Está en la página 1de 12

international journal of refrigeration 31 (2008) 12531264

available at www.sciencedirect.com

w w w . i i fi i r . o r g

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

An advanced nonlinear switched heat exchanger


model for vapor compression cycles using
the moving-boundary method
Thomas L. McKinley1, Andrew G. Alleyne*
Department of Mechanical Science and Engineering, University of Illinois at Urbana-Champaign,
1206 West Green Street, MC-244, Urbana, IL 61801, USA

article info

abstract

Article history:

Simulation is commonly used to develop control and diagnostic algorithms. Because vapor

Received 7 July 2007

compression cycles are essentially heat management devices, transient and steady-state

Received in revised form

heat exchanger performance must be predicted accurately. However, for hardware- and

19 December 2007

software-in-the-loop purposes, the simulation must run in real-time. To reconcile these

Accepted 30 January 2008

competing needs, a new lumped parameter or moving-boundary heat exchanger model

Published online 8 February 2008

was created. Accuracy concerns are addressed by including finned surfaces, nonlinear
air temperature distributions, and non-circular passages. In its current form, the model

Keywords:

is applicable to single pass, cross-flow heat exchangers. The mathematical basis of the

Refrigeration system

model is given and shown to be consistent with integral forms of the energy and continuity

Compression system

equations. Although known to be more computationally efficient than finite volume

Modelling

models, moving-boundary models become singular and fail under certain conditions. To

Heat exchanger

address this shortcoming, particular attention was focused on algorithms for switching be-

Transient state

tween different representations and rezoning wall temperatures. Robustness to changing

Research

flow regimes is demonstrated through simulation test cases, and model application to
a chiller system is shown. As such, the model provides improved accuracy, robustness,
and operating range while maintaining real-time capability.
2008 Elsevier Ltd and IIR. All rights reserved.

Nouvelle modelisation non lineaire pour un echangeur de


chaleur pour les cycles a` compression de vapeur faisant appel
a` une methode de limites mobiles
Mots cles : Syste`me frigorifique ; Syste`me a` compression ; Modelisation ; Echangeur de chaleur ; Regime transitoire ; Recherche

* Corresponding author. Tel.: 1 217 244 9993; fax: 1 217 244 6534.
E-mail addresses: tmckinl2@uiuc.edu (T.L. McKinley), alleyne@uiuc.edu (A.G. Alleyne).
1
Tel.: 1 812 372 5506.
0140-7007/$ see front matter 2008 Elsevier Ltd and IIR. All rights reserved.
doi:10.1016/j.ijrefrig.2008.01.012

1254

international journal of refrigeration 31 (2008) 12531264

Nomenclature
a0 , a1 , a2 empirical constants in the void fraction correlation
[dimensionless]
refrigerant cross-sectional flow area [m2]
ACR
air-to-structure surface area [m2]
ASA
refrigerant-to-structure surface area [m2]
ASR
air specific heat at constant pressure [J kg1 K1]
cPA
wall (structure) specific heat [J kg1 K1]
cW
fraction of air-to-structure surface area on fins
FFIN-A
[dimensionless]
fraction of refrigerant-to-structure surface area on
FFIN-R
fins [dimensionless]
refrigerant saturated liquid enthalpy (at pressure P)
hf
[J kg1]
refrigerant saturated vapor enthalpy (at pressure P)
hg
[J kg1]
average refrigerant enthalpy for zone j [J kg1]
hj
outlet refrigerant enthalpy from zone j [J kg1]
hj-OUT
inlet refrigerant enthalpy (boundary condition to
hR-IN
the model) [J kg1]
outlet refrigerant enthalpy from the heat
hR-OUT
exchanger [J kg1]
j
subscript for zone number (1 superheated,
2 two-phase, 3 sub-cooled)
gain in the enthalpy pseudo state equation,
Kh
currently set to five [s1]
gain in wall temperature pseudo state equation,
KT
currently set to five [s1]
gain in the mean void fraction pseudo state
Kg
equation, currently set to five [s1]
wall thermal conductivity [W m1 K1]
kW
refrigerant passage length [m]
LR
refrigerant mass for zone j [kg]
mj
_A
air mass flow rate (boundary condition to the
m
model) [kg s1]
_ R-IN
inlet refrigerant mass flow rate (boundary
m
condition to the model) [kg s1]
_
mR-OUT outlet refrigerant mass flow rate (boundary
condition to the model) [kg s1]
wall (structure) mass [kg]
mW
_ 12
refrigerant mass flow rate from zone 1 to zone 2
m
[kg s1]
_ 23
refrigerant mass flow rate from zone 2 to zone 3
m
[kg s1]
NTU
number of transfer units [dimensionless]
P
refrigerant pressure [Pa]
air-to-structure heat transfer rate for zone j [W]
Q_ Aj
Q_ Rj
structure (wall)-to-refrigerant heat transfer rate
for zone j [W]
inlet air temperature (boundary condition to the
TA-IN
model) [K]

1.

Introduction

Vapor compression cycles are widely used to transfer heat to


or from people and equipment. In accomplishing that goal,

TA-OUT j
TRj
TWj
TWTj
t
tW
URj
uj
Vj
_j
W
x
x0
x~
x~OUT
x_
aA
aRj
g
g
gTOT

hFA
hFRj
zj

zmin
dz
rf
rg
rj
n12
n23
mf
mg

outlet air temperature for zone j [K]


average refrigerant temperature for zone j [K]
average wall temperature for zone j [K]
wall temperature being transported across the
rightmost boundary of zone j due to rezoning [K]
time [s]
wall thickness [m]
structure (wall)-to-refrigerant overall heat transfer
coefficient for zone j [W m2 K1]
refrigerant internal energy for zone j [J kg1]
refrigerant volume for zone j [m3]
boundary work for refrigerant zone j [W]
state vector
state vector at time t 0 (initial condition)
vapor quality [dimensionless]
vapor quality exiting the two-phase zone
[dimensionless]
time derivative of the state vector
average air-side heat transfer coefficient
[W m2 K1]
average refrigerant-side heat transfer coefficient
for zone j [W m2 K1]
local void fraction [dimensionless]
mean void fraction [dimensionless]
equilibrium mean void fraction for complete
condensation from saturated vapor to saturated
liquid [dimensionless]
air-side fin efficiency [dimensionless]
refrigerant-side fin efficiency for zone j
[dimensionless]
fraction of heat exchanger length covered by zone
j, also called normalized zone length
[dimensionless]
minimum normalized zone length before
switching occurs [dimensionless]
normalized length of excess liquid volume in zone
2 [dimensionless]
refrigerant saturated liquid density (at pressure P)
[kg m3]
refrigerant saturated vapor density (at pressure P)
[kg m3]
average refrigerant density for zone j [kg m3]
velocity of the interface between zone 1 and zone 2
[m s1]
velocity of the interface between zone 2 and zone 3
[m s1]
refrigerant saturated liquid absolute viscosity (at
pressure P) [Pa s]
refrigerant saturated vapor absolute viscosity (at
pressure P) [Pa s]

certain design constraints must be met. For example, to prevent compressor failure, the refrigerant must enter as a superheated vapor and exit at acceptable discharge temperature.
Also, for system stability, the refrigerant must enter the

