International Journal of Heat and Mass Transfer 48 (2005) 941–955

www.elsevier.com/locate/ijhmt

Two-phase flow in high-heat-flux micro-channel heat sink

for refrigeration cooling applications:
Part II—heat transfer characteristics
Jaeseon Lee, Issam Mudawar *

Boiling and Two Phase Flow Laboratory, School of Mechanical Engineering, Purdue University, West Lafayette, IN 47907, United States

Received 12 July 2004; received in revised form 10 September 2004

Available online 13 November 2004

Abstract

This paper is the second of a two-part study concerning two-phase flow and heat transfer characteristics of R134a in
a micro-channel heat sink incorporated as an evaporator in a refrigeration cycle. Boiling heat transfer coefficients were
measured by controlling heat flux (q00 = 15.9  93.8 W/cm2) and vapor quality (xe = 0.26  0.87) over a broad range of
mass velocity. While prior studies point to either nucleate boiling or annular film evaporation (convective flow boiling)
as dominant heat transfer mechanisms in small channels, the present study shows heat transfer is associated with dif-
ferent mechanisms for low, medium and high qualities. Nucleate boiling occurs only at low qualities (xe < 0.05) corre-
sponding to very low heat fluxes, and high fluxes produce medium quality (0.05 < xe < 0.55) or high quality (xe > 0.55)
flows dominated by annular film evaporation. Because of the large differences in heat transfer mechanism between the
three quality regions, better predictions are possible by dividing the quality range into smaller ranges corresponding to
these flow transitions. A new heat transfer coefficient correlation is recommended which shows excellent predictions for
both R134a and water.

2004 Elsevier Ltd. All rights reserved.

Keywords: Micro-channels; Flow boiling; Refrigeration

1. Introduction their work was focused entirely on single-phase heat

sinks, recent research efforts have shifted to two-phase
The pioneering work by Tuckerman and Pease [1] in heat sinks, which offer significant advantages compared
the early 1980s spawned unprecedented interest in the to their single-phase counterparts. A key drawback of
use of micro-channel heat sinks as a means for dissipat- single-phase heat sinks is the large (and often detrimen-
ing large amount of heat from small, high-flux devices in tal) temperature gradient in the device resulting from a
a variety of computer and aerospace applications. While stream-wise rise in the coolant temperature. Two-phase
heat sinks capitalize upon latent heat exchange, which
both increases the convective heat transfer coefficient
*
Corresponding author. Tel.: +1 765 494 5705; fax: +1 765 inside the heat sink and helps maintain a more uniform
494 0539. device surface temperature, dictated mostly by the
E-mail address: [email protected] (I. Mudawar). coolantÕs saturation temperature.

0017-9310/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2004.09.019
942 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

Nomenclature

Bo boiling number Xvv Martinelli parameter for laminar liquid—

C two-phase multiplier parameter laminar vapor flow
Co convection number xe thermodynamic equilibrium quality
cp specific heat z stream-wise coordinate
dh hydraulic diameter
E correlation constant Greek symbols
f friction factor b ratio of channel depth to width
Fr Froude number g fin efficiency
G mass velocity [kg/m2 s] l viscosity
g gravitational acceleration q density
Hch micro-channel height r surface tension
Htc distance from thermocouple to base of / two-phase pressure drop multiplier
micro-channel
h local heat transfer coefficient [W/m2 K]; Subscripts
enthalpy [J/kg] 3 three-sided wall heating
k thermal conductivity [W/m K] 4 four-sided wall heating
L length b micro-channel base wall
m fin parameter c halfway location along micro-channel
Mw molecular weight cor correlation
N correlation constant; number of experimen- exp experiment
tal data f saturated liquid
Nu Nusselt number fg property difference between saturated vapor
P pressure [Pa] and saturated liquid
Pr reduced pressure fo liquid only
Pr Prandtl number g saturated vapor
q00 heat flux through heat sink base area [W/m2 in channel inlet
or W/cm2] lam laminar
Re Reynolds number based on channel hydrau- nb nucleate boiling
lic diameter out channel outlet
S correlation constant pred prediction
T temperature s solid (copper)
DTsub subcooling sat saturated
v specific volume [m3/kg] sp single-phase
Wch micro-channel width tp two-phase
Ww half-width of copper sidewall tur turbulent
We Weber number w wall condition.
X Martinelli parameter
Xvt Martinelli parameter for laminar liquid—
turbulent vapor flow

