International Communications in Heat and Mass Transfer 38 (2011) 1349–1353

Contents lists available at SciVerse ScienceDirect

International Communications in Heat and Mass Transfer

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

A numerical and experimental investigation of flow maldistribution in a

micro-channel heat sink☆
G. Kumaraguruparan a, R. Manikanda Kumaran b, T. Sornakumar a, T. Sundararajan b,⁎
a
Dept. of Mechanical Engg, Thiagarajar College of Engineering, Madurai, 625015, India
b
Dept. of Mechanical Engg, Indian Institute of Technology Madras, Chennai, 600036, India

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

Available online 9 September 2011 A study of flow mal-distribution in U-type micro-channel configuration is presented. Numerical simulations
indicate that flow deceleration and associated pressure recovery in the inlet header lead to flow separation
Keywords: and recirculation which cause oscillations in channel-wise mass flow distribution. Increase in flow resistance
Micro-channels by decrease in channel depth, width or number of channels or increase in channel length, results in a more
Mal-distribution uniform distribution. Mal-distribution increases at high flow rate or low viscosity due to the dominance of
Flow separation
inertial phenomena. Experiments performed on a 25-channel setup illustrate that small manufacturing var-
Inlet header
iations in channel dimensions introduce random fluctuations in flow distribution.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction types of flow manifolds and the associated mal-distribution patterns

Micro-channels are employed in the cooling of microelectronics,
offshore gas/fuel processing, gas turbine blade cooling, PEM-fuel
cells and in high temperature nuclear reactors [1,2]. Channels
whose widths lie in the range of a few microns to a few hundred mi-
crons [3] are generally termed as micro-channels. The high volumet-
ric heat dissipation rates encountered in many modern electronic
devices necessitate the use of large heat transfer areas, which could
be achieved by resorting to coolant flow through micro-channels
[4,5]. Pressure drop and heat transfer studies of micro-channel flow
have been performed in the recent past, using single channels or mul-
tiple parallel channels. The single phase experiments on single micro-
channels [6–8] reveal that the experimental and theoretical friction
factor data are in good agreement and there is no early transition to
turbulent regime. But, experiments on multiple micro-channels ex-
hibit inconsistent results with regard to friction factor variation and
flow transition [9,10], which are attributed to mal-distribution effects
not being properly accounted for. Flow mal-distribution can occur
due to size variations between the channels because of manufactur-
ing tolerances or due to changes in the viscosity of the fluid or be-
cause of poor manifold design [11]. The effect of viscosity variation
on mal-distribution is incorporated by introducing a viscosity ratio
in the correlations for friction factor and Nusselt number [12]. The
have been discussed by Bajura and Jones [13]. Pan et al. [14] have
shown that in order to achieve a uniform flow distribution, the
micro-channel depth, width and pitch should be made as small as
possible; also, the channel length should be large. It has been ob-
served that the optimal channel configuration depends on the mass
flow rate and Reynolds number of flow [15,16].
Although a few studies are available in literature which explain
the effects of channel geometry on flow mal-distribution in micro-
channels, the detailed mechanisms which cause mal-distribution
have not been identified. Further, the influence of factors such as
the number of channels, fluid viscosity and manufacturing tolerances
has not been highlighted. In the present study, an experimental and
numerical study is performed to systematically investigate the causes
for flow mal-distribution and to highlight the effects of parameters
such as the number of channels, type of fluid used and manufacturing
tolerances.
2. Mathematical model and numerical solution procedure
Subjected to continuum, steady, incompressible and laminar (for
the small flow rates considered here) flow considerations, the equa-
tions that govern the three dimensional flow in the micro-channels
can be written in flux form suitable for finite volume discretization
over a cell, using Cartesian tensor notation, as follows:

☆ Communicated by A.R. Balakrishnan. Continuity equation:

⁎ Corresponding author at: Thermodynamics and Combustion Engineering Lab,
Department of Mechanical Engineering, Indian Institute of Technology, Madras, Chennai,


600036, India.
∮ Vj ·nj ·dA ¼ 0 ð1Þ
E-mail address: tsundar@iitm.ac.in (T. Sundararajan).

