Experimental Thermal and Fluid Science 28 (2004) 97–103

www.elsevier.com/locate/etfs

Liquid flow friction factor and heat transfer coefficient

in small channels: an experimental investigation
a,c,* a,b a,b a
Bruno Agostini , Barbara Watel , Andr
e Bontemps , Bernard Thonon
a
CEA-GRETh, 17 rue des Martyrs, 38054 Grenoble, France
b
LEGI/GRETh, Universit
e Joseph Fourier, 17 rue des Martyrs, 38054 Grenoble, France
c
ADEME, 2 Square Lafayette, 49004 Angers, France

Abstract

This article presents friction factor and heat transfer coefficient experimental results obtained with a liquid flow of R134a in
rectangular mini-channels. Two test sections made of aluminium multi-port extruded (MPE) tubes (channels dimensions 1.11
1.22
mm and 0.73
0.72 mm) were tested. Mass flux ranges from 65 to 2900 kg/m2 s and heat flux from 210 to 49 700 W/m2 . The
emphasis has been put on metrology problems and how they influence experimental results in mini-channels. Literature correlations
for large tubes were found to predict our results reasonably well.
Ó 2003 Elsevier Inc. All rights reserved.

1. Introduction convection heat transfer in MPE industrial mini-chan-

nels with R134a. This article presents single-phase flow
New environmental policies for global warming pre- results. The study of boiling heat transfer is in progress.
vention are based on a drastic reduction of greenhouse
effect gases emissions. One solution is the miniaturiza-
tion of heat exchangers thanks to mini-channels 2. Experiment description
(hydraulic diameter about 1 mm) whose increased heat
transfer coefficients allow a reduced fluid inventory and 2.1. The test loop
a better thermal efficiency. They are already widely used
in vehicles air conditioning condensers and are bound to Fig. 1 shows the test loop used for this study. It is
be used in evaporators and domestic climatisation. made of two distinct circuits, the main circuit with the
Results found in literature about heat transfer in R134a flow where the test section is inserted and a sec-
mini-channels ([1–3]) for single-phase flow are often ondary cooling circuit with 0 °C glycol–water mixture to
contradictory and operating conditions change from one cool the fluid heated in the test section. The main ele-
study to another so that comparisons are difficult. De- ments of the R134a circuit are a centrifugal pump
tailed information on geometry like tube roughness and (Micropump, 220 V, 200 W, 10–100 l/h), a counter flow
hydraulic diameter uniformity or fluid distribution are heat exchanger (Vicarb 11 plates) for the R134a cooling,
often missing which could explain a discrepancy in re- a liquid tank and a coriolis mass flowmeter (Rosemount
sults. Performances of mini-channels are under-esti- Micromotion CMF 025, 0–1090 kg/h) with a 4–20 mA
mated by classical correlations such as Chen [4], Shah [5] transmitter. Heat is extracted out of the glycol–water
or Liu and Winterton [6] established for large diameter mixture thanks to a R22 cooling unit.
tubes. Thus a test loop has been set-up to study forced
2.2. The test section

Fig. 2 shows the test section and its instrumentation.

Both ends are equipped with 90° manifolds for the fluid
*
Corresponding author. distribution. The tube diameter used for these manifolds
E-mail address: [email protected] (B. Agostini). is ten times the mini-channels hydraulic diameter in
0894-1777/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved.
doi:10.1016/S0894-1777(03)00027-X
98 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103

Nomenclature

Latin letters Subscripts

Dh hydraulic diameter, m atm atmosphere
e wall thickness, m bulk bulk flow
h channel height, m exp experimental
L tube heated length, m f friction
Lth thermal entry length, m fl fluid
l channel width, m G global
M _ total mass flow rate, kg/s g gravity
m_ mass flux, kg/m2 s i inlet
p pressure, bar o outlet
Dp pressure loss, bar sing singular
q_ heat flux, W/m2 w wall
S internal heated surface, m2
Dimensionless numbers
T temperature, K
f friction factor
DTsub sub sub-cooling, K
N number of tubes
Dz total tube length, m
Nu Nusselt number
Greek letters Pr Prandtl number
a heat transfer coefficient W/m2 K Re ¼ m_  Dh =ll Reynolds number
c ¼ h=l aspect ratio
n singular pressure loss coefficient
k thermal conductivity, W/m K
l dynamic viscosity, Ns/m2

resistive heating
U P
U : voltmeter I T safety valve
I : ammeter
Tw
∆P
P
M : mass flowmeter T Tfl,i Tfl,o T
∆P
∆P : differential pressure Tfl,i electrode Tfl,o
test section electrode thermocouple

outlet manifold
P : pressure tap flat tube
inlet manifold

T : temperature
T

Tw,0 Tw,1
W

external loop
cooler

glycolwater mixture
0 ˚C
1 2
rotameters

R134a loop T
L=690 mm
A-A
tank

M
thermocouple

liquid pump
H

Fig. 1. The test loop.

