You are on page 1of 12

International Journal of Heat and Mass Transfer 74 (2014) 448459

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage:

Experimental analysis of heat transfer of supercritical uids in plate heat

Pourya Forooghi , Kamel Hooman
Queensland Geothermal Energy Centre of Excellence, School of Mechanical and Mining Engineering, The University of Queensland, QLD 4072, Australia

a r t i c l e

i n f o

Article history:
Received 23 January 2014
Received in revised form 7 March 2014
Accepted 18 March 2014

Plate heat exchanger
Convection heat transfer

a b s t r a c t
Heat transfer of a supercritical refrigerant with highly variable properties close to pseudo-critical temperature was experimentally investigated in plate heat exchangers. Two different plate corrugation angles
(30 and 60) were examined while the Reynolds and the Prandtl number range from 800 to 4200 and
3.2 to 4.2, respectively. The results are found to be different from those obtained using classical DittusBoelter type correlations. Two possible effects were investigated: effect of wall-to-bulk property
ratio and that of buoyancy. The former was found to be important and was accounted for in the correlation using the correction factor proposed by Jackson and Hall. The latter was found not to be signicant
for corrugation angle of 60. For corrugation angle of 30, however, buoyancy effects were found to have
some inuence, yet majority of the data points are found to be within 15% of those predicted using the
2014 Elsevier Ltd. All rights reserved.

1. Introduction
On account of their compactness and high heat transfer coefcients, plate-type heat exchangers (PTHEs) have been increasingly
used in various industries in the past decades [13]. With improvements in manufacturing techniques and invention of novel designs,
high pressure and temperature uids can be pumped through
PTHEs [3,4]. Queensland Geothermal Energy Centre of Excellence
(QGECE) has been considering PTHE as a favorite candidate for
being used in the development of binary geothermal power cycles.
An area of major focus in QGECE is the study of power cycles with
supercritical working uids to bring about higher energy conversion efciencies for geothermal energy resources.
The term supercritical uid is used in this paper to address a
uid with a pressure higher than its critical pressure. At any supercritical pressure, there is never two distinguishable liquid and vapor phases in equilibrium. What happens instead is a gradual
transition from high-density liquid-like uid to low-density gaslike uid with an increase in the temperature. With temperature
close to pseudo-critical temperature (Tpc), the rate of this decrease
in density intensies leading to very high thermal expansion coefcients untypical to most single phase uids. Moreover, at a supercritical pressure, specic heat is considerably higher in the vicinity
of Tpc. Pseudo-critical temperature itself depends on the pressure,
Corresponding author. Tel.: +49 721 60845880.
E-mail address: (P. Forooghi).
0017-9310/ 2014 Elsevier Ltd. All rights reserved.

and approaches critical temperature as the pressure tends to critical

pressure [5].
Heat transfer of a supercritical uid with rapid changes in density and specic heat, like what described above, can be different
from that of normal uids. It has been known since long time ago
that heat transfer of supercritical uids in straight tubes does not
follow the prediction of conventional heat transfer correlations
when uids temperature approaches the pseudo-critical temperature [69]. It is partly due to the fact that temperaturedependent thermophysical properties may be considerably
different near the wall compared to those at bulk temperature.
Obviously, new correlations were called for containing corrections
for the unusually high wall-to-bulk ratios of density and specic
heat. A number of such correlations have been suggested in the
literature for turbulent heat transfer in circular pipes [6,812].
Although most of the correlations were derived based on experiments on a specic uid most commonly CO2 or water the
correction factor expressions are very similar to each other. Jackson, therefore, proposed a semi-empirical correlation, according
to which the Nusselt number of a variable-property uid ow is
equal to that of the same uid ow with properties evaluated
at the bulk temperature using a constant property correlation
corrected by two correction factors representing variations of
density and specic heat [9,13]:


C p;b




P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459


Latin symbols
area [m2]
correlation constant []
effective wall-temperature correction factor []; Eq.
specic heat [kJ/kg K]
hydraulic diameter [m]
mass velocity [kg/m2 s]
gravity acceleration [m/s2]
q q q gd3

m, n

Grashof number; b b l w h
enthalpy [kJ/kg]
convective heat transfer coefcient [kW/m2 K]
data point index []
plate length [m]
correlation exponents []
mass ow rate [kg/s]
total number of data points []
Nusselt number; HTCd


Prandtl number; lkC P

heat transfer power [kW]


Reynolds number; Gd


Richardson number; Gr
standard deviation []
temperature [C]
students statistical factor [];
_  CP
thermal capacity [kJ/s K]; m
overall heat transfer coefcient [kW/m2 K]
coordinate in ow direction [m]

Greek symbols
correlation exponents []
corrugation angle []

e p R T b C p dT =T b  T w . Exponents a1 and a2 were sugwhere C
gested to be equal to the values used in correlation of Krasnoshchekov and Protopopov [10], which are variables themselves. In a
simpler version of their correlation, Jackson and Hall [9] proposed
constant values of 0.5 and 0.3 for a1 and a2, respectively.
A correlation of kind Eq. (1), however, may not be adequate to
predict heat transfer to or from a supercritical uid ow. In such
a uid ow, buoyancy force may affect the ow eld in a range
of Reynolds number where buoyancy is negligible in typical uid
ows. Most notable occurrence of this phenomenon is reported
for turbulent heat transfer in vertical pipes. It has been observed
that in such ows, with an increase in buoyancy forces, heat transfer is impaired, for upward ow direction, and is enhanced for
downward ow direction [1416]. For a laminar ow the converse
is true. The reason is explained as the deformation of velocity prole due to the effect of buoyancy force leading to a reduction or
enhancement of shear stress (depending on the ow direction) in
a region of ow where turbulence production is concentrated. Such
a change in the level of turbulence production is reected by a
change in the local heat transfer coefcient, which is highly sensitive to the amount of turbulence diffusivity. A detailed description
of the underlying physics can be found in [1720]. One would
expect this phenomenon be geometry-dependent but as all
above-mentioned references studied supercritical uid ow
through vertical pipes, it is hard to extend use of the existing
experimental results for plate heat exchangers. Recently, Forooghi




