Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Bioinspiration & Biomimetics

PAPER You may also like


- Biomimicking interfacial fracture behavior
Bioinspired design and optimization for thin film of lizard tail autotomy with soft
microinterlocking structures
wearable and building cooling systems Navajit S Baban, Ajymurat Orozaliev,
Christopher J Stubbs et al.

- A minimally designed soft crawling robot


To cite this article: Jonathan Grinham et al 2022 Bioinspir. Biomim. 17 015003 for robust locomotion in unstructured pipes
Wenkai Yu, Xin Li, Dunyu Chen et al.

- DIII-D research advancing the physics


basis for optimizing the tokamak approach
to fusion energy
View the article online for updates and enhancements. M. E. Fenstermacher, for the DIII-D Team:,
J. Abbate et al.

This content was downloaded from IP address 14.139.187.71 on 12/01/2024 at 09:52


Bioinspir. Biomim. 17 (2022) 015003 https://doi.org/10.1088/1748-3190/ac2f55

PAPER

Bioinspired design and optimization for thin film wearable and


R E C E IVE D
31 January 2021 building cooling systems
R E VISE D
1 October 2021
AC C E PTE D FOR PUBL IC ATION
Jonathan Grinham1 , 2 , 3 , ∗ , Matthew J Hancock4 , Kitty Kumar3 , Martin Bechthold1 , 2 , 3 ,
13 October 2021 Donald E Ingber3 , 5 , 6 and Joanna Aizenberg3 , 6 , 7
1
PUBL ISHE D Harvard Graduate School of Design, United States of America
16 December 2021 2
Harvard Center for Green Buildings and Cities, United States of America
3
Wyss Institute for Biologically Inspired Engineering, Harvard University, United States of America
4
Veryst Engineering, LLC, United States of America
5
Vascular Biology Program and Department of Surgery, Boston Children’s Hospital and Harvard Medical School,
United States of America
6
Harvard John A Paulson School of Engineering and Applied Sciences, United States of America
7
Department of Chemistry and Chemical Biology, Harvard University, United States of America

Author to whom any correspondence should be addressed.
E-mail: jgrinham@gsd.harvard.edu

Keywords: bioinspired, numerical optimization, microfluidic, heat exchange, buildings, wearable

Abstract
In this work, we report a paradigmatic shift in bioinspired microchannel heat exchanger
design toward its integration into thin film wearable devices, thermally active surfaces in buildings,
photovoltaic devices, and other thermoregulating devices whose typical cooling fluxes are below
1 kW m−2 . The transparent thermoregulation device is fabricated by bonding a thin corrugated
elastomeric film to the surface of a substrate to form a microchannel water-circuit with bioinspired
unit cell geometry. Inspired by the dynamic scaling of flow systems in nature, we introduce
empirically derived sizing rules and a novel numerical optimization method to maximize the
thermoregulation performance of the microchannel network by enhancing the uniformity of flow
distribution. The optimized network design results in a 25% to 37% increase in the heat flux
compared to non-optimized designs. The study demonstrates the versatility of the presented
design and architecture by fabricating and testing a scaled-up numerically optimized heat
exchanger device for building-scale and wearable applications.

1. Introduction the environment. In many of these examples, the rate


of heat exchange is controlled dynamically through
In nature, organisms use fluid flows for thermal and the dilation or contraction of vascular smooth mus-
chemical exchange, transport of nutrients and waste, cle cells, which increase or decrease available surface
and parts of the immune system. The fluid mechan- area and blood flow [4].
ics of vascular networks in warm-blooded organ- Nature’s ability to regulate heat using multifunc-
isms are well studied for thermal regulation [1]. Bulk tional, hierarchical, and dynamic flow structures has
blood flows are carried through large arteries that generated numerous biologically-inspired fluid flow
then branch into successively smaller structures, ulti- architectures in microfluidic devices. To maintain a
mately arriving at capillary beds within tissues. These constant shear stress in blood vessels, Murray’s Law
hierarchical structures minimize the hydraulic resis- provides widely recognized bio-mimetic design rules
tance of bulk flows while maximizing the surface area for the hierarchical scaling relationship of daugh-
available for heat and mass transport at the capillary- ter and parent branches in vascular flow systems
scale. In some animals, capillary beds near the surface [5, 6]. These scaling principles have been confirmed
of appendages or other external structures, such as through extensive studies of fluid transport in ani-
ears of elephants [2] or beaks of toucans [3], provide mals [7] and have been translated to microfluidics
a high surface area exchange element capable of reg- [8, 9]. Scaling rules have also been applied to opti-
ulating core body temperatures by shedding heat to mize bulk flow and diffusion in volumetric branching

© 2021 IOP Publishing Ltd


Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 1. Mapping of heat flux (y) relative to scale of working area (x) for single phase microchannel heat exchangers including
commercial and defense electronic cooling (X) [21–28], laser diodes () [24, 29, 30], mobile air conditioning () [31–33],
concentrated photovoltaics (+) [35–39], wearable devices and medical devices (◦) [41–45], photovoltaic panels () [36, 52],
and thermally active building devices () [56–60]. The numeric subscripts are provided for each data citation.

structures [10, 11]. For example, modeling construc- [35–39], to name a few examples. For these tradi-
tal structures provides an optimal method for cool- tional applications, the microchannels are built into
ing a volume from a single flow source [12, 13]. The or onto the surface of a rigid device with a cooling
constructal method is shown to be optimal for plate area on the order of 1 to 10 cm2 to cool with a heat
configurations [14]. Experimentally, branching, or flux greater than 1000 kW m−2 (100 W cm−2 ), sum-
bifurcated, designs show improved fluid flow distri- marized in figure 1. Early studies reported cooling
bution compared to channel arrays with a single dis- with heat fluxes up to 790 W cm−2 , or 7900 kW m−2
tributor [15]. Computational fluid dynamics research [24] and future devices are projected to cool with
confirms that bifurcated designs improve thermal dis- heat fluxes up to 3 kW cm−2 , or 30 000 kW m−2
tribution and reduce required pumping power com- [23]. To accommodate such high heat fluxes, these
pared to conventional channel designs [16]. traditional microchannel devices are constructed
Other research that may not be considered bioin- from materials that exhibit high thermal conductivity
(k  20 W (m K)−1 ), are good electrical insulators
spired uses the electric circuit analog, which provides
(ρe  1018 μΩ cm), and operate at high temperatures
a simplified method of modeling fluid flow in chan-
(Tmax ∼ 120◦ C) [40].
nels and distributors [17, 18]. Similar methods are
In contrast, thin, transparent, and possibly flexi-
used to optimize flow distribution using network and
ble, heat exchangers required for modern technolo-
lattice geometry designs [19]. Moreover, multi-scale,
gies, such as wearable devices and medical devices
numbering up methods show improved flow unifor-
[41–51], photovoltaic panels [36, 52–55], thermally
mity in microchannel chemical reactors [20]. Across active building devices [56–62], and sky cooling pan-
these and other examples, researchers show that to els [63], only have heat fluxes below 1 kW m−2
achieve high heat flux and cooling efficiency, heat over a larger area, on the order of 1 m2 (figure 1).
exchanger designs should address both uniform tem- For such cooling devices, the material requirements
perature distribution and flow distribution along with are completely different from traditional devices. For
low thermal resistivity and low hydraulic resistivity example, these devices may require high flexibility
[21]. (E  1 GPa) and high transmissivity in the visible
Using these and similar design methods, spectrum (T > 95%), typically resulting in a low ther-
microchannel heat exchangers have been widely mal conductivity material (k  1 W (m K )−1 ) [40].
applied for cooling of commercial and defense Because the system requirements for these devices
electronics [21–28], laser diodes [24, 29, 30], mobile address low heat fluxes over large surface areas (see
air conditioners [31–34], and compact photovoltaics far lower right in figure 1), this distinct cluster of