e

l e=0.3 mm
N channels W
order to suppress fluid distribution problems. The whole
item was thermally insulated with foam. The test section Fig. 2. The test section.
itself is made of two functional parts: an adiabatic sec-
tion for the flow hydrodynamic entry length, ( ), and a undertook our own dimensions measurements with an
heating zone placed between two pairs of electrodes electron microscope after polishing in a resin matrix.
brazed on the tube to perform a joule effect heating, ( ). Table 1 shows results averaged on several different
The electrodes are connected to a power supply (up to channels. Uncertainties are twice the standard deviation
2800 A) with two pairs of 240 mm2 cross-section copper (95% of data in this interval).
cables. As the manufacturers’ data on the channels For wall temperature measurements two thermo-
geometry were not accurate enough for this study we couples (type E, 0.5 mm diameter, 0 °C reference) are
 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103 99

Table 1
Hydraulic diameter measurements
Tube l (mm) h (mm) Dh (mm) c N W (mm)
1 1.11  0.02 1.22  0.01 1.17  0.05 1.1  0.03 32 48
2 0.73  0.01 0.72  0.01 0.77  0.05 0.98  0.03 18 18

fixed on the heated part of the tube, 6 cm after the first Table 2
electrode and 6 cm before the second. They are cali- Operating conditions
brated with a Rosemount 162-CE platinum resistance Tube 1 Tube 2
thermometer in the range 0–35 °C. The temperatures are Friction factor
then calculated with second order polynomials. In m_ (kg/m2 s) 50–600 155–2800
DTsub;i (K) 5–23 45–65
melting ice the thermocouples measure 0  0:1 °C so
pe (bar) 4–5 15–19
that the global temperature measurements accuracy is Flow Ascendant Horizontal
estimated at 0.1 K.
Heat transfer coefficient
The entrance and exit manifolds have pressure and
m_ (kg/m2 s) 65–600 95–2900
temperature taps to measure the fluid pressure and q_ (W/m2 ) 210–2500 940–49 700
temperature. We used an absolute pressure sensor DTsub;i (K) 7–18 45–65
(Rosemount type II) at the entrance calibrated from 1 to pe (bar) 9 15–19
7 bar with a 25 mbar accuracy, a first differential Flow Ascendant Ascendant
pressure sensor (Kent-Deltapi-K) between the test sec-
tion entrance and exit calibrated from 0 to 400 mbar Table 3
with a 1 mbar accuracy, a second differential pressure Measurements and uncertainties
sensor (Rosemount) calibrated from 0 to 75 mbar with a Range Error
0.15 mbar accuracy and two type K thermocouples to m_ (kg/m2 s) 50–2800 1.5–9%
measure the inlet and oulet fluid temperature calibrated q_ (W/m2 ) 210–49 700 3–6%
from 0 to 35 °C with a 0.1 °C accuracy. The three T (°C) 0–55 0.1 °C
pressure sensors are calibrated with a DRUCK pressure DP (mbar) 4–1700 0.3–13%
Dh (mm) 1.17 and 0.77 4–7%
measurement standard. The flexible tubes between the Dz (mm) 1050 and 1100 0.5%
pressure taps and the pressure sensors were not heated
because the working pressure was far from the satura-
of 20 as advised by Moffat [7]. Table 3 summarises the
tion pressure at atmospheric temperature. The current
different measurements and their total uncertainties
intensity is measured by a shunt in the power supply
(taking into account statistical errors). For uncertainties
used for joule effect heating (4 A). The voltage is
calculations the Moffat [8] and Kline and Mc Klintock
measured directly between the two electrodes (1 mV).
[9] methods were used.
The calibrated coriolis mass flowmeter uncertainty was
ð0:1%  M _ þ 0:054Þ.
4. The friction factor

