International Communications in Heat and Mass Transfer 75 (2016) 177–182

Contents lists available at ScienceDirect

International Communications in Heat and Mass Transfer

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

Generating optimal topologies for heat conduction by heat flow

paths identification☆
Baotong Li, Jun Hong ⁎, Xiangyang Tian
State Key Laboratory for Manufacturing Systems Engineering, School of Mechanical Engineering, Xi'an Jiaotong University, Xi’an, 710049, P.R. China

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

Available online 16 April 2016 Knowing how heat “flows” inside a structure is significant for designers to make sure that the structure will per-
form its intended heat conduction function properly. This paper describes the principle and the process of defin-
Keywords: ing an optimal topology for heat conduction by interpreting the heat flow paths obtained from a standard finite
Heat conduction element analysis. First, a simple equilibrium equation is introduced to characterize heat flow paths along which
Heat flow paths the heat transfer rate remains constant as it traverses the conduction domain. Then, mathematical properties of
Topology optimization
the path plots are discussed in detail, providing designers with an insight of what kind of path distribution is the
Paths distribution
best for heat conduction. A brand new optimizer is developed to sketch the most effective topology by homoge-
nizing the spacing of heat flow paths across the conduction domain. Finally, the optimization procedure is illus-
trated in a directive and descriptive way, where the well-known dichotomic constructal tree is selected as the
initial layout to be tested. Unlike the conventional methods, the new approach needs neither the sensitivity anal-
ysis nor the modification of the existing finite element programs, it is very straightforward to use in practice.
© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction overall thermal resistance is minimized for heat conduction. Despite

In electronic engineering or micro-electronic instruments, the issue
of how to effectively cool the heat generating volume or surface has
attracted more and more attention. One common and cost-effective so-
lution is based on inserting certain quantity of high conductivity mate-
rial in the heating region to collect and discharge the generated heat to a
heat sink by heat conduction. The cooling effectiveness is dependent not
only on the quantity and quality of high conductivity material but also
on the topological arrangement of high conductivity link inside the
heat generating region. At this background, the planning of heat con-
duction paths is a fascinating problem for designers. However, it is
also a difficult problem to fully harness the cooling potential of the
added high conductivity material, especially when the thermal loading
condition is complicated.
When pressed with engineering puzzles, human often draw guid-
ance and inspiration from the natural world. A well-known case is the
constructal theory proposed by Bejan [1–3], in which tree-like networks
are assembled iteratively to form an optimal configuration so that the
☆ Communicated by W.J. Minkowycz.
⁎ Corresponding author.
E-mail address: jhong_email@163.com (J. Hong).
its great application potential, most of the cooling channels predicted
from the constructal theory bifurcate regularly and rigidly, which may
be geometrically far from the true optimal layout. With the increasing
variety of electronic chips and diversities of cooling requirements, flex-
ible design of heat conduction paths is needed. Since the idea of layout
optimization has been presented as a problem of searching for the opti-
mal material distribution [4,5], topology optimization has become the
most powerful tool for structure design. In recent decades, topology op-
timization has been developed for more and more complex scenarios. In
general, there are two broad categories of topology optimization used
for heat conduction design. One is the microstructure based approach,
like the homogenization design method (HDM) [6,7], the solid isotropic
microstructures with penalization (SIMP) method [8–10], etc. The other
is the geometry approach, like the evolutionary structural optimization
(ESO) method [11,12], the bidirectional ESO (BESO) method [13], the
level set (LST) method [14–16], the cellular automation (CA) method
[17], and so on. More details about the topology optimization for heat
conduction design can be found in [18].
Although the literature on heat conduction optimization and the
technologies that underpin it are growing fast, there is little available
on how a designer creates the conduction topology by optimizing the
distribution pattern of heat flow paths. In fact, a well-designed heat
flow paths can deliver better topologies for heat conduction. The suc-
cessful analysis on heat flow paths in topology optimization would