2
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 2. Examples of distribution and exchange architectures in nature: (A) rete mirabile structure in vertebrates; (B) vein
structure in leaves. (C) Schematic of a bioinspired design with a network of channels (width w) arrayed in a diamond-shaped unit
cell geometry (side length δ). (D) Schematic of a synthetic device with component and material call-outs. (E) Schematic section
through network channels (width w and height h).

applications demands novel design strategies. To date, arbitrary channel network designs. Here we demon-
such strategies have not been proposed due to a lim- strate the method on the diamond-shaped unit cell
ited understanding of fluid flow and heat transfer design explored in phase one as well as on square
behavior for small-scale flows in the related new mate- lattice structures. The results suggest that uniform
rial palette [64]. flow may be achieved by successively sizing each
In this work, we report the design of transparent channel’s width in the network and that flow uni-
microchannel heat exchange devices fabricated from formity can accordingly improve temperature unifor-
a highly transparent (T ∼ 95% [65, 66]), stretch- mity and cooling performance.
able polymer (E = 0.57–3.7 MPa) with a bioin-
spired diamond-shaped unit cell design that applies
to a broad set of applications with surface areas on 2. Channel sizing study
the order of 1 m2 . To achieve this, we first opti-
mize the unit cell geometry of the microchannel net- In the first phase of the study, we establish siz-
work architecture for low heat fluxes on a 125 mm × ing relationships and characteristic values for key
125 mm device by experimentally studying the corre- design parameters by employing measures of heat
lation between heat flux and channel dimensions, i.e. flux, heat exchange effectiveness, and thermal resis-
width and length. Second, we optimize fluid flow uni- tance to benchmark the cooling performance of heat
formity over larger surface areas using a novel com- exchangers with microchannel networks composed of
putational method inspired by dynamic vasodilation diamond-shaped unit cells.
and vasoconstriction used to increase and decrease
flow in blood vessels [4]. Said computational method 2.1. Heat exchanger architecture
comprises an automated algorithm implemented in To design the heat exchanger we adopt an architec-
MATLAB® that iteratively runs finite element sim- ture similar to the network designs used in natural
ulations in COMSOL Multiphysics® and applies to and mammalian systems for material distribution and

3
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Table 1. Parameter values tested. Parameters include: channel height h, channel width
w, unit cell spacing δ, channel area ratio φ, and channel spacing to width ratio δ/w.

h (μm) w (mm) δ (mm) φ (%) δ/w (-)

70 0.5 2, 5, 10, 20, 80 58, 26, 14, 7, 2 4, 10, 20, 40, 160
70 1.0 10, 20, 40, 80 26, 14, 7, 4 10, 20, 40, 80
70 2.0 8, 20, 80 58, 26, 7 4, 10, 40

exchange. Examples include the rete mirabile’s net- inlet and outlet [68, 69],
like arterial and venous structures which function
ṁcp (Toutlet − Tinlet )
as high surface area counter-current exchangers in qflow = , (2)
many vertebrate cardiovascular systems (figure 2(A) A
[1]) and the complex venation used for fluid trans- where ṁ is the mass flow rate of liquid through the
port in plant leaves (figure 2(B) [67]). Compared device, cp is the heat capacity of the liquid, A is the
with single, parallel, and serpentine channel designs, plan area of the device (0.0156 m2 ), and Tinlet and
the diamond-shaped unit cell design mitigates device Toutlet are the temperatures at the inlet and outlet,
failure due to blockage. Each channel in the dia- respectively. Tests are conducted by first heating the
mond network carries the same or similar flow, unlike suspended heat exchanger device (with a heat lamp
in other designs where low flow rate channels are placed directly above the device) without flow to
present merely to mitigate blockage (e.g. the square bring the surface temperature to a steady state (T∞ ),
network design considered later). Note that some net- measured between 38 and 40 ◦ C in our experiment.
work designs that appear square have inlet and out- Then, room temperature water (Tinlet = 20 ◦ C) is
let ports at opposing corners; such designs are effec- pumped through the microchannel network at flow
tively diamond networks [19]. The heat exchanger rates of 2, 6, and 10 ml min−1 . The typical Reynolds
design considered in this first phase of the study com- numbers are in the range 1  Re  300 where
prises a network of channels (width w) forming a Re = υDν h and υ is the average flow velocity in one
unit cell (side length δ) with a diamond-shaped layout channel, Dh is the hydraulic diameter, and ν is
(figure 2(C)). Each device is fabricated by bonding the kinematic viscosity of water. The outlet water
a thin PDMS layer containing molded channel voids temperature (Toutlet ) was measured to provide a mea-
(channel height h) to a glass sub-layer forming the surement of qflow from equation (2). In our exper-
microchannel network (figures 2(D) and (E)). iment, water traverses an open circuit and empties
To design an optimal diamond-shaped into a waste reservoir to maintain a constant inlet
microchannel network with uniform flow dis- water temperature Tinlet throughout the experiment.
tribution and heat transfer behavior, we study the In the preferred system, we envision the outflowing
effects of the channel area ratio (φ) on heat flux and liquid would flow through another heat exchanger
thermal resistance. φ is defined as the ratio of the to reduce its temperature to a desired level prior to
plan area of a microchannel segment to the portion reintroduction into the array.
of the plan area of the device cooled by said segment The heat exchange effectiveness (E) is the ratio of
(figure 2(C)) and is given by the heat removed by the fluid flow (qflow ) described
√ above in equation (2) to the heat applied to the
2 2wδ − 2w2 device by the lamp (qlamp ) described below in
φ= × 100%. (1)
δ2 equations (7)–(10). Effectiveness ranges from 0 to 1,
where E = 1 indicates that all heat is removed by the
By design of the diamond pattern, the value of φ for fluid flow. Effectiveness is defined by
each unit cell is the same as that for the global channel
network. Unless otherwise specified, the panel designs qflow
E= . (3)
herein have a plan area of 80 mm × 80 mm (device qlamp
area 125 mm × 125 mm) and h = 70 μm. Table 1 lists
the parameter values tested. Figure 3(A) shows the heat flux removed by the flow
(qflow ) calculated in equation (2) from measured inlet
and outlet water temperatures for diamond chan-
2.2. Results nel network designs with different w and φ values
We employ multiple measures to characterize the (tested values listed in table 1). Figure 3(A) also shows
cooling performance of our prototypical microchan- the corresponding heat exchange effectiveness for the
nel heat exchangers: the steady state heat flux, the heat same tested values. Because the heat flux of the lamp is
exchange effectiveness, and thermal resistance. The constant, the inter-sample heat flux and effectiveness
heat flux removed by the flow (qflow ), i.e. cooling flux, trends are the same. Therefore, a higher qflow indicates
of a microchannel heat exchanger has been previously a higher rate of heat removal and a more effective heat
described using the energy balance between the flow exchanger design.

4
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 3. (A) Heat flux removed by the flow (qflow ) calculated in equation (2) (left y-axis) and heat exchange effectiveness (E,
right y-axis) vs channel area ratio φ. (B) Total thermal resistance Rtherm , calculated in equation (4), vs channel area ratio φ. (C)
Sample thermal image showing area used for surface temperature average, as well as typical reference locations for measuring
Tinlet and Tinlet (measurements taken when inlet/outlet tubes were covered in insulation, which was removed to generate the image
shown here). (D) Measured average steady state surface temperature T̄ s for devices with channel width w = 0.5 mm and room
temperature water (20 ◦ C) flowing at 6 ml min−1 . In (A) and (B), dashed lines show power-law fit for w = 0.5 mm, while error
bars indicate the standard deviation for at least three independent measurements.