1255

international journal of refrigeration 31 (2008) 12531264

expansion device as a liquid. Hence, heat exchanger design


and control clearly play central roles in both system performance and reliability.
Computer simulation is commonly used to accelerate development, and this paper focuses on modeling for controls
and diagnostics purposes. As discussed by Bendapudi and
Braun (2002), heat exchanger modeling approaches range
from finite volume/finite difference to moving-boundary
methods, including techniques which are a hybrid of the
two. Many control applications (e.g. hardware-in-the-loop/
software-in-the-loop simulation, embedded system emulators) require that the code run in real-time. While finite volume or finite difference models are fully capable of
predicting system transients (Cullimore and Hendricks, 2001;
MacArthur and Grald, 1989; Eborn et al., 2005; Limperich
et al., 2005), Grald and MacArthur (1992) showed that the
lumped parameter or moving-boundary method is
much faster. Even though real-time capability using a finite
volume technique has been demonstrated by Rossi and Braun
(1999), the faster speed of the moving-boundary method has
caused it to become the method of choice for controls purposes (Rasmussen, 2006; Willatzen et al., 1998; He et al.,
1995; Leducq et al., 2003; Cheng and Asada, 2006; Jensen and
Tummescheit, 2002)
The primary reason for the speed difference is that the
moving-boundary method divides the heat exchanger into
a minimum number of zones (at most three), representing
regions where refrigerant is in the superheated vapor, saturated mixture, and sub-cooled liquid phases. Because refrigerant-side heat transfer coefficient and refrigerant density
vary substantially between these regions, zone lengths
must be resolved to accurately predict total heat transfer
rate, refrigerant mass, and refrigerant pressure. The moving-boundary method accomplishes this by directly predicting zone lengths. In contrast, the finite volume method
divides the heat exchanger into a number of stationary elements, and mesh resolution must be fine enough to resolve
superheated and sub-cooled zone lengths. Bendapudi et al.
(2005) showed that for a fixed, uniform mesh, approximately
15 elements are needed to achieve mesh independent
results, a fivefold increase from the moving-boundary technique. They also showed that the associated increase in
the number of states causes the finite volume code to run
a factor of 24 times slower than the corresponding
moving-boundary method (Bendapudi, 2004). While these
findings have not yet been shown to be universal, they are
not atypical.
Although attractive for controls applications (Rasmussen,
2006; Willatzen et al., 1998; He et al., 1995; Leducq et al., 2003;
Cheng and Asada, 2006; Jensen and Tummescheit, 2002) there
are still critical areas for improvement of existing movingboundary model formulations. Most importantly, many
models assume a fixed number of zones. Therefore, if the
volume of a zone becomes much smaller than the others,
the governing equation set becomes singular and a unique
solution cannot be found. This severely limits applicability
of the approach because the simulation fails. Secondly, published models are often based on the assumption of smooth
circular refrigerant tubes, thereby not accounting for design
features that enhance heat exchanger effectiveness.

To address these issues, this paper presents a new moving-boundary heat exchanger model for sub-critical vapor
compression cycles. For the sake of brevity, we present a version for condensers, although development of an analogous
one for evaporators is straightforward. Table 1 compares
this new formation with those in the literature. Accuracy
concerns of previous models are addressed by including
finned surfaces, nonlinear air temperature distributions,
and non-circular passages. In addition, robustness concerns
are met by dynamically switching between two- and threezone models as required. Despite additional complexity, the
model runs in real-time. Collectively, these changes are a significant step forward in accuracy, robustness, and operating
range of moving-boundary models.

2.

Modeling approach

Fig. 1 portrays the heat exchanger model from a mathematical


viewpoint. The time varying inputs or boundary conditions
are air mass flow rate, air inlet temperature, refrigerant inlet
and outlet mass flow rates, and refrigerant inlet enthalpy.
The model would be part of a vapor compression cycle simulation, so the inputs would be provided by models of other
components (e.g. compressors, expansion valves, pipes, junctions). Conditions within the heat exchanger at each instant
of time are represented by the following state vector:
x h1

h3

z1

z2

TW1

TW2

TW3

g

(1)

Referring again to Fig. 1, the time derivative of the state vector


is a function ( f ) of the state vector (x), the model inputs, and
numerous physical parameters. The solution is advanced in
time by integrating the derivative from known initial conditions (x(0)). Model outputs provide an interface to other system components and the user, and can be envisioned as
a function ( g) of the state vector. For example, air outlet temperatures, refrigerant pressure, and outlet refrigerant enthalpy can be considered model outputs.
To derive functions f and g, the heat exchanger is divided
into control volumes or zones (see Fig. 2). Interface boundaries between zones vary over time and are tracked by the
model hence, the term moving-boundary method. As described in the literature (Grald and MacArthur, 1992; Rasmussen, 2006; Willatzen et al., 1998; He et al., 1995; Leducq
et al., 2003; Cheng and Asada, 2006; Jensen and Tummescheit, 2002; Bendapudi, 2004), the governing equations are
derived by considering conservation of mass and energy
in each control volume. Simplifying assumptions are as
follows:
- The refrigerant flow is one-dimensional, compressible,
and unsteady.
- The refrigerant pressure is uniform within the heat exchanger. Therefore, the momentum equation is not
needed.
- Slip flow in the two-phase region can be modeled adequately through a void fraction correlation.
- The air entering the condenser has uniform velocity, temperature, and pressure.