3. Operating conditions and procedure 4.1. Results

Table 2 shows operating conditions for the study of The total pressure loss through the test section is:
the friction factor and heat transfer coefficient. The mass
Dpexp ¼ Dpf þ Dpsing þ Dpg ð1Þ
flow rate range was chosen so that we could observe the
transition from laminar to turbulent regime. The R134a with
thermo-physical properties were calculated with the m_ 2 Dz
REFPROP 6.1 software. Pressure dependence of these Dpf ¼  4f  ð2Þ
2ql Dh
properties was not taken into account around the
working pressure since it is not sensitive with sub-cooled Dpg ¼ ql  g  Dz ð3Þ
liquid. For the heat transfer study the heat flux was
chosen so that the fluid temperature rise should be m_ 2
maximal and the exit sub-cooling maintained high en- Dpsing ¼ n ð4Þ
2ql
ough (5 K) to improve the accuracy. The wall tem-
yielding
perature was kept under Tsat . After a 3 h long
stabilisation period signals from sensors are monitored Dh 2ql  Dh
4f þ n  ¼ 2 ðDpexp  ql  g  DzÞ: ð5Þ
through a HP3421 data logger and averaged by groups Dz m_  Dz
100 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103

Table 4 0.11
4f and Re global uncertainties Dh = 0.77 mm
0.1 Shah & London
DRe=Re Dð4f þ n  Dh =DzÞ=ð4f þ n  Dh =DzÞ Blasius
Tube 1 10% 15% 0.09
Tube 2 4% 7%
0.08

0.07

ξ ⋅ Dh / ∆z + 4f
In the vertical configuration (tube 1) the liquid height
was calculated with the pressure loss without flow. Table 0.06
4 shows final uncertainties on Re and 4f . These are
mainly due to errors on the channels dimensions. The
pressure losses through the straight sections of the 0.05
manifolds have been neglected because of their large
diameter compared to the mini-channels (0.3% of the
0.04
total pressure loss).
Figs. 3 and 4 represent the left hand term in Eq. (5) as
a function of the Reynolds number. The Shah and
London [10] correlation for a laminar flow in a rectan- 0.03
gular channel, 500 1000 5000
Re
4f  Re ¼ 96ð1  1:3553  c þ 1:9467  c2  1:7012
Fig. 4. 4f þ n  Dh =Dz versus Re (Dh ¼ 0:77 mm).
c3 þ 0:9564  c4  0:2537  c5 Þ; ð6Þ
and the Blasius correlation for a turbulent flow
2
Dh 2ql ðl  hÞ  N 2
4f ¼ 0:316  Re0:25 ; ð7Þ 4f ¼ Dpf   ð10Þ
Dz M_2
are also represented.
These results are very sensitive to the accuracy on the D4f Dl Dh DDh Dl
¼2 þ2 þ 7 : ð11Þ
hydraulic diameter. Considering only geometric uncer- 4f l h Dh l
tainties yields: Thus a 3% uncertainty on l and h generates a 21%
2lh uncertainty on 4f in the laminar regime.
Dh ¼ ð8Þ
lþh
4.2. Analysis
DDh Dl Dh Dðl þ hÞ Dl
¼ þ þ 3 ð9Þ
Dh l h lþh l Our measurements on tube 2 show that classical lit-
erature results on friction factors in large tubes are still
usable in mini-channels. However tube 1 measurements
demonstrate that singular pressure losses were not neg-
0.13
D h = 1.17 mm ligible as for tube 2. These are difficult to evaluate be-
0.12 Shah & london
Blasius
cause of high manifold to channel flow area ratios which
0.11 were not found in literature. Furthermore, the laminar
0.1 and turbulent regimes clearly appear and the transition
0.09 occurs for 1800 < Re < 2000. An uncertainties weighted
linear regression was performed on the friction factor
ξ ⋅ Dh / ∆z + 4f

0.08 and leaded to results gathered in Table 5. These data

show that the singular pressure loss coefficient for tube 1
0.07
can be estimated at n  20.
0.06

5. The heat transfer coefficient

0.05
5.1. Data reduction

0.04 A power balance applied on our data showed that a

500 1000 2000 3000 4000 correction taking into account heat transfer with the sur-
Re
roundings was necessary to obtain power losses below
Fig. 3. 4f þ n  Dh =Dz versus Re (Dh ¼ 1:17 mm). 10% for all data and 5% for most of them in the case
 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103 101