Despite these attributes, dissipating very large heat A key challenge to implementing this cooling scheme is
fluxes can lead to unacceptably high device temperatures. the relatively poor understanding of two-phase flow and
This is due to both the convective resistance of the cool- heat transfer phenomena in a micro-channel, let alone
ant itself, as well as the heat diffusion resistance associ- refrigerant-cooled heat sinks. The first part of the present
ated with the metallic structure of the heat sink, device study [2] examined the pressure drop characteristics of
and intermediate bonding agent. Avoiding high temper- R134a in a two-phase heat sink and demonstrated the
atures in high-flux temperature-sensitive devices can be practical benefits of incorporating a heat sink as an evap-
achieved by greatly reducing the temperature of the cool- orator in a refrigeration cycle. The present study explores
ant, i.e. by refrigeration. A simple means to accomplish- the heat transfer characteristics of the same heat sink.
ing this goal is to incorporate the heat sink as an The dominant heat transfer mechanism inside a small
evaporator in a standard refrigeration cycle. This is pre- channel remains an open question. Dispute over this issue
cisely the system examined in the present study. is evident in several recent studies concerning boiling heat
Table 1
Summary of prior relevant mini- and micro-channel studies
Author(s) Geometry Fluid(s) q00 [W/cm2] Flow rate xe htp [W/m2K] Remarks
2
Lazarek and Black (1982) [3] d=3.1 mm, L=12.3, 24.6 cm, R113 1.4–38.0 125–750 kg/m s <0–0.60 2000–10,000 htp = f(q00 )
semi-circular, single nucleate boiling dominant
htp independent of xe, G
Wambsganss et al. (1993) [4] d=2.92 mm, L = 36.8 cm, R113 0.88–9.08 50–300 kg/m2 s 0–0.90 1500–7000 htp=f(q00 )

J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955
circular, single nucleate boiling dominant
htp independent of xe,G
Bower and Mudawar (1994) [5] d = 2.54, 0.51 mm, L=2.86 cm, R113 3–256 45, 64 ml/min <0–CHF
circular parellel 17 micro, 3 mini
Tran et al. (1996) [6] d = 2.46 mm (circ.), R12 3.6–12.9 44–832 kg/m2 s 0–0.96 2000–10,000 htp = f(q00 )
dh = 2.40 mm (rect.), nucleate boiling dominant
L = 87.0 cm, single htp independent of xe,G
Kew and Cornwell (1997) [7] d = 1.39–3.69 mm, L = 50.0 cm R141b 0.97–9.0 188–1480 kg/m2 s <0–0.95 1000–7000 htp = f(q00 , xe),
circular, single htp increases with xe
nucleate and convective
boiling important
Ravigururajan (1998) [8] dh = 0.425 mm, L = 2.052 cm, R124 8.0–65.0 75–250 ml/min 0–0.5 2000–25,000 htp = f(q00 ,xe),
rectangular, 54 parallel htp increases with xe
Yan and Lin (1998) [9] d = 2.0 mm, L = 10.0 cm, circular, R134a 0.5–2.0 50–200 kg/m2 s 0.1–0.90 2000–6000 htp = f(q00 , xe, Tsat)
28 parallel
Bao et al. (2000) [10] d = 1.95 mm, L = 27.0 cm, R11, R123 0.5–20.0 50–1800 kg/m2 s <0–0.90 3000–17,000 htp / q00 0.735
circular, single nucleate boiling dominant
htp independent of xe,G
Mehendale and Jacobi dh = 0.8 mm, L = 7.4 cm, R134a 0–0.34 0–0.5 g/s 0.07–0.25 0–9000 htp = f(q00 )
(2000) [11] rectangular, 52 parallel nucleate boiling dominant
htp independent of xe,G
Lee and Lee (2001) [12] dh = 0.784, 6.67 mm, L = 30.0 cm, R113 0.3–1.58 50–200 kg/m2 s 0.15–0.75 1000–5000 htp = f(xe)
rectangular, single convective boiling dominant
htp increases with xe,
q00 effect minor
Lin et al. (2001) [13] d = 1.1 mm, L = 50.0 cm, R141b 0.18–7.2 510 kg/m2 s <0–1.0 1000–6000 htp = f(q00 , xe)
circular, single nucleate and convective boiling
important
Yu et al. (2002) [14] d = 2.98 mm, L = 91.0 cm, Water– 10.0–30.0 50–200 kg/m2 s 0–1.0 8000–42,000 htp = f(q00 )
circular, single ethylene nucleate boiling dominant
glycol htp independent of xe,G
Warrier et al. (2002) [15] dh = 0.75 mm, L = 30.74 cm, FC84 0–5.99 557–1600 kg/m2 s 0.03–0.5 htp = f(q00 , xe)
rectangular, 5 parallel
Wen et al. (2004) [16] dh = 1.33 mm, L = 28.0, 24.8 cm, Water 2.6–16.0 57–211 kg/m2 s <0–0.3 14,000– 29,000 htp / q00 0.44
rectangular, single nucleate and convective boiling
important
Huo et al. (2004) [17] d = 2.01, 4.26 mm, circular, single R134a 1.3–15.0 100–500 kg/m2 s 0–0.9 8000–42,000 htp = f(q00 , xe)

943
nucleate and convective
boiling important
944 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