0735-1933/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.icheatmasstransfer.2011.08.020
1350 G. Kumaraguruparan et al. / International Communications in Heat and Mass Transfer 38 (2011) 1349–1353

of 0.5%) beyond a mesh of about 1,065,558 cells. Hence, for all the
Nomenclature
subsequent studies, a grid with 1,065,558 cells has been adopted.
Simulations have been carried out until the residues fall below
A Area
1 × 10 −6 for all flow variables and below 1 × 10 −4 for continuity equa-
d Depth
tion, during the iterative solution procedure.
g Acceleration due to gravity
L Length
m Total flow rate 3. Experimental setup and measurement procedure
n Unit normal
N Number of channels The 25-channel heat sink set up employed in the experimental
Δp Pressure drop study is shown in Fig. 1. For each channel of this set up, the width
t Time and depth have been measured using a profilometer at five different
V Volume axial locations. At each location, the dimensions are measured three
W Width times and averaged. It is to be noted, however, that there are size var-
xi, xj Cartesian coordinate direction iations (up to about 5%) along the length of each channel and also be-
tween channels, due to manufacturing tolerances. A transparent
Greek symbols acrylic sheet along with a slotted mild steel cover is fixed over the
p Density micro-channels to visualize the flow inside the channels and the
σ, τ Normal and Shear stress header. Water is used as the working fluid in the experiments. In
order to quantify the mal-distribution encountered in the micro-
Subscripts channel setup, water passing through each channel is collected indi-
C Channel vidually without employing an outlet header.
Cell Computational cell Prior to the start of experiments, high pressure air is passed
through the micro-channel heat sink, which is immersed in water.
After ensuring that there is no air leak, KMnO4-water solution is
sent through the setup to guarantee that the fluid flows only through
Momentum equation:
the micro-channels, without spilling over the channels. De-ionized,


 de-gassed water from the reservoir is pumped with the help of 0.5
∂Vi
ρ·Vcell · þ ρ·∮ Vj ·nj ·Vi ·dA ¼ ρ·gi ·Vcell þ ∮ σij nj ·dA ð2Þ HP self priming mono-set pump, through a 5 μm water filter. The
∂t
water flow rate is controlled with the help of valves provided near
the outlets. When flow is stabilized, the water from each channel is
Where the stresses are given by
collected with the measuring jar for 3 to 5 min, and the mass of the
! water is measured by an electronic balance with a maximum error
∂Vi ∂Vj of about 0.2%. For a fixed flow rate, the experiment has been repeated
σij ¼ −pδij þ μ þ
∂xj ∂xi a few times on different days to ensure the repeatability of the data

with δij denoting the Kronekar delta and i, j denoting Cartesian direc-
tional indices. Due to the low Knudsen number range (Kn b 0.001) of
Table 1
the present study, no additional terms have been included to repre- Predicted pressure drop (Δp) and Mal-distribution factor (MF) for various cases
sent micro-phenomena in the governing equations and boundary studied.
conditions.
Sl. No. Values Pressure drop Maldistribution
The geometrical dimensions of the channels along with the type of
(Δp) between factor (MF) in%
the fluids and their mass flow rates employed for the simulation are inlet and
summarized in Table 1. A three-dimensional model of the micro- outlet/Pa
channel geometry is created using the software Gambit 2.1 and the Default case wc = 0.5 mm, dc = 5 mm, N = 25, 942 3.06
governing equations of flow are solved using the finite volume Lc = 150 mm, Fluid = Water
based commercial CFD solver, Fluent 6.2. A segregated implicit solver m = 0.007 kg/s
with SIMPLE pressure-correction algorithm has been adopted to com- Wc (mm) 0.05 808,782 0.11
0.075 240,161 0.10
pute the flow field. The boundary conditions applied for the analysis 0.1 101,576 0.15
are: prescribed mass flow rate at the header inlet; no-slip conditions 0.3 3938 0.80
at the walls of the header and the channels; at the header exit, the dc (mm) 0.1 1,172,970 0.02
pressure is specified as atmospheric. For the default geometry, the ef- 0.5 21,104 0.45
1 6410 0.85
fect of mass flow rate on the channel-wise flow distribution has been
2.5 1973 1.69
studied. Although majority of the cases studied here correspond to Lc (mm) 50 377 9.96
the laminar regime, in a few cases, the Reynolds number is of the 100 661 5.46
order of 4000. In such cases, the one equation Spalart–Allmaras (SA) Wf (mm) 0.5 945 2.79
turbulence model has been adopted to compute the flow pattern. 1 945 3.52
N 5 4706 0.98
The SA model has been employed due its relative simplicity and effec-
10 2305 1.04
tiveness in capturing the turbulent features of boundary layer domi- 15 1543 1.23
nated internal flows. 20 1165 2.45
Grid sensitivity for the numerical predictions has been investigat- m (kg/s) 0.0207 3346 8.20
0.0562 13,301 25.54
ed using various structured non-uniform meshes, with number of
0.0780 21,904 24.88
cells ranging from 615,646 to 1,366,790. Finer grids have been used Fluids Ethylene glycol 11,972 0.08
within the channels in order to capture the maldistribution effects Air 95,352 69.82
and the flow features accurately. The predicted channel-wise normal- m = 0.0597 45099.61 32.12
ized flow rates for different meshes indicate that the numerical re- m = 0.0798 74467.89 35.8435
m = 0.0956 102643.36 37.9433
sults are insensitive to grid refinement (with a maximum deviation
 G. Kumaraguruparan et al. / International Communications in Heat and Mass Transfer 38 (2011) 1349–1353 1351