Table 5
Comparison of friction results with literature
Tube ð4f þ n  ðDh =DzÞÞ
Re, Re < 1800 ð4f þ n  ðDh =DzÞÞ
Re0:25 , Re > 2000 Shah and London Blasius
0:25
1 ð58  12Þ þ ð22  2Þ  Re ð0:419  0:02Þ þ ð18  1Þ  Re 57 0.316
2 57  11 0:347  0:004 57 0.316

of tube 1. Thus the wall heat flux for tube 1 measure- 0.22
ments was calculated by
0.2
U I
q_ ðzÞ ¼  aatm ðTw ðzÞ  Tatm Þ; ð12Þ
S 0.18

aatm being a global heat transfer coefficient taking into 0.16

account natural convection, radiation, and insulator

∆Tw [˚C]
thickness. It is estimated with the power balance which 0.14
leads to coherent values. For tube 2 aatm is negligible
because with a higher working pressure we achieved 0.12
heat fluxes high enough to reduce the uncertainty.
The thermocouples are fixed on the external tube 0.1
surface while the internal wall temperature is needed to
0.08
calculate the heat transfer coefficient at wall/R134a
interface. We solved the heat equation to estimate the 0.06
difference between these two temperatures, finally yield-
ing DT  0:01 K. Since the temperature difference is less 0.04
than the uncertainty on temperature measurements, the 1000 2000 5000
measured wall temperature is taken as the internal wall Re
temperature. We used the corrected heat flux of Eq. (12)
Fig. 5. Total uncertainity on Tw (Dh ¼ 0:77 mm).
to calculate the fluid temperature yielding:
Z z
q_ ðzÞ  S
Tfl ðzÞ ¼ Tfl;i þ dz: ð13Þ
0
_
M  cp ðzÞ  L 30
Dh = 1.17 mm
Shah & London
5.2. Results 20
Gnielinsky
Dittus Boelter
The global Nusselt number is calculated as follows:

aG  D h _
q
NuG ¼ ; aG ¼ ; 10
kfl DTlm 9
kfl ð0Þ þ kfl ðLÞ
Nu

8
kfl ¼ : ð14Þ 7
2
6
DTlm being defined as 5
ðTw;0  Tfl;0 Þ  ðTw;1  Tfl;1 Þ
DTlm ¼
 ð15Þ 4
T Tfl;0
ln Tw;0 w;1 Tfl:1
3
The laminar–turbulent transition occurs for 1800 <
Re < 2000 as with the friction factor but the transition is
2
smoother for tube 2. As shown on Fig. 5 the uncertainty 300 1000 4000
on Tw is multiplied by 3 in the transition region certainly Re
because of flow instabilities. Figs. 6 and 7 show the
Fig. 6. Global Nusselt number (Dh ¼ 1:17 mm).
global Nusselt number compared to literature correla-
tions for large tubes. In the turbulent regime our data
are well predicted by the Gnielinsky correlation:
4f ¼ ð1:82  logðReÞ  1:64Þ2 ; ð17Þ
ðf =2Þ  ðRe  1000Þ  Pr 6 5
Nu ¼ pffiffiffiffiffiffiffiffi ð16Þ with 2300 < Re < 10 , 0:6 < Pr < 10 . The Dittus–
1 þ 12:7 f =2ðPr2=3  1Þ Boelter correlation is also represented for comparison.
102 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103

Dh = 0.77 mm viscosity variation from the tube wall to the bulk flow is
50 Shah & London negligible: lw =lbulk < 1:05. Table 7 shows the global
Dittus Boelter uncertainties on Nu and Re.
Gnielinsky

Table 7
Nu and Re uncertainties
10 DRe=Re DNu=Nu
Tube 1 4.7% 7–28%
Nu

5 Tube 2 7.8% 8.5–11%

30
Garimella & al.
Adams & al.
Peng & al.
1 20

500 1000 5000 10000

Re
10
Fig. 7. Global Nusselt number (Dh ¼ 0:77 mm). 9
Nu 8
7
The Shah and London [10] correlation for uniform 6
heat flux heating,
5
Nu ¼ 8:235ð1  2:0421  c þ 3:0853  c2  2:4753  c3 4
þ 1:0578  c4  0:1861  c5 Þ; ð18Þ
3
is within experimental errors for tube 1 but over predicts
them for tube 2 when Re < 1000. Further investigations
with a varying heat flux for a fixed mass flux are in 2
300 1000 4000
progress to check this phenomenon. The thermal entry
Re
length may be calculated by Lth =Dh ¼ 0:0431  Re  Pr
[11] supposing it is still valid for mini-channels. Table 6 Fig. 8. Comparison with mini-channel correlations (Dh ¼ 1:17 mm).
shows that for tube 1 up to half the tube length is in the
thermal entry length and up to four for tube 2. This
could explain a slight increase of Nu with Re in the Garimella & al.
laminar regime, but not as much as observed for tube 2. 50 Adams & al.
Peng & al.
The fluid temperature increase is less than 10 °C
along the test section for tube 1 so that the Reynolds
and Prandtl numbers variation along the test section is
small, on the contrary the temperature increase can
reach 50 °C for tube 2 and the variations along the test
10
section are:
Nu