Table 2
Contrasting features of present study compared with prior small channel boiling
Present study Most prior studies
Application channel geometry Compact heat sink (cold plate) Compact heat exchanger
231 lm · 713 lm L = 2.53 cm dh mostly > 1 mm L  10 cm
Heat flux High Low
Input heat flux Uniform across flat surface By external convection
Ratio of acceleration to 20–45% Small to negligible
total pressure drop
Working fluid Environmentally-friendly refrigerant Ozone-depleting refrigerants

transfer in mini- and micro-channels. Table 1 shows As indicated in part 1 of this study [2], the micro-chan-
researchers are divided into two groups. The first shares nels were formed by cutting 53 of 231 lm wide and
the view that nucleate boiling is dominant and therefore 713 lm deep micro-slots into the 25.3 · 25.3 mm2 top
dictates overall heat transfer inside the channel, citing surface of an oxygen-free copper block. A transparent
the local heat transfer coefficient is dependent on heat cover plate formed a top insulating surface for the
flux, but not mass velocity or quality [3,4,6,10,11,14]. micro-channels and provided optical access to the two-
The second group shares the observation that the local phase flow inside the micro-channels. A fiberglass plastic
heat transfer coefficient is a function of quality and mass housing surrounded the copper block and provided inlet
velocity in addition to wall heat flux [7–9,12,13,15–17]. and outlet plenums that ensured uniform flow distribu-
Experimental results by this second group generally show tion between micro-channels as well as convenient loca-
the local heat transfer coefficient decreases with increas- tions for inlet and outlet pressure and temperature
ing quality. This points to annual film evaporation (some- measurement. Three cartridge heaters were embedded
times called convective boiling) as the dominant heat in the copper block and a type-K thermocouple inserted
transfer mechanism. The discrepancy between the two halfway along the micro-channels was used to infer the
groups may be explained by the different coolants, oper- micro-channel base temperature at the same stream-wise
ating conditions and channel sizes used by different inves- location. Additional details concerning the construction
tigators yielding different dominant mechanisms. This of the micro-channel test module and measurement
important issue will be discussed later in this paper. uncertainty are available in part 1 of this study [2].
As indicated earlier, the primary purpose of the pre-
sent paper is to explore the two-phase heat transfer char- 2.2. Operating conditions
acteristics of a refrigerant-cooled micro-channel heat
sink. Table 2 points out the unique focus of this study The refrigeration loop shown in Fig. 1 supplied a two-
compared to prior small channel refrigerant studies. phase R134a mixture at the inlet to the micro-channel
While much of the existing refrigeration literature con- heat sink. This mixture was formed by throttling the flow
cerns long (L > 10 cm) mini-channels (d > 2 mm), the by a manual metering valve situated upstream of the evap-
present study concerns a short heat sink (L = orator. Since the throttling process is isenthalpic, the
25.27 mm) utilizing parallel micro-channels (231 lm · evaporatorÕs inlet enthalpy was set equal to the enthalpy
713 lm). Furthermore, the heat sink examined in this determined from the temperature and pressure measured
study serves as a high-heat-flux evaporator (cold plate at the inlet to the throttling valve. The evaporator inlet
for electronic or power devices) subjected to a fairly uni- vapor quality was determined from this enthalpy value
form heat flux, while the vast majority of prior small- and the measured evaporator inlet pressure. Mass flow
channel refrigeration research concerns low heat flux rate was determined from the volumetric flow rate meas-
evaporators that are heated by external convection. It ured by the rotameter and density of subcooled liquid
is a primary goal of the present study to develop a fun- based on the temperature and pressure measured just
damental understanding of dominant mechanisms, and downstream of the rotameter. The evaporator outlet
recommend an accurate method for predicting the heat quality was determined from the following energy
transfer performance of this unique cooling scheme. balance:
4q00 L
xe;out  xe;in ¼ : ð1Þ
Gd h hfg
2. Experimental methods Experimental operating conditions spanned the fol-
lowing ranges: inlet quality of xe,in = 0.001  0.25, outlet
2.1. Apparatus and measurement techniques quality of xe,out = 0.49  superheat, mass velocity of G =
127  654 kg/m2 s, heat flux of q00 = 159  938 kW/m2
Fig. 1 shows the micro-channel heat sink incorpo- (15.9–93.8 W/cm2), and inlet pressure of Pin = 1.44 
rated as an evaporator in an R134a refrigeration loop. 6.60 bar.
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 945

Fig. 1. Schematic of refrigeration loop.

3. Determination of heat transfer coefficient

Fig. 2 shows a unit control volume consisting of a

single micro-channel and half of the surrounding copper
walls. A simplified fin model of the copper walls is used
to construct an energy balance for this control volume.
Equating the heat influx to the unit cell through the bot-
tom solid wall to the heat efflux by flow boiling along the
micro-channel bottom and side walls (since the top wall
is adiabatic) yields [18]
q00 ðW ch þ 2W w Þ ¼ htp ðT w;b  T f;sat ÞðW ch þ 2gH ch Þ; ð2Þ
where g is the fin efficiency. Since the top wall is adia-
batic, the fin efficiency is given by [19]
tanhðmH ch Þ
g¼ ; ð3Þ
mH ch
where m is the fin parameter defined as
sffiffiffiffiffiffiffiffiffiffiffi
htp
m¼ : ð4Þ
ks W w
The fin base temperature, Tw,b, was calculated using
the assumption of one-dimensional heat diffusion be-
tween the plane of the thermocouple embedded in the
copper block and the plane containing the fin base.
q00 H tc
T w;b ¼ T tc  : ð5Þ
ks
The fluid temperature, Tf,sat, in Eq. (2) is the bulk
temperature halfway along the micro-channel based on
pressure at the same location determined from the Fig. 2. Micro-channel unit cell.
946 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

measured inlet pressure and the pressure drop correla- steep for large heat fluxes but approaches zero at the
tion scheme presented in part 1 of this study [2]. Calcu- lowest heat flux of q00 = 15.9 W/cm2. This trend points
lations revealed that, excepting superheated evaporator to htp values becoming constant (independent of xe,c)
outlet conditions, Tf,sat, was close to the mean value of had even lower heat flux values been tested. As indicated
the measured inlet and outlet temperatures. earlier, the compressor precluded operating at such low
Eqs. (2) and (3) are therefore used to determine htp heat fluxes.
since all other parameters are easily measured or
calculated. 4.2. Flow visualization results

