexity with finer that consist of four fins each. Two of these segments running parallel
topological details come together with increased computational costs. to each other are shown in Fig. 4. Throughout the pass, these segments
A full optimization of the entire core is not realistic, as small fin shapes are distributed in a pattern similar to the distribution of serrated fins.
relative to the whole core would require significant computational The main variation between the segments is reduction in size of the
resources. To obtain a complex topology while keeping costs low, a middle fin pair in the direction of the flow, as can be seen in Fig. 3(b).
narrow 2D section between the inlet and the outlet of a single pass is The final segment closest to the outlet differentiates itself from the rest
selected as the design domain, as indicated in Fig. 3(a). The full pass by having a single fin in the middle rather than the two. This is due to
is then obtained by repeating the obtained topology periodically along the fluid temperature increasing toward the outlet, therefore fins closer
the width of the pass. This approach assumes that there is no variation to the inlet will transfer more heat than ones at the back. For the final
in flow profile along the width of the pass, which in practice is not segment, further decrease in fin size would result in sub-optimal fin
accurate, leading to sub-optimal topologies for a single pass. However, efficiency, and the optimization combines the two middle fins into a
improved performance compared to traditional fin designs is possible single fin.
as those designs also apply the same assumption [22]. For the purpose
of comparison, the initial shape in Fig. 3(a) represents the common 3.2. Core design
serrated fin design and is used as the basis for normalization of results.
Because the objective of this study is to evaluate lightweight heat The dimensionless flow parameters listed in the previous section
exchangers, the volume constraint 𝑉𝑚𝑎𝑥 for the fluid is set to 0.9. The were originally based on a flow pass with a characteristic length of
center of mass constraint is set to the center of the design domain, 12 𝑙𝛺 . 3.08 [mm] resulting in a fin section width of 2.82 [mm] and fin height
The energy loss constraint multiplier 𝜂 is set to 2, meaning two times of 3.41 [mm]. Because this scale is difficult to manufacture with high
the energy loss of the initial shape in Fig. 3(a). The design domain 𝛺 is degree of accuracy, the final design was brought up to a characteristic
20 × 0.5 with symmetry boundary conditions applied to the boundaries length of 23.1 [mm], leading to a fin section width of 21.15 [mm] and
𝛤sym . The total analysis domain is further extended by the inlet and fin height of 25.575 [mm]. The resulting pass topology has a width of
outlet domains with a size 1 × 0.5 where optimization does not occur. 211.5 [mm] and length of 221.5 [mm].
The boundary conditions applied to the model are: To design a 3D model using the 2D result of topology optimization
in Fig. 3(c), the boundary line, defined by the 0.5 value of the char-
𝜕𝑢𝑥 𝜕𝑇 acteristic function 𝛾, is extruded in the 𝑧 direction by 25.575 [mm] to
𝑢𝑥 = 1, 𝑢𝑦 = 0, 𝑇 = 0 on 𝛤in , 𝑢𝑦 = 0, = 0, = 0 on 𝛤sym ,
𝜕𝑦 𝜕𝑦 form a surface mesh of the fin wall.
𝝈 ⋅ 𝐧 = 0, 𝐪 ⋅ 𝐧 = 0 on 𝛤out For the core, a simple 3 pass design was selected, were the middle
pass would be evaluated. The choice of a 3 pass configuration was due
(22)
to size constraints of the 3D printer. The outer two passes would contain
For the performed optimization, Re = 500, Pr = 4.04, 𝛽 = 7.05 are the secondary fluid used to heat the primary fluid in the middle pass.
calculated from targeted performance parameters. The parameters for To reduce the molding complexity, the fin shape of the two outer paths
Pr and 𝛽 are derived by considering water of 20 ◦ C being heated by is that of a simple straight wall, defined as the plain fin shape. The fluid

5
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 6. Heat transfer experiment; (a) Experiment setup, (b) Test unit.