DRe DPr
2:5% < < 21% and  1:3% < < 9%: 5
Re Pr
ð19Þ
However the Prandtl number variation with fluid tem-
perature is within error bars. Furthermore the dynamic

Table 6 1
Thermal entry lengths
L=Dh Lth =Dh
Re ¼ 300 Re ¼ 1500 500 1000 5000 10000

Tube 1 610 50 246

Re
Tube 2 903 50 246
Fig. 9. Comparison with mini-channel correlations (Dh ¼ 0:77 mm).
 B. Agostini et al. / Experimental Thermal and Fluid Science 28 (2004) 97–103 103

5.3. Comparison with mini-channels correlations [3] correlation best predicts our data among mini-
channels correlations. In the laminar regime, the Shah
Figs. 8 and 9 show the correlations established by and London [10] law was able to predict our measure-
Peng and Peterson [1], Adams et al. [2] and Garimella ments in tube 1 but failed with tube 2, especially for
et al. [3] in mini-channels compared to our data. The lowest Reynolds numbers. However a longitudinal
Peng et al. correlation over predicts our data but was conductive heat flux in aluminium as low as 0.2 W can
devised for Dh < 0:75 mm and shorter channels with a explain this result so that we are currently working on
higher L=Dh . The Adams et al. correlation in the tur- this explanation. Further investigations on boiling heat
bulent regime seems not more reliable than the Gnie- transfer in such channels are in progress on both tubes.
linsky correlation. The best results are obtained with the
Garimella et al. correlation although operating condi-
tions are slightly different, especially for the aspect ratio.
References

6. Conclusions [1] X. Peng, G. Peterson, Convective heat transfer and flow friction
for water flow in microchannels structures, International Journal
of Heat and Mass Transfer 39 (12) (1996) 2599–2608.
This experimental campaign convinced us that the [2] T. Adams, S. Abdel-Khalik, S. Jeter, Z. Qureshi, An experimental
study of mini-channels should put the emphasis on investigation of single-phase forced convection in microchannels,
uncertainties. Small uncertainties on the channels International Journal of Heat and Mass Transfer 41 (6–7) (1998)
dimensions imply high errors on the friction factor so 851–857.
[3] S. Garimella, W. Dowling, M.V.D. Veen, J. Killion, The effect of
that dimensions should be known better than 1%. For
simultaneously developing flow on heat transfer in rectangular
heat transfer studies the thermal insulation was a major tubes, Heat Transfer Engineering 22 (6) (2001) 12–25.
concern and power balances better than 10% were ob- [4] J. Chen, Correlation for boiling heat transfer to saturated fluids in
tained with temperatures corrections for heat transfer convective flow, Industrial Engineering and Chemistry Process
with the surroundings for tube 1, that is why we used a Design and Development 5 (3) (1966) 322–329.
[5] M. Shah, A new correlation for heat transfer during boiling
far higher working pressure for tube 2 to achieve higher
flow through pipes, Transactions of the ASHRAE 82 (1976) 66–
heat fluxes. In the data reduction process both the heat 86.
flux and local fluid temperature are affected by uncer- [6] Z. Liu, R. Winterton, A general correlation for saturated and
tainties on the power dissipated in the test section so subcooled flow boiling in tubes and annuli, based on a nucleate
that the Nusselt number in the laminar regime is quite pool boiling equation, International Journal of Heat and Mass
Transfer 34 (11) (1991) 2759–2766.
sensitive to the power balance.
[7] R. Moffat, Contributions to the theory of single-sample uncer-
The laminar–turbulent transition occurs around tainty analysis, Journal of Fluids Engineering 104 (1982) 250–261.
Re ¼ 2000 as expected for both friction factor and heat [8] R. Moffat, Using uncertainty analysis in the planning of an
exchange coefficient measurements. Friction factors experiment, Journal of Fluids Engineering 107 (1985) 173–182.
were found to be correctly predicted with classical [9] S. Kline, F. McClintock, Describing uncertainties in single-sample
experiments, Mechanical Engineering (1953) 3–8.
relations for tube 2 but were slightly under estimated for
[10] R. Shah, A. London, Laminar Flow Forced Convection in Ducts,
tube 1 because of singular pressure losses. We showed Academic Press, 1978.
that the Gnielinsky correlation is still convenient in [11] J. Taine, J.-P. Petit, Transferts thermiques, Dunod Universit
e,
mini-channels when Re > 2000 and the Garimella et al. 1989.