International Journal of Multiphase Flow 36 (2010) 284–292

Contents lists available at ScienceDirect

International Journal of Multiphase Flow

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / i j m u l fl o w

Multi-dimensional simulation of thermal non-equilibrium channel flow

D.P. Schmidt a,*, S. Gopalakrishnan a, H. Jasak b
a
Mechanical and Industrial Engineering, 160 Governors Dr., University of Massachusetts, Amherst, MA 01003, United States
b
University of Zagreb, Croatia

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

Article history: The Homogenous Relaxation Model (HRM) is used to study thermal non-equilibrium, two-phase flows
Received 29 December 2008 with flash-boiling and condensation. Typically, such non-equilibrium phase-change models have been
Received in revised form 16 September studied in one-dimensional flow, but the goal of the present work is to create and utilize a multi-dimen-
2009
sional CFD implementation. The simulations are able to handle general polyhedral meshes, an important
Accepted 27 November 2009
Available online 11 December 2009
convenience for irregular channel or nozzle shapes. The model is applied to flash-boiling flow in short
channels and validated against experimental measurements. The simulations predict the multi-dimen-
sional features that have been observed in the past in experiments. Nozzle choking is also observed in
Keywords:
Flash-boiling
the calculations.
Flashing Ó 2009 Elsevier Ltd. All rights reserved.
Nozzle
Two-phase
CFD

1. Introduction is very small and is not significant when compared to the liquid
density. Thus little energy transfer is required to produce vapor.
When a hot fluid has a vapor pressure that falls between the up- However, in the nozzles that will be considered in the present
stream and downstream pressure in a nozzle, the discharge of the work, equilibrium assumptions would lead to predictions of unre-
nozzle may be sensitive to the effects of interphase heat transfer. alistically high velocities, on the order of a 1000 m/s.
This heat transfer will take place on small length scales and will In contrast, for hot liquid the phase change is more like a boiling
be affected by interfacial and turbulent dynamics. Neither the de- process. The difference between the saturated vapor density and
tails of the small-scale temperature fluctuations, the amount of saturated liquid density decreases at higher temperature. Conse-
interfacial area, nor the small scale velocity features are known. quently, the liquid must provide more energy per unit volume of
Despite these complexities, the limits of thermal equilibrium and vapor. Thus flashing nozzle simulations require additional model-
frozen flow have been useful for very long and very short nozzles, ing of finite-rate heat-transfer processes. Further distinctions are
respectively. An intermediate closure that addresses the finite rate provided by Sher et al. (2008), who reviewed and categorized typ-
of heat transfer between phases would provide wider applicability ical modeling approaches. Classic studies by Wallis (1980), Fauske
to nozzle geometries. If the analyses could further be extended to (1965), Henry and Fauske (1971) and Moody (1965) have explored
multiple dimensions, then multi-dimensional CFD techniques the role of thermal non-equilibrium in a variety of channel geom-
could be applied to studying flash-boiling nozzles. etries. In an interesting bridge between the two regimes, Vortmann
The rate of heat transfer and its role as a limiting factor in phase et al. (2003) have modeled cavitation with a return-to-equilibrium
change depends largely upon the temperature of the fluid. Pres- approach.
sure-driven phase change can be viewed as a spectrum with cavi- Kato et al. (1994) presented an analysis that indicates when ther-
tation at the cold end of the spectrum and flash-boiling at the hot mal effects limit bubble growth. Kato et al. numerically integrated
end. In some cavitating flows, the time scales of heat transfer can the Rayleigh–Plesset equation and the energy equation. For a bound-
be assumed to be much faster than the time scales governing accel- ary condition at the phase interface, Kato calculated the rate of en-
eration due to pressure (Knapp et al., 1970). Consequently, for ergy transferred out of the liquid by conduction, as the interface
small, high-speed cavitating flows, thermal equilibrium assump- produced vapor. The vapor production gave the growth rate of the
tions have produced successful cavitation models (Schmidt et al., bubble, and thus the wall velocity. One of their main observations
1999b). Under such conditions, the vapor density of the cold fluid was the significance of Jakob number and the change in governing
phenomena over the lifetime of a growing bubble.
* Corresponding author. Tel.: +1 413 545 1393. Mach number effects are another phenomenon thought to play
E-mail address: schmidt@ecs.umass.edu (D.P. Schmidt). an important role in the flashing of superheated fluids. Simões