http://dx.doi.org/10.1016/j.icheatmasstransfer.2016.04.016
0735-1933/© 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
178 B. Li et al. / International Communications in Heat and Mass Transfer 75 (2016) 177–182

require a well-established supporting theory, an efficient path-plotting
strategy and a proficient topology identification algorithm. With all
these taken into consideration, this paper addresses the creation of
heat conduction topology from a different perspective, the perspective
of heat flow paths identification. The design idea takes inspiration
from an analogy of streamlines in hydromechanics [19,20]. Such a sim-
The requirement for defining the heat flow paths is actually to
create a heat tube in which there is no contribution to the heat
transfer rate on its boundaries AB and CD. For boundary AB, this re-
quires
Z B

ple analogy appears to provide insights on the way of how a structure Q  n dA ¼ 0 ð5Þ
A
conducts heat, and facilitate the suppression of redundant features in
topology optimization, which can be considered as a great improve-
That is,
ment compared with other optimization methods.
The layout of this paper is as follows. We first describe the funda-
0 2 31 0 2 31
mental theory of heat flow paths in Section 2. Mathematical properties Z B h i i nx
of the path plots are then explained in detail. A new evolutionary algo- @ qx qy qz  4 j 5A  @½ i j k   ny 5AdA ¼ 0
4 ð6Þ
rithm is developed for heat conduction design. Key details about the al-
A k nz
gorithm are provided in Section 3. Numerical examples are presented in Eq. (6) is achieved if the tangent lines of every points on bound-
Section 4. The main findings of the research work are concluded in aries AB and CD are parallel to the corresponding heat flux vector
Section 5. Q. This means the heat flow paths can be obtained by tailoring the
local angular orientation of the heat tube boundaries. A numerical in-
2. Fundamental theory tegration scheme is employed to plot the contours through the
pointing vector field. Unlike the commonly used fourth-order
Suppose that an arbitrary linear thermal conductor occupies domain Runge–Kutta algorithm, the approach used in this research is not
Ω, where Dirichlet boundary ΓT (i.e., temperature boundary, T0) and time independent but space independent, which means the time
Neumann boundary Γq (i.e., heat flux boundary, q) are pre-specified. A discretization, Δ t, is replaced by the scalar spatial discretization, Δ l,
heat ‘stream tube’ is introduced, which inherits properties similar to in order to represent a small increment along the heat flow path, as
that of a fluid stream tube, for instance, no heat (fluid) should cross shown in Eq. (7):
the tube boundaries such that the equilibrium is satisfied over the tube.
∂x ∂y ∂z
Φa ¼ Φb ð1Þ V¼ iþ  jþ k ð7Þ
∂l ∂l ∂l

where Φa and Φb are the heat transfer rate at the section AC and BD,
respectively.
An arbitrary segment of unit area taken from the tube boundary is
shown in Fig. 1, in which the vector n is the normal vector of the bound-
ary and is given by
n ¼ nx  i þ ny  j þ nz  k and n2x þ n2y þ n2z ¼ 1
where i, j, and k are the unit vector in the X, Y, and Z directions,
respectively.
The heat flux Q at every point in the domain Ω is expressed as
Q ¼ qx  i þ qy  j þ qz  k
where V is the pointing vector field defined over the conduction
domain Ω.
The field of pointing vectors can be first defined by averaging the
elemental values of heat flux onto the nodes embedded in the finite
element mesh. Then the vector at any arbitrary point in the domain
Ω can be obtained by relating the point with an element and then in-
ð2Þ
terpolating from the corresponding nodal values. Suppose that Pi(xi,
yi, zi) is the initial point, a new point Pi + 1(x i + 1, y i + 1, z i + 1) can be
positioned by the modified Runge–Kutta scheme. The coordinates
of the new point ζi + 1(ζ = x, y, z) is a weighted sum of ζi and dζj
(j = 1, 2, 3, 4). There are four values of dζj; each new value is calcu-
lated at a location that depends on the previous one, as shown in
ð3Þ Eqs. (8) and (9):