Consistent cooling performance trends are tested (figure 3(A), table 1). Rtherm includes the mea-
observed for each of the flow rates tested. For sured thermal resistance of the device (Rtherm,device ),
each channel width w, as the channel area ratio φ which accounts for the fluid convection and material
increases, so too does the heat flux removed by the conductance and the measured thermal resistance to
flow, qflow , and the effectiveness, E; however, the rate the environment (Rtherm,env ), which accounts for the
of increase decreases with increasing φ. In particular, air convection at the surfaces and radiant transfer to
for each flow rate, qflow and E increase two-fold as φ and from the environment. The measured values are
increases from 4% to 14%; however, qflow and E only defined by
increase by 5% as φ increases from 14% to 58% and
as channel spacing to width ratio (δ/w) decreases Rtherm = Rtherm,device + Rtherm,env (4)
from 20 to 4, respectively. This suggests that, for the
applications outlined in the introduction, suitable T̄ s − Tinlet
Rtherm,device = (5)
heat exchange designs are those that have δ/w of Aqlamp
order 20, so that the corresponding φ is of order T̄ s − Tlab
15%. The law of diminishing returns applies beyond Rtherm,env = , (6)
Aqlamp
those limits, i.e. it is not necessary to have densely
packed channels. This limit may also have fabrication where T̄ s is the average steady-state surface tempera-
benefits. Smaller channels may be more difficult to ture with water flowing, Tinlet is the inlet water tem-
fabricate and a larger area between channels allows perature, Tlab is the laboratory’s air temperature, A is
for more area to bond the elastomer containing the the plan surface area of the device, and qlamp is the heat
channel array to the substrate, likely increasing yield flux on the device generated by the lamp. Figure 3(C)
during fabrication. shows T̄ s measurements collected using infrared (IR)
To extend the cooling performance characteriza- imaging and averaged spatially across the channel
tion further, figure 3(B) shows the approximate mea- array. Figure 3(D) shows the measured average sur-
sured total thermal resistance (Rtherm ) for the devices face temperature reached at steady state for select

5
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

samples. The value of qlamp in equations (2), (4)–(5) nels scales as 1/h3 , where h is the channel height, dou-
was not measured. However, since the steady state sur- bling the channel height reduces the hydraulic resis-
face temperature (T∞ ) without flow was measured tance eight-fold, which was previously shown for a
for each test, we can approximate heat energy input similar device design [62]. This strategy was also pre-
of the lamp by approximating the heat transfer rate viously discussed for similar devices with low gravity-
to the environment from the device (conservation of driven flow rates appropriate for building applica-
energy) as follows: tions [68]. We too use larger channel heights in the
larger devices reported in section 3.4. Also, similar to
qlamp = htotal (T∞ − Tlab ) (7)
natural flow structures, like the rete mirabile, a plural-
ity of small channels distributed over an area improve
htotal = hconv,top + hconv,bot + hrad,top + hrad,bot (8)
thermal resistance while reducing the flow rate per
hrad = σ(T2∞ + T2lab )(T∞ + Tlab ) (9) channel, reducing the pressure drop and the required
pumping power.
NuL kair This concludes the results from the first phase of
hconv = . (10)
L our study, which establishes metrics (heat flux, effec-
Here A is the plan surface area of the device, h total is the tiveness, and thermal resistance) to benchmark cool-
total heat transfer coefficient, and Tlab is the ambient ing performance and suitable characteristic values for
air temperature. An approximation of htotal is found design parameters (width-to-spacing ratio w/δ and
by calculating the radiant heat transfer coefficients on area ratio φ) of heat exchangers with diamond-shaped
each side of the device (hrad,top , hrad,bot ) in equation (9) microchannel networks.
using the Stefan–Boltzmann constant (σ) and the
emissivity of the device surface (). An approximation
of the convective heat transfer coefficients on each
side of the device (hconv,top , hconv,bot) is found from 3. Numerical optimization of heat
equation (10) using a previously described calcula- exchanger
tion for the convective heat transfer coefficient of a
heated or cooled surface at an arbitrary angle [70, 71]. In this section, we report results from the second
The result is htotal ∼ 20 W m−2 K−1 , which is appro- phase of our study on microchannel heat exchang-
priate based on a survey of radiant heating and cool- ers. As outlined in the introduction, good tempera-
ing systems that have similar constraints [72]. From ture and flow distribution are important factors for
this approximation, we estimate qlamp = 445W m−2 , a improved heat exchange performance. Figure 3(C)
reasonable approximation given the qflow values mea- shows the temperature non-uniformity across a typi-
sured in figure 3(A). Note that the approximation of cal diamond-shaped microchannel network, resulting
Rtherm characterizes the cooling performance of device in a ‘V’-shaped isotherm oriented laterally across the
geometries within our sample data, and is not meant array with observable hot spots at the downstream
for comparison with other devices. corners. To address this temperature non-uniformity
Returning to figure 3(B), the diminishing returns and study potential thermal performance improve-
trends for Rtherm are similar to those observed for ments generated by a more uniform flow distribution,
qflow and E: reductions in thermal resistances begin we introduce a novel numerical optimization strat-
to wane beyond a channel area ratio of φ = 15%. egy inspired by the dynamic vasodilation and vaso-
Rtherm decreases by 75% as φ increases from 4% to constriction used to increase and decrease the flow
14%; however, Rtherm decreases less than 15% as φ in skin blood vessels [4]. The numerical optimization
increases from 14% to 58%. Again, the law of dimin- routine narrows or widens channel segments to max-
ishing returns applies at channel spacing-to-width imize the flow uniformity across the array. As a result,
ratios δ/w beyond order 20 and φ beyond order 15%. the outer channels corresponding to longer flow paths
Finally, Rtherm introduces a common optimization are widened, while the central channels correspond-
trade-off in nature and engineering between ther- ing to shorter flow paths are narrowed, increasing
mal resistance and hydraulic resistance. Flow systems the uniformity of the total flow resistance along any
with small channel cross-sections tend to have lower path through the array. We also present the results
thermal resistance but larger hydraulic resistance and of a complementary experimental study to confirm
associated pumping power. Hydraulic resistance is an the improved cooling performance of the numerically
important design constraint for microfluidic devices optimized designs.
[73, 74]. However, earlier studies showed that since As described in the introduction, multiple exam-
hydraulic resistance is sensitive to channel cross- ples in nature use hierarchical flow structures that
sectional dimensions, the latter may be increased by successively size vessels within branching structures
small amounts to reduce the hydraulic resistance and from large vessels for bulk fluid flow to smaller ves-
pumping power below desired levels without signif- sels for diffusion. The distributors and collectors
icantly affecting the thermal resistance [62, 68]. For of the devices presented are simple versions of this
example, since the hydraulic resistance of wide chan- hierarchical channel sizing. When optimizing larger