passes are separated by a 1 [mm] thick plate. Finally, adding side-bars 4.1. j-f method
and parting sheets, the result is a plate-fin heat exchanger core with
dimensions 231.5 × 221.5 × 54.575 [mm]. The core after molding is
shown in Fig. 5. In the case of the j-f method, the dimensionless factors 𝑗 and 𝑓 are
In addition to the core using optimized topology for the middle obtained using the expressions in Eq. (23) for a small segment of a flow
pass, a core using the initial shape from Fig. 3(a) is also molded for path [22].
evaluation and comparison. ( )
2
2
𝐺𝑚
ℎ (Pr) 3 𝐿
4. Performance evaluation methods 𝑗= , 𝛥𝑝 = 4𝑓 (23)
𝐺𝑚 𝑐𝑝 𝐷ℎ 2𝜌

The heat exchanger core performance is evaluated using the total For a plate-fin heat exchanger, this segment consists of several fins
heat transfer between the two fluids passing through the core and the in a single pass. The heat transfer coefficient 𝛼 and pressure loss 𝛥𝑝 are
pressure drop that occurs within the core. To obtain these values, this
obtained for a given fluid and flow regime through a CFD analysis of the
study employs three different methods: (1) Chilton–Colburn j-factor
analogy combined with the Fanning friction factor to calculate the NTU flow path segment. The 𝑗 and 𝑓 factors are then applied for the entire
method (j-f method), (2) Numerical analysis of the full heat exchanger core to estimate its performance using the NTU method. To apply the
model (CFD method), (3) Experimental measurements of pressure and NTU method for a heat exchanger utilizing topologically optimized fin
temperature values (Experimental). shapes, the calculation of the total effective heat transfer surface area

6
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 7. Pressure loss experiment; (a) Experiment setup, (b) Test unit.

needs to take into account the variation between fin geometry as: 4.2. CFD method
∑
𝑛
𝐴 = 𝐴𝑃 + 𝜂𝑖 𝐴𝑆𝑖 , (24) In practice, performing a CFD analysis of the entire heat exchanger
𝑖=1 core, where all fins are modeled directly, results in an analysis with
where the primary heat transfer surface is calculated using the charac- unfeasible computational costs. This is generally due to the scale of
teristic function of the topology: the fins compared to the core and constructing a good quality mesh
to accurately model flow and heat transfer leads to a large number of
𝐴𝑃 = 2𝑆𝐿2 𝛾d𝛺. (25) cells required.
∫𝛺
The core model used in this study consists of only three passes
The secondary heat transfer surface 𝐴𝑆𝑖 and fin efficiency 𝜂𝑖 are cal- and the passed are scaled to make them larger than passes in reg-
culated for each unique fin topology 𝑖. Unique fin topology refers to fins ular cores. This allows for a full core CFD analysis with somewhat
obtained by optimization, while fins obtained during pass construction acceptable computational costs. The CFD model used to analyze to flow
by periodically repeating the optimized topology are not unique. For was a segregated flow-energy solver with the standard 𝑘 − 𝜖 model
unique fins, the cross section area is calculated using the characteristic used for turbulent flow. The total number of cells used for both fluid
function as: domains and the solid domain is 58800000. Conducted on a AMD
𝐴𝐶𝑆𝑖 = 𝑆𝐿2 (1 − 𝛾)d𝛺. (26) EPYC [email protected] GHz using 96 cores and 16 GB Ram per core, a single
∫𝛺𝑖 analysis required 48 h to converge.
The fin efficiency is obtained as:
( ) 4.3. Experimental
tanh 𝑚𝑖 2𝑏
𝜂𝑖 = , (27) To verify the results of the numerical analyses, experiments to
𝑚𝑖 2𝑏 obtain the cores heat transfer and pressure loss were conducted. The
were: heat transfer experiment was conducted using the same setup as the
√
optimization: water flowing through the topologically optimized pass,
ℎ𝑃𝑖
𝑚𝑖 = . (28) heated by a secondary fluid at almost constant temperature. Lubricating
𝑘𝑠 𝐴𝐶𝑆𝑖
oil was selected as a secondary fluid, and is pumped at a high rate to
Here, 𝑃𝑖 is the perimeter of the cross section 𝐴𝐶𝑆𝑖 , 𝑏 is the fin height, minimize temperature drop. The experiment setup is shown in Fig. 6(a)
and 𝑘𝑠 is the thermal conductivity of the fin material. and the test unit in Fig. 6(b).