channel inclination angle []

plate thickness [m]
thermal conductivity [W/m K]
viscosity [kg/m s]
density [kg/m3]

constant property
f, f
uid index
Ethylene Glycol
heat transfer
data point index
mean value
variable property
heat transfer coefcient
log mean temperature difference
not wall-temperature corrected
plate type heat exchanger
QGECE Queensland Geothermal Energy Centre of Excellence
test heat exchanger
wall-temperature corrected

and Hooman, have reported numerical studies of this phenomenon

for inclined pipes [21], and corrugated channels [22], which would
be specically useful in the study of plate heat exchangers. This issue will be discussed in Section 3.3.
A number of experimental researches on heat transfer of supercritical uids in vertical and horizontal ducts have been published
in the past decades. Watts and Chou [15] measured heat transfer
coefcient of heated supercritical water owing in vertical pipes,
and developed an empirical correlation. Their correlation accounts
for both physical effects discussed above (wall-to-bulk property
variation and buoyancy). More recently, a similar correlation was
developed by Bae and co-workers based on extensive experiments
they carried out on heated supercritical CO2 in vertical pipes and
annuli [16,23,24]. As mentioned before, although the correlations
are developed for different uids, the suggested correction factors
are fairly similar and indicate on generality of the analysis. Jiang
et al. [25,26] and Kim and Kim [27] also performed experiments
on supercritical heat transfer in vertical pipes under heating condition to provide further evidence on the effect of property variation
on heat transfer coefcient. Liao and Zhao [28] studied a similar
problem in horizontal tubes with diameters between 0.5 and
2.16 mm, and reported the inuence of buoyancy in these tubes.
A number of reports are available in the literature for heat transfer
of supercritical uids owing in horizontal tubes or tube bundles
being cooled by another uid owing outside tubes [12,29,30].
These reports mostly concerned about nding correction factors


P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

for accounting for wall-temperature properties and did not discuss

any buoyancy effect.
As described above, the studies on turbulent heat transfer in uids with highly variable properties are mainly focused on simple
geometries most commonly circular pipe and, expectedly, care
has to be taken when it comes to use the existing correlations for a
much more complicated geometry like a PTHE. A PTHE is a heat exchanger composed of several patterned plates stacked together.
The gaps between plates act as ow passages; hot and cold streams
are distributed between the passages in such a way that different
streams are in contact with two sides of a plate acting as a (usually
thin) heat transfer surface. It is a common practice to use corrugated plates in PTHEs. Usually the corrugation directions of adjacent plates are in opposite angle in order to generate a complex
ow pattern which guarantees adequate mixing (see Fig. 1).
Attempts for nding reliable heat transfer correlations for
PTHEs with corrugated plates have started as early as 60s. In one
of the earliest, and yet most systematic, studies, Okada et al.
[31], investigated different geometries by changing corrugation angles and pitch-to-depth ratio of the plates. They found heat transfer
correlations of the Dittus and Boelter type to be suitable for their
experimented heat exchangers. Kumar [32] and Thonon [33] tried
to develop their own heat transfer correlations for single-phase
working uids to come up with similar type of correlation, but
slightly different in constants. Muley and Manglik [34] tried to obtain a generalized correlation, which can be applied to a range of
geometries. In their correlation, the parameters are functions of
corrugation angle and surface enlargement factor (the ratio of real
to projected surface area). Apart from geometry-dependent parameters, their correlation too has a Dittus and Boelter type. A more
theoretical approach to a generalized formula was taken by Martin
[35], and consequently Dovic et al. [36], in which existing formulas
for some basic geometries were combined together to derive an
equation suitable for plate heat exchangers. Flow in a straight conduit and ow in a 2D channel between two wavy plates were the
two basic geometries used in their approach. These geometries
were chosen because they were believed to represent main ow
patterns inside a plate heat exchanger. Such ideas are supported
by studies of ow patterns inside passages of plate heat exchangers
with corrugated plates. As an example, Focke et al. [37,38] found
different ow patterns in plate heat exchangers and argued that
the corrugation angle of the plate is the determining geometric
parameter. One can, as a general rule, conclude from all above
mentioned studies that the corrugation angle is, by far, the most

Fig. 1. Schematic gure of a passage between two adjacent cross-corrugated plates

in a plate heat exchanger (narrow lines represent corrugation lines of the lower

important geometric parameter in the heat transfer behavior of

plate heat exchangers.
A complete review of the literature reveals that none of the
existing correlations for heat transfer in plate heat exchangers is
proven to work where rates of change in thermophysical properties particularly density and specic heat are signicant, like
heat transfer of supercritical uids in the vicinity of pseudo-critical
temperature. The present experimental study is an attempt to ll
this gap in the literature and provide practical guidelines to be
used in applications with the possibility of occurrence of such a

2. Experiments
2.1. Test facility
Fig. 2b schematically shows the test facility used for this
experimental study. There were two loops, which formed the test
facility; the refrigerant loop or the main loop in which the working uid ows and the Glycol loop or heating loop in which
Ethylene Glycol ows. 98%-pure Peruoro-butane (Molecular formula: C4F10; CAS#: 355-25-9; critical pressure: 2.32 MPa; critical
temperature: 113.2 C) was used as the refrigerant in this study
due to its low critical pressure compatible with the available
equipment. This uid shows expected trends in thermophysical
properties near its critical point (see Fig. 3). The refrigerant and
Ethylene Glycol were circulated in their (separate) loops by
means of a positive displacement pump and a centrifugal pump,
respectively. Mass ow rate of the refrigerant was measured
using a Coriolis owmeter. Before the refrigerant is pumped, it
had to be depressurized. This was done by means of an adjustable
pressure-reduction regulator. The refrigerant cooled down in a
cooler, in which water at ambient temperature was used as the
coolant. A regenerator is added to the refrigerant loop to provide
a suitable heat balance for desired temperatures. For both cooler
and regenerator, similar plate heat exchangers different from
the test heat exchanger described below were used. In order
to reduce the uctuations of mass ow rate, an accumulator lled
with compressed N2 was installed before the inlet of the refrigerant pump. This accumulator was also used to adjust the overall
volume of the refrigerant in different working conditions, to avoid
a need for recharging the loop each time. It was, in particular,
necessary because of the strong sensibility of density to temperature for supercritical working uids leading to dramatic change
in the volume once heating temperature was changed. A
costume-designed electrical oil heater was used to heat up
Ethylene Glycol to a controlled outlet temperature. The maximum
temperature of Ethylene Glycol was always kept below 180 C, in
order to avoid boiling.
The main part of the test loop was the test heat exchanger
(THE), in which the refrigerant received heat from hot Ethylene
Glycol owing in the heating loop. Two commercial brazed plate
heat exchangers, only different in their corrugation angle, were
used in turn as THEs, both of which similarly consist of 10 plates
forming 9 passages (5 in hot side and 4 in cold side). THE was installed in the loop so that the mean ow direction was vertical. The
specications of THEs are presented in Table 1. At all four ports of
the THE, temperatures were measured using four RTDs with Pt100
elements. Pressure was also measured at both inlet and outlet of
refrigerant side using high-accuracy pressure transducers with 0
3 MPa range and accuracy of 0.15%. THE and all connecting pipes
between the measurement spots were thermally insulated using
glass-ber wraps (with a thickness of at least 5 mm) in order to
avoid any heat loss and unwanted heat transfer that might affect
the results.