6
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

channel arrays (see section 3.4), we use bifurcated dis- channel segment in the array from the target, in
tributor and collector manifolds to maintain a tighter this case the global average, Qtarget = Qtotal /Ny .
distribution of optimized channel widths in the array. (b) Given a set of channel segment widths, a
two-dimensional geometry is generated by a
MATLAB® script using COMSOL® ’s Livelink
for MATLAB® . Some geometry cleanup is nec-
3.1. Numerical optimization of channel network
design essary to remove small faces or edges where the
channel segments meet. The geometry is then
This section introduces a novel numerical approach
to optimize the uniformity of flow distribution across meshed, followed by a laminar flow simulation
the diamond cell and square lattice channel net- in COMSOL® . Flow rates at each channel seg-
works of interest, each of which is characterized ment (Qchannel ) are evaluated and reported to the
by a Nx × Ny array of channel segments (figure 4). MATLAB® script.
During optimization, the width of each channel (c) The width (w) of each segment is updated as
segment in a given network is iteratively adjusted w new = wold × Qtarget /Qchannel , where Qtarget is
(thereby adjusting the segment’s flow resistance) the target average flow rate across a channel seg-
to minimize the difference between the flow rate ment. Therefore, if the flow rate in a segment is
through said segment and the global average flow greater than target Qtarget , its width is reduced to
rate Qtarget = Qtotal /Ny , where Qtotal is the total flow increase its flow resistance and reduce the flow
rate through the device. The iterative optimization rate, and vice versa. If needed, channel width
routine is implemented in MATLAB® and COM- restrictions may be imposed here, which may
SOL Multiphysics® ; salient MATLAB® code and limit the achievable flow uniformity.
COMSOL® files are provided in the SI (https:// (d) Steps 2 and 3 are repeated until the maximum
stacks.iop.org/BB/17/015003/mmedia) so that oth- relative difference between channel segment flow
ers may apply the optimization to a wider range rates and the target, max(|Qchannel /Qtarget | − 1)
of channel flow and heat exchange applications7 . (which by definition equals the maximum change
Laminar flow simulations are performed in COM- in channel segment width when scaled by the
SOL Multiphysics® . The flow is governed by the new width, max(|w new − w old |/w new )) is below
Navier–Stokes equations, which incorporate flow a certain threshold. In cases reported herein,
continuity and conservation of momentum. To said threshold depends on the mesh resolution
save memory and expedite the simulations, the (details below).
two-dimensional shallow-channel approximation is
employed, which is valid for wide channels. We To shorten run time, the optimization routine
assume the fluid (water at 20◦ C) is incompressible begins with a coarse mesh and progresses to medium
and solve for the steady-state flow. No-slip bound- and then fine meshes. Precise meshing details may be
ary conditions are applied on all channel walls. The found in the code provided in the SI. The numer-
flow rate Qtotal is specified at the device inlet and zero ical optimization of the 16 × 16 diamond cell net-
dynamic pressure is specified at the outlet. The chan- work began with the initial non-optimized design in
nel height is uniform across the device, to facilitate figure 4(A) which has a maximum relative flow devi-
manufacturing, and is set between 100 μm (for reg- ation max(|Qchannel /Qtarget − 1|) of 80%. After 5, 2,
ular sized devices) and 500 μm (for large arrays, see and 4 iterations with the coarse, medium, and fine
section 3.4). The channel spacing is also constant, resolution meshes, respectively, said relative flow rate
δ = 10 mm. The distributor and collector widths of deviation was less than 0.68%, 0.34% and 0.06%. The
the simulated diamond and square channel networks numerically optimized design is shown in figure 4(B).
are 5 mm. Using the stated minimum number of iterations for
The iterative numerical optimization algorithm is each mesh resolution required 1.5 h in total run time
as follows: on six Intel® Xeon® 3.50 GHz cores. Optimizing
the 12 × 12 square lattice network in figure 4(C)
(a) The widths of the channel segments in the
proceeded in a similar manner, though convergence
array are initialized by the non-optimized value,
was slower. The maximum relative flow deviation of
w = 1 mm. This is a suitable choice given the
the non-optimized design is 63%. Similar maximum
discussion in section 2 since the corresponding
relative flow deviations, 0.73%, 0.48%, 0.07% were
channel area ratio is φ  15%. The objective is
achieved for coarse, medium, and fine mesh resolu-
to minimize the deviation of flow rate across each
tions with 25, 16, and 29 iterations, respectively. Said
7 Note that the code provided in the SI is an updated version of iterations required a total of 9 h and produced the
that used for this manuscript which improves the convergence to optimized design shown in figure 4(D).
an optimized design, especially involving the medium and fine
Figure 4 shows the predicted velocity magnitude
mesh resolutions. The new code produces similar optimized dia-
mond networks, but the optimized square networks have narrower and relative deviation from the average flow rate
channel segment width ranges. (Qchannel /Qtarget − 1) for the initial and numerically

7
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 4. (Left) Initial and optimized channel network designs and resulting velocity magnitudes predicted by two-dimensional
shallow-channel flow simulations at a total flow rate of 5 ml min−1 (Right) Relative deviation between flow rate through a
particular channel segment and the global average for (A) initial 16 × 16 diamond cell channel network with 1 mm width for all
channel segments in the array, (B) optimized 16 × 16 diamond cell channel network with widths of channel segments in the array
optimized to maximize flow uniformity, (C) initial 12 × 12 square lattice network with 1 mm width for all channel segments in
the array, (D) optimized 12 × 12 square lattice network with widths of channel segments in the array optimized to maximize flow
uniformity. Relative deviations (right, z-axis) for (A) and (C) are three-orders of magnitude larger than those for (B) and (D).
Channel heights are 100 μm.