a 4. Validation of numerical predictions

Channels Lu and Wang [17] carried out simulations to investigate the mal-
distribution patterns in an I-type micro-channel configuration. Re-
sults predicted by the present model for the same configuration
with a header inlet velocity of 0.2 m/s are compared in Fig. 3. The pre-
dicted average channel inlet velocities of the present model compare
well with those of [17]. In particular, similar oscillatory variations are
observed in both the results. Since some of the dimensions such as
header depth were not reported in Ref [17], it was not possible to
consider an exactly identical channel configuration. Notwithstanding
such differences, good match is observed between the two sets of pre-
dictions in Fig. 3.
Header
5. Results and discussion

The experimental results obtained for the actual micro-channel

setup illustrate the fact that due to size variations along the channel
b length and between the channels, significant random fluctuations
occur in the channel-wise flow distribution (refer Fig. 2). Therefore,
Filter in order to systematically analyze the effects of various geometric
and operational parameters on mal-distribution, uniform channels
have been considered in the numerical simulations. Different param-
eters have been varied one at a time as listed in Table 1, to investigate
their effects on maldistribution.

5.1. Features of micro-channel flow (default configuration)

In order to highlight the flow phenomena which are responsible

Pump
Fig. 1. Experimental setup of micro-channel heat sink used in the study.
measured. It is seen in Fig. 2 that the experimental data for different
runs are quite consistent. A statistical analysis carried out on the
mal-distribution data shows that the average uncertainty in the
mass flow rate measurement (for 95% confidence level) is ±2.034%
(with variation in the range of 0.75% to 5% for individual channels).
As compared to the numerical predictions, the experimental data ex-
hibit additional random fluctuations. Probably due to random varia-
tions in the channel dimensions during manufacturing.
1.4
1.2
Normalized mass flow rate
for the undulations observed in channel-wise flow rates, the velocity
contours in the inlet header have been plotted in Fig. 4a. It is noticed
that the flow field in the inlet header is not uniform due to the pres-
ence of recirculation zones in the vicinity of some channel inlets. In
general, due to flow diversion through the channels, the average ve-
locity of flow decreases in the direction away from the header inlet,
which results in static pressure rise similar to the case of a diffuser
flow. While velocity values in the central part of the header are
high, the velocities near the channel inlets are low due to wall fric-
tion. The low velocity flow in the peripheral region of the header is
unable to flow against the adverse pressure gradient beyond a certain
distance, and hence flow separation gives rise to the formation of a
recirculation zone. Recirculation bubbles are found to occur near the
channel inlets of 8 to 11 and 14 to 17, which are highlighted in the fig-
ure. In some cases, even the entry of the flow from the header to the
channel occurs in the reverse direction, as indicated by the arrows. In
Fig. 4b, it is observed that static pressure value increases in the inlet
header from the input end to the dead end of the header, with undu-
lations occurring due to reverse flow. On the other hand, the static
0.10

0.09
Ref(24)
1.0 0.08 Present study
0.07
Velocity / ms-1

0.8 0.06
0.6 0.05
Experimental-Run 1
0.04
0.4 Experimental-Run 2
Experimental-Run 3 0.03
0.2 Experimental-Run 4 0.02
Numerical
0.01
0.0
5 10 15 20 25 0.00
0.00 0.01 0.02 0.03 0.04 0.05 0.06
Channel number
Header length / m
Fig. 2. Repeatability of experimentally measured flow distributions and comparison
with numerical flow distribution for m = 0.0597/kg s−1. Fig. 3. Validation for predicted channel inlet velocity distribution.
1352 G. Kumaraguruparan et al. / International Communications in Heat and Mass Transfer 38 (2011) 1349–1353

pressure variation in the outlet header smoothly decreases in the flow 1.020
direction, because the overall flow rate as well as the flow velocity in- 0.05mm
1.015