P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459


Fig. 2. Test facility (a) and its schematic diagram (b).

Table 1
Specication of test heat exchangers.
Corrugation angle
Number of plates (including end plates)
Total heat transfer area
Cross section area (cold/hot)
Plate spacing
Plate material
Passage hydraulic diameter
Maximum working pressure
Maximum /minimum working temperature

Fig. 3. Properties of Peruoro-buthane at 2.55 MPa.

2.2. Data collection and reduction

As mentioned before, two plate heat exchangers (with corrugation angles of 30 and 60) were used as THEs. For each, three

30, 60
0.226 m2
8.64/10.8  104 m2
2 mm
3.2 mm
4.5 MPa
100/150 C

different cases with regard to ow arrangement and directions

were studied (see Table 2 for a summary of all six cases). In each
case, refrigerant mass-ow-rate and Ethylene Glycol temperature
varied to create a range of working conditions. Each working condition leads to one data point in the experiments. Ethylene Glycol
mass-ow-rate was constant for all data points within one case,
and, in general, this variable was set to be considerably larger than
that of refrigerant. It guarantees the convective thermal resistance
of the refrigerant side to be the dominant one, thus the overall heat
transfer coefcient of THE would be mainly sensitive to changes in
the refrigerant side, which was the purpose of this study. It also
leads for the hot stream temperature range to be considerably
smaller than that of the cold stream. Pressure of the main
loop was measured at both ends of THE and was controlled by


P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

Table 2
Flow arrangement and directions in different cases.


Flow arrangement

Direction of refrigerant

Glycol mass-ow-rate (kg/s)











the adjustable valve located downstream of the outlet port so that

its value remains constant during measurements of one data point.
The pressure variations during the whole experiment were also
slight; the working pressure was 2.55 0.1 MPa for all points.
The pressure drop through heat exchanger found to be negligible.
Totally, 90 data points were recorded in all six cases. For every data
point, data logging started after the system reached a steady state
condition, i.e. no time-dependency in the measured quantities was
observed except for random uctuations. The total duration of data
collection for each data point was a few times more than the time
it takes for the system to reach steady state. During the data collection time of every single data point, numerous measurements (at
least 100) were made to reduce the uncertainty due to uctuations
as will be further discussed in the error analysis section. A
summary of working conditions during the whole experiments is
presented in Table 3.
For each data point, total heat transfer rate of THE is determined
using the rst law of thermodynamics from the measured refrigerant mass-ow-rate as well as inlet and outlet enthalpies (as
functions of temperatures):

_ R hout  hin R :
Q exp m

Average overall heat transfer coefcient of the heat exchanger can

be found from Q(exp) and the measured temperatures at four ports
of the heat exchanger:


Q exp

AHT  DT mean

Overhead line sign j indicates an area-averaging over the

whole heat transfer surface area;

UxdA=AHT ;


where U(x) indicates local overall heat transfer coefcient which is

inverse-linearly proportional to the overall thermal resistance (dened as the of summation of convective thermal resistances of both
refrigerant and Ethylene Glycol sides plus conductive thermal resistance of the plate), i.e.;


1=HTCR x 1=HTCG x 2 =kplate

Table 3
Working conditions of test loop.

Main loop
High pressure (THEa working pressure)
Low pressure
THE inlet temperature
THE outlet temperature

2.452.65 MPa
<0.35 MPa
0.020.1 kg/s
3075 C
84128 C

Heating loop
Maximum temperature (THE inlet)
Heating power

120160 C
0.171.38 kg/s
<10 kW

Test heat exchanger.

Obviously, only U can be directly determined from the experimental data. Precisely speaking, convective heat transfer coefcients
(HTCs) can vary in both directions (on a plate), but in this paper,
they are considered only variable in the ow direction (x), which
can be considered a 1D analysis.
In Eq. (3), DTmean is the mean temperature difference of the two
streams in the heat exchanger. Assuming 1D variation in temperatures, it can be shown that

DT mean 

DT 2  DT 1
DT 2

1=TC R  1=TC G

DT 1 1=TC R 1=TC G

where subscripts 1 and 2 refer to the two ends of the heat exchanger and negative and positive signs must be used for parallel- and
counter-ow, respectively. TC stands for thermal capacity, which
is determined as

_  CP :
TC m

Since specic heat (CP) of a uid is a function of temperature,

thermal capacities are, in general, temperature-dependent. TCR
and TCG in Eq. (6) are local values. The average values of thermal
capacity TC R and TC G  can be directly found from the following

_  hout  hin =T out  T in :