7
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 8. Performance obtained using j-f factors; (a) Pressure loss, (b) Heat transfer.

Table 1 fin and the Serrated fin, the Serrated fin 2 was designed to have the same
Performance comparison with serrated fin. pressure drop as the Topology fin at Re = 500. The Serrated fin 2 shape is
Reynolds 500 750 1000 1500 2000 the same as Serrated fin, but the fin number normal to the flow direction
Topology fin 𝛥𝑝 232.9% 204.1% 186.8% 173.8% 163.7% is increased to match the Topology fin pressure drop.
Serrated fin 2 𝛥𝑝 231.7% 231.7% 231.7% 231.7% 231.7% The results for pressure drop show that at Re = 500 the Topology fin
Topology fin 𝑄 116.8% 116.3% 117.6% 120.5% 122.2% has 2.32 times higher pressure drop which is higher than the 2 times
Serrated fin 2 𝑄 108.1% 107.5% 106.9% 106.2% 105.7% energy loss constraint used in optimization. This is due to differences in
pressure and energy loss if the flow profile is not fully developed. The
pressure drop for the Serrated fin 2 is the same as for the Topology fin by
For the pressure drop test, flow occurs only in the topologically design. By increasing the Reynolds value, the pressure drop of Serrated
optimized pass under isothermal conditions. It was also identified fin 2 stays 2.32 times the Serrated fin as both have the same fin shape.
that the pressure drop using water is low enough for pressure sensor However, the Topology fin shows a slower increase in pressure drop with
accuracy to significantly impact measurements. increased Reynolds value. This more gradual increase in pressure drop
The reduce the impact of sensor accuracy, the fluid was changed to can be explained by the more aerodynamic shape of the Topology fin
air with the corresponding sensors. All performance calculations, other which have a well defined leading and trailing edges. The rectangular
shape of Serrated fin and Serrated fin 2 mean that increasing flow rates
than the topology optimization, are adjusted to consider air in the case
will result in greater flow separation and vortexes around the sharp
of pressure drop. The experiment setup for the pressure drop is shown
edges, adding to the pressure drop.
in Fig. 7(b) and the test unit in Fig. 7(b).
For heat transfer at Re = 500, the Topology fin shows a higher
value than the Serrated fin 2 while having the same pressure drop.
5. Performance evaluation results Additionally, the advantage of the Topology fin over the serrated fins
increases with increased Reynolds values. The improved heat transfer
Performance evaluation is conducted for flow rates defined by Re = of the Topology fin comes from the fin pattern distribution. The indi-
500, 750, 1000, 1500, 2000, for a total of 5 test points. This approach vidual fins are offset such that they split the fluid stream to a greater
allows to test how the topology behaves at values of Re other than the degree than the two serrated fins, resulting in a more distributed heat
one it was optimized for. exchange. Furthermore, the two serrated fins maintain the same fin
The performance values obtained by the j-f method are shown in shape downstream, while the Topology fin has slightly larger upstream
Fig. 8 and a point by point comparison with Serrated fin is given in Ta- fins than downstream fins. As the temperature gradients are larger
ble 1. The models used for evaluation are: the Topology fin obtained by upstream, those fins contribute to the heat exchange more. Therefore,
optimization, Serrated fin used as the initial shape of the optimization, the same volume of fins is more efficiently distributed.
and Serrated fin 2. As the energy loss constraint during optimization is To verify the results of the j-f method, the results of the experiment
set to 2, resulting in difference in pressure drop between the Topology and the CFD method are imposed in Fig. 9, with a point by point

8
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 9. Performance comparison; (a) Pressure loss, (b) Heat transfer.