Normalized mass flow rate

0.075mm
creases towards the header outlet. It is evident that the flow separa-
1.010 0.1mm
tion and recirculation occurring in the inlet header are primary 0.3mm
responsible for the flow mal-distribution between the channels. 1.005 0.5mm

5.2. Effects of various parameters on flow maldistribution 1.000

0.995
The effects of parameters such as the channel width (Wc) and fluid
viscosity on flow maldistribution are depicted in Figs. 5–6 by plotting 0.990
the normalized channel flow rate with respect to the channel num- 0.985
ber. The trends seen in Figs. 5–6 and Table 1 can be understood
from the overall concept outlined here. The flow distribution de- 0.980
5 10 15 20 25
velops as a result of the relative flow resistances experienced by the
fluid on its path through the inlet header, micro-channel and the out- Channel number
let header. As discussed already, the flow separation and recirculation
Fig. 5. Effect of channel width on flow distribution.
phenomena in the inlet header and the viscous flow through the
channels seem to play a predominant role in deciding the extent of
flow mal-distribution for the U-type configuration considered. Thus, can be shown that flow mal-distribution decreases with decrease in
a relative comparison of flow resistances of the inlet header and the channel depth or increase in channel length because of increase in
channel can explain the effect of each parameter on mal-distribution. channel flow resistance (Table 1). The channel pitch has only a
Fig. 6 shows the effect of the channel width on maldistribution. minor influence on the flow maldistribution.
Flow mal-distribution is less for low channel widths (b 0.1 mm) and The pressure variation in a flow is contributed by both inertial and
it becomes considerable for larger widths. As the channel width de- viscous phenomena. For instance, the pressure variations in the inlet
creases, the flow resistance of the channel increases (relative to that and the outlet headers are the consequences of static pressure rise or
of inlet header) which leads to better flow uniformity. Similarly, it drop as per the Bernoulli's equation, in the regions of flow deceleration

Flow inlet

b 1.02305e+5 1.0144e+5
1.02300e+5
outlet header pressure/Pa

1.0142e+5
Inlet header presure/Pa

1.02295e+5
1.0140e+5
1.02290e+5
1.02285e+5 1.0138e+5
1.02280e+5
1.0136e+5
1.02275e+5
1.0134e+5
1.02270e+5
1.02265e+5 1.0132e+5
0.00 0.01 0.02 0.03 0.04 0.05 0.06
Header length /m

Fig. 4. (a) Contours of velocity magnitude in the inlet header (m = 0.007/kg s−1); (b) Static pressure variation along the inlet and outlet headers at m = 0.007 kg s−1.
 G. Kumaraguruparan et al. / International Communications in Heat and Mass Transfer 38 (2011) 1349–1353 1353

a 1.4 6. Conclusions

1.3 Water
The study indicates that in a U-type micro-channel configuration,
Normalized mass flow rate
Air
1.2 Ethylene glycol static pressure increases or decreases in the flow direction due to flow
deceleration or acceleration, in the inlet and outlet headers respec-
1.1 tively. Because of adverse pressure gradients in the inlet header,
1.0 flow separation and recirculation occur, which result in non-
uniform undulating flow variation between the channels. Dominance
0.9 of inertial phenomena in the inlet header results in greater maldistri-
0.8 bution, while larger flow resistance in the channels gives rise to a
more uniform flow distribution. At smaller channel width or depth
0.7 or larger channel length, flow distribution is more uniform because
0.6 of increase in the channel flow resistance. On the other hand, for an
5 10 15 20 25 increase in the overall flow rate or for less viscous fluid, mal-
Channel number distribution increases because of the dominance of inertial effects
within the inlet header. Experimental measurements on a microchan-
nel heat sink setup show that even small variations in the channel di-
b 1.020 mensions (of the order of 5%) due to manufacturing tolerances,
Water impose significant random fluctuations in the channel-wise distribu-
1.015
Normalized mass flow rate

Ethylene glycol tion of flow rates.