Flow visualization was used to better understand the

4. Results and discussion aforementioned heat transfer coefficient trends. As indi-
cated earlier, the transparent cover plate provided direct
4.1. Two-phase heat transfer coefficient optical access to the boiling phenomena occurring with-
in the micro-channel. The clarity of video and still
Fig. 3 shows the variation of the measured heat images was limited by the small micro-channel size,
transfer coefficient htp at z = L/2 with quality at the same lighting requirements, and internal reflection within the
stream-wise location (obtained by substituting L in Eq. inner walls of the micro-channels. Fig. 4 shows an image
(1) by L/2) for different values of heat flux. This quality of the flow in two adjacent micro-channels along with a
was varied experimentally by reducing mass flow rate schematic rendering of the same image at each of three
(through increased throttling upstream of the evapora- values of quality halfway along the channel. Notice the
tor) while maintaining a constant heat flux. To maintain slight differences in boiling pattern between the adjacent
reliable compressor operation, only values in excess of micro-channels. As explained in part 1 of this study [2],
xe,c = 0.25 could be tested. Fig. 3 shows htp decreases these differences are the result of a mild parallel channel
with increasing vapor quality, a basic feature of annular instability causing slight flow pattern variations among
evaporative boiling, not nucleate boiling as was sug- the micro-channels. This form of instability was more
gested in some of the earlier studies. Another notewor- pronounced at lower quality values.
thy feature of the same data is the large magnitude of Fig. 4(a) shows a combination of slug flow and to a
htp (2000–50,000 W/m2 K), which is far superior to values lesser extent bubbly flow for xec = 0.39. Small bubbles
typically attained with mini- and macro-channels. exhibited violent mixing in the liquid slugs between the
Fig. 3 also shows the influence of heat flux on htp is elongated bubbles. There are two possible sources for
very strong up to quality values around xe,c = 0.55, be- these small bubbles. The first is entrance inclusion since
yond which there is a great diminution in the heat flux the flow enters the evaporator as a two-phase mixture.
effect. For xe,c < 0.55, the slope of htp versus xe,c is very The second is nucleation within the micro-channels,

Fig. 3. Variation of local heat transfer coefficient for R134a with thermodynamic equilibrium quality for different heat fluxes.
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 947

Fig. 4. Two-phase flow pattern and schematic renderings of R134a flow at (a) xe,c = 0.39, q00 = 31.64 W/cm2, G = 235.2 kg/m2 s; (b)
xe,c = 0.53, q00 = 31.6 W/cm2, G = 163.3 kg/m2 s, and (c) xe,c = 0.68, q00 = 31.6 W/cm2, G = 128.2 kg/m2 s.

which can be easily detected for bubbles nucleating from Notice the two elongated bubbles in the lower micro-
the sidewalls. Previous studies with a water-cooled channel separated by a very thin liquid ridge, tending
micro-channel heat sink showed bubbly flow could to coalesce into a single longer bubble or transition to
rarely be sustained [18]. In fact, annular flow was the annular flow.
dominant regime for most medium- and high-flux condi- Fig. 4(c) shows fully developed annular flow with a
tions. This drastic difference between two-phase flow very thin liquid film covering the micro-channel walls
patterns for R134a and water is explained by the low at xe,c = 0.68. The film appears too thin in certain loca-
surface tension of R134a producing far smaller bubbles tions to maintain adequate cooling. This is evident in the
than in water [20]. Very small bubbles are far more apt lower of the two micro-channels were dryout might be
to travel discretely along a micro-channel before coalesc- occurring locally, and where droplet deposition from
ing into long bubbles, thus enabling the bubbly and slug the vapor core appears very sparse. These observations
flow regimes to be more prevalent in refrigerant micro- point to xe 0.55 as a lower boundary for local dryout,
channel flow. and perhaps critical heat flux (CHF) for refrigerant
Fig. 4(b) shows the flow pattern at a quality value micro-channel flow.
close to the transitional value discussed earlier in con- Interestingly, Tran et al. [6] detected transition from
junction with the variation of htp with xe,c, which was slug to annular flow around xe 0.6–0.7, compared to
the upper quality boundary for a strong heat flux effect. xe 0.25–0.35 for macro-channels. The present study
This slug flow pattern consists mainly of elongated bub- points to transition values smaller than those of Tran
bles with very few small bubbles within the liquid slugs. et al., but greater than for macro-channels. Clearly,
948 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