1256

international journal of refrigeration 31 (2008) 12531264

Table 1 Comparison of lumped parameter models


Model
capability
Non-circular tubes
Fin effects
Structure thermal
resistance
Nonlinear air temp
distribution
Upwind treatment
for wall rezoning
Variable number of
zones (switching)
Condenser
Evaporator

Grald and
MacArthur
(1992)

He
et al. (1995)

Willatzen
et al. (1998)

Jensen and
Tummescheit
(2002)

Leducq
et al. (2003)

Bendapudi
(2004)

Rasmussen
(2006)

This
paper

YES
NO
NO

NO
NO
NO

NO
NO
NO

NO
NO
NO

YES
NO
NO

NO
NO
NO

NO
NO
NO

YES
YES
YES

NO

NO

NO

NO

YES

NO

NO

YES

NO

NO

NO

YES

NO

NO

NO

YES

NO

NOa

YES

NO

NO

YES

NO

YES

NO
YES

YES
YES

NO
YES

NO
YES

YES
YES

YES
YES

YES
YES

YES
NO

a Switching capability for the He model is described by Cheng and Asada (2006).

- The air flow is quasi-steady and incompressible.


- The heat exchanger is in a single pass, cross-flow
arrangement.
- Structure internal energy can be adequately represented
using a constant specific heat and a single wall temperature for each of the three-zones (superheated, two-phase,
and sub-cooled).
- Conduction along the heat exchanger axis is negligible.
- Average air-side and refrigerant-side heat transfer coefficients are adequate to predict heat transfer rates.
The following sections describe the governing equations
for a condenser handling refrigerant and moist air. Under
nearly all operating conditions, the refrigerant entering the
condenser is a superheated vapor. This is true because the
inlet condition to the compressor must be a superheated
vapor, and the compression process (with its associated pressure and entropy increase) increases superheat. Assuming at
least some condensation, the outlet condition ranges from
a saturated liquidvapor mixture to a sub-cooled liquid.
This is modeled using either a two-zone (superheated and

two-phase zones) or a three-zone (superheated, two-phase,


and sub-cooled zones) representation. The authors also
derived a single zone representation, but it is omitted for
the sake of brevity.

2.1.

Governing equations: air flow

The air flow is assumed to be incompressible, with uniform


inlet temperature, pressure, and velocity over the condenser

mR-IN

1LR
z

hR-IN

QR1

Superheated
Zone
m12

TW1
QA1

h2g

QR2
Physical Parameters

2LR

ACR ,ASA, ASR,...

TW2

Two-Phase
Zone

QA2

Inputs

mA
TA-IN

m23

P
x(0)

hR-OUT

mR-IN
mR-OUT

h2f

TA-OUTj

3LR

QR3
Sub-Cooled
Zone

Outputs
mR-OUT

hR-IN
Fig. 1 Mathematical model.

Fig. 2 Heat exchanger schematic.

TW3
QA3

1257

international journal of refrigeration 31 (2008) 12531264

face. We also assume the specific heat is constant and the


flow is unmixed. In many applications, the mass and
specific heat of the air are much smaller than that of the
structure or the refrigerant. Therefore, energy storage in
the air can be neglected. Under these constraints, the air
temperature profile from the inlet to the outlet of the heat
exchanger at each instant of time can be derived from the
equations for steady, one-dimensional flow. Furthermore,
if the heat transfer coefficient is constant over the width
of the heat exchanger, the inlet and outlet temperatures
are related by


(2)
TA-OUTj TWj TA-IN  TWj expNTU

the temperature crossing the boundary is that of the zone


itself. In a coordinate system attached to the heat exchanger, normalized interface velocities are as follows:
v12
v23

dz1
dt

(8)



dz
dz1 dz2

v12 2
dt
dt
dt

(9)

Knowing these velocities, temperatures are assigned as


follows:
If v12 > 0 :

TWT1 TW2

(10a)

If v12  0 :

TWT1 TW1

(10b)

where
NTU

aA ASA 1  FFIN-A 1  hFA 


_ A cPA
m

(3)

The heat transfer rate from the air to the structure, for each
zone, can be calculated from


_ A zj cPA TA-IN  TA-OUTj
(4)
Q_ Aj m

2.2.

Governing equations: structure

Heat transfer normal to the heat exchangers z-axis (see Fig. 2)


is caused by air-side and refrigerant-side convection as well as
conduction along fins. Since the surface area is many times
greater than the cross-section area, conduction along the
length of the heat exchanger can be neglected.
Assuming the wall specific heat is constant, conservation
of mass and energy can be combined to derive state equations
for the average wall temperature of each zone. For a threezone representation:


1 Q_ A1  Q_ R1
dz
dTW1
(5)
TWT1  TW1 1
z1 cW mW
dt
dt

1 Q_ A2  Q_ R2
z2

cW mW

(11a)

If v23  0 :

(11b)

TWT2 TW2

This technique is analogous to the upwind or donor cell method


used in computational fluid dynamics. By summing Eqs. (5)(7)
for arbitrary values of interface velocities, it can be shown that
the method possesses the conservative property.
The wall-to-refrigerant heat transfer rate for each zone is
computed from


(12)
Q_ Rj zi URj ASR TWj  TRj
Taking into account thermal resistances of the wall and fins,
and assuming a uniform heat transfer coefficient over each
zone, the overall heat transfer coefficient is given by
URj

1
tW
1

h
kW 1  FFIN-R aRj 1  FFIN-R 1  h

(6)
(7)

Terms containing time derivatives of normalized zone


lengths represent energy transport due to boundary movement or wall rezoning. The temperatures within these terms
must be chosen carefully so that energy is conserved on an
integral basis. By that, we mean the time derivative of total
wall energy (calculated using the weighted average of the
wall temperatures of each zone) must equal the total net
heat transfer rate into the wall (calculated by summing the
net heat transfer rate into each zone). Such a method is
said to possess the conservative property.
To accomplish this objective, the temperature transported between neighboring zones is chosen based on the
interface velocity. Specifically, when the rightmost boundary
of a zone moves to the right, the temperature crossing the
boundary is that of the neighboring zone to the right. Correspondingly, when the rightmost boundary moves to the left,

(13)
i
FRj

For a two-zone representation, the sub-cooled zone is inactive. In that case, Eq. (7) is replaced with the following pseudo
state equation:
KT TW2  TW3




dz1 dz2
dz
dz
dTW2

TWT2  1 TWT1  TW2 2

dt
dt
dt
dt
dt




1 Q_ A3  Q_ R3
dz1 dz2
dTW3

TW3  TWT2
z3 cW mW
dt
dt
dt

If v23 > 0 TWT2 TW3

dTW3
dt

(14)

This equation causes the wall temperature for the inactive


sub-cooled zone to track the wall temperature of the active
two-phase zone. Its function is to provide a reasonable initial
condition, should the representation change from two-zones
to three-zones.