8
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

optimized channel network designs. The first col- case, though we felt that the two-dimensional opti-
umn also shows the channel segment width changes. mization was adequate for this study. Secondly, the
In the initial (non-optimized) designs with constant channel network geometries are optimized for one
channel segment width, the inlet-to-outlet flow paths particular Reynolds number (i.e. flow rate, fluid vis-
via the central region are shorter than those through cosity, and channel height), and may not perform
the outer regions. Therefore, the flow resistance is optimally for other Reynolds numbers. This may
less (more) and the channel segment flow rates are be particularly relevant near the inlet and outlet.
more (less) for flow paths through the central (outer) Lastly, the channel width optimization may produce
regions. The numerical optimization widens chan- channels that are too wide or narrow to fabricate or
nels in the outer regions, especially near the corners, that intersect, limiting the achievable maximum uni-
and narrows the central channel segments, especially formity. In this case, the master inlet and outlet could
near the ports, to decrease (increase) the flow resis- be replaced by bifurcating structures (see section 3.4).
tance and direct more (less) flow to outer (central)
channels. The second column shows the relative flow 3.2. Experimental validation of optimized and
deviation Qchannel /Qtarget − 1 for each channel seg- non-optimized channel network designs
ment in the array. The flow uniformity increases dra- We fabricated and tested the non-optimized and
matically between non-optimized and optimized dia- numerically optimized channel network designs to
mond and square channel arrays. The maximum flow assess performance improvements. To demonstrate
non-uniformity max(|Qchannel /Qtarget − 1|) of the ini- the scalability of our optimization approach, we also
tial diamond and square network designs are 80% performed a preliminary study on a heat exchanger
and 63%, respectively. In the numerically optimized design with a larger surface area. We first tested one
designs, said non-uniformity is 0.06% and 0.07% for pair of optimized and non-optimized designs from
the diamond and square channel networks, respec- each main type of channel network considered herein,
tively. In the next section, we study experimentally the diamond and square. The distributor and collector
effect of said flow uniformity optimization on surface widths of the non-optimized systems were 2 mm,
temperature distribution. Note that the optimization while those of the optimized systems were chosen as
condition in step 3 could be replaced by one related to 5 mm, which, for the designs considered here, yield
the surface temperature above each channel segment, reasonable optimized channel segment widths in the
if heat transfer were solved in addition to the flow. arrays. Note that very wide channels may lack struc-
All designs simulated herein are symmetric with tural integrity, while very narrow channels may com-
respect to the x-axis, the line connecting master inlet plicate fabrication. We measured and compared the
and outlet. The non-optimized designs (which form temperature uniformity of these optimized and non-
the starting point of the numerical optimizations) are optimized designs. For additional comparison, we
also symmetric with respect to the y-axis, which is fabricated and tested serpentine, parallel, and branch-
perpendicular to the x-axis halfway between the mas- ing channel network designs (with 1 mm channel
ter inlet and outlet. Though the optimized designs widths and 10 mm channel spacing), which are com-
seem qualitatively (visually) symmetric with respect monly used in industrial and scientific applications
to the y-axis, they are not precisely so due to inertial (see review in section 1). Experiments were run with
effects (e.g. the streamlines for flow from a narrow a constant static temperature (T∞ ) between 38◦ C and
channel segment to a wider segment are not identi- 40◦ C. Room temperature water (Tinlet = 20◦ C) was
cal to those for flow moving in the reverse direction). pumped through each microchannel network at a
Moreover, the numerical optimization algorithm was flow rate of 6 ml min−1 (Reynolds number Re 20)
designed for and validated on the full channel net- and 30 ml min−1 for the larger designs (Reynolds
work designs (i.e. without reducing the computa- number Re 20).
tional domains using the x-axis symmetry) in order
that it may also be applied to non-symmetric cases in 3.3. Results for diamond, square, and common
the future. channel network designs
A number of issues may reduce the unifor- Channels in each heat exchanger design were visual-
mity of the flow rates across the numerically opti- ized by filling the channels with a graphitic carbon
mized channel networks. The accuracy of the two- suspension (figures 5(A), 6(A) and 8(A)). The tem-
dimensional flow simulations (which rely on the perature distributions are reported in figures 5(B),
shallow channel approximation) depends on the 6(B) and 8(B) for each channel network design tested.
size of the channel aspect ratios. Following opti- To help visualize the lateral (y) variation in surface
mization, some channel segments may not have temperature distribution, the 22.5◦ C isotherm (i.e.
widths much larger than height, reducing the accu- 2.5◦ C warmer than the inlet water temperature 20◦ C)
racy of the flow simulation and therefore the chan- is plotted as a point above at each channel segment in
nel network design optimization. A similar itera- the array (figures 5(B), 6(B) and 8(B)). Said isotherm
tive numerical technique involving three-dimensional is also plotted separately for each design in a Cartesian
finite element simulations could be pursued in this grid (figures 5(C), 6(C) and 8(C)).

9
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 5. Temperature uniformity analysis for optimized and non-optimized diamond and square channel network designs.
(A) Channel geometries are visualized by filling channel arrays with a graphitic carbon suspension solution. (B) Thermal images
of the top of each heat exchanger device show the effect of cooling by the channel arrays. Magenta dots indicate position of 22.5◦ C
isotherm on the top device surface above channels in the array. (C) Axial position of 22.5◦ C isotherm across the channel array vs
lateral (y) position in the array. The x and y grid spacings are 10 mm.

Figure 5 parts (B) and (C) show that the tem- temperature distributions in the optimized devices.
perature distributions of the optimized diamond and Moreover, the ‘V’-shape of the temperature profile
square network designs are more laterally uniform of the non-optimized designs likely indicates that the
across the channel array. In particular, temperature flow rates are higher in the central channels near the
profiles in the lateral (y) direction have shallow ‘U’ inlet and lower in the outer channels near the cor-
shapes in the optimized designs, compared to sharper ners, as predicted by our numerical simulations. The
‘V’ shapes in the non-optimized designs. The ‘U’ vs sharp temperature increases observed in figure 5(B)
‘V’-shaped profiles indicate that the numerical opti- near the outer channels (forming the outer edges of
mization of flow uniformity results in more uniform the observed ‘U’-shaped profiles) are the result of heat

10
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 6. Temperature uniformity analysis for branching, parallel, and serpentine channel networks typical of industrial heat
exchangers. All geometries have constant channel spacing δ = 10 mm and width w = 1 mm. (A) Channel geometries are
visualized by filling channel arrays with a graphitic carbon suspension solution. (B) Thermal images of the top of each heat
exchanger device show the effect of cooling by the channel arrays. Magenta dots indicate position of 22.5◦ C isotherm on the top
device surface above channels in the array. (C) Axial position of 22.5◦ C isotherm across the channel array vs lateral (y) position in
the array. The x and y grid spacings are 10 mm.

diffusing from the non-cooled solid-phase perimeter temperature deviations (T) and define temperature
of device. uniformity [75] by
Figure 6 shows the temperature distributions
for serpentine, parallel, and branching channel net- Temperature Uniformity
 
work designs, which are commonly used in industry. Max(T) − Min(T)
The serpentine design has the most uniform surface = 1− × 100%.
Mean(T)
temperature distribution (flat isotherm) while the
(11)
parallel and branching designs have ‘V’-shaped
isotherms. The maximum, mean, and minimum temperatures
To further quantify and compare the uniformity in equation (11) are taken over all channel segments
of the temperature distributions across channel net- except those on the outer boundaries adjacent to non-
work designs, we calculate the deviation of surface cooled regions.
temperatures (figures 5(C), 6(C) and 8(C)) from their The temperature uniformity for different designs
maximums above intersection points where channel are reported in figure 7. The non-optimized diamond
segments cross the axial centerline (x) of the array. and square channel network designs have tempera-
The deviations for channel segments that are the ture uniformities of 52 ±6% and 46 ±7%, respec-
same distance from the axial centerline of the net- tively. The optimized diamond and square channel
work are then averaged. We then scale the averaged network designs have temperature uniformities of

11
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 7. Experimentally measured heat flux and temperature uniformity for each heat exchanger design tested.

77 ±5% and 74 ±5%, respectively. This shows that diamond and square network designs also have aver-
the optimized diamond and square channel network age surface temperatures (T̄ s ) that are 2◦ C and 1◦ C
designs have temperature uniformities that are 25% lower, respectively, than their non-optimized coun-
and 28% higher, respectively, than non-optimized terparts. The total thermal resistances (Rtherm ) of
designs. The optimized diamond and square network the optimized diamond and square channel network
designs have higher temperature uniformity com- designs, defined in equation (4), are also lower than
pared to the parallel and branching designs; how- those for the non-optimized designs. These findings
ever, they have a lower uniformity compared to may be rationalized by noting that the regions near
the serpentine design, which has a single channel the collectors of the optimized diamond and square
and a near-uniform temperature profile (figure 7). designs are cooler than those of the corresponding
non-optimized designs (i.e. the temperatures at the
The latter result is expected based on intuition and
downstream corners are cooler). This suggests that in
previous literature. Though the serpentine design has
the optimized designs, more heat is being removed by
optimal uniformity, it suffers from a pressure drop on
the higher flow rates in the channels near the periph-
the order of 40-times that for branching designs [33],
ery. Lastly, the optimized diamond and square chan-
has a longer flow path and thus requires higher flow
nel network designs have heat fluxes around 10% to
rates to avoid thermally overworking the fluid, and 46% higher than those of the serpentine, parallel, and
would also be more severely affected by a clog. branching designs (figure 7).
To compare the cooling performance between
channel network designs, we calculate the heat flux
3.4. Larger heat exchangers
removed by the flow (qflow ) defined in equation (2)
Figure 8 demonstrates the efficacy of the numerical
from the flow rate and water inlet (Tinlet ) and design optimization method on larger surface area
outlet (Toutlet ) temperatures (figure 7). The non- heat exchangers (300 mm × 300 mm) by apply-
optimized diamond and square channel network ing the method to a 40 × 40 diamond-shaped unit
designs have heat fluxes qflow of 129 ± 4 W m−2 and cell design (we considered 16 × 16 arrays in earlier
137 ± 3 W m−2 , respectively. The optimized diamond sections). The non-optimized geometries have typi-
and square channel network designs have heat fluxes cal parameters from section 2.1 with δ = 10 mm and
of 178 ± 10 W m−2 and 171 ± 5 W m−2 , respec- w = 1 mm. Unlike the smaller devices considered
tively. Thus, the flows in the optimized diamond and in earlier sections, the laminated fabrication method
square channel networks remove 37% and 25% more employed for the larger devices produces a channel
heat per unit time than their non-optimized counter- height of h = 500 μm. More critically, we used a hier-
parts. Consistent with these findings, the optimized archical sizing approach and designed the device with