0301-9322/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijmultiphaseflow.2009.11.012
 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292 285

Moreira and Bullard (2003) modeled high-speed jets emanating the effects of bubble nuclei was necessary. Their modified model
from short nozzles, where expansion waves formed downstream was able to reliably match mass flow rate measurements in nozzles
of a liquid core. They applied the solution of a Chapman–Jouguet with length-to-diameter ratios from 0.3 to 3.6.
wave to the process of flash-boiling and predicted choked flow In the current work, we choose not to rely on detailed models of
downstream of the wave. interfacial area, convection coefficient, and temperature field in the
Empirical observations are also essential. In experiments such turbulent, two-phase heat-transfer process. Given the nearly
as Reitz (1998), the mass flow rate through a short nozzle was intractable complexity of the detailed heat-transfer process, it is
clearly a function of upstream liquid temperature. As the temper- pragmatic to rely on an empirical model that encapsulates the
ature of the upstream liquid approached the vapor temperature at physics in simple correlation. The initial nuclei size and number
the upstream conditions, mass flow rate decreased. When heated density are not usually available, nor is the assumption of spherical
to a point just below the upstream vapor temperature, the flow bubbles always justifiable. As an alternative means of closure, the
rate dropped abruptly. Kim and O’Neal (1993) made observations Homogenous Relaxation model (Downar-Zapolski et al., 1996) is
of refrigerants flashing in short tubes. Another phenomena that employed. Like most proposed closures for two-phase channel
can occur in slightly subcooled flows are condensation shocks, as flow, this model was originally developed for flow in one-dimen-
observed in experiments by Yu et al. (1987). sion and has been mostly explored only for one-dimensional sce-
However, the complex physics are only the first obstacle to cre- narios. Duan et al. (2006) employed the Homogenous Relaxation
ating CFD simulations of phase change. Depending on the speed Model in simulating the evolution of external multi-dimensional
and size of the channel flow, the rate of heat transfer can range flow in a Lagrangian particle simulation. However, the correlation
from slow, e.g. the thermal equilibrium limit, to very fast, namely used by Duan et al. is several orders of magnitude slower than
the frozen-flow limit. When the rate of phase change is extremely the original correlation of Downar-Zapolski et al. The present work
fast, numerical stiffness problems can occur. Unless an implicit explores the flashing nozzle behavior in multiple dimensional
model of heat transfer is closely coupled to conservation of mass channel flow by constructing an Eulerian computational fluid
and momentum equations, the resulting scheme may be limited dynamics code around the Homogenous Relaxation Model.
to very small time steps. For application to transient, three-dimen- There is some reason to believe that such an extension of a one-
sional flow, severe stability constraints would render an explicit dimensional closure to multiple dimensions could be possible. In
model prohibitively expensive. the only example known to the authors of multi-dimensional
CFD calculations of internal flashing flow, Minato et al. (1995) used
a simple one-dimensional non-equilibrium two-phase flow analy-
2. Approach sis to close a two-fluid, two-dimensional, model of flashing flow.
Their approach was quite computationally expensive, limiting
Once it is decided to pursue modeling of thermal non-equilib- their investigation to extremely coarse meshes. This initial study
rium, one must next decide whether to employ a full two-phase has garnered no attention from other researchers (as measured