Table 2 small drop in temperature for the secondary fluid is observed in the
Performance comparison j-f method with experiment.
direction perpendicular the flow of the primary fluid. Temperature
Reynolds 500 750 1000 1500 2000 drops in the secondary fluid in the direction parallel to the flow of the
Topology fin 𝛥𝑝 124.6% 126.0% 126.5% 127.7% 131.3% primary fluid are minimal. This temperature distribution is sufficiently
Serrated fin 𝛥𝑝 128.1% 108.1% 97.5% 90.2% 89.0%
approaches the ideal constant temperature boundary condition used
Topology fin 𝑄 86.3% 87.6% 96.4% 103.3% 107.5% during topology optimization.
Serrated fin 𝑄 94.9% 95.5% 102.5% 105.4% 108.7%

6. Conclusion
Table 3
Performance comparison j-f method with cfd.
The primary goal of this study is investigate methods of utilizing
Reynolds 500 750 1000 1500 2000
topology optimization in order to derive a practical heat exchanger
Topology fin 𝑄 102.6% 106.7% 105.0% 105.8% 105.9% design with performance that is comparable to traditional design meth-
ods. The main problem addressed is the problem of scale, were topology
optimization of a heat exchanger, due to computational costs, can-
comparison in Table 2. The CFD method is only applied to obtain the not provide a sufficient level of detail of the internal structure for a
heat transfer inside the Topology core due to analysis costs. For the design to be practical. To overcome this problem, this study designs
pressure drop, the Topology fin j-f and experimental results show a a plate-fin heat exchanger by utilizing fin topology obtained through
deviation between 25%–32%, while the Serrated fin deviates between localized topology optimization in 2D. The exchanger is then designed
11%–28%. In the case of heat transfer, the deviation is noticeably by assembling multiple optimized sections into a single fluid pass.
lower, 7%–14% for the Topology fin and 5%–9% for the Serrated fin. To evaluate the design and determine its practicality, the NTU
In both cases, one of the know causes of deviation is instrumentation method used commonly to estimate the performance of heat exchangers
accuracy due to the low fluid flow rates in the scaled up pass, which was modified to consider variations in fin shapes obtained by opti-
necessitated the use of air for the pressure loss evaluation. Additionally, mization. The total heat transfer and pressure drop estimated by the
for pressure drop, surface roughness and potential irregularities due modified NTU method has a deviation of about 5% compared to values
to the 3D printing process may affect the result of the experiment, calculated by a full CFD analysis of the entire heat exchanger. This
while not being considered in the numerical evaluation. The better result shows that the modified method can provide highly accurate es-
agreement between the CFD method and j-f, shown in Table 3, further timations of the performance, with only a fraction of the computational
indicates potential differences between the numerical models and the costs.
manufactured core. The second step was to compare the performance values to those
The results of the CFD analysis, shown as the temperature distri- of the traditional serrated fin design. The comparison of calculated
bution and the velocity streamlines, are shown in Figs. 10 to 11. A pressure drop between the considered fin designs showed that the

9
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

Fig. 10. Core CFD temperature distribution for hot and cold pass.