where qx, qy, and qz are the value of heat flux in the X, Y, and Z directions,  
respectively. ∂ζ  ∂ζ 
dζ 1 ¼ Δl dζ 2 ¼ Δl
The heat transfer rate at the arbitrary plane can be obtained by inte- ∂l ζ n ∂l ζ n þdζ 1
  2
ζ ¼ ðx; y; zÞ ð8Þ
grating the heat flux: ∂ζ  ∂ζ 
dζ 3 ¼  Δl dζ 4 ¼  Δl
Z ∂l dζ 2
ζnþ ∂l ζ n þdζ 3
2

Φ¼ Q  n dA ð4Þ

1
ζ iþ1 ¼ ζ i þ ðdζ 1 þ 2dζ 2 þ 2dζ 3 þ dζ 4 Þ ζ ¼ ðx; y; zÞ ð9Þ
where the dot indicates the vector dot product. 6

Fig. 1. Construction of the heat flow paths.

 B. Li et al. / International Communications in Heat and Mass Transfer 75 (2016) 177–182 179

Fig. 2. Optimization of 2-D heat tube with variable cross-sectional area.

3. Identification of heat conduction topology where

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3.1. Mathematical properties 1  0 2
F¼ 1 þ f ðxÞ ð17Þ
k  g ðxÞ
To explore the relationship between the conduction efficiency and qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 0 2
the path distribution pattern, an arbitrary segment of a certain heat G ¼ g ðxÞ 1 þ f ðxÞ ð18Þ
tube is cut out from the conduction domain, where the curve L refers
to the boundary of the heat tube staring from point A(x1,y1) to point Substituting Eqs. (17) and (18) into Eq. (16), we can get Eq. (19):
B(x2,y2), while the curve L′ is another boundary of the heat tube, as
shown in Fig. 2. 1
The thermal resistance of the selected heat tube is given by λk− ¼0 ð19Þ
g 2 ðxÞ
Z x2
Rtube ¼ 1=½k  AðxÞ dl ð10Þ Eq. (19) is completely unrelated with the geometric function f(x)
x1 and such equation is true when and only when g(x) is a constant func-
tion. That means the most efficient path for heat conduction is a path of
where k is the thermal conductivity, and A(x) is the cross sectional area. constant cross-sectional area starting from the point of heat sources to
Suppose that the heat tube is of unit thickness, the thermal resis- the point of heat sinks. The thermal equilibrium is satisfied locally
tance can be rewritten as along any segment of the heat tube boundaries, thus we have
Z qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
x2
1  0 2 ΔΦ ¼ g ðxÞ  jQ ðx; yÞj ð20Þ
Rtube ¼ 1 þ f ðxÞ dx ð11Þ
x1 k  gðxÞ
The distribution uniformity of the heat flow paths indicates
where g(x) is the normal distance from the curve L to curve L′, f(x) is the changing heat flux Q, if the path narrows, the ratio of the heat
geometric function of the curve L, and f′(x) is the first-order derivative transfer rate to the cross-sectional area gives the higher heat
of the function f(x). flux and therefore leads to a higher temperature gradient which
The heat conduction design can be considered as tailoring the shape is difficult for conduction. In a sense, the distribution uniformity
of heat tube boundaries (heat flow paths) to obtain the minimum ther- should be a critical feature to be optimized in the heat conduc-
mal resistance if the overall volume of the heat tube is fixed, which is tion design.
shown as follows:
3.2. Evolutionary algorithm
Design objective: Rtube → min ð12Þ
As described above, design improvement for heat conduction will be
Design constraints: V tube −V 0 ¼ 0 and g ðxÞN0 ð13Þ achieved if the distribution of heat flow paths is homogenized. To get
this goal, the following algorithm is developed using the basic concept
Design variables: g ðxÞ ð14Þ of the bi-directional evolutionary structural optimization method
(BESO) [21], in which elements within the design domain can either
where V0 is the heat tube volume, which is a constant value, as shown in be added or reduced.
Eq. (15). Step_1: The maximum allowable domain is specified. Relative parame-
Z qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ters are defined including the maximum Kmax and the mini-
x2  0 2
g ðxÞ 1 þ f ðxÞ dx ¼ V 0 ð15Þ mum Kmin of the thermal conductivity, the required volume