Table 3
Previous correlations for boiling in small channels
No. Author(s) Correlation Remarks
 
1 Chen (1966) [21] Nu3 Macro-channel
htp ¼ ðEhsp þ Shnb Þ
Nu4 correlation
kf
hsp ¼ Nu ; Nulam ¼ constðRefer to Eq: (7) and (8)Þ;
dh
Nutur ¼ 0:023Re0:8 0:4
f Pr f

!
k 0:79
f c0:45 0:24
p;f vg
hnb ¼ 0:00122 DT 0:24 0:75
sat DP sat
r0:5 l0:29
f h0:24
fg vf
0:49

 1:78  
1 Ref E1:25
E¼ 1 þ 0:5 ; S ¼ 0:9622  0:5822tan1
X 6:18 104

h i0:5 Gð1  xe Þd h lf cpf

X ¼ ðdp=dzÞf =ðdp=dzÞg ; Ref ¼ ; Prf ¼
lf kf
 
Shah (1982) [22,23] Nu3
2 htp ¼ MaxðE; SÞhsp Macro-channel
Nu4 correlation
kf
hsp ¼ Nu ; Nulam ¼ constðRefer to Eq: (7) and (8)Þ;
dh
Nutur ¼ 0:023Re0:8 0:4
f Pr f

For 1:0 < N ; S ¼ 1:8=N 0:8 ; E ¼ 230Bo0:5 for Bo > 3 105

or E ¼ 1 þ 46Bo0:5 for Bo < 3 105

For 0:1 < N 6 1:0; S ¼ 1:8=N 0:8 ; E ¼ FBo0:5 expð2:74N 0:1 Þ

For N 6 0:1; S ¼ 1:8=N 0:8 ; E ¼ FBo0:5 expð2:47N 0:15 Þ;

F ¼ 14:7 for Bo P 11 104 or F ¼ 15:43 for Bo < 11 104

N ¼ Co for Frf P 0:04 or N ¼ 0:38Frf0:3 Co for Frf < 0:04;

 0:8  0:5
1  xe vf v2 G2
Co ¼ ; Frf ¼ f
xe vg gd h
 
Nu3
3 Liu and Winterton (1991) [24] htp ¼ ððEhsp Þ2 þ ðShnb Þ2 Þ0:5 Macro-channel
Nu4 correlation

kf
hsp ¼ Nu ; Nulam ¼ constðRefer to Eq: (7) and (8)Þ;
dh
Nutur ¼ 0:023Re0:8 0:4
f Pr f

  0:35
vg
E¼ 1 þ xe Prf 1
vf
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 949

Table 3 (continued)
No. Author(s) Correlation Remarks

0:55
hnb ¼ 55P 0:12
r ðlog10 ðP r ÞÞ M 0:5
w q
000:67
;
1
S ¼ ð1 þ 0:055E0:1 Re0:16
fo Þ

If Frf 6 0:05 replace E by EFr0:12Fr

f
f
and S by SFr0:5
f

kf
4 Lazarek and Black (1982) [3] htp ¼ Nu3 ½30Re0:857
fo Bo
0:714
dh Small channel
 correlation
vg 0:4 2
5 Tran et al. (1996) [6] htp ¼ Nu3
Nu4 8:4 105 ðBo2 Wefo Þ0:3 vf ; Wefo ¼ vf Gr d h Small channel
correlation
  Small channel
Nu3
htp ¼ ðEhsp Þ correlation
6 Lee and Lee (2001) [12] Nu4
kf
hsp ¼ Nu ; Nulam ¼ constðRefer to Eqs: (7) and (8)Þ;
dh
Nutur ¼ 0:023Re0:8 0:4
f Prf

C 1
E ¼ 10:3b0:398 /0:598
f ; /2f ¼ 1 þ þ ; C ¼ 6:185 102 Re0:726
fo
X X2

h i
X 2 ¼ ðdp=dzÞf =ðdp=dzÞg


vg 0:2 Small channel
7 Yu et al. (2002) [14] htp ¼ Nu3
Nu4 6:4 106 ðBo2 Wefo Þ0:27 vf
correlation
 
Nu3
8 Warrier et al. (2002) [15] htp ¼ ðEhsp Þ
Nu4

kf
hsp ¼ Nu ; Nulam ¼ constðRefer to Eqs: (7) and (8)Þ;
dh
Nutur ¼ 0:023Re0:8 0:4
f Prf

E ¼ 1:0 þ 6Bo1=16 þ f ðBoÞx0:65

e ; f ðBoÞ ¼ 5:3ð1  855BoÞ

vapor quality is not the sole indicator for regime transi- rectangular geometry of the present study using the fol-
tion, and other parameters such surface tension and lowing relation:
channel shape and dimensions must influence transition
behavior. Nu3
htp ¼ htp;cor ; ð6Þ
Nu4

5. Assessment of prior correlations where htp,cor is the value predicted from a correlation,
and Nu3 and Nu4 are the single-phase Nusselt numbers
Several popular macro-channel correlations and re- for laminar flow with three-sides and four-sides wall
cently recommended small-channel correlations were heating, respectively [25].
examined for accuracy in predicting the present heat
Nu3 ¼ 8:235ð1  1:883b þ 3:767b2  5:814b3
transfer data. These correlations are summarized in
Table 3. Since these correlations concern circular chan- þ 5:361b4  2:0b5 Þ ð7Þ
nels with uniform circumferential heating, predictions
had to be adjusted for the three-sided wall heating and and
950 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

Fig. 5. Comparison of R134 a heat transfer coefficient data with predictions based on (a) Chen [21], (b) Shah [22,23], (c) Lazarek and
Black [3], (d) Liu and Winterton [24], (e) Tran et al. [6], (f) Lee and Lee [12], (g) Yu et al. [14], and (h) Warrier et al. [15].