12
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Figure 8. Temperature uniformity analysis for large surface area optimized and non-optimized diamond-shaped unit cell
designs. The non-optimized geometries have typical parameters from section 2.1 with δ = 10 mm and w = 1 mm. (A) Channel
geometries are visualized by filling channel arrays with a graphitic carbon suspension solution. (B) Thermal images of the top of
each heat exchanger device show the effect of cooling by the channel arrays. Magenta dots indicate position of 22.5◦ C isotherm on
the top device surface above channels in the array. (C) Axial position of 22.5◦ C isotherm across the channel array vs lateral (y)
position in the array. The x and y grid spacings are 15 mm.

bifurcated distributor and collector channels with 3%, compared to the smaller devices considered
5 mm widths (figure 8(A)). Had we instead used sin- in section 3.3. While the material composition dif-
gle large channels for the distributor and collector, as fers between the large and small devices considered
in the devices considered in section 3.3, the numeri- herein, this finding suggests multiple points of con-
cal optimization routine would have produced chan- sideration for future research. As discussed above, the
nels of widths well outside our large array fabrication smaller devices have larger perimeter-to-area ratios
limits (0.5 mm to 4 mm). Even with said bifur- compared to the larger devices. Since the flow rate
cated distributor and collector, we still had to impose optimization directs more flow to the periphery in
said limits on the optimized channel widths, which the devices considered herein, the smaller devices are
reduced the flow uniformity in the few columns of able to remove a larger proportion of heat from the
channel segments nearest the distributor and collec- periphery compared to the larger devices. Also, recall
tor. Channel segments that were more than three
that even with bifurcated distributor/collector, we still
columns away from the distributor and collector had
had to impose channel segment width restrictions on
relative flow rate deviations of less than 0.15%.
the large devices, which limited the level of flow rate
As in section 3.3, we ran a temperature uniformity
uniformity. We could have added additional bifur-
analysis on the non-optimized and optimized designs
cations, but wanted to minimize the space taken up
(figures 8(B) and (C)). Similar to the optimized
by the distributor/collector. Such restrictions were
diamond and square designs in section 3.3, the tem-
perature distribution of the 40 × 40 optimized dia- not needed on the smaller devices, due to size and
mond channel network is more laterally uniform fabrication method, and hence the smaller devices
across the channel array resulting in a shallow ‘U’- were able to achieve higher flow uniformity. More-
shaped temperature profile compared with a ‘W’- over, the bifurcated distributor/collector improved
shaped temperature profile in the non-optimized the flow rate uniformity and thermal performance
design. The optimized diamond channel network of the non-optimized large device, resulting in a less
design has a 3.5-fold higher temperature unifor- pronounced improvement by optimizing the array of
mity (equation (11)) compared to the non-optimized the large device. That said, the 3.5-fold higher tem-
design (figure 7). perature uniformity shown in the optimized device
For the larger devices, numerical optimization suggests that the optimization to the array still pro-
leads to a smaller increase in heat flux (qflow ), around vides a meaningful improvement in flow uniformity.

13
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

This suggests that, for flow and temperature unifor- 5. Materials and methods
mity, optimizing the interior channels of a device may
reduce the need for multiple bifurcations, in turn Polydimethylsiloxane (PDMS, Sylgard 184, Dow
reducing the device area used for bifurcated distrib- Corning) layers (2 mm thick) were molded on an
utor and collector channels. Additionally, the 3.5-fold original master template, fabricated with mirror-
higher temperature uniformity and limited cooling finish vinyl (Sign Warehouse, 70 μm thick) by scribe-
improvement may indicate that the mass flow rate cutting (Graphtec cutting plotter CE5000-60), and
across the larger devices may be too low, resulting in a layered on a flat plastic surface. The PDMS layers were
thermally overworked fluid. Finally, the larger chan- bonded to 3 mm glass and 2 mm polydimethylsilox-
nel height changes the aspect ratio and reduces the ane layers by first treating with a corona arc discharge
corresponding Nusselt numbers, which may reduce system (ETP, BD-20), then holding at 70◦ C overnight.
the convective heat transfer rate. Large devices (figure 8) are laminated using 3 mm
glass plates bonded to a laser cut 500 μm adhesive
foamed acrylic (3M VHB) interior layer.
4. Conclusion A peristaltic pump (Stenner SVP1) was calibrated
and used to control the water flow rate through the
In this work we presented an experimentally derived
device channels. In study phase one, a 150 W incan-
sizing rule for microchannel networks with bioin-
descent lamp was placed approximately 50 cm above
spired diamond-shaped unit cell designs. This
the sample heat exchange device, to heat its top sur-
design strategy is intended for new, non-traditional
face. The heat exchange device was suspended 75 mm
microchannel devices that cool large surface areas at
above the work surface. Prior to cooling, the top sur-
relatively low heat fluxes in wearable, medical, and
face of the device was heated up to a static (i.e. flow
building industry applications. Our experimental
off) temperature between 38◦ C to 40◦ C. In study two,
characterization of these channel networks shows
the device was placed on a silicone heat pad (Omega
that the channel segment width to spacing ratio
SRFRA-3/10). Prior to cooling, the top surface of
(w/δ) may be taken on the order of 20 or greater,
the device heated up to 38◦ C. Thermal IR measure-
or equivalently, a channel area ratio (φ) greater than
ments were made using an FLIR SC 5600 camera,
approximately 15%. In other words, more closely
and 5 K-type thermocouples (Omega 5TC-TT-K-20-
packed channels do not appreciably improve heat
36 ±0.75%) were attached to the PDMS and glass
exchanger performance. Fabrication strategies related
surfaces to verify and calibrate IR temperature mea-
to the w/δ scale factor should also be considered in
surements. Inlet and outlet water temperatures were
future design work. For example, since the minimal
measured using 2 K-type thermocouples (Omega
suitable value for w/δ is approximately 20, larger
5TC-TT-K-20-36 ±0.75%) and collected using a data
channel widths and lengths may be used, within fab-
logger (GL Data Logger ±0.01%). Image processing
rication limitations. Likewise, as the area of elastomer
software (WebPlotDigitizer) was used to generate the
between channels increases, the total surface area
isotherm on the top device surface above the chan-
for bonding increases, increasing the yield during
nel array with a temperature of 2.5◦ C above the inlet
fabrication and reducing the risk of delamination at
water temperature. To visualize the channel network
higher flow rates.
geometry, a graphitic carbon suspension (Speedball
We also reported a novel, bioinspired numerical
Super Black Free Flowing India Ink) was pumped by
method for optimizing channel flow rate uniformity
a syringe pump (Harvard Apparatus, PhD Ultra).
and therefore heat exchanger cooling performance.
The numerical method was applied to two geome-
tries (diamond and square channel networks) in the Author credit
present work, and we demonstrated that the method
is sufficiently flexible to be generalized to larger heat Jonathan Grinham: conceptualization, methodology,
exchangers. We fabricated and tested optimized and formal analysis, investigation, writing-original draft,
non-optimized diamond and square channel network visualization. Matthew Hancock: conceptualization,
designs and compared their temperature uniformity methodology, simulation development in COMSOL
and cooling performance. The increased tempera- Multiphysics® with MATLAB® , Formal analysis,
ture uniformity of the optimized designs increases investigation, writing-original draft, visualization.
the heat flux by 25% to 37% compared to non- Kitty Kummar: conceptualization, methodology,
optimized designs. We also showed experimentally writing-review and editing. Martin Bechthold:
that the optimized designs have better cooling per- writing-review and editing, supervision. Donald
formance relative to conventional industry channel E Ingber: writing-review and editing, supervision.
network designs fabricated from the same elastomer Joanna Aizenberg: writing-review and editing,
composite. supervision.