TC m

The integral in Eq. (6) can be found merely with knowledge of the
port temperatures. It is possible because CP is only a function of
temperature for every uid (when the variation of pressure is not
dramatic). Temperature of each uid can be found from that of
the other uid, at every point, through rst law of thermodynamics.
As a result, the whole integrant can be stated as a function of one
variable (temperature) only if the variation of CP with temperature
is known. Therefore, it is possible to calculate DTmean directly from
the experimental data.
Eq. (6) can be reduced to the well-known log-mean-temperature-difference (LMTD) formula, if CR and CG are not variable, which
is not the case here, especially for the refrigerant. Ease of use of
LMTD formula, however, may be tempting to use it instead of the
above approach, which requires some extra effort for numerical
integration. To obtain an idea about how much error the simplied
approach might introduce to the analysis, separate calculations
have been done for all data points using both approaches and the
results are presented for the percentage of error for LMTD
approach in Fig. 4. It is observed that the error is not negligible
especially when the temperature of refrigerant tends to the pseudo-critical temperature, where rate of variation in specic heat is
All physical properties in this study were determined based on
NIST database using REFPROP software [39].
2.3. Error analysis
Error analysis in this study is based on the well-established
multiple-sample analysis approach presented in [40]. The overall
uncertainty of a quantity is equal to root-sum-square combination

P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459


2.3.2. Uncertainty in mass-ow-rate

Based on manufacturers specication, the instrument error for
the Coriolis owmeter is 0.1% of the absolute measured value.
However at the beginning, some severe uctuations were
observed in the measurement of mass-ow-rate. The uctuations
were reduced to some degree by minimizing the vibrations in the
system. In the main experiments, the uctuations varied from
point to point, with standard deviation being always less than
7%. The remaining uctuations possibly originate in the function
of the reciprocal positive displacement pump. Fixed error combined with the uctuations was used into determine the total
uncertainty in mass-ow-rate for each data point. The biggest
obtained value of uncertainty for a data point was 1.5% of the
absolute value.

Fig. 4. Error in calculation of mean temperature reference using LMTD as a function

of refrigerant outlet temperature.

of uncertainties due to all sources of error.1 The contribution of random uncertainties, however, can be reduced to a high extent by
using the mean value out of an increased number of independent
measurements. As mentioned before, for each data point, more than
100 measurements have been made to reduce this contribution as
far as possible. Accordingpto
Moffat [40], uncertainty due to random
errors is equal to t  S= N where S is standard deviation of all
measurements and N is the number of measurements. t is Students
multiplier, which is approximately 2 for N > 60.
As explained before, there are two types of measured quantities
in the present study: mass-ow-rate and temperature. All
uncertainties introduced to the experiments necessarily stem from
errors in the measurements these two quantities.

2.3.1. Uncertainty of temperature

The RTDs used for measuring temperatures have been already
calibrated with tolerance of 0.35 C. This can be regarded as the
xed component of the instrument error for all measurements
through the course of experiments. Fixed errors due to system
may be caused by numerous factors. Attempts has been made to
eliminate all of them as much as possible; the test heat exchangers
and all pipes in between temperature sensors were insulated to
prevent any unwanted heat transfer in thermocouples or any heat
loss. Besides, RTDs are placed as close to the heat exchangers ports
as possible. Another source of xed error could be temperature
variation at different point of a cross-section where the measurement takes place. This error was ruled out by doing an auxiliary
experiment in which the intrusion of RTD is changed a few times;
no systematic error was observed in this auxiliary experiment.
Random errors in temperature measurement can be simply
determined for every data point based on the observed scatter in
the measurements. This scatter varied from one data point to another, but the standard deviation was always smaller than 0.1 C.
Combining this with the instrument error mentioned above with
an assumption that the xed errors due to system are negligible
the overall uncertainty in temperature measurements would be
less than 0.4 C.

Unless specied otherwise, uncertainty, in this paper, is based on 95%

probability (20 to 1 odds in favor of the real value being within the specied interval).

2.3.3. Uncertainty in Q(exp) and U exp

As explained in the previous section, Q(exp) and U exp are calculated based on the measured values of temperature and massow-rate. In order to evaluate the uncertainty in each calculated
variable, it is necessary to determine its sensitivity to all measured variables rst. The values of sensitivity are equal to partial
derivative of the calculated variable with respect to the measured
variable. They are straightforward to obtain from the equations
already presented except for DTmean, for which computerized
analysis introduced in [40] was employed. Once the sensitivity
is known, the uncertainties in Q(exp) and U exp can be determined
based on uncertainties in the measured values of temperature
and mass-ow-rate, which were already discussed. For each
point, the analysis is based on all measurements for that data
It must be mentioned here that the errors which are xed during measurements of a single data point may have a variable part
when different data points are taken into account since for recording every new data point, changes had to be made to the test facility. Fixed errors of the instrument may vary when the value of the
actual measurable quantity changes. On the other hand, it is possible that system errors, which are xed for one data point, vary
from one data point to another. Note that six different cases have
been studied, for each of which parts of pipe work and insulation
should have been be redone. Moreover, variables such as pump frequency, accumulator pressure and heater outlet temperature were
different in different data points. These variable parts of xed errors, if exist, cannot be smoothed out by averaging among the
measurements of a single data point. Therefore, if a variable has
to be calculated based on various data points, some data scatter
may arise. This issue will be further discussed in the next section.
3. Results and discussion
3.1. Test results
Fig. 5 presents the values of heat transfer rate (Q(exp)) plotted
against refrigerant mass-ow-rate for all data points, distinguished
by the case numbers. For all cases, Q(exp) increases monotonically
with mass-ow-rate. It must be mentioned that inlet temperatures
of refrigerant and Ethylene Glycol are not identical for different
points. According to Fig. 5, no big difference is observed between
cases with parallel- and counter-ow arrangements; it is reasonable since, as mentioned before, the hot stream temperature is almost constant, compared to that of the cold stream.
Fig. 6 presents the obtained values of overall heat transfer coefcient U exp for all data points, plotted against average Reynolds
number dened as


G  dh



P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

Fig. 5. Heat transfer power plotted against refrigerant mass-ow-rate for both test heat exchangers. Error bars indicate on 20 to 1 uncertainty interval.