solution with separate transport equations or a pseudo-fluid ap- by subsequent citations) and has not prompted any further studies
proach where the mixture of phases is represented by a continuous in this area in the 14 intervening years since its publication. Given
density variable. The former approach offers complete generality, the limited computational resources of the time, the ability to cal-
including separate velocities for each phase, while the latter ap- culate two-dimensional flashing flow, even on their very coarse
proach offers relative simplicity and expediency. For an example mesh, is most remarkable.
of one-dimensional modeling using separate conservation equa- The present work investigates the potential of extending the
tions for each phase, see Boure et al. (1976). one-dimensional HRM approach to multiple dimensions, for use
For the current investigation, the pseudo-fluid approach was as closure of a multi-dimensional CFD code. In contrast to Minato,
employed. Though the inclusion of slip has been shown to be we neglect interphase slip on the sub-grid scale and use a pseudo-
important by Moody (1965) and by Henry and Fauske (1971) in fluid approach, saving the computational cost of solving separate
one-dimensional analyses, the pseudo-fluid approach still allows momentum equations. The success of this approach will offer
relative velocity between the phases on the resolved scales in mul- new possibilities for multi-dimensional simulation of flash-boiling
ti-dimensional CFD. For this reason, a no-slip model is less restric- flow. The challenge will be constructing a stable coupling between
tive in higher dimensions than in one-dimension. For example, an the HRM closure and the basic conservation equations.
annular flow might have low speed vapor surrounding a high
speed liquid core, which can be resolved with a no-slip model in
3. Derivation of governing equations
two-dimensions. Some of the limitations of this sub-grid no-slip
assumption will be investigated in the results.
In as much as possible, the flash-boiling flow simulation pre-
A benefit of the pseudo-fluid approach without assumption of
sented here relies on basic conservation laws. Given the assump-
slip is that no explicit model for interphase drag is required. By tak-
tion of no-slip within a cell, the pseudo-fluid approach produces
ing the limit of infinitely fast momentum exchange, one avoids the
the same basic conservation laws as for a single fluid. These are gi-
numerical problems of very high-drag rates and tight coupling be-
ven below for conservation of mass, momentum, and energy. In the
tween phase velocities, such as high computational cost and prob-
following equations, the variable / represents the mass flux and s
lems with numerical instability. The main risk of using the pseudo-
is the stress tensor. In the present study, only laminar flow is con-
fluid approach is that interphase momentum transfer will be over-
sidered, but the stress tensor does include Stokes’ hypothesis for
predicted.
treating the second coefficient of viscosity.
Given the assumption of no sub-grid slip, the emphasis then
shifts to the thermal non-equilibrium modeling. A successful @q
example of such an approach is the work of Valero and Parra þr/¼0 ð1Þ
@t
(2002), who employed an ”Equal Velocity Unequal Temperature” @ qU
!
!
for modeling one-dimensional critical two-phase flow. They closed þ r  ð/UÞ ¼ rp þ r  s ð2Þ
@t
the basic conservation equations using a model of heat and mass
transfer from spherical bubbles. They investigated their model pre- The energy equation is included, even though it is of little signifi-
dictions for short nozzles and found that a modification to include cance in the current work. All the simulations in the current study
286 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292