Fig. 11. Core CFD velocity vector field streamlines.

optimized fin has equal or equivalently lower pressure drop compared a consequence, optimizing small sections to obtain higher level of
to serrated designs. The total heat transfer in the optimized fin is also detail and assembling those sections into a larger structure is a valid
improved compared to serrated designs. For a typical heat exchanger, design approach when using topology optimization for heat exchangers.
an increase in total heat exchange comes at a cost of pressure loss Additionally, the optimized fin maintained its improved performance
increase. The optimized fin showed that for a higher increase in total for flow parameters outside of those used in the optimization. Because
heat transfer, the corresponding increase in pressure loss is lower than of this, designs obtained by topology optimization may be used in heat
the serrated fin. In conclusion, it is shown that improvements obtained exchangers that have multiple operating specifications.
by localized 2D topology optimization are maintained even when the However, problems are encountered when attempting to manufac-
design is expanded to a scale of the entire heat exchanger core. As ture the core using current 3D printer technology. The need to replicate

10
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

topology with sufficient accuracy results in a scaled up fin model that is These equations assume uniform a fin shape, however, in the case
not suitable for compact heat exchangers. 3D printer ability to generate of topology optimization, 𝐴𝑓 and 𝜂𝑓 are replaced by 𝐴𝑆𝑖 and 𝜂𝑖 as
small scale structures with almost continuous variation of shape and discussed in Section 4.1.
thickness is essential in order to manufacture compact designs. Another After obtaining the 𝑁𝑇 𝑈 , the exchanger effectiveness is obtained as
problem encountered was the surface roughness that resulted in devia- a function 𝜖 = 𝜖 (𝑁𝑇 𝑈 , 𝐶 ∗ ). For a cross-flow plate-fin heat exchanger,
tion of experimental results compared to analysis results. This increased the relation is interpolated from an experimentally derived curves
surface roughness is due to small grains of meta solidifying along the found in most heat exchanger textbooks [22].
laser path. The complex internal structure of heat exchangers limits the
use of traditional removal methods. Furthermore, these methods would References
locally alter the topology of the design, affecting performance.
Therefore, improvements in 3D printing technology are needed to [1] J. Taborek, G. Hewitt, N. Afgan, Heat exchangers–theory and practice, in: Inter-
national Center for Heat and Mass Transfer Symposium, Hemisphere Publishing
fully utilize topology optimization in the case of heat exchangers. In
Corporation, Washington, DC, 1983.
terms of topology optimization, effects on topology of surface smooth- [2] S. Sanaye, H. Hajabdollahi, Thermal-economic multi-objective optimization
ing performed on 3D printed parts need to be investigated, and the of plate fin heat exchanger using genetic algorithm, Appl. Energy 87
optimization needs to be adjusted. The assumption of uniform flow (6) (2010) 1893–1902, http://dx.doi.org/10.1016/j.apenergy.2009.11.016, URL
at the inlet is a common one. In practice however, it does not hold, https://www.sciencedirect.com/science/article/pii/S0306261909005030.
[3] R. Song, M. Cui, Single- and multi-objective optimization of a plate-fin heat