2.3.
Governing equations: refrigerant
(three-zone representation)
The governing equations for the refrigerant are now given, assuming three-zones are present. Only the final forms of these
equations are given in this section, although details concerning their derivation are included in Appendix. For the superheated zone, the continuity and energy equations are as
follows:
_ 12
_ R-IN
dz1 z1 vr1 dP z1 vr1 dh1
m
m

dt r1 vP dt r1 vh1 dt
r1 ACR LR r1 ACR LR

(15)



_ R-IN hR-IN  h1
hg  h1
Q_ m
dh1 1 dP
_ 12 R1


m
r1 dt r1 ACR LR z1
r1 ACR LR z1
dt

(16)

1258

international journal of refrigeration 31 (2008) 12531264

(15)(18) still apply. However, continuity and energy equations


for the two-phase zone become

Also
h1

hR-IN hg
2

(17)


Differentiating


dh1 1 dhR-IN dhg
1 dhR-IN 1 vhg dP

2
2 dt
2 vP dt
dt
dt
dt

(18)

For the two-phase zone, the continuity and energy equations


are as follows:
_ 23
_ 12
dz2 z2 vr2 dP
z vr dg
m
m


2 2
dt r2 vP dt r2 ACR LR r2 ACR LR r2 vg dt






hf  h2
hg  h2
vh2 1 dP
_
_ 12


m23 
m
r2 ACR LR z2
vP r2 dt r2 ACR LR z2
Q_ R2
vh2 dg

vg dt r2 ACR LR z2

(19)

20

Wedekind et al. (1978), who pioneered the use of mean void


fraction for lumped parameter models, demonstrated that
for fixed inlet and outlet vapor qualities, the time derivative
of this parameter can be neglected (Beck and Wedekind,
1981). However, as is discussed below, it is advantageous to include it for switching purposes. Specifically, for the three-zone
representation we use
vgTOT dP
dg
 Kg g  gTOT
dt
vP dt

(21)

This equation was derived by noting that the equilibrium


value of mean void fraction is a function of pressure, and including a gain for relaxation to that value. The gain was set
to five (equivalent to a time constant of 200 ms) as a compromise between equilibrium tracking and run time. This value is
not necessarily universal, and may depend on heat exchanger
dynamics and time derivatives of model inputs.
For the sub-cooled zone, the continuity and energy equations are as follows:
_ 23
_ R-OUT
dz1 dz2 1  z1  z2 vr3 dh3
m
m

r3
dt
dt
vh3 dt
r3 ACR LR r3 ACR LR


(22)

hf  h3
dh3
1 dP
_ 23 

m
r3 dt
r3 ACR LR 1  z1  z2
dt
_ R-OUT hR-OUT  h3
Q_ R3  m

r3 ACR LR 1  z1  z2

_ 12
_ R-OUT
dz1 z2 vr2 dP
z vr dg
m
m



2 2
dt r2 vP dt r2 ACR LR r2 vg dt
r2 ACR LR





hg  h2
vh2 1 dP
vh dg
_ 12 2


m
vP r2 dt r2 ACR LR z2
vg dt
_ R-OUT h2-OUT  h2
Q_ R2  m

r2 ACR LR z2

(24)

25

Since the sub-cooled zone is inactive (constant length):


dz1 dz2

0
dt
dt

(26)

Two-phase zone outlet enthalpy (in Eq. (25)) is determined


from pressure and mean void fraction. More details on this
procedure are given in Appendix.
In contrast with the three-zone representation, the time
derivative of mean void fraction can be significant. However,
its value is determined by the aggregation of the continuity
and energy equations instead of Eq. (21). This is detailed in
the following section.
Since the sub-cooled zone is inactive, Eqs. (22) and (23) are
replaced with the following pseudo state equation:


dh3
Kh hf  h3
dt

(27)

This causes the sub-cooled zone enthalpy to track saturated


liquid enthalpy, ensuring a reasonable initial condition
should the representation change from two-zones to threezones.

3.

Switching criteria

The pseudo state equations allow the representations to share


a common state vector (Eq. (1)). Therefore, switching only entails selecting the equation set to be solved at the next time
step.
Specifically, a switch from three- to two-zone representation is triggered when
1  z1  z2 z3 < zmin

(28)

And also

23

In Eq. (22), the liquid-phase refrigerant was assumed to be incompressible. However, changes in density with enthalpy are
included, so the model accounts for thermal expansion. In Eq.
(23), the outlet enthalpy is equal to h3 when the outlet flow rate
is positive. When the outlet flow rate is negative, the enthalpy
is a boundary condition provided by the neighboring downstream component.

2.4.
Governing equations: refrigerant (two-zone
representation)
The governing equations for the two-zone representation are
now presented. Since the superheated zone is unaffected, Eqs.

dz1 dz2 dz3




<0
dt
dt
dt

(29)

These conditions can be stated as, The normalized subcooled zone length is nearly zero and continuing to decrease.
It was found that the three-zone representation could easily
handle normalized sub-cooled zone lengths of 0.01, but near
0.001 the linear equation solver used in the solution procedure
(see the following section) issued a warning message regarding matrix singularities. The switching threshold needed to
lie within this range, and as a starting point 0.005 was chosen.
Although the threshold is dependent on numerical precision
and solution algorithms, clearly 0.005 is small compared to total heat exchanger length.
Switching from two- to three-zone representation occurs
when
dz z2 gTOT  g > zmin

(30)

1259

international journal of refrigeration 31 (2008) 12531264

and also
dg
<0
dt

Known boundary conditions


(mA,TA-IN,mR-IN,mR-OUT,hR-IN)

(31)