are run under adiabatic conditions and simulations proceed until a A slightly different fit is suggested for upstream pressures above 10
steady-state is reached. Hence, total enthalpy will be constant in bar, as given by Eq. (10).
these limits. However, in order to guarantee time–accuracy, an
equation for energy or enthalpy is required. The following form is H ¼ H0 aa /b ð10Þ
used, neglecting the kinetic energy of the fluid, viscous energy dis-
The dimensionless pressure /, defined in Eq. (11), differs from the
sipation, and conduction.
definition in Eq. (9) by including the critical pressure pc . The coeffi-
ð@ qhÞ @p ! cient values in the high-pressure correlation, Eq. (10) are
þ r  ð/hÞ ¼ þ U rp ð3Þ
@t @t H0 ¼ 3:84  107 [s], a ¼ 0:54, and b ¼ 1:76. Both correlations
Eqs. (1)–(3) are not a closed system of equations. In single-phase will be explored in this work.
flow, an equation of state would be required. However, where  
p p
non-equilibrium heat transfer governs much of the flow dynamics, / ¼  sat  ð11Þ
pc  psat 
there is no equation of state that would suffice. The two-phase mix-
ture represented by the pseudo-fluid assumption is not in thermo- In the present study, the flow at the channel inlets were pure liquid.
dynamic equilibrium. As explained above, our hypothesis is that a With no vapor present, the phase change timescale would be un-
relaxation to equilibrium would be an appropriate model for closing bounded and vaporization would never begin. To avoid numerical
the equations. For this purpose, we employ the Homogenous Relax- overflow and to provide a means of treating boiling incipiency, a
ation Model. very small lower bound of 1015 was applied. In all likelihood, dis-
The Homogenous Relaxation Model is based on a linearized solved gasses could provide an incipient void fraction in excess of
expansion proposed by Bilicki and Kestin (1990). The general mod- this value. As will be discussed in the conclusions section, this is a
el form originates with refrigeration modeling by Einstein (1920). likely area for future study.
It has been used by numerous others for one-dimensional two- If the continuity equation is used for solving for mixture density
phase flow. The model represents the enormously complex process and conservation of momentum is used for velocity, then Eq. (4) is
by which the two phases exchange heat and mass. The model form primarily responsible for determining the pressure. In contrast to
determines the total derivative of quality, the mass fraction of incompressible or low-Mach number Navier–Stokes solvers, the
vapor. current model does not seek a pressure that projects velocity into
Dx x  x consistency with the continuity equation. Instead, we solve for the
¼ ð4Þ pressure that satisfies the chain rule and employs the continuity
Dt H
equation indirectly. Through the chain rule, the pressure responds
Eq. (4) describes the exponential relaxation of the quality, x, to the to both compressibility and density change due to phase change.
equilibrium quality, x , over a timescale, H. The equilibrium quality The behavior of pressure is seen to be both hyperbolic and para-
is a function of the enthalpy and the saturation enthalpies at the lo- bolic, while the phase-change model appears as a source term.
cal pressure, as given by Eq. (5) with bounds at zero and unity. In order to provide close coupling with velocity, the momentum
h  hl equations and continuity equation are combined with Eq. (4) to
x ¼ ð5Þ provide a pressure equation. The procedure starts with conserva-
hv  hl
tion of mass and momentum, Eqs. (1) and (2), respectively. The
The quality, the mass fraction of vapor, is calculated from each cell’s next step is to discretize the momentum equation. This discretiza-
void fraction, a for densities falling inside the saturation dome. tion can take many forms, but they can all be represented generally
aqv using the form of Eq. (12).
x¼ ð6Þ
q
ap U p ¼ HðUÞ  rp ð12Þ
The void fraction in the two-phase region is, in turn, a function of
the local density as well as the saturated vapor and liquid densities This expression represents the discrete equation applied to each cell
at the local pressure. in the domain. The subscript p refers to the point of interest using
the notation of Ferziger and Peric (2002). The H operator represents
ql  q
a¼ ð7Þ convection and diffusion as discretized equation coefficients multi-
ql  qv plied by neighboring velocities plus source terms. The coefficient ap
The timescale in Eq. (4) is empirically fit to data describing is the coefficient term of the matrix of velocity equations.
flashing flow of water in long, straight pipes. The work of Dow- The chain rule can also be used to express the total derivative of
nar-Zapolski et al. (1996) provides two correlations, one recom- density, as in Eq. (13). The chain rule stands in place of the typical
mended for relatively high pressures, above 10 bar, and a equation of state, since this is a simulation of non-equilibrium
different correlation for lower pressures. In the low-pressure form, fluid. Note that for thermodynamic non-equilibrium, density is a
for upstream pressures below 10 bar, the best-fit values suggested function of three variables: pressure, quality, and enthalpy (Bilicki
by Downar-Zapolski et al. for flashing water appear in Eq. (8). The and Kestin, 1990).
empirical parameters include H0 and the two exponents. These   
Dq @ q Dp @ q Dx @ q Dh
values are H0 ¼ 6:51  104 [s], a = 0.257, and b = 2.24. ¼  þ  þ  ð13Þ
Dt @p x;h Dt @x p;h Dt @h p;x Dt
a b
H ¼ H0 a w ð8Þ
Currently, the last term in Eq. (13) is neglected due to the near-isen-
The variable a represents the volume fraction of vapor and w is a thalpic nature of the adiabatic channel flows currently considered.
dimensionless pressure difference between the local static pressure The first term on the right side represents a contribution to the den-
and the vapor pressure, as defined in Eq. (9). The absolute value is sity change due to two-phase compressibility. This two-phase com-
used in the present work since the pressure in the domain can fall pressibility is calculated as a mass average of the two single-phase
below the saturation pressure. compressibilities. This term could be significant in transonic flow. In
  cases where the two-phase compressibility is not significant, this
p  p
w ¼  sat  ð9Þ term can be omitted, which offers the advantage of producing a
psat  symmetric matrix for the discretized pressure equations. Calcula-
 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292 287