Fig. 6. Overall heat transfer coefcient plotted against refrigerant average Reynolds number for both test heat exchangers. Error bars indicate on 20 to 1 uncertainty interval.

where G is mass velocity (mass ow rate divided by total passage

area) and dh is the hydraulic diameter of the passage. The average
Reynolds number is obtained by averaging among local values of
Reynolds number obtained from a 1D analysis which will be
explained later.
Although, as expected, U exp shows a generally increasing trend
with Reynolds number, a big deal of irregularity is also observed
for both heat exchangers. It is, as already discussed, believed to
be due to the strong dependence of physical properties on temperature. As such, a detailed investigation is called for.
As discussed in the introduction, heat transfer correlations of
Dittus and Boelter type are widely used for PTHEs. Similarly, in this
study a correlation of the same type is used, i.e.

ant side, which is the dominant side in terms of heat transfer resistance. Therefore, either a 0D approach must be used in which some
averaging practice is required or a 1D one, which accounts for the
variation of temperature, and thereby variations of other parameters. The former approach would lead to some extra uncertainty
in the results, in spite of its simplicity. Therefore, the latter approach was chosen for this study, for which the following heat balance equation was numerically solved for both refrigerant and
Ethylene Glycol streams in the entire heat transfer surface area:

Nu C  Ren Pr m :

where positive and negative signs refer to the ow directions.

Refrigerant ow direction is always assumed positive, so the sign
for Ethylene Glycol is positive and negative for parallel- and counter-ow, respectively. Here, x0,f denotes the inlet position, which
is always zero for f = R; for f = G it is zero or L for parallel- and counter-ow arrangements, respectively. The equations of the two
streams are coupled by heat transfer term in the right hand side. Local heat transfer coefcient, U(x), can be found from Eq. (5), in


However, it was observed, as will be shown later, that the above

correlation cannot satisfactorily reect the physics of the problem;
then the idea of using a correction factor of Jackson and Hall type
emerges, i.e. correcting Nusselt number from Eq. (10) using wall
temperature correction factor from Eq. (1). The result would be

Nu C  Ren Pr m

C p;b





To avoid unnecessary complication of the problem, the simplied

values of 0.5 and 0.3 are used for a1 and a2 in the present study.
If one wishes to use a correlation of above types, it must be noted
that Reynolds and Prandtl numbers as well as the property ratios
vary considerably along the heat exchanger, at least for the refriger-

_ f C P;f

dT f
Ux  T f x  T f 0 x;
f ; f R; G; f f 0 ; T f x0;f T f ;in ;

HTCf x Nuf x 




Nuf(x) can be determined based on local parameters using either

Eq. (10) or Eq. (11) depending on which type of correlation is being
examined. Since both sides of the heat exchanger are identical,


P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

same correlation can be used for both sides (f = R, G). It is worth

mentioning that, the convective thermal resistance of the refrigerant side, has always been found to be at least an order of magnitude
larger than the other two thermal resistances, and, in general, U
could be suitably approximated by HTCR. Obviously, all thermophysical properties appearing in the solution are functions of local
In order to solve Eq. (12), heat transfer surface area have been
one-dimensionally discretized using 500 grid points, which guarantees mesh independence with less than 0.01% error when compared to results obtained on a 1000 grid points. First order
forward difference scheme was used for the temperature derivative and a time-marching iterative method was adopted by adding
an articial transient term to the left hand side of the equation. The
validity of the solution was checked in comparison with the analytical expression for mean temperature difference (Eq. (6)). The nal
results are also in a reasonable agreement with other correlations
available in the literature as will be discussed later.
The parameters n, m and C were found by minimizing a target
function total deviation of the correlation from experimental
data, which is straightforward apart from the fact that a differential
equation had to be solved rather than calculating an algebraic
expression for heat transfer coefcient.

[31]. This is quite satisfactory noting the difference between available experimental results in the literature obtained for different
heat exchangers. Furthermore, the present correlation, although
obtained for a different working uid from those of other works,
matches well with other correlations, which is an approval of the
generic nature of our analysis.
In order to investigate the wall temperature correction, it is necessary to compare the scattering intervals of both correlations with
regards to experimental data. To do that, for every data point, the
value of constant C, which exactly leads to the experimental heat
transfer coefcient U exp for that point, is obtained. Obviously,
the value of C reported in Table 4, for either WC or NWC case,
should be the mean of all single-data-point values of C for that
case. To distinguish between mean and single-data-point C, respectively, subscripts mean and i are used:

C mean

Optimum values for (n, m) were found to be almost the same for
both not wall temperature corrected (NWC) and wall temperature corrected (WC) correlations, i.e. Eqs. (10) and (11), respectively. These values are (0.74, 0.35) for b = 30 and (0.71, 0.35) for
b = 60. The value of constant C is however different for the two
correlations being 0.09 (WC) and 0.076 (NWC) for b = 30 and
0.187 (WC) and 0.165 (NWC) for b = 60. A summary of the parameters is presented along with the values reported by other experimentalists in Table 4. To better compare the present correlation
with other correlations their predicted variation of Nusselt number
with Reynolds number are plotted all in the same graph (Fig. 7). A
constant Prandtl number equal to 3.6, which is the average refrigerant Prandtl number for the present study, was used in these
graphs. Both WC and NWC cases were shown in Fig. 7; since the
other works do not account for any wall-to-bulk ratio of density
and/or specic heat, the points for WC correlation in these graphs
are calculated by neglecting the correction factors in Eq. (11), i.e.
qw 2
equating C p
with unity. The prediction of WC case lies

perfectly within those of other correlations, while for NWC correlation, some underestimation is observed. Considering WC to be
the correct correlation (this issue will be discussed in depth later),
the present results are in best agreement with those of Okada et al.