If mean void fraction is below the equilibrium value for complete condensation from saturated vapor to liquid, the term
inside parentheses in Eq. (30) will be positive. This means
there is excess liquid volume in the two-phase zone. The value
on the left-hand side of Eq. (30) is the normalized length of the
excess liquid volume. So, these conditions can be stated as,
The mean void fraction indicates there is noticeable excess
liquid volume inside the two-phase zone and it continues to
accumulate. As a starting point, the amount of excess liquid
which can be tolerated was chosen to be the same as the minimum value for switching from three-zone to two-zone
representations.
Eq. (30) explains why the three-zone representation allows
mean void fraction to evolve over time through Eq. (21). Originally, we simply used the equilibrium value. Eq. (30) shows
that this causes a jump in mean void fraction when switching
to a three-zone representation. This jump leads to the loss of
a corresponding amount of liquid refrigerant. By using Eq. (21),
we ensure mean void fraction is continuous and therefore
refrigerant mass is conserved. This is another unique feature
of our formulation.
Since refrigerant charge is an important parameter for vapor compression cycles, mass conservation is a very desirable
property. For the sake of robustness, we allow deactivated
zone lengths to stay near 0.005 and thereby prevent singularities when switching from a two-zone representation back to
a three-zone representation. However, since deactivated
zones do not contribute to the total heat transfer experienced
by the air or refrigerant, this is equivalent to reducing heat
transfer area. Although this introduces an error in the computed total heat transfer rate, it is no more than 0.5%. As discussed below, the expected accuracy of the code is 5%, so
this is not a significant contributor to total model error.

Known state vector, x


(eqn (1))
Calculate properties
of air and refrigerant
Calculate transfer rates
(eqn (2)-(4), (12)-(13))

3
Zones?

YES

NO

Form Ay=B
(eqn (15),(16),(18)-(23))

Form Ay=B
(eqn (15),(16),(18),(24),(25))

Solve for y
(eqn (32))

Solve for y
(eqn (33))

Solve for:
(eqn (26), (27))

d 2
dt

dh3
dt

dTWj
Solve for:
(eqn (5)-(11), (14)) dt

dTWj
Solve for:
(eqn (5)-(11)) dt

Check for switch to


2-zones (eqn (28),(29))

Check for switch to


3-zones (eqn (30),(31))

Integrate state vector


forward in time
Repeat

4.

Solution procedure

Fig. 3 outlines the solution procedure, with some of the more


salient features detailed below. It can be implemented in any
software package or language which is capable of solving linear equation sets and ordinary differential equations.
Knowing the state vector at each instant of time, all
thermo-physical properties and their partial derivatives can
be calculated through property relationships. By the state postulate, this requires knowledge of two independent, intensive
properties, which are as follows:
- Superheat zone: pressure and enthalpy.
- Two-phase zone: pressure and mean void fraction.
- Sub-cooled zone: pressure and enthalpy.
Heat transfer coefficients for each zone are then computed. For the superheated and sub-cooled zones, a Colburn
modulusReynolds number correlation is used along with
the assumption that mass flow rates in the superheated
and sub-cooled zones are the inlet and outlet mass flow

Fig. 3 Solution procedure flow chart.

rates, respectively. The heat transfer coefficient in the twophase zone is found empirically. Fin efficiencies are determined from well known equations (Incropera et al., 2007).
In turn, the wall-to-refrigerant heat transfer is calculated
from Eq. (12). Air-to-wall heat transfer is computed from
a Colburn modulusReynolds number correlation, fin efficiency equations, and Eqs. (2)(4).
For the three-zone representation, Eqs. (15), (16), and (18)
(23) form a set of eight linear equations (Ay B) in the vector:

y

dz1
dt

dz2
dt

dP dh3
dt dt

dh1
dt

dg
dt

T
_ 23
_ 12 m
m

(32)

Note that as the length of the sub-cooled zone 1  z1  z2


approaches zero, terms in Eqs. (22) and (23) approach zero or
infinity. These cause matrix singularities which force us to
switch to a two-zone representation when sub-cooled zone
length is still a small, positive value.

1260

international journal of refrigeration 31 (2008) 12531264

The vector y includes the derivatives of six states, among


them the interface velocities. This allows Eqs. (5)(11) to be
solved for the time derivatives of wall temperatures. Hence,
all state derivatives are known and the solution can be integrated in time.
For the two-zone representation, Eqs. (15), (16), (18), (24), and
(25) form a set of five linear equations (Ay B) in the vector:

T
dh1 dg
(33)
y dz1 dP
_ 12
m
dt dt
dt dt
This vector includes the derivatives of four states. Eqs. (26)
and (27) are then used to calculate two more state derivatives.
Since interface velocities are known at this point, Eqs. (5), (6),
(8)(11), and (14) are solved for the time derivatives of the wall
temperatures. Hence, all state derivatives are known and the
solution can be integrated in time.

5.

Model integrity check

Model integrity is the consistency of the computer simulation


and the governing equations, and was checked in three
ways. First, hand calculations were used to ensure steadystate and transient predictions agreed with the governing
equations.
Second, steady-state model results were compared to those
of an independent one-dimensional, finite volume code. The

geometry, which is proprietary to the sponsor of this work,


was a plate type heat exchanger similar to that of Table 3.
For a given heat exchanger design point, lumped parameter
model predictions of total heat transfer rate and refrigerant
outlet temperature agreed within 3% and 2.1  C, respectively,
of one-dimensional code results. Therefore, the movingboundary model is consistent with the one-dimensional
code. A much more extensive evaluation of models and test
data was completed by Bendapudi (2004), who for a shelland-tube heat exchanger showed agreement within 5%
between moving-boundary models, finite volume models,
and test data over a wide range of operating conditions for
refrigerant pressures and temperatures.
Pressure drop through the condenser was predicted by the
one-dimensional code to be 0.2% of the absolute inlet pressure, which corresponds to a change in saturation temperature of only 0.35  C. Therefore, for this particular design and
set of boundary conditions, the assumption of uniform pressure within the heat exchanger did not significantly compromise model accuracy. However, this conclusion would have
to be evaluated on a case-by-case basis.
Third, consistency of results with integral forms of the continuity and energy equations was checked through several
test cases. Table 2 summarizes the test suite, which includes
both two- and three-zone representations. After exciting the
model with different waveforms, the predicted mass and internal energy were compared to values from the integral

Table 2 Check of model results vs integral continuity and energy equations


Test case 1

Test case 2

Test case 3

Test case 4

Test case 5

Test case 6

Number of zones

Three

Three

Three

Two

Two

Two

BC waveform
Air mass flow rate
Inlet air temperature
Inlet ref mass flow rate
Outlet ref mass flow rate
Inlet ref enthalpy