tions where the compressibility was neglected are explicitly men- The ratio of the off-diagonal terms to diagonal terms that ap-
tioned below. pear in Eq. (12) have dimensions of velocity and can be thought
If we subtract conservation of mass, Eq. (1), from Eq. (13) then of as a velocity field prior to pressure projection, as indicated in
the left side gives an expression for velocity divergence at the new Eq. (18).
time step.
 
@ q Dp @ q Dx H
qr  U ¼  þ  ð14Þ U ¼ ð18Þ
@p x;h Dt @x p;h Dt ap

Using Eq. (12) and U p in place of U, the momentum equation can be

coupled with the chain rule to produce an equation for pressure. This velocity is interpolated to face centers to produce a flux field,
     / , that is used in Eq. (20).
@ q @p @ q H 1 @ q
þ ðU  r pÞ þ q r   q r rp þ With multiple PISO iterations, the non-linearity of the momen-
@p x;h @t @p x;h ap ap @x p;h tum equation can be accommodated. However, the phase-change
Dx model presents an additional challenge: the last term in Eq. (15),

Dt representing the effects of the phase-change model is highly
¼0 ð15Þ non-linear and strongly dependent on pressure. As a shorthand,
we define this term as M in Eq. (19). Using linearization, the PISO
This is a mixed-character transient convection/diffusion equation. iterations also provide secant method iterations for semi-implicitly
The transmission of pressure waves, which is essential for any com- including the pressure, as shown in Eq. (20). The superscripts k and
pressible flow calculation, is allowed by the transient and convec- k þ 1 indicate the previous and current PISO iteration, respectively.
tive terms in the equation while the pressure is kept in range and
is damped by the Laplacian
 term. For low-Mach number flows the   

terms containing @@pq can be dropped. The terms were retained @ q x  x
M ð19Þ
x;h
for some calculations but dropped for other calculations since they @x p;h H
  
change mass flow rate very little but slow the rate of solver conver- 
1 @ q @ qp
þ r q pkþ1
U þ qr  /
gence. Without the compressibility terms, the linear system is sym- q @p  x;h @t
metric and can be solved with approximately half the cost of the full
1 @M kþ1
system of equations. Results with and without the compressibility  qr rpkþ1 þ Mðpk Þ þ ðp  pk Þ ¼ 0 ð20Þ
ap @p
terms are reported in the next section.

4. Numerical approach Typically, two to five PISO/secant iterations were employed, each
requiring solution of the pressure equation. Without the compress-
An attractive feature of this pressure equation is that most of ibility terms, the linear system for pressure is symmetric and is
the terms are linear in p plus the model in Eq. (1) can be inserted solved using a diagonal incomplete Cholesky preconditioned conju-
directly into the last term. In the limit of constant density, an gate-gradient method. With the full pressure equation, a diagonal
incompressible formulation is recovered. Schmidt et al. (1999a) incomplete LU preconditioned bi-conjugate gradient is used. The
used a similar idea (but neglecting the derivative of density with non-orthogonal parts of the Laplacian are handled with a deferred
respect to pressure) in a two-step projection method on a stag- correction approach that also benefits from the multiple iterations
gered mesh approach. The implementation on a staggered mesh if the computational mesh is highly skewed (Jasak, 1996). Once
was well-suited for their two-dimensional structured grid solver. Eq. (20) has been solved, the pressure field is used to correct the
In order to facilitate the application of the current model to fluxes and the time step is completed. The pressure must also be
three-dimensional solutions with unstructured, polyhedral mesh updated in the momentum equation. This is done by reconstructing
support, the current implementation will use a collocated variable the face-based pressure gradients into a cell-centered gradient. This
approach. reconstruction process can produce spurious out-of-plane velocities
The first step in each time step is the solution of conservation of in two-dimensions that are discarded.
mass, Eq. (16). This is done implicitly. This approach produces a set of equations that are solved on an
arbitrary polyhedral mesh in two and three-dimensions. The
@q
þ r  ð/v qÞ ¼ 0 ð16Þ underlying framework is provided by OpenFOAM (Weller et al.,
@t
1998), which permits rapid construction of CFD codes in an ob-
Here, the volumetric flux, /v , is based on the velocity field from the ject-oriented framework that includes a wide choice of discretiza-
previous time step, interpolated to cell faces. The new value of den- tion schemes. The current implementation is stable enough for
sity from Eq. (16) is interpolated to cell faces and a new mass flux / two-dimensional calculations and has been used in a few three-
is calculated. Next, the thermodynamic variables such as void frac- dimensional calculations. However, the cases where experimental
tion, quality, and compressibility are updated using the new value data are available for validation are all two-dimensional. Properties
for density. were evaluated from lookup tables generated with Lemmon et al.
As in the PISO algorithm (Issa, 1986), the velocity field is pre- (2007). Viscosity was calculated from a volume-weighted average,
dicted using a lagged pressure, indicated by the superscript n. an assumption frequently used in pseudo-fluid CFD simulation
The equation for this predicted velocity, U 0 , is given in Eq. (17). La- (Kubota et al., 1989; Chen and Heister, 1994; Schmidt et al.,
ter, when pressure is updated, the additional contribution from the 1999b; Senocak and Shyy, 2002).
change in pressure will be used as a corrector to the velocity field. All variables were located at cell centers, except for interpolated
fluxes located at cell faces. For all calculations, the flux terms were
ð@ qU 0 Þ
þ r  ð/U 0 Þ ¼ rpn þ r  ðlrU 0 Þ ð17Þ evaluated using a Gamma TVD scheme and the Laplacian is discret-
@t ized according to the recommendations of Jasak et al. (1999). Fixed
Eq. (17) represents three linear systems of equations, one for each pressure boundary conditions and zero gradients of velocity were
component of velocity, and is solved implicitly with the pressure typically used at inlets and exits, while the walls were treated as
gradient acting as an explicit source term, in the form of Eq. (12). no-slip. Courant numbers for time-stepping ranged up to 1.2.
288 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292