14
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

Acknowledgments [16] Ramos-Alvarado B, Li P, Liu H and Hernandez-Guerrero A


2011 CFD study of liquid-cooled heat sinks with
microchannel flow field configurations for electronics, fuel
This research received funding from the Wyss Insti- cells, and concentrated solar cells Appl. Therm. Eng. 31
tute for Biologically Inspired Engineering at Harvard 2494–507
University. [17] Ajdari A 2004 Steady flows in networks of microfluidic
channels: building on the analogy with electrical circuits C.
R. Phys. 5 539–46
Data availability statement [18] Oh K W, Lee K, Ahn B and Furlani E P 2012 Design of
pressure-driven microfluidic networks using electric circuit
analogy Lab Chip 12 515–45
The data that support the findings of this study are [19] Tondeur D, Fan Y, Commenge J-M and Luo L 2011
available upon reasonable request from the authors. Uniform flows in rectangular lattice networks Chem. Eng.
Sci. 66 5301–12
[20] Saber M, Commenge J-M and Falk L 2009 Rapid design of
ORCID iDs channel multi-scale networks with minimum flow
maldistribution Chem. Eng. Process. 48 723–33
Jonathan Grinham https://orcid.org/0000-0001- [21] Ohadi M M, Choo K, Dessiatoun S and Cetegen E 2013
Next Generation Microchannel Heat Exchangers
6374-2800 (SpringerBriefs in Thermal Engineering and Applied Science)
Matthew J Hancock https://orcid.org/0000-0001- (New York: Springer)
9820-3620 [22] Smakulski P and Pietrowicz S 2016 A review of the
Donald E Ingber https://orcid.org/0000-0002- capabilities of high heat flux removal by porous materials,
microchannels and spray cooling techniques Appl. Therm.
4319-6520 Eng. 104 636–46
Joanna Aizenberg https://orcid.org/0000-0002- [23] Ebadian M and Lin C 2011 A review of high-heat-flux heat
2343-8705 removal technologies J. Heat Transfer 133 110801
[24] Mundinger D, Beach R, Benett W, Solarz R, Krupke W,
Staver R and Tuckerman D 1988 Demonstration of
References high-performance silicon microchannel heat exchangers for
laser diode array cooling Appl. Phys. Lett. 53
[1] Frost P G H, Siegfried W R and Greenwood P J 1975 1030–2
Arterio-venous heat exchange systems in the Jackass [25] Chen N C J, Felde D K and Yoder G L 1996 Thermal analysis
penguin Spheniscus demersus J. Zool. 175 231–41 of two-phase microchannel cooling Report No.
[2] Phillips P K and Heath J E 1992 Heat exchange by the pinna CONF-961105-5 Oak Ridge National Laboratory
of the African elephant (Loxodonta africana) Comp. [26] Jaeseon Lee I and Mudawar I 2009 Low-temperature
Biochem. Physiol., A 101 693–9 two-phase microchannel cooling for high-heat-flux thermal
[3] Tattersall G J, Andrade D V and Abe A S 2009 Heat management of defense electronics IEEE Trans. Compon.
exchange from the toucan bill reveals a controllable vascular Packag. Technol. 32 453–65
thermal radiator Science 325 468–70 [27] Shao B, Sun Z and Wang L 2007 Optimization design of
[4] Tansey E A and Johnson C D 2015 Recent advances in microchannel cooling heat sink Int. J. Numer. Methods Heat
thermoregulation Adv. Physiol. Educ. 39 139–48 Fluid Flow 17 628–37
[5] Murray C D 1926 The physiological principle of minimum [28] Liu D and Garimella S V 2007 Flow boiling heat transfer in
work: I. The vascular system and the cost of blood volume microchannels J. Heat Transfer 129 1321–32
Proc. Natl Acad. Sci. 12 207–14 [29] Roy S K and Avanic B L 1996 A very high heat flux
[6] Murray C D 1926 The physiological principle of minimum microchannel heat exchanger for cooling of semiconductor
work: II. Oxygen exchange in capillaries Proc. Natl Acad. Sci. laser diode arrays IEEE Trans. Compon., Packag., Manuf.
12 299–304 Technol. 19 444–51
[7] LaBarbera M 1990 Principles of design of fluid transport [30] Glukhikh I V, Polikarpov S S, Frolov S V, Volkov A S and
systems in zoology Science 249 992–1000 Privezentsev V V 2010 Cooling of silver bullet laser diode
[8] Barber R W and Emerson D R 2008 Optimal design of submodules Tech. Phys. 55 855–9
microfluidic networks using biologically inspired principles [31] Qi Z, Zhao Y and Chen J 2010 Performance enhancement
Microfluid. Nanofluid. 4 179–91 study of mobile air conditioning system using microchannel
[9] Emerson D R, Cieślicki K, Gu X and Barber R W 2006 heat exchangers Int. J. Refrig. 33 301–12
Biomimetic design of microfluidic manifolds based on a [32] Garimella S 2003 Innovations in energy efficient and
generalised Murray’s law Lab Chip 6 447–54 environmentally friendly space-conditioning systems Energy
[10] Cohn D L 1954 Optimal systems: I. The vascular system 28 1593–614
Bull. Math. Biophys. 16 59–74 [33] Han Y, Liu Y, Li M and Huang J 2012 A review of
[11] Cohn D L 1955 Optimal systems: II. The vascular system development of micro-channel heat exchanger applied in
Bull. Math. Biophys. 17 219–27 air-conditioning system Energy Procedia 14
[12] Bejan A 1997 Constructal-theory network of conducting 148–53
paths for cooling a heat generating volume Int. J. Heat Mass [34] Peng Q and Du Q 2016 Progress in heat pump air
Transfer 40 799–816 conditioning systems for electric vehicles-a review Energies
[13] da Silva A K, Lorente S and Bejan A 2004 Constructal 9 240
multi-scale tree-shaped heat exchangers J. Appl. Phys. 96 [35] Escher W, Ghannam R, Khalil A, Paredes S and Michel B
1709–18 2010 Advanced liquid cooling for concentrated photovoltaic
[14] Cho K-H, Lee J, Ahn H S, Bejan A and Kim M H 2010 Fluid electro-thermal co-generation 2010 3rd Int. Conf. Thermal
flow and heat transfer in vascularized cooling plates Int. J. Issues in Emerging Technologies Theory and Applications
Heat Mass Transfer 53 3607–14 (ThETA) pp 9–17
[15] Liu H, Li P, Van Lew J and Juarez-Robles D 2012 [36] Royne A, Dey C J and Mills D R 2005 Cooling of
Experimental study of the flow distribution uniformity in photovoltaic cells under concentrated illumination: a
flow distributors having novel flow channel bifurcation critical review Sol. Energy Mater. Sol. Cells 86
structures Exp. Therm. Fluid Sci. 37 142–53 451–83