Square
Constant
Constant
Constant
Constant

Constant
Constant
Square
Square
Constant

Constant
Constant
Constant
Constant
Sawtooth

Square
Constant
Constant
Constant
Constant

Constant
Constant
Square
Square
Constant

Constant
Constant
Constant
Constant
Sawtooth

Conservation checks
Refrigerant mass (kg)
Initial condition
Amplitude (model)
Amplitude (integral equation)
Error (% of initial condition)

0.09870
0.00000
0.00000
0.0004

0.09870
0.03000
0.03000
0.0000

0.09870
0.00000
0.00000
0.0030

0.02870
0.00001
0.00000
0.0254

0.02870
0.00600
0.00600
0.0002

0.02870
0.00000
0.00000
0.0096

Refrigerant int. energy (kJ)


Initial condition
Amplitude (model)
Amplitude (integral equation)
Error (% of initial condition)

11.5308
0.5832
0.5809
0.0201

11.5308
2.8281
2.8240
0.0354

11.5308
0.1185
0.1268
0.0719

5.2966
0.2555
0.2494
0.1164

5.2966
0.7645
0.7708
0.1186

5.2966
0.1374
0.1379
0.0104

Wall int. energy (kJ)


Initial condition
Amplitude (model)
Amplitude (integral equation)
Error (% of initial condition)

218.0421
21.7485
21.8380
0.0410

218.0421
3.4757
3.4757
0.0000

218.0421
12.0449
12.1536
0.0499

221.6947
20.4739
20.5580
0.0379

221.6947
9.8552
9.8552
0.0000

221.6947
21.1603
21.1603
0.0000

Wall ref int. energy (kJ)


Initial condition
Amplitude (model)
Amplitude (integral equation)
Error (% of initial condition)

229.5729
22.3311
22.4177
0.0377

229.5729
6.2648
6.2631
0.0007

229.5729
12.1611
12.1859
0.0108

226.9913
20.7288
20.8028
0.0326

226.9913
10.6196
10.6263
0.0029

226.9913
21.2936
21.2946
0.0005

1261

international journal of refrigeration 31 (2008) 12531264

forms of the continuity and energy equations. Peak-to-peak


amplitudes in mass and energy were compared, and error
was expressed as a percentage of the initial value. This normalization was chosen because it is comparable to error criteria used by the variable time-step RungeKutta integrator.
Model error is non-zero for several reasons, among them
integration errors, inconsistency in property relationships
(e.g. internal energy vs enthalpy), and round-off. Nevertheless, errors were always less than 0.15% and most of the
time less than 0.01%. Therefore, no significant errors are apparent in the code or the governing equations.

trace (see Fig. 4) shows that the model switched representations eight times. As expected for this periodic input, zone
length and refrigerant pressure also varied periodically.
Shortly after switching to a three-zone representation, zone
length transitions to 0.01, since excess liquid volume in the
two-phase zone is appended to the sub-cooled zone by the action of Eq. (21). A corresponding test case with the initial conditions in the two-zone rather than the three-zone regime
provided similar results.

7.
Model stability check

Model stability was checked by holding air and refrigerant inlet conditions constant, but varying refrigerant outlet flow rate
sinusoidally to force repeated switching. Boundary conditions, initial conditions, and physical parameters for this
test case are described in Table 3. The sub-cooled zone length

Table 3 Model stability test case


Flow arrangement
Pass arrangement
Construction
Boundary conditions

Cross-flow
Single, unmixed
Plate type
Air

R-134a

Inlet flow rate (kg s1)


Inlet temperature ( C)
Inlet pressure (kPa)
Inlet humidity ratio (kg kg1)
Outlet mean flow rate (kg s1)
Outlet flow rate
amplitude (peak-to-peak, kg s1)
Outlet flow rate frequency (rad s1)

1.000
41
100
0.0468
1
0

0.060
70
1500
NA
0.060
0.006

Physical parameters

Air-side

Ref-side

Wall properties
Mass (kg)
Thickness (mm)
Specific heat (kJ kg1 C1)
Thermal conductivity (kW m1 C1)

2.679
1.149
Not reqd.
1.000
8.797E02
8.775E04
6.727
2.906
Offset strip fin (both sides)
0.854
0.659
0.102
4.763
5.51
1.143

0.152
0.953
9.45
3.175

3.835
0.406
0.875
0.173

Correlations
Single-phase heat transfer
correlation ()
Two-phase heat transfer
coefficient (kW m2 C1)
Void fraction correlation ()
a Fin length 1/2  passage height.

Sub-Cooled Zone Length


Refrigerant Pressure
0.10

1600

0.08

1500

0.06

1400

0.04

1300

0.02

1200

Manglik and Bergles (1995)

0.00
10.53

50

100

150
200
Time (s)

250

Zivi (1964)

Fig. 4 Sub-cooled zone length and refrigerant


pressure model stability test case.

1100
300

Pressure (kPa)

Hydraulic diameter (mm)


Passage length (m)
Cross-sectional area (m2)
Total surface area (m2)
Fin type ()
Fraction of surface area
on fins (in decimal)
Fin thickness (mm)
Fin lengtha (mm)
Fin density (fin cm1)
Fin offset length (mm)

The condenser model was applied in a sample chilled water


system using R-134a refrigerant, a waterpropylene glycol
mixture as a coolant, and moist air as a heat sink (see Fig. 5).
Electronic expansion valve voltage was varied using a PI controller with feedback on chiller coolant outlet temperature
and a set point of 11.2  C. Compressor speed was varied by another PI controller to balance compressor and expansion valve
flows. The system was allowed to stabilize for 30 s, then load
decreased by 25%. Actuator inputs and system output are
shown in Fig. 6.
The entire system shown in Fig. 5 was modeled using the
Thermosys Matlab/Simulink toolbox (Alleyne et al., 2007).
Models for vapor compression cycle components are detailed
by Rasmussen (2006). Previous versions of Thermosys
employed Rasmussens condenser models, which were only
capable of a fixed number of zones. As shown in Table 1, the
model described in this paper is a significant improvement.
As mentioned previously, real-time simulation is a necessity for HIL and SIL simulations. Run time is a function of
the system being modeled (e.g. number of states, eigenvalues),
computer hardware, computer software, integration error
bounds, and programming technique. In this case, running
under Matlab 7.1 on a Pentium M (1.86 GHz clock rate) with

Normalized Zone Length

6.