Nu4 ¼ 8:235ð1  2:042b þ 3:085b2  2:477b3 The first three correlations in Table 3 are popular for
4 5
determining the two-phase heat transfer coefficient for
þ 1:058b  0:186b Þ: ð8Þ macro-channels. These correlations include a single-
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 951

phase term, hsp, based on turbulent liquid flow prevalent transfer coefficient in refrigerant-cooled micro-channel
in large channels. The Dittus–Boelter relation used for heat sinks.
hsp is a key drawback to using these three correlations As indicated earlier, the compressor of the refrigera-
for micro-channel flow, where the liquid is almost invar- tion system used in the present study precluded testing at
iably laminar. The laminar liquid flow is the reason be- low qualities. In order to both extend the quality range
hind using Eqs. (7) and (8) when evaluating hsp in the of the new correlation and demonstrate its validity for
three macro-channel correlations. drastically different fluids, recent micro-channel water
The five remaining correlations in Table 3 were all data by Qu and Mudawar [18] were examined along
developed for smaller channels. Notice that correlations with the new R134a data. Fig. 6 shows the variation
4, 5 and 7 assume heat transfer is dominated by nucleate of the two-phase heat transfer coefficient data with qual-
boiling, while correlations 6 and 8 are based on a form ity for the water data. Interestingly, the water data were
similar to that of macro-channel correlations. obtained by modulating heat flux for a constant mass
Fig. 5 compares the measured heat transfer coeffi- velocity, while the present R134a data were generated
cient to predictions based on each of the correlations by regulating flow rate for a constant heat flux.
from Table 3. The predictive accuracy of a correlation A new technique was sought that could correlate
was measured by the mean absolute error, defined as both the water and R134a data based on the Martinelli
 parameter, X, and account for micro-channel effects not
1 X j htp;pred  htp;exp j
MAE ¼ 100 : ð9Þ represented in prior correlations. Exhaustive inspection
N htp;exp
of both the water and R134a databases showed two
Fig. 5 shows all eight correlations yield poor predic- additional dimensionless parameters were required to
tions of the present data. The macro-channel correla- adequately correlate all the data. They are the boiling
tions generally overpredict the data in the high quality number, Bo, and the liquid Weber number Wefo. Figs.
region (region of low htp values), and underpredict in 7 and 8 show variations of X2, Bo and Wefo with quality
the low quality region. The three small channel correla- and heat flux for R134a and water, respectively, the lat-
tions based on nucleate boiling (correlations 4, 5 and 7) ter being more concentrated in the low xe,c region. Over-
and two remaining small channel correlations (6 and 8) all, these three parameters show different trends for
all show poor predictions, evidenced by their large MAE different quality ranges. Therefore, better predictions
values. are possible by dividing the overall quality range into
smaller ranges corresponding as closely as possible to
the aforementioned flow transitions and data trends.
6. New correlation Three quality ranges provided adequate coverage of
both the water and R134a databases. The first is a low
The above assessment of prior correlations points a quality range of xe = 0  0.05. This low heat flux range
need for a new tool for accurate prediction of the heat is most closely associated with bubble nucleation and in-

Fig. 6. Variation of local heat transfer coefficient for water (Qu and Mudawar [18]) with thermodynamic equilibrium quality for
different mass velocities.
952 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

Fig. 8. Variations of (a) Martinelli parameter, (b) boiling

Fig. 7. Variations of (a) Martinelli parameter, (b) boiling number, and (c) Weber number with quality for water (Qu and
number, and (c) Weber number with quality for R134a. Mudawar [18]).

cludes 50 water data points and none of the R134a data. corresponds mostly to bubbly/slug flow, where the heat
These data show rapid and irregular variation of the transfer data are no longer uniquely determined by the
heat transfer coefficient with quality as shown in Fig. Martinelli parameter, but are sensitive to Bo and Wefo
6. Fig. 8(a) shows the Martinelli parameter in this range as well. The R134a and water data exhibit different
is uniquely determined by quality irrespective of flow trends with quality in this range. While the heat transfer
rate or subcooling, and shows the sharpest decrease with coefficient decreases with increasing quality for both flu-
increasing quality. ids, this decrease is far more pronounced for R134a, Fig.
The second quality range is xe = 0.05  0.55, which 3, than for water, Fig. 6. Those differences may be par-
includes 83 R134a and 157 water data points. This range tially explained by the different values and trends of Bo
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 953