5. Results and validation pressible flow, the pressure would be expected to be extremely
low downstream of this corner, due to the separated flow and
All the test cases used for validation and study were channel the constraint of a divergence-free velocity field. However, with
flows containing water. Though the simulation is transient, the flashing flow, the decrease in pressure creates an increase in the
experiments were always steady-state, so all calculations were rate of phase change. The flashing of the liquid creates a positive
run until both the inflow and outflow had stabilized. The typical velocity divergence that allows the contraction to occur with a rel-
flash-boiling experiment is a straight channel with a sharp inlet. atively small dip in pressure behind the inlet corner.
The sharp corner creates the potential for a separated flow with The contours of mass fraction are visible in the upper half of
strong two-dimensional flow features. For convenience, these Fig. 3. Because of the extreme density ratio between the liquid
kinds of flows were simulated as axisymmetric flow. To avoid and vapor, the vapor formed near the corner does not correspond
imposing boundary conditions where sharp gradients would be to a significant portion of the liquid mass. This result is consistent
present, a plenum was added to both the inlet and outlet side of with the decision not to include a separate momentum equation
the channel, as shown in Fig. 1. The addition of the plenum pro- for the vapor phase, since transported momentum would be pro-
vides some separation distance between the imposed boundary portional to vapor mass. The tiny mass fraction of vapor does not
conditions and the region of interest. An unstructured quadrilateral seem to warrant a separate momentum equation.
mesh was used throughout the whole domain. The contours of volume fraction in Fig. 3 show this rapid vapor
A high temperature, high pressure test case was taken from Tik- generation at the inlet corner. The two-phase density in the com-
honenko et al. (1978), who explored critical flow of hot water in putational domain ranges from the initial saturated liquid density
various pipes with a sharp inlet. These experiments include data down to a value of 1:5 kg=m3 . The sharp corner induces a phase
from pressure taps placed along the length of the pipe. In this sim- change around the outer periphery of the flow. This vapor remains
ulation, a channel with a 25 mm diameter and 250 mm length was as an outer sheath for the length of the nozzle, as previously de-
simulated. The inlet conditions were saturated water at 4 MPa and scribed in experimental studies (Sher et al., 2008).
the downstream pressure was specified to be one atmosphere. This radial density and velocity profiles are interesting features
First, this test case was used to check grid independence of the of a multi-dimensional CFD study of flashing nozzles. It serves as
solution. Even though a perfectly sharp corner represents a singu- an example of macroscopic interphase slip, where the liquid core
larity, the flow should show an acceptably low sensitivity to the moves with one velocity at the inner radius in the nozzle and the
mesh resolution in order for the results to be useful. The nozzle vapor could move with a different velocity near the nozzle walls.
was meshed using a coarse mesh of 3500 cells and a finer mesh Between x/D of 2 and 8, very little change occurs in the axial
with 15 thousand cells. A comparison of predicted wall pressures direction. The pressure gradient is minimal and there is little
from the several calculations can be seen in Fig. 2. Both the high- change in the radial density or velocity profile. However, near
pressure and low-pressure correlations (see Eqs. (8) and (10)). the nozzle exit plane, a dramatic change occurs, as shown in Fig. 4.
The coarser mesh is sufficiently close to the fine mesh result, such Figs. 2 and 4 indicate that part of the pressure drop across the
that the modeling assumptions are a larger source of error than the nozzle occurs at the inlet, followed by a relatively flat pressure re-
discretization error. gion, and then a second pressure drop at the exit. As the pressure