Sample application

1262

international journal of refrigeration 31 (2008) 12531264

Ambient Air

Comp Speed
Valve Input
Chiller Outlet Temp

Receiver

Compressor

Exp
Valve

Liquid

13.0

1.0

12.5

0.8

12.0

0.6

11.5

0.4

11.0

0.2

10.5

0.0

20

40

100

10.0
120

Fig. 6 Sample system inputoutput response to step


change in chiller load.

Exp
Tank

Load
Fig. 5 Sample chilled water system schematic.

1 GB RAM (782 MHz access rate), the simulation of the entire


system runs 1.56 times faster than real-time. The integrator
was a common RungeKutta method, and the computer was
a common laptop PC. The only special programming technique was a custom interpolation function for calculating
refrigerant thermo-physical properties. This function used
the well known shape function method used in finite element
codes (Segerlind, 1984).
Fig. 7 shows normalized zone lengths during the run.
Sub-cooled zone length peaked at 0.28 near t 54 s. The
expansion valve is closing at this time and valve input leads
compressor speed input. As a result, more refrigerant is
entering and condensing in the condenser than leaves it,
and a plug of liquid refrigerant forms. To accurately model
this operating condition, a three-zone representation is
required. If a two-zone model were used instead, the twophase heat transfer coefficient would extend over the subcooled region and the outlet enthalpy would be limited to
the saturated liquid value.
Near t 70 s, the sub-cooled region is deactivated. This is
due to overshoot in valve and compressor inputs, causing
more refrigerant to leave the condenser than is supplied and
therefore condensed. As a result, the plug of liquid refrigerant
is drained away. If a three-zone model were used, the simulation would abort at this time due to a matrix singularity.

However, our switched model overcomes this obstacle, breaking the trade-off between accuracy and robustness.

8.

Conclusions

This paper describes an advanced, nonlinear, moving-boundary heat exchanger model for vapor compression cycles. It includes several improvements to model accuracy and stability,

Sub-Cooled Zone
Superheated Zone
0.40
0.35

Normalized Zone Length

Pump

80

Time (s)

Receiver
Chiller

60

Temperature (C)

Normalized Compressor Speed


Normalized Exp Valve Voltage

Condenser

1.2

0.30
0.25
0.20
0.15
0.10
0.05
0.00

20

40

60

80

100

Time (s)
Fig. 7 Sample system zone length response to step
change in chiller load.

120

international journal of refrigeration 31 (2008) 12531264

yet still runs in real-time. Original techniques for switching


were presented, and by including fins and non-circular flow
passages it is applicable to both finned tube bundle and plate
type heat exchangers. As described in Appendix, the model
formulation is also general enough to accommodate a wide
range of void fraction correlations. Although the Zivi (1964)
correlation was used in the sample application, different correlations (ASHRAE, 2005) can be used by simply changing the
function used to calculate g. The same is true of heat transfer
coefficients by simply changing the function used, a wide
range of correlations and geometries can be represented. In
either case, there is no need to reformulate the governing
equations.
Model predictions agree well with steady-state one-dimensional code results, showing that overall thermal resistance is
accurately represented. Assuming thermal capacitances (e.g.
refrigerant and structure masses, specific heats, latent heats)
are also adequately represented, the model will give accurate
transient predictions. In fact, moving-boundary techniques
have already been shown to provide reasonably accurate transient predictions (Grald and MacArthur, 1992; Rasmussen,
2006; Leducq et al., 2003; Bendapudi, 2004; Alleyne et al.,
2007). Nevertheless, further experimental validation against
both steady-state and transient data is planned.
Test cases demonstrate model integrity and stability. The
importance of switched representations was illustrated in
a sample application.
In its current form, the model is specific to condensers. Future work will also include development of a corresponding
evaporator model and extension to start-stop cycles.

Appendix
In this section, more information is provided on the derivation
of the governing equations for the refrigerant and implementation of the mean void fraction correlation.
The governing equations for the refrigerant were derived
from a control volume analysis of each zone. This is detailed
below for the superheated zone, and an analogous approach
was taken for the two-phase and sub-cooled zones. The continuity equation is
_ 12
_ R-IN  m
m

dm1
d
dr
dz
r1 ACR LR z1 ACR LR z1 1 r1 ACR LR 1
dt
dt
dt
dt
(34)

In the superheated zone, density can be considered a function


of pressure and enthalpy. By the chain rule:
dr1 vr1 dP vr1 dh1

dt
vP dt vh1 dt

(35)

Substituting Eq. (35) into Eq. (34) and rearranging terms yields
Eq. (15).
Applying continuity to the two-phase zone yields an analogous expression to Eq. (34), but this time density is considered to be a function of pressure and mean void fraction. By
the chain rule:
dr2 vr2 dP vr2 dg

dt
vP dt vg dt

(36)

1263

Combining Eq. (36) with the equivalent of Eq. (34) yields Eqs.
(19) and (24).
Applying continuity to the sub-cooled zone results in an
analogous expression to Eq. (34). Density is a function of pressure and enthalpy, but the relationship to pressure is weak.
The chain rule becomes
dr3 vr3 dP vr3 dh3 vr3 dh3

y
dt
vP dt vh3 dt vh3 dt

(37)

Combining Eq. (37) with the equivalent of Eq. (34) yields


Eq. (22).
The energy equation for the superheated zone is


d
d
d
P
_ 1 m
_ R-IN hR-IN  m
_ 12 hg m1 u1 m1 h1 
m1
Q_ R1 W
dt
dt
dt
r1
(38)
Substituting the definitions of boundary work and density into
this equation gives
dV1
d
d
_ R-IN hR-IN  m
_ 12 hg m1 h1  PV1
m
Q_ R1  P
dt
dt
dt

(39)

Applying the chain rule to the right hand side:


dV1
dh
dm1
dV1
dP
_ R-IN hR-IN  m
_ 12 hg m1 1 h1
m
P
 V1
Q_ R1  P
dt
dt
dt
dt
dt
(40)
Because the boundary work term appears on both the left and
right hand sides, it can be cancelled. Also
V1 ACR LR z1

(41)

Combining Eqs. (40), (41), and (34) yields Eq. (16).


The energy equation for the two-phase zone is derived in
an analogous manner, with the exception that the enthalpy
is a function of pressure and mean void fraction. Therefore,
by the chain rule:
dh2 vh2 dP vh2 dg

dt
vP dt
vg dt

(42)

A similar approach for the sub-cooled zone results in Eq. (23).