x1 fraction (β b 1), as well as the rejection ratio (0 ≤ RR ≤ 1) and
inclusion ratio (0 ≤ IR ≤ 1).
This is a variational problem, and the Euler function is constructed as Step_2: Establish the finite element model and carry out a linear analy-
follows: sis. The distribution uniformity of the heat flow paths is speci-
  fied as the optimization criterion. The magnitude of the heat
∂F ∂G d ∂F ∂G flux at the centroid of each element is extracted (denoted as
þλ − þλ 0 ¼0 ð16Þ
∂g ∂g dx ∂g 0 ∂g | Qi |) and used to evaluate the paths distribution uniformity.
180 B. Li et al. / International Communications in Heat and Mass Transfer 75 (2016) 177–182
Fig. 3. Simulation to confirm the straightening effect of heat flow paths.
Meanwhile, the one with the maximum magnitude is selected
as a benchmark (denoted as |Q|max).
Step_3: Elements with high conductivity material (Kmax) are selected
as the starting physical domain. Elements will be removed
from the uneven distributed regions if they satisfy Eq. (21), in
which the material is driven to Kmin forcing the paths to distrib-
ute homogeneously in a new FEA.
 distributed across the frontier between high and low conductivity mate-
jQ i j
jQ j max
≤RR
Step_4: During the computation, the selected elements fall into
two groups, one at Kmax and the other at Kmin . After sever-
al iterations, if a state is reached where no element in the
domain can satisfy Eq. (21), the rejection ratio would be
updated by Eq. (22) for the optimization process to
continue.
RR ¼ RR þ IR
Step_5: Update the FE model and carry out a finite element analysis
again. The heat flux vector is updated for each element and
used to form an ordered list according to their magnitudes.
The material of elements that are assigned with the minimum
conductivity (Kmin) is driven to Kmax from the top of the list
until the required fraction (ξ = 0.005) is reached.
Step_6: Step 2 through step 5 are repeated until the volume constraint
is reached.
Fig. 4. Layout 1 evolution from the 1st to the 150th step.
3.3. Test for straightening effect of heat flow paths
In fact, the suggested algorithm is capable of straightening the
curved heat flow paths so as to make it possible for minimizing the in-
efficient heat conduction mechanisms. To confirm this characteristic, a
simple case is conducted in this part. The design domain is a rectangle
of 100 mm × 60 mm. The FE model is constructed by Plane 55 elements
with the tool of ANSYS APDL. The mesh grid size is defined as
0.5 mm × 0.5 mm. A constant heat input q (0.25 W) is specified at
boundary Γin, a temperature T = 0 °C is imposed at boundary Γout, and
the boundary Γwall is set as adiabatic. The blue region refers to elements
with the maximum thermal conductivity (Kmax =200 W/(m∙K)), while
the green region refers to those with the minimum thermal conductiv-
ity (Kmin =0.5 W/(m∙K)). At the very beginning, the conduction path is
designed to follow a curved trajectory (an inappropriate conduction to-
pology), and then the algorithm is applied to identify the optimal con-
duction topology without changing the material volume fraction
(β=30%).
Fig. 3 illustrates the changing history of the heat flow paths. It can be
observed that the optimizer changes the layout of the highly conductive
material link by homogenizing the spacing of heat flow paths and con-
verges toward the final configuration, which turns out to be a straight-
ened conduction path. In this way, the heat flux becomes equally
ð21Þ rials. As a result, a decrease of the maximum temperature is expected.
Although the most efficient topology for heat conduction is a
straightened path, heat flow paths are usually forced to follow curved
trajectories dictated by the domain geometry or the distribution of ther-
mal loads. It can be expected that homogenizing heat flow paths where