It can be shown that the ratio of Ci to Cmean is the ratio of heat

transfer coefcient obtained in experiments to that predicted by
the correlation, or

C mean

3.2. Effect of wall-to-bulk property ratios

U exp 

U corr i


In Fig. 8, all values of Ci obtained using both WC and NWC correlations are presented. Ideally, there must be no scatter in these values. It would be the case if an ideal correlation was used, i.e. a
correlation that captures the physics of the problem in full details.
In this sense, the scatter in Fig. 8 could be considered a result of
conceptual error. There is yet another possible source of data scatter in Fig. 8 which could be the result of what addressed in the error
analysis section as the variable part of xed errors. Although care
has been taken to block any source of xed error, except for the
inevitable instrument error, it is always possible that unknown error sources emerge during the experiment. Such an error, if exists,
however, must be almost the same for both heat exchangers since
the whole experimental procedure is exactly similar for all cases
presented in Table 2. Therefore, any difference in the amount of
scatter for the two heat exchangers can only be due to the conceptual error not experimental error.
It is observed in Fig. 8 that, for b = 60, the standard deviation of
data scatter can be signicantly reduced from 12.6% to 4.6% by
including wall temperature correction. For b = 30, however, the
reduction is from 11.4% to 9.7%, which can hardly be called a meaningful improvement. In view of the above, one may argue that
using a heat transfer correlation, which accounts for wall-to-bulk
property variations, can bring about better results for b = 60 and
not for b = 30. This deduction will be further investigated in the
following paragraphs.

Table 4
Correlation parameters; comparison of the present results with other reports.



Kumar [84]
Muley and Manglik [99]
Okada et al. [72]
Thonon [95]










Kumar [84]
Muley and Manglik [99]
Okada et al. [72]
Thonon [95]


P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

Fig. 7. Comparison of the present correlation for Nusselt number with other works for both corrugation angles in the studied range of Reynolds number.

Fig. 8. Scattering of experimental C constant obtained from no wall corrected (up) and wall corrected (down) correlations. For each case, data are plotted against both
refrigerant average Reynolds number and Prandtl number.

In Fig. 8, data were plotted against both the average Reynolds

and Prandtl numbers to observe no signicant dependency. A
signicant trend, however, can be found when data points are
plotted against a third variable, which is the value of wall

temperature correction itself as indicated by Fig. 9. The variable

appearing on abscissa in this gure is the effective value of wall
temperature correction applied in each data point, which was
calculated as

P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459


Fig. 9. Experimental C constant obtained from no wall corrected (left) and wall corrected (right) correlations plotted against effective wall temperature correction factor.



Ren Prm

eC p
C p;b


Ren Pr m



In fact, CF measures how much, for each data point, heat transfer
coefcient is corrected because of wall-to-bulk property ratios. It
is clearly observed that, in NWC case, the value of constant C is proportional to CF. For both corrugation angles b = 30, 60, the numerical value of the slope is of the order of magnitude of one although
the trend line is somewhat steeper for b = 60. Comparing WC and
NWC results, it can be argued that, when the employed heat transfer correlation does not include the required correction due to wallto-bulk property ratio (NWC), the required correction emerges in
the value of Ci. In other word, there is a physical variable on which
the heat transfer coefcient depends but is not accounted for by the
heat transfer coefcient. For WC case, i.e. with a correlation of type
Eq. (11), this dependence is already accounted for, so C shows no
dependence on CF. It is observed that the trend lines in the right
hand side picture are also not perfectly horizontal. This is most
probably because of the use of a simplied correction factor with
constant exponents. As mentioned in the introduction, there are
more elaborated correction equations available in the literature,
among which the right choice can be made depending on the
expected accuracy. The important nding of Fig. 9 is existence of
a dependence on wall-to-bulk property ratio, which can be satisfactorily eliminated by use of a right correction equation.
3.3. Effect of buoyancy
It was shown in the previous section that, for both plate geometries, use of a heat transfer correlation, which contains a wall
temperature correction expression can remove, to a satisfactory
extent, the dependency of error on the property ratios, which
would appear if a Dittus and Boelter type correlation was used.
However, as both Figs. 8 and 9 suggest, the amount of data scatter
is only reduced by this correction for b = 60 and not for b = 30.
When WC correlation is used, the experimental data scatters within around 20% of the mean value for b = 30 which is more than 10%
bigger than that of b = 60 (considering root sum square combination of uncertainties, the extra uncertainty is approximately 15%).
As explained in the previous section, this difference in the amount
of scatter should be a result of some conceptual error. It suggests
the existence of a physical effect for b = 30 which is absent in
b = 60. It was mentioned in the introduction that buoyancy can
be able to affect the heat transfer coefcient in supercritical uids
with similar density variations to that of this problem, and its
effect is also highly geometry-dependent.

Forooghi and Hooman [21,22], numerically investigated the possible effect of buoyancy on the ow patterns and heat transfer that
can arise in a plate heat exchanger. They estimated that, in an inclined conduit, the effect of buoyancy on turbulent heat transfer
is most severe around 50% in the worst scenario for c angles
smaller than 10, where c is the angle of the conduit with the vertical direction. This effect decays as c increases and is negligible
for c larger than 50. On the other hand, studies of ow pattern inside plate heat exchangers suggest that, for both b = 30 and 60, the
dominant regime of ow inside a plate heat exchanger is ow along
furrows, which can be regarded analogous to ow in conduits that
make an angle equal to b with the vertical direction [35,38]. Based
on this fact, the effect of buoyancy in the heat exchangers with
either b angle can be roughly estimated by what Forooghi and Hooman [21] reported for the same c angle. The result would be a negligible effect for b = 60 and one with around 15 to 20% inuence for
b = 30. The gures approximately match the value of extra data
scatter observed for b = 30 in the present study. It must be reminded that the buoyancy effect could be positive or negative
depending on the ow direction being downward or upward.
To better investigate the effect of buoyancy, values of Ci for all
data points, distinguished by the direction of ow, are plotted
against Richardson number in Fig. 10. For the test heat exchanger
with b = 60, all points are within approximately 10% of the mean
value. For the one with b = 30, one out of four to ve points lie out
of this accuracy interval, all of which belong to Richardson numbers higher than a certain threshold. For the latter heat exchanger,
data scattering generally increases with Richardson number.
Richardson number is widely used as a measure of buoyancy
forces relative to viscous forces. In this work it is dened as:




qb qb  qw  g  d3h


Although Fig. 10 provides some general information about the effect

of buoyancy, it is hard to identify a clear trend. It can be due to at
least two factors; rst, the range of Reynolds number chosen for this
study is very close to the laminarturbulence transition region. It is
generally hard, if possible, to dene a clear transition threshold in
the plate heat exchangers, mainly due to the complicated and multi-pattern nature of ow. Experiments of Focke et al., for example,
suggest ranges of 150 < Re < 600 for b = 60and 1000 < Re < 3000
for b = 30, none of which being clear-cut [38]. Hence, for the problem under consideration, probably some data points are in the transition regime, where the behavior of turbulent heat transfer is
considerably complicated and, as a matter of fact, all previous
works on supercritical uid ow considered fully turbulence ows.