and further improvements may be gained by implementing global flow
exchanger with offset strip fins adopting the genetic algorithm, Appl. Therm. Eng.
parameter variation onto localized optimization. 159 (2019) 113881, http://dx.doi.org/10.1016/j.applthermaleng.2019.113881,
URL https://www.sciencedirect.com/science/article/pii/S1359431118364196.
Declaration of competing interest [4] Y. Xue, Z. Ge, X. Du, L. Yang, On the heat transfer enhancement of plate fin
heat exchanger, Energies 11 (6) (2018) http://dx.doi.org/10.3390/en11061398,
URL https://www.mdpi.com/1996-1073/11/6/1398.
The authors declare that they have no known competing finan-
[5] K. Boukhadia, H. Ameur, D. Sahel, M. Bozit, Effect of the perforation design
cial interests or personal relationships that could have appeared to on the fluid flow and heat transfer characteristics of a plate fin heat ex-
influence the work reported in this paper. changer, Int. J. Therm. Sci. 126 (2018) 172–180, http://dx.doi.org/10.1016/
j.ijthermalsci.2017.12.025, URL https://www.sciencedirect.com/science/article/
Data availability pii/S1290072917313571.
[6] M. Samadifar, D. Toghraie, Numerical simulation of heat transfer enhancement
in a plate-fin heat exchanger using a new type of vortex generators,
The data that has been used is confidential. Appl. Therm. Eng. 133 (2018) 671–681, http://dx.doi.org/10.1016/j.
applthermaleng.2018.01.062, URL https://www.sciencedirect.com/science/
Appendix. 𝝐-NTU method article/pii/S1359431117363068.
[7] J. Deaton, R. Grandhi, A survey of structural and multidisciplinary continuum
This a brief explanation of the 𝜖-NTU method used to estimate heat topology optimization: Post 2000, Struct. Multidiscip. Optim. 49 (1) (2014) 1–38,
http://dx.doi.org/10.1007/s00158-013-0956-z.
transfer rate in heat exchangers. The total heat transfer rate between [8] T. Borrvall, J. Petersson, Topology optimization of fluid in Stokes flow, Internat.
two fluids inside a heat exchanger can be calculated as: J. Numer. Methods Fluids 41 (1) (2003) 77–107, http://dx.doi.org/10.1002/fld.
( ) 426.
𝑞 = 𝜖𝐶𝑚𝑖𝑛 𝑇𝐻𝑖𝑛 − 𝑇𝐶𝑖𝑛 (A.1) [9] N. Aage, T. Poulsen, A. Gersborg, O. Sigmund, Topology optimization of large
scale Stokes flow problems, Struct. Multidiscip. Optim. 35 (2) (2008) 175–180,
where 𝜖 is the heat exchanger effectiveness. This dimensionless param- http://dx.doi.org/10.1007/s00158-007-0128-0.
eter is a function of the 𝑁𝑇 𝑈 , 𝐶 ∗ and flow arrangement. The heat [10] E. Dede, Optimization and design of a multipass branching microchannel heat
capacity rate ratio 𝐶 ∗ is defined as: sink for electronics cooling, J. Electron. Packag. 134 (4) (2012) 041001, http:
//dx.doi.org/10.1115/1.4007159.
𝐶𝑚𝑖𝑛
𝐶∗ = (A.2) [11] C. Ingram, G. Pingen, Multi-layer, pseudo 3D thermal topology optimization of
𝐶𝑚𝑎𝑥 heat sinks, in: ASME International Mechanical Engineering Congress and Exposi-
tion, Proceedings, Vol. 7, IMECE, 2012, http://dx.doi.org/10.1115/IMECE2012-
where 𝐶𝑚𝑖𝑛 is the lower value of the thermal capacity between the two
93093.
fluids, and 𝐶𝑚𝑎𝑥 is the higher. The Number of transfer units, or 𝑁𝑇 𝑈 [12] J.H. Haertel, K. Engelbrecht, B.S. Lazarov, O. Sigmund, Topology optimization
is defined as: of a pseudo 3D thermofluid heat sink model, Int. J. Heat Mass Transfer 121
𝑈𝐴 (2018) 1073–1088, http://dx.doi.org/10.1016/j.ijheatmasstransfer.2018.01.078.
𝑁𝑇 𝑈 = (A.3) [13] P. Mario, K. Tsuguo, F. Kenichiro, Optimization of fin topology for heat transfer
𝐶𝑚𝑖𝑛