temperature gradients are high against the material frontier will sup-
port the emergence of conduction topology toward high temperature
region and will locally facilitate the heat conduction.
ð22Þ
4. Case study
To demonstrate the optimization procedure in a directive and de-
scriptive way, the constructal-theory network is selected as the initial
layout (Layout 1) to be evolved. The design procedure of the Layout 1
will not be repeated here, detailed explanation can be found in [1].
Note that Layout 1 is a constructal tree, whereas Layout 2 has the
same definition in terms of k (200 W/(m ∙ K)) and β (20%) as that of
the Layout 1 but initialized randomly instead of the classical dichotomic
 B. Li et al. / International Communications in Heat and Mass Transfer 75 (2016) 177–182
Fig. 5. Layout 2 evolution from the 1st to the 150th step.
tree type. The comparison between Layouts 1 and 2 will highlight the
impact of initial layout on the final conduction topology calculated.
Figs. 4 and 5 give the evolution of both the conduction topology and
heat flow paths for Layouts 1 and 2, respectively.
For a certain initial layout of the high conductivity material, the sug-
gested algorithm leads to a primary V-shaped topology, which is similar
to those produced by Dirker and Meyer [22] and Errera and Bejan [23].
Heat flow paths in the final obtained topologies distribute more
homogeneous compared to those in the initial designs, indicating the
V-shaped topology is the most effective layout to heat conduction for
a square domain with a small exposed isothermal boundary as regarded
in this case. Final topologies obtained for Layouts 1 and 2 are somewhat
similar in global but diverse in detail. The final maximum temperature
of Layout 2 is 10.36% higher than that of the Layout 1. This might be be-
cause the initial Layout 1 is already a near-optimized result by the
constructal theory, whereas Layout 2 is an unspecified collection of
isolated conductive isles. Even so, the final maximum temperature of
Layout 2 is 8.95% lower compared to that of the initial maximum tem-
perature of Layout 1. This means the suggested algorithm is at least as
efficient as the constructal theory in this case.
5. Conclusions
This paper explores the challenges faced in designing an optimal to-
pology for heat conduction. A procedure for graphical representation of
heat flow paths is developed and demonstrated, highlighting relative
importance of various parts of the domain in heat conduction. Based
on this, a brand new optimizer is developed to sketch the most effective
layout to reduce the maximum temperature level. By homogenizing the
heat flow paths distribution, the high conductivity material evolves
step-by-step and forms at convergence a configuration, where the tem-
perature gradients at the frontier between high and low conductivity
materials are equalized. The unique features of the suggested method
include its capability of creating a distinct heat conduction topology
without mathematical sophistication. Since there is no optimality con-
dition imposed in the solution process, the approach may not give an
ideal optimum solution in a strict mathematical point of view; however,
it provides paths toward optimum designs. The heat flow paths can be
obtained from a standard finite element analysis, and optimization
with the suggested algorithm can be conducted as an external loop to
any standard finite element package. Therefore, it is easy to be used
for the design of conductive cooling schemes in various thermal engi-
neering scenarios.
181
Acknowledgments
The work reported in this paper is supported by the National Natural
Science Foundation of China [51405377] and the Hi-Tech Research and
Development Program of China [2012AA040701].
References