P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459

Fig. 10. Experimental C constant obtained from plotted against Richardson number.

Thus, it is hard to expect the theory to be perfectly reproduced by

the present experiments. In addition to that, even in the experiments dealing with straight pipes [14,16,41,42], similar scatter is
observed within the experimental points. In other words, the effect
of buoyancy on turbulent supercritical uid ows is in general one
with some inherent uncertainty.
It is worth mentioning that b = 60, is presently, by far, the most
widely used corrugation angles for plate heat exchangers, for
which one can comfortably ignore the effect of buoyancy; this is
most probably the case for all corrugation angles larger than 50.
For b = 30, the present data are enough to raise the concern of possible deviations from what one would expect using the available
correlations in the literature.
For corrugation angles smaller than 30 (including the case of
combined corrugation angle with one angle smaller than 30)
experimental proof is required; for this case numerical observations [21,22] suggest possibility of more severe deviations.
4. Conclusion
Experiments have been done on two plate-type heat exchangers
with corrugation angles of b = 30, 60 to investigate the performance of this type of heat exchangers in the working conditions
where density and specic heat both strongly depend on temperature, like in supercritical uids. The ranges of Reynolds and Prandtl
number are 8004200 and 3.24.2, respectively. It was found that:
In a PTHE, Whenever a working uid exchanges heat in conditions where the rates of variations of its thermophysical properties (specic heat and density) with temperature are high, it is
expected that heat transfer correlations not accounting for the
effect of wall-to-bulk property ratios (for example those of Dittus and Boelter type), fail to make satisfactory predictions. A
specic example of such uids is a supercritical uid in a pressure slightly larger than its critical pressure when either/both of
bulk or/and wall temperatures is/are close to the pseudo-critical
To tackle the above mentioned problem, a two-component heat
transfer correlation is suggested: the rst component is the normal heat transfer correlation that would be used if thermophysical properties were constant. This correlation should, however,
be multiplied by a correction factor of Jackson and Hall type
(see Eq. (1)). The correlations found in this research are:

Nu 0:187  Re0:71 Pr 0:35

Nu 0:09  Re0:74 Pr 0:35

qw 0:3
; b 60 ;
C p;b
qw 0:3
; b 30 ;
C p;b

but any other correlation obtained through the above two-component approach can bring about satisfactory results. This idea
can be, in particular, useful because a specic correlation developed for a specic design may work the best for that design.
For a plate heat exchanger with corrugation angle of 30 there
might be up to 20% deviation in results due to the effect of
buoyancy. For the corrugation angle of 60 the buoyancy effects
are negligible.
Conict of interest
None declared.
This research paper was made possible through substantial
technical support from Mr Jason Czapla. Authors would also like
to appreciate help from staff of the School of Mechanical and
Mining Engineering Workshops in University of Queensland, in
particular, safety ofcer, Mr Hugh Russell.
[1] Z.H. Ayub, Plate heat exchanger literature survey and new heat transfer and
pressure drop correlations for refrigerant evaporators, Heat Transfer Eng. 24
(5) (2003) 316.
[2] S. Kakac, H. Liu, Heat Exchangers: Selection, Rating and Thermal Design,
second ed., CRC Press, 2002.
[3] M.M. Abu-Khader, Plate heat exchangers: recent advances, Renew. Sustain.
Energy Rev. 16 (2012) 18831891.
[4] J.E. Hesselgreaves, Compact Heat Exchangers: Selection, Design and Operation,
Elsevier Science & Technology Books, 2001.
[5] R.B. Duffey, I.L. Pioro, Experimental heat transfer of supercritical carbon
dioxide owing inside channels (survey), Nucl. Eng. Des. 235 (2005) 913924.
[6] A.A. Bishop, R.O. Sandberg, L.S. Tong, Forced convection heat transfer to water
at near-critical temperature and supercritical pressure, in: American Society of
Chemical Engineers- International Chemical Engineers Symposium No. 2,
[7] B. Shiralkar, P. Grifth, The effect of swirl, inlet conditions, ow direction, and
tube diameter on the heat transfer of uids at supercritical pressure, ASME J.
Heat Transfer 92 (3) (1970) 465474.
[8] H.S. Swenson, J.R. Carver, C.R. Kakarala, Heat transfer to supercritical water in
smooth-bore tube, ASME J. Heat Transfer 87 (1965) 477484.
[9] J.D. Jackson, W.B. Hall, Forced convection heat transfer to uids at supercritical
pressures, in: S. Kakac, D.B. Spalding (Eds.), Turbulent Forced Convection in
Channels and Bundles, Hemisphere Publishing, 1979, pp. 563611.
[10] E.A. Krasnoshchekov, V.S. Protopopov, Experimental study of heat exchange in
carbon dioxide in the supercritical rabge at high temperature drops, Teploz.
Vysok. Temp. 4 (3) (1966).
[11] S. Mokry, I.L. Pioro, A. Farah, K. King, S. Gupta, W. Peiman, P. Kirilov,
Development of supercritical water heat-transfer correlation for vertical bare
tubes, Nucl. Eng. Des. 241 (2011) 11261136.
[12] X.L. Huai, S. Koyama, T.S. Zhao, An experimental study of ow and heat
transfer of supercritical carbon dioxide in multi-port mini channels under
cooling conditions, Chem. Eng. Sci. 60 (12) (2005) 33373345.
[13] J.D. Jackson, A semi-empirical model of turbulent convective heat transfer to
uids at supercritical pressure, in: ASME Conference Proceedings, 2008(48167),
2008, pp. 911921.