within lightweight plate-fin heat exchangers, J. Thermophys. Heat Transfer 35
where 𝑈 is overall heat transfer coefficient and 𝐴 is the total heat (4) (2021) 677–685, http://dx.doi.org/10.2514/1.T6121.
transfer area. The inverse of the overall thermal conductance, 𝑈 𝐴 is [14] A. Fawaz, Y. Hua, S. Le Corre, Y. Fan, L. Luo, Topology optimization of heat
exchangers: A review, Energy 252 (2022) 124053, http://dx.doi.org/10.1016/
the overall thermal resistance. Thus, the 𝑁𝑇 𝑈 can be obtained by
j.energy.2022.124053, URL https://www.sciencedirect.com/science/article/pii/
calculating: S0360544222009562.
𝑅𝐻𝑓 𝑅𝐶𝑓 [15] E.M. Dede, S.N. Joshi, F. Zhou, Topology optimization, additive layer
1 1 1
= ( ) +( ) + 𝑅𝑊 + ( ) +( ) (A.4) manufacturing, and experimental testing of an air-cooled heat sink,
𝑈𝐴 𝜂𝑂 ℎ𝐴 𝐻 𝜂𝑂 𝐴 𝐻 𝜂𝑂 ℎ𝐴 𝐶 𝜂𝑂 𝐴 𝐶 J. Mech. Des. 137 (11) (2015) http://dx.doi.org/10.1115/1.4030989,
111403, arXiv:https://asmedigitalcollection.asme.org/mechanicaldesign/article-
The first and fourth term on the right hand side are the convective
pdf/137/11/111403/6800273/md_137_11_111403.pdf.
thermal resistances for the hot and cold fluid, respectively. The second [16] S. Zeng, B. Kanargi, P.S. Lee, Experimental and numerical investigation of
and fifth terms are the fouling resistances for the individual fluids, and a mini channel forced air heat sink designed by topology optimization,
the third term is the thermal resistance of the wall between the two Int. J. Heat Mass Transfer 121 (2018) 663–679, http://dx.doi.org/10.1016/
fluids. The convective heat transfer coefficient ℎ is calculated for each j.ijheatmasstransfer.2018.01.039, URL https://www.sciencedirect.com/science/
article/pii/S0017931017336414.
fluid as:
[17] H. Li, X. Ding, F. Meng, D. Jing, M. Xiong, Optimal design and thermal modelling
𝑗𝐺𝑚 𝑐𝑝 for liquid-cooled heat sink based on multi-objective topology optimization: An
ℎ= 2
(A.5) experimental and numerical study, Int. J. Heat Mass Transfer 144 (2019) 118638,
(Pr) 3 http://dx.doi.org/10.1016/j.ijheatmasstransfer.2019.118638, URL https://www.
The overall surface efficiency 𝜂𝑂 is calculate from the fin efficiency sciencedirect.com/science/article/pii/S0017931019320289.
[18] L.C. Høghøj, D.R. Nørhave, J. Alexandersen, O. Sigmund, C.S. Andreasen, Topol-
𝜂𝑓 as:
ogy optimization of two fluid heat exchangers, Int. J. Heat Mass Transfer 163
𝐴𝑓 ( ) (2020) 120543, http://dx.doi.org/10.1016/j.ijheatmasstransfer.2020.120543,
𝜂𝑂 = 1 − 1 − 𝜂𝑓 (A.6) URL https://www.sciencedirect.com/science/article/pii/S0017931020334797.
𝐴

11
M. Petrovic et al. Applied Thermal Engineering 218 (2023) 119232

[19] M.A. Salazar de Troya, D.A. Tortorelli, J. Andrej, V.A. Beck, Three-dimensional [21] K. Yaji, T. Yamada, S. Kubo, K. Izui, S. Nishiwaki, A topology optimization
topology optimization of heat exchangers with the level-set method, 2021, arXiv method for a coupled thermal–fluid problem using level set boundary expres-
e-prints arXiv:2111.09471. sions, Int. J. Heat Mass Transfer 81 (2015) 878–888, http://dx.doi.org/10.1016/
[20] T. Yamada, K. Izui, S. Nishiwaki, A. Takezawa, A topology optimization method j.ijheatmasstransfer.2014.11.005.
based on the level set method incorporating a fictitious interface energy, Comput. [22] K. Thulukkanam, Heat Exchanger Design Handbook, in: Dekker Mechani-
Methods Appl. Mech. Engrg. 199 (45) (2010) 2876–2891, http://dx.doi.org/10. cal Engineering, CRC Press, 2000, URL https://books.google.co.jp/books?id=
1016/j.cma.2010.05.013. G52EfFF4uQYC.

12