[1] A. Bejan, Constructal-theory network of conducting paths for cooling a heat gener-
ating volume, Int. J. Heat Mass Transf. 40 (1997) 799–816.
[2] A. Bejan, Constructal theory: from thermodynamic and geometric optimization to
predicting shape in nature, Energy Convers. Manag. 39 (1998) 1705–1718.
[3] A. Bejan, M.R. Errera, Deterministic tree network for fluid flow: geometry for
minimal flow resistance between a volume and one point, Fractals 5 (1997)
685–695.
[4] M.P. Bendsøe, N. Kikuchi, Generating optimal topologies in structural design using
homogenization method, Comput. Methodods Appl. Mech. Eng. 71 (1988) 197–224.
[5] M.P. Bendsøe, O. Sigmund, Topology Optimization—Theory, Methods, and Applica-
tions, Springer, Berlin Heidelberg, 2004.
[6] S.W. Zhou, Q. Li, Computational design of microstructural composites with tailored
thermal conductivity, Numer. Heat Trans. Appl. 54 (2008) 686–708.
[7] N. de Kruijf, S.W. Zhou, Q. Li, Topological design of structures and composite mate-
rials with multiobjectives, Int. J. Solids Struct. 44 (2007) 7092–7109.
[8] Y.C. Zhang, S.T. Liu, Design of conducting paths based on topology optimization,
Heat Mass Transf. 44 (2008) 1217–1227.
[9] A. Takezawa, G.H. Yoon, S.H. Jeong, Structural topology optimization with strength
and heat conduction constraints, Comput. Methods Appl. Mech. Eng. 276 (2014)
341–361.
[10] X.J. Liu, C. Wang, Y.H. Zhou, Topology optimization of thermoelastic structures using
the guide-weight method, Sci. China Technol. Sci. 57 (2014) 968–979.
[11] Q. Li, G.P. Steven, Y.M. Xie, O.M. Querin, Evolutionary topology optimization for tem-
perature reduction of heat conducting fields, Int. J. Heat Mass Transf. 47 (2004)
5071–5083.
[12] R. Ansola, E. Vegueria, J. Canales, Evolutionary optimization of compliant mecha-
nisms subjected to non-uniform thermal effects, Finite Elem. Anal. Des. 57 (2012)
1–14.
[13] T. Gao, W.H. Zhang, J.H. Zhu, Topology optimization of heat conduction problem in-
volving design-dependent heat load effect, Finite Elem. Anal. Des. 44 (2008)
805–813.
[14] C. Zhuang, Z. Xiong, H. Ding, A level set method for topology optimization of heat
conduction problem under multiple load cases, Comput. Methods Appl. Mech.
Eng. 196 (2007) 1074–1084.
[15] K. Yaji, T. Yamada, S. Kubo, K. Izui, S. Nishiwaki, A topology optimization method for
a coupled thermal-fluid problem using level set boundary expressions, Int. J. Heat
Mass Transf. 81 (2015) 878–888.
[16] G.X. Jing, H. Isakari, T. Matsumoto, Level set-based topology optimization for 2D heat
conduction problems using BEM with objective function defined on design-
dependent boundary with heat transfer boundary condition, Eng. Anal. Boundary
Elem. 61 (2015) 61–70.
[17] R. Boichot, L.G. Luo, Y.L. Fan, Tree-network structure generation for heat conduction
by cellular automaton, Energy Convers. Manag. 50 (2009) 376–386.
[18] K. Maute, Topology Optimization in Structural and Continuum Mechanics, Springer,
Vienna, 2014 389–407.
182 B. Li et al. / International Communications in Heat and Mass Transfer 75 (2016) 177–182

[19] B. Nouanesengsy, T.Y. Lee, H.W. Shen, Load-balanced parallel streamline generation
on large scale vector fields, IEEE Trans. Vis. Comput. Graph. 17 (2011) 1785–1794.
[20] C. Rossl, H. Theisel, Streamline embedding for 3D vector field exploration, IEEE
Trans. Vis. Comput. Graph. 18 (2012) 407–420.
[21] O.M. Querin, G.P. Steven, Y.M. Xie, Evolutionary structural optimization (ESO) using
a bidirectional algorithm, Eng. Comput. 15 (1998) 1031–1048.
[22] J. Dirker, J.P. Meyer, Topology optimization for an internal heat conduction cooling
scheme in a square domain for high heat flux applications, J. Heat Trans.-T. ASME
135 (2013) 1110101–11101010.
[23] M.R. Errera, A. Bejan, Deterministic tree networks for river drainage basins, Fractals
6 (1998) 245–261.