Table 4
Present correlation scheme
xe Correlation Data MAE (%)
0.267
0–0.05 htp = 3.856X hsp,f 50 water data points 11.6
ðdp=dzÞf Nu3 k f
X2 ¼ ; hsp;f ¼
ðdp=dzÞg dh
!0:5  0:5  0:5
lf 1  xe mf
X mm ¼
lg xe mg
!0:5    
ff Re0:25
g 1  xe 0:5 mf 0:5
X mt ¼
0:079 xe mg
Gxe d h
Reg ¼
lg
0.05–0.55 htp ¼ 436:48Bo0:522 Wefo
0:351 0:665
X hsp;f 83 R134a data points 11.9
00 2
q mf G d h
Bo ¼ ; Wefo ¼ 157 water data points
Ghfg r
0.55–1.0 htp=max{(108.6X1.665hsp,g),hsp,g} 28 R134a data points 16.1
Nu3 k g
hsp;g ¼ for laminar gas flow
dh
hsp;g ¼ 0:023Re0:8 0:4
g Prg for turbulent gas flow

and Wefo for the two fluids. Comparing Figs. 7(b) and mid range includes the effects of Bo and Wefo as well.
Fig. 8(b) shows drastic differences in the magnitude of Overall, convection to liquid is important for both
Bo, brought about mostly by a much smaller latent heat the low and mid quality regions, while convection to
of vaporization for R134a compared to water. Fig. 7(c) vapor becomes important for the high quality region.
shows far greater Wefo values for R134a compared to For the latter, the low viscosity of R134a vapor yields
water, Fig. 8(c), because of the much smaller surface ten- vapor Reynolds numbers corresponding to turbulent
sion of R134a. Furthermore, the large pressure changes flow at high-heat-flux conditions despite the small
in the R134a experiments produced appreciable prop- hydraulic diameter of the micro-channel. Thus, the sin-
erty changes. Incorporating Bo and Wefo in the new cor- gle-phase vapor term in the high quality correla-
relation is anticipated to account for the drastic property
variations that are far more accentuated in a micro-
channel than in a macro-channel.
The last quality range is the liquid deficient region of
xe = 0.55  1.0, consisting of predominantly annular
flow and local dryout. This includes 28 R134a data
points but none of the water data. Fig. 7(a) shows a con-
vergence of Martinelli number values for different heat
fluxes. In this region, the thin annular film is gradually
evaporated, sustained only by droplets entrained in the
vapor core. Once the majority of droplets are depleted,
evaporation results in monotonic thinning and eventual
dryout of the liquid film. With this dryout, the heat
transfer coefficient is greatly reduced as heat begins to
be transferred directly to the vapor. Interestingly, the
Martinelli parameter approaches zero as quality goes
to unity. Since dryout is followed by convection to
vapor, a minimum value for the heat transfer coefficient
correlation was set equal to that for convective heat
transfer to pure vapor.
Table 4 summarizes the new correlations for the
three quality regions. The low and high quality regions Fig. 9. Comparison of heat transfer coefficient data for R134
are based solely on the Martinelli parameter while the and water with predictions based on new correlation scheme.
954 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955