1.010
1.005 References

1.000
0.995
0.990
0.985
0.980
5 10 15 20 25
Channel number
Fig. 6. (a) Predicted flow distribution for different fluids used in the study; (b) Predicted
flow distribution for water and ethylene glycol.
or acceleration. When the fluid viscosity is relatively high, these inertial
effects are subdued and as a result, the effects of flow separation and
recirculation are suppressed. As seen from Fig. 6a, b, flow mal-
distribution is less for the more viscous fluids of ethylene glycol and
water, as compared to air; in fact, flow distribution for water is better
than that for air but poorer than that of ethylene glycol. It can be
shown that, the flow becomes more uniform as the channel length in-
creases, due to the relatively larger contribution of channel flow resis-
tance, as compared to the inlet header resistance.
For different operating and geometric conditions a maldistribution
factor (MF) defined as {(mmax − mmin)/muni} x 100 is calculated, and
listed in Table 1. Here, mmax, mmin and muni denote the maximum
flow rate, minimum flow rate and uniform flow rate for the channel-
wise flow distribution. The MF value for the default case is 3.06%. How-
ever, at higher total flow rates, its value is higher due to stronger inertial
effects. Smaller MF indicates better flow distribution through the chan-
nels. The value of MF increases with increase in channel width, channel
depth and the number of channels. It is evident that in a parallel channel
configuration, equivalent resistance will decrease with increase in the
number of channels and hence, mal-distribution increases. The value
of MF decreases with decrease in channel length and fluid viscosity, as
explained earlier. The total pressure drop values for different cases are
also listed in Table 1.
[1] L.M. Pua, S.O. Rumbold, Industrial micro-channel devices — where are we today?
Proc. First International Conference on Micro-channels and Mini-channels,
ICMM2003-1101, New York, 2003, pp. 773–780.
[2] F. Yilin, L. Lingai, Recent applications of advances in micro-channel heat exchangers
and multi-scale design optimization, Heat Transfer Engineering 29 (5) (2008)
461–474.
[3] M. Gad- el- hak, The MEMS Hand book, CRC publishers, Boca Raton, Florida, 2006.
[4] C.B. Sobhan, S.V. Garimella, A Comparative analysis of studies on heat transfer and
fluid flow in micro-channels, Microscale Thermophysical Engineering 5 (2001)
293–311.
[5] D.B. Tuckerman, R.F.W. Pease, High performance heat sinking for VLSI, IEEE Electron
Device Letters EDL- 2 (1981) 126–129.
[6] P. Wu, W.A. Little, Measurement of friction factors for the flow of gases in very
fine channels used for microminiature Joule–Thomson refrigerators, Cryogenics
23 (5) (1983) 273–277.
[7] J.C. Harley, Y. Huang, H.H. Bau, J.N. Zemel, Gas flow in micro-channels, Journal of
Fluid Mechanics 284 (1995) 257–274.
[8] K. Vijayalakhsmi, K.B. Anoop, H.E. Patel, P.V. Harikrishna, T. Sundararajan, S.K.
Das, Effects of compressibility and transition to turbulence on flow through
micro-channels, International Journal of Heat and Mass Transfer 52 (9–10)
(2009) 2196–2204.
[9] X.F. Peng, G.P. Peterson, Convection heat transfer and flow friction for water flow
in micro-channel structures, International Journal of Heat and Mass Transfer 39 (12)
(1996) 2599–2608.
[10] J.M. Cuta, C.E. McDonald, A. Shekarriz, Forced convection heat transfer in parallel
channel array micro-channel heat exchanger, Advances in Energy Efficiency
Heat/Mass Transfer Enhancement, ASME HTD 338 (1996) 17–23.
[11] R.K. Shah, D.P. Sekulic, Flow Maldistribution and Header Design (Chapter 12) in
Fundamentals of Heat Exchanger Design, John Wiley and Sons, New Jersey, 2003.
[12] G.R. Putnam, W.M. Rohsenow, Viscosity induced non uniform flow in laminar
flow heat exchangers, International journal of heat and mass transfer 28 (5)
(1985) 103-038.
[13] R.A. Bajura, E.H. Jones, Flow distribution manifolds, ASME Journal of Fluids Engi-
neering 98 (1976) 654–666.
[14] M. Pan, Z. Dehuai, T. Yong, C. Dongqing, CFD-based study of velocity distribution
among multiple parallel micro-channels, Journal of Computers 4 (11) (2009)
1133–1138.
[15] E.R. Delsman, A. Peirik, M.H.J.M. De Croon, G.J. Kramer, J.C. Schouten, Micro-channel
plate geometry optimization for even flow distribution at high flow rates, Trans.
IChemE, Part A, Chemical Engineering Research and Design 82 (A2) (2004) 267–273.
[16] S.H. Choi, S. Shin, Y.I. Cho, The effect of the Reynolds number and width ratio on
the flow distribution in manifolds of liquid cooling modules for electronic packaging,
International Communication in Heat and Mass Transfer 20 (1993) 607–617.
[17] M.C. Lu, C.C. Wang, Effect of the inlet location on the performance of parallel-
channel cold plate, IEEE Transactions on Component and Packaging Technologies
29 (1) (2006) 30–38.