P. Forooghi, K. Hooman / International Journal of Heat and Mass Transfer 74 (2014) 448459
[14] J.D. Jackson, W.B. Hall, Inuence of buoyancy on heat transfer to uids owing
in vertical tubes under turbulent condition, in: S. Kakac, D.B. Spalding (Eds.),
Turbulent Forced Convection in Channels and Bundles, Hemisphere
Publishing, 1979, pp. 613640.
[15] M.J. Watts, C.T. Chou, Mixed convection heat transfer to supercritical pressure
water, in: International Heat Transfer Conference 3, Munchen, 1982, pp. 495
[16] Y.Y. Bae, H.Y. Kim, D.J. Kang, Forced and mixed convection heat transfer to
supercritical CO2 vertically owing in a uniformly-heated circular tube, Exp.
Therm. Fluid Sci. 34 (2010) 12951308.
[17] T. Aicher, H. Martin, New correlations for mixed turbulent natural and forced
convection heat transfer in vertical tubes, Int. J. Heat Mass Transfer 40 (15)
(1997) 36173626.
[18] V.A. Kurganov, A.G. Kaptilny, Velocity and enthalpy elds and eddy
diffusivities in a heated supercritical uid ow, Exp. Therm. Fluid Sci. 5 (4)
(1992) 465478.
[19] V.A. Kurganov, A.G. Kaptilnyi, Flow structure and turbulent transport of a
supercritical pressure uid in a vertical heated tube under the conditions of
mixed convection. Experimental data, Int. J. Heat Mass Transfer 36 (13) (1993)
[20] W.S. Kim, S. He, J.D. Jackson, Assessment by comparison with DNS data of
turbulence models used in simulations of mixed convection, Int. J. Heat Mass
Transfer 51 (2008) 12931312.
[21] P. Forooghi, K. Hooman, Numerical study of turbulent convection in inclined
pipes with signicant buoyancy inuence, Int. J. Heat Mass Transfer 61 (2013)
[22] P. Forooghi, K. Hooman, Effect of buoyancy on turbulence convection heat
transfer in corrugated channels a numerical, Int. J. Heat Mass Transfer 64
(2013) 850862.
[23] Y.-Y. Bae, H.-Y. Kim, Convective heat transfer to CO2 at a supercritical pressure
owing vertically upward in tubes and an annular channel, Exp. Therm. Fluid
Sci. 33 (2) (2009) 329339.
[24] H. Song, Investigations of a printed circuit heat exchanger for supercritical CO2
and water, Kansas State University, Manhattan, Kansas, 2006.
[25] P.X. Jiang, R.F. Shi, C.R. Zhao, Y.J. Xu, Experimental and numerical study of
convection heat transfer of CO2 at supercritical pressures in vertical porous
tubes, Int. J. Heat Mass Transfer 51 (2008) 62836293.
[26] P.X. Jiang, Y. Zhang, Y.J. Xu, R.F. Shi, Experimental and numerical investigation
of convection heat transfer of CO2 at supercritical pressures in a vertical tube
at low Reynolds numbers, Int. J. Therm. Sci. 47 (2008) 9981011.


[27] D.E. Kim, M.H. Kim, Experimental study of the effects of ow acceleration and
buoyancy on heat transfer in a supercritical uid ow in a circular tube, Nucl.
Eng. Des. 240 (10) (2010) 33363349.
[28] S.M. Liao, T.S. Zhao, Measurement of heat transfer coefcients from
supercritical carbon dioxide owing in horizontal mini/macro channels,
ASME J. Heat Transfer 124 (2002) 413420.
[29] C. Dang, E. Hihara, In-tube cooling heat transfer of supercritical carbon dioxide.
Part 1. Experimental measurement, Int. J. Refrig. 27 (2004) 736747.
[30] C.H. Son, S.J. Park, An experimental study on heat transfer and pressure drop
characteristics of carbon dioxide during gas cooling process in a horizontal
tube, Int. J. Refrig. 29 (2006) 539546.
[31] K. Okada, M. Ono, T. Tomimura, T. Okuma, H. Konno, S. Ohtani, Design and heat
transfer characteristics of a new plate heat exchanger, Heat Transfer Jpn. Res. 1
(1) (1972) 9095.
[32] H. Kumar, The plate heat exchanger: construction and design, in: 1st UK
National Conference on Heat Transfer, Leeds, 1984, pp. 12751284.
[33] B. Thonon, R. Vidil, C. Marvillet, REcent research and developments in plate
heat exchangers, J. Enhanced Heat Transfer 2 (12) (1995) 149155.
[34] A. Muley, R.M. Manglik, Experimental study of turbulent ow heat transfer
and pressure drop in a plate heat exchanger with chevron plates, ASME Heat
Transfer 121 (1999) 110117.
[35] H. Martin, A theoretical approach to predict the performance of chevron-type
plate heat exchangers, Chem. Eng. Process. 35 (4) (1996) 301310.
[36] A. Dovic, B. Palm, S. Svaic, Generalized correlations for predicting heat transfer
and pressure drop in plate heat exchanger channels of arbitrary geometry, Int.
J. Heat Mass Transfer 52 (2009) 45534563.
[37] W.W. Focke, P.G. Knibbe, Flow visualization in parallel-plate ducts with
corrugated walls, J. Fluid Mech. 165 (1986) 7377.
[38] W.W. Focke, J. Zacharides, I. Oliver, The effect of the corrugation inclination
angle on the thermohydraulic performance of plate heat exchangers, Int. J.
Heat Mass Transfer 28 (8) (1985) 14691479.
[39] <>, in.
[40] R.J. Moffat, Describing the uncertainities in experimental results, Exp. Therm.
Fluid Sci. 1 (1988) 317.
[41] Z.H. Li, P.X. Jiang, C.R. Zhao, Y. Zhang, Experimental investigation of convection
heat transfer of CO2 at supercritical pressures in a vertical circular tube, Exp.
Therm. Fluid Sci. 34 (2010) 11621171.
[42] A. Bruch, A. Bontemps, S. Colasson, Experimental investigation of heat transfer
of supercritical carbon dioxide owing in a cooled vertical tube, Int. J. Heat
Mass Transfer 52 (2009) 25892598.