15
Bioinspir. Biomim. 17 (2022) 015003 J Grinham et al

[37] Reddy K S, Lokeswaran S, Agarwal P and Mallick T K 2014 [56] Price Industries 2011 Engineer’s HVAC Handbook: A
Numerical investigation of micro-channel based active Comprehensive Guide to HVAC Fundamentals 1st edn
module cooling for solar CPV system Energy Procedia 54 (Winnipeg: Price Industries Limited)
400–16 [57] British Standards Institution 2004 BS EN 14240:2004:
[38] Bahaidarah H M S, Baloch A A B and Gandhidasan P 2016 Ventilation for Buildings—Chilled Ceilings—Testing and
Uniform cooling of photovoltaic panels: a review Renewable Rating (International Organisation for Standardization)
Sustainable Energy Rev. 57 1520–44 [58] Stetiu C 1998 Radiant cooling in US office buildings:
[39] Wang Y, Ding G-F and Fu S 2007 Highly efficient manifold towards eliminating the perception of climate-imposed
microchannel heatsink Electron. Lett. 43 978–80 barrier PhD Thesis Lawrence Berkeley National Laboratory
[40] Ashby M F 2011 Materials Selection in Mechanical Design [59] Chow T-T, Li C and Lin Z 2010 Innovative solar windows
4th edn (Burlington, MA: Butterworth-Heinemann) for cooling-demand climate Sol. Energy Mater. Sol. Cells 94
[41] Bartkowiak G, Dabrowska A and Włodarczyk B 2015 212–20
Construction of a garment for an integrated liquid cooling [60] Chow T-T, Li C and Lin Z 2011 Thermal characteristics of
system Text. Res. J. 85 1809–16 water-flow double-pane window Int. J. Therm. Sci. 50 140–8
[42] Ernst T C and Garimella S 2013 Demonstration of a [61] American Society of Heating Refrigerating and
wearable cooling system for elevated ambient temperature Air-Conditioning Engineers 2009 ASHRAE Handbook
duty personnel Appl. Therm. Eng. 60 316–24 Fundamentals (Atlanta, GA: American Society of Heating,
[43] Sun Y and Jasper W J 2015 Numerical modeling of heat and Refrigeration and Air-Conditioning Engineers)
moisture transfer in a wearable convective cooling system [62] Grinham J, Craig S, Ingber D E and Bechthold M 2020
for human comfort Build. Environ. 93 50–62 Origami microfluidics for radiant cooling with small
[44] Yazdi M M and Sheikhzadeh M 2014 Personal cooling temperature differences in buildings Appl. Energy 277
garments: a review J. Text. Inst. 105 1231–50 115610
[45] Song G-S and Seo J T 2009 Relationship between ambient [63] Zhao D, Aili A, Zhai Y, Lu J, Kidd D, Tan G, Yin X and Yang
temperature and heat flux in the scrotal skin Int. J. Androl. R 2019 Subambient cooling of water: toward real-world
32 288–94 applications of daytime radiative cooling Joule 3 111–23
[46] Epstein Y, Shapiro Y and Brill S 1986 Comparison between [64] Khan M G and Fartaj A 2011 A review on microchannel
different auxiliary cooling devices in a severe hot/dry heat exchangers and potential applications Int. J. Energy Res.
climate Ergonomics 29 41–8 35 553–82
[47] Drost M K and Friedrich M 1997 Miniature heat pumps for [65] Stankova N et al 2015 Optical properties of
portable and distributed space conditioning applications polydimethylsiloxane (PDMS) during nanosecond laser
IECEC-97 Proc. 32nd Intersociety Energy Conversion processing Appl. Surf. Sci. 374 96–103
Engineering Conf. vol 2 pp 1271–4 [66] Tomholt L, Geletina O, Alvarenga J, Shneidman A V, Weaver
[48] Nah P K, Pradhan C K, Nag A, Ashtekar S P and Desai H J C, Fernandes M C, Mota S A, Bechthold M and Aizenberg
1998 Efficacy of a water-cooled garment for auxiliary body J 2020 Tunable infrared transmission for energy-efficient
cooling in heat Ergonomics 41 179–87 pneumatic building façades Energy Build. 226 110377
[49] Gregory R P, Cooke T, Middleton J, Buchanan R B and [67] Carvalho M R, Losada J M and Niklas K J 2018 Phloem
Williams C J 1982 Prevention of doxorubicin-induced networks in leaves Curr. Opin. Plant Biol. 43 29–35
alopedia by scalp hypothermia: relation to degree of cooling [68] Hatton B D, Wheeldon I, Hancock M J, Kolle M, Aizenberg
Br. Med. J. 284 1674 J and Ingber D E 2013 An artificial vasculature for adaptive
[50] Bulow J, Friberg L, Gaardsting O and Hansen M 1985 thermal control of windows Sol. Energy Mater. Sol. Cells 117
Frontal subcutaneous blood flow, and epi- and 429–36
subcutaneous temperatures during scalp cooling in normal [69] Bergman T L, Lavine A S, Incropera F P and Dewitt D P
man Scand. J. Clin. Lab. Invest. 45 505–8 2011 Fundamentals of Heat and Mass Transfer (New York:
[51] Ekwall E M, Nygren L M L, Gustafsson A O and Sorbe B G Wiley)
2013 Determination of the most effective cooling [70] Rohsenow W M, Hartnett J P and Cho Y I 1998 Handbook
temperature for the prevention of chemotherapy-induced of Heat Transfer 3rd edn (New York: McGraw-Hill)
alopecia Mol. Clin. Oncol. 1 1065–71 [71] Lienhard J H IV and Lienhard J H V 2016 A Heat Transfer
[52] Hasanuzzaman M, Malek A B M A, Islam M M, Pandey A K Textbook Version 2.04 4th edn (Cambridge, MA: Phlogiston
and Rahim N A 2016 Global advancement of cooling Press)
technologies for PV systems: a review Sol. Energy 137 [72] Shinoda J, Kazanci O B, Tanabe S-I and Olesen B W 2019 A
25–45 review of the surface heat transfer coefficients of radiant
[53] Radziemska E 2003 The effect of temperature on the power heating and cooling systems Build. Environ. 159 106156
drop in crystalline silicon solar cells Renew. Energy 28 1–12 [73] Stone H A, Stroock A D and Ajdari A 2004 Engineering
[54] Zondag H A, de Vries D W, van Helden W G J, van flows in small devices: microfluidics toward a lab-on-a-chip
Zolingen R J C and van Steenhoven A A 2002 The thermal Annu. Rev. Fluid Mech. 36 381–411
and electrical yield of a PV-thermal collector Sol. Energy 72 [74] Squires T M and Quake S R 2005 Microfluidics: fluid
113–28 physics at the nanoliter scale Rev. Mod. Phys. 77 977–1026
[55] Coventry J S 2005 Performance of a concentrating [75] Commenge J-M, Saber M and Falk L 2011 Methodology for
photovoltaic/thermal solar collector Sol. Energy 78 multi-scale design of isothermal laminar flow networks
211–22 Chem. Eng. J. 173 541–51

16

You might also like