Next, some details regarding the mean void fraction are
provided. A number of different correlations relating local
void fraction and local vapor quality are of the form (see ASHRAE, 2005):
!a1
!a2
!a3 #1
"
rg
1  x~
mf
(43)
g 1 a0
x~
rf
mg
In the case of the Zivis correlation, a0 1, a1 1, a2 0.67, and
a3 0. Eq. (43) can be written in an even more general form as
g gx~; P

(44)

Now define a new coordinate j which is parallel to z (see


Fig. 2), with its origin at the start of the two-phase zone, a value
of one at the end of the two-phase zone, and increasing linearly along the heat exchanger. Since the refrigerant cross-sectional flow area is assumed to be constant, the mean and local
void fractions are related by
Z 1
g
gx~j; Pdj
(45)
0

To calculate the integral, we need a relationship between


vapor quality and j. At the start of the two-phase zone the
quality equals one, and at the end it equals x~OUT . Although

1264

international journal of refrigeration 31 (2008) 12531264

any appropriate relationship between these two points can be


assumed, in this case we assume this relationship is linear.
That is to say
x~ 1 j$x~OUT  1

(46)

Eqs. (44)(46) define a relationship between the mean void


fraction, the refrigerant pressure, and the outlet vapor fraction. For the sake of computational efficiency, we create tables
relating the three parameters before starting the simulation.
Since the refrigerant pressure and mean void fraction are
known at each instant of time, the tables are used to find
x~OUT . The outlet enthalpy from the two-phase zone is then determined from
hR-OUT hf 1  x~OUT hg

(47)

Note that this approach can be applied for a range of void


fraction correlations and vapor quality distributions, without
affecting the governing equations in the body of the report.

references

Alleyne, A., Rasmussen, B., Keir, M., Eldredge, B., 2007. Advances
in energy systems modeling and control. In: Proceedings of
the ACC. IEEE, New York, NY, USA, pp. 43634373.
ASHRAE, 2005. ASHRAE Handbook: Fundamentals, SI ed.
American Society of Heating, Refrigerating, and AirConditioning Engineers, Atlanta, GA, USA.
Bendapudi, S., Braun, J.E., 2002. A Review of Literature on
Dynamic Models of Vapor Compression Equipment. Technical
Report HL2002-8. Ray W. Herrick Laboratories, Purdue
University.
Bendapudi, S., Braun, J.E., Groll, E.A., 2005. Dynamic model of
a centrifugal chiller system model development, numerical
study, and validation. ASHRAE Trans. 111, 132148.
Bendapudi, S., 2004. Development and evaluation of modeling
approaches for transients in centrifugal chillers. PhD thesis,
Purdue University.
Beck, B.T., Wedekind, G.L., 1981. A generalization of the system
mean void fraction model for transient two-phase evaporating
flows. J. Heat Transfer 103, 8185.
Cullimore, B.A., Hendricks, T.J., 2001. Design and transient
simulation of vehicle air conditioning systems. SAE Paper
2001-01-1692.
Cheng, T., Asada, H.H., 2006. Nonlinear observer design for
a varying-order switched system with application to heat
exchangers. In: Proceedings of the ACC. IEEE, Minneapolis,
MN, USA, pp. 28982903.

Eborn, J., Tummescheit, H., Prol, K., 2005. Airconditioning


a modelica library for dynamic simulation of AC Systems. In:
Proceedings of the Fourth International Modelica Conference,
March 78, Hamburg-Harburg, Germany, pp. 185192.
Grald, E.W., MacArthur, J.W., 1992. A moving-boundary
formulation for modeling time-dependent two-phase flows.
Int. J. Heat Fluid Flow 13 (3), 266272.
He, X.D., Liu, S., Asada, H., 1995. Modeling of vapor compression
cycles for advanced controls in HVAC systems. In:
Proceedings of the ACC, vol. 5. IEEE, Seattle, WA USA, pp.
36643668.
Incropera, F.P., Dewitt, D.P., Bergman, T.L., Lavine, A.S., 2007.
Fundamentals of Heat Transfer, sixth ed. John Wiley & Sons,
New York, NY, USA, pp. 137162.
Jensen, J.M., Tummescheit, H., 2002. Moving boundary models for
dynamic simulations of two-phase flows. In: Proceedings of
the Second International Modelica Conference, Mar 1819,
Oberpfaffenhofen, Germany, pp. 235244.
Limperich, D., Braun, M., Schmitz, G., Prol, K., 2005. System
simulation of automotive refrigeration cycles. In: Proceedings
of the Fourth International Modelica Conference, March 78,
Hamburg-Harburg, Germany, pp. 193199.
Leducq, D., Guilpart, J., Trystram, G., 2003. Low order dynamic
model of a vapor compression cycle for process control
design. J. Food Process. Eng. 26, 6791.
MacArthur, J.W., Grald, E.W., 1989. Unsteady compressible twophase flow model for predicting cyclic heat pump
performance and a comparison with experimental data. Int. J.
Refrigeration 12 (1), 2941.
Manglik, R.M., Bergles, A.E., 1995. Heat transfer and pressure drop
correlations for the rectangular offset strip fin compact heat
exchanger. Experimental Therm. Fluid Sci. 10, 171180.
Rossi, T.M., Braun, J.E., 1999. A real-time transient model for air
conditioners. In: Proceedings of the IIR 28th Congress of
Refrigeration (Refrigeration into the Third Millennium),
Sydney, Australia, September 1924.
Rasmussen, B., 2006. Dynamic Modeling and Advanced Control of
Air Conditioning and Refrigeration Systems. PhD thesis,
University of Illinois at Urbana-Champaign.
Segerlind, L.J., 1984. Applied Finite Element Analysis, second ed.
John Wiley & Sons, New York, NY, USA, pp. 6164.
Willatzen, M., Pettit, N.B.O.L., Ploug-Sorensen, L., 1998. A general
dynamic simulation model for evaporators and condensers in
refrigeration. Part I: Moving-boundary formulation of twophase flows with heat exchange. Int. J. Refrigeration 21 (5),
398403.
Wedekind, G.L., Bhatt, B.L., Beck, B.T., 1978. A system mean void
fraction model for predicting various transient phenomena
associated with two-phase evaporating and condensing flows.
Int. J. Multiphase Flow 4, 97114.
Zivi, S.M., 1964. Estimation of steady-state steam void-fraction by
means of the principle of minimum entropy production. J Heat
Transfer 86, 247252.

También podría gustarte