tion must allow for both laminar or turbulent vapor References

flow.
Table 4 shows the effect of the Martinelli parameter is [1] D.B. Tuckerman, R.F.W. Pease, High-performance heat
important for each of the three quality ranges. The pre- sinking for VLSI, IEEE Electron Dev. Lett. 2 (1981) 126–
sent correlations show the heat transfer coefficient is 129.
[2] J. Lee, I. Mudawar, Two-phase flow in high-heat-flux
proportional to the Martinelli parameter raised to a pos-
micro-channel heat sink for refrigeration cooling applica-
itive exponent, whereas prior macro-channel correla-
tions: Part I—Pressure drop characteristics, Int. J. Heat
tions employ a negative exponent for the same Mass Transfer (in press).
parameter. [3] G.M. Lazarek, S.H. Black, Evaporative heat transfer,
Fig. 9 shows excellent agreement between predictions pressure drop and critical heat flux in a small vertical tube
based on the new correlation scheme and both the with R-113, Int. J. Heat Mass Transfer 25 (1982) 945–959.
R134a and water data. An overall MAE of 12.26% indi- [4] M.W. Wambsganss, D.M. France, J.A. Jendrzejczyk,
cates excellent predictive capability, especially with most T.N. Tran, Boiling heat transfer in a horizontal small-
of the data falling within a ±30% error range and cap- diameter tube, J. Heat Transfer 115 (1993) 963–972.
turing the correct data trend. [5] M.B. Bowers, I. Mudawar, High flux boiling in low flow
rate, low pressure drop mini-channel and micro-channel
heat sinks, Int. J. Heat Mass Transfer 37 (1994) 321–332.
7. Conclusions [6] T.N. Tran, M.W. Wambsganss, D.M. France, Small
circular- and rectangular-channel boiling with two refrig-
Experiments were performed to explore the heat erants, Int. J. Multiphase Flow 22 (1996) 485–498.
transfer characteristics of a micro-channel heat sink [7] P.A. Kew, K. Cornwell, Correlations for the prediction of
boiling heat transfer in small-diameter channels, App.
incorporated as an evaporator in a refrigeration cycle.
Therm. Eng. 17 (1997) 705–715.
Flow visualization studies at different quality values [8] T.S. Ravigururajan, Impact of channel geometry on two-
were used along with heat transfer coefficient data to de- phase flow heat transfer characteristics of refrigerants in
velop a new correlation scheme for micro-channel flow microchannel heat exchangers, J. Heat Transfer 120 (1998)
that was validated for both R134a and water. Key find- 485–491.
ings from the study are as follows: [9] Y.-Y. Yan, T.-F. Lin, Evaporation heat transfer and
pressure drop of refrigerant R-134a in a small pipe, Int. J.
(1) Two-phase heat transfer in micro-channel heat sinks Heat Mass Transfer 41 (1998) 4183–4194.
is associated with different mechanisms for low, [10] Z.Y. Bao, D.F. Flectcher, B.S. Haynes, Flow boiling heat
transfer of Freon R11 and HCFC123 in narrow passages,
medium and high quality flows. Bubbly flow and
Int. J. Heat Mass Transfer 43 (2000) 3347–3358.
nucleate boiling occur only at low qualities
[11] S.S. Mehendale, A.M. Jacobi, Evaporative heat transfer in
(xe < 0.05) corresponding to very low heat fluxes. mesoscale heat exchangers, ASHRAE Transactions: Sym-
High fluxes produce medium quality (0.05 < posia, Dallas, TX, 2000, pp. 445–452.
xe < 0.55) or high quality (0.55 < xe < 1.0) flows [12] H.J. Lee, S.Y. Lee, Heat transfer correlation for boiling
(depending on flow rate), where the heat transfer is flows in small rectangular horizontal channels with low
dominated by annular film evaporation. Because aspect ratios, Int. J. Multiphase Flow 27 (2001) 2043–2062.
of the large differences in heat transfer mechanism [13] S. Lin, P.A. Kew, K. Cornwell, Two-phase heat transfer to
between the three quality regions, better predictions a refrigerant in a 1 mm diameter tube, Int. J. Refrige. 24
are possible by dividing the quality range into smal- (2001) 51–56.
[14] W. Yu, D.M. France, M.W. Wambsganss, J.R. Hull,
ler ranges corresponding to these flow transitions.
Two-phase pressure drop, boiling heat transfer, and critical
(2) While the Martinelli parameter commonly used in
heat flux to water in a small-diameter horizontal tube, Int.
macro-channel correlations is important to each J. Multiphase Flow 28 (2002) 927–941.
of the three quality ranges, accurate correlation of [15] G.R. Warrier, V.K. Dhir, L.A. Momoda, Heat transfer
data for different coolants requires incorporating and pressure drop in narrow rectangular channels, Exp.
the effects of boiling number and Weber number Therm. Fluid Sci. 26 (2002) 53–64.
for the medium quality range. [16] D.S. Wen, Y. Yan, D.B.R. Kenning, Saturated flow
(3) A new three-range two-phase heat transfer coeffi- boiling of water in a narrow channel: time-averaged heat
cient correlation is recommended which shows excel- transfer coefficient and correlations, App. Therm. Eng. 24
lent predictive capability for both R134a and water. (2004) 1207–1223.
[17] X. Huo, L. Chen, Y.S. Tian, T.G. Karayiannis, Flow
boiling and flow regimes in small diameter tubes, App.
Therm. Eng. 24 (2004) 1225–1239.
Acknowledgment [18] W. Qu, I. Mudawar, Flow boiling heat transfer in two-
phase micro-channel heat sinks—I. Experimental investi-
The authors are grateful for the support of the Office gation and assessment of correlation methods, Int. J. Heat
of Naval Research (ONR) for this research. Mass Transfer 46 (2003) 2755–2771.
 J. Lee, I. Mudawar / International Journal of Heat and Mass Transfer 48 (2005) 941–955 955

[19] F.P. Incropera, D.P. Dewitt, Fundamentals of Heat and [23] M.M. Shah, Chart correlation for saturated boiling heat
Mass Transfer, fifth ed., Wiley, New York, 2002. transfer: equation and further study, ASHRAE Trans. 88
[20] S. Mukherjee, I. Mudawar, Pumpless loop for narrow (1982) 185–196.
channel and micro-channel boiling, ASME J. Electron. [24] Z. Liu, R.H.S. Winterton, A general correlation for
Packaging 125 (2003) 431–441. saturated and subcooled flow boiling in tubes and annuli,
[21] J.C. Chen, Correlation for boiling heat transfer to based on a nucleate pool boiling equation, Int. J. Heat
saturated fluids in convective flow, I & EC Process Des. Mass Transfer 34 (1991) 2759–2766.
Develop. 5 (1966) 322–329. [25] R.K. Shah, A.L. London, Laminar flow forced convection
[22] M.M. Shah, A new correlation for heat transfer during in ducts: a source book for compact heat exchanger
boiling flow through pipes, ASHRAE Trans. 82 (1976) 66– analytical data, Advances in Heat Transfer, Academic
86. Press, New York, 1978, Supplement 1.