The results for static pressure along the pipe wall in Fig. 2 also drops further below the vapor pressure of 4 MPa, the rate of phase
permit a comparison of the two correlations, as well as revealing change increases. The nature of the timescale correlation provided
some of the internal flow features. The experimentally-measured by Downar-Zapolski et al. (1996) also captures the effect of
pressure shows a slight local minimum near the inlet corner due increasing interfacial area for phase change, due to the dependence
to the separated flow. As the computational results will show, on vapor volume fraction. So, with the creation of vapor, the rate of
the liquid forms a vena contracta downstream of the inlet. Once phase change is further increased. Note how the timescale shown
past the vena contracta, the pressure partially recovers and then in Fig. 5 correlates with the creation of vapor. By conservation of
drops precipitously at the exit. mass, the drop in density is accompanied by an increase in axial
The computational results follow the expected trend. There is a velocity. Conservation of momentum then indicates that pressure
local pressure minimum on the wall just downstream of the inlet drops further. This pressure drop, in turn, feeds back into the flash-
corner in the simulations. The pressure recovers slightly and then ing process. The pressure finally reaches the downstream value
remains nearly constant along the nozzle length until just up- just outside of the nozzle.
stream of the nozzle exit. The pressure at the nozzle exit decreases The anticipation that the flashing flow process would continue
dramatically due to rapid flashing near the exit plane. However, just beyond the nozzle motivated the decision to place the compu-
both correlations under-predict the rate of flash-boiling, though tational boundary downstream of the nozzle. However, the model
the low-pressure correlation is especially far off. The high-pressure does not account for the presence of non-condensible gasses, such
correlation produces pressures that are much closer to the experi- as occur in air. Fortunately for this case, it appears from the veloc-
mental measurements. ity field that air is not entrained into the near-exit flow. The strong
In addition to predicting pressure, the computational results favorable pressure gradient near the nozzle exit discourages the
show other features of interest, such as pressure, velocity, density, counter-flow of air and produces no recirculating flow in the exit
and the rate of change of quality. These results can be used to ex- plenum of the computational domain.
plain the nozzle behavior under these flow conditions. The ability of the model to predict choking was also investi-
As in both single-phase and cavitating nozzles, the flow sepa- gated. The high liquid temperature and low downstream pressure
rates off of the sharp inlet corner (Schmidt, 1997). In an incom- suggests that the flow should be choked (Fauske, 1965). Computa-

Fig. 1. A typical two-dimensional computational domain.

 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292 289

1
0.9
0.8
0.7
0.6

p/p0
0.5
0.4 Experiment
Coarse Mesh - incompressible -High pressure correlation
0.3 Fine Mesh - incompressible -High pressure correlation
Coarse mesh - incompressible - Low pressure correlation
0.2 Fine Mesh - incompressible - Low pressure correlation
Compressible - Low pressure correlation
0.1
0
0 2 4 6 8 10
x/D
Fig. 2. Static pressure versus position at the wall for saturated water at 4 MPa discharging through a 25 mm tube with L/D = 10

Fig. 3. Predicted vapor mass fraction and volume fraction in the 4 MPa saturated water experiment of Tikhonenko. The domain has been reflected around the axis of
symmetry so that two fields can be shown simultaneously.

Fig. 4. Predicted pressure and equilibrium mass fraction 

x in the 4 MPa saturated water experiment of Tikhonenko.

tionally, this is indeed the case. The above simulation was re-run duces the pressure drop across the nozzle by 2.6%. The computed
with the downstream pressure set to two atmospheres, which re- mass flow rate changed by less than 0.03%.
290 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292

Fig. 5. Predicted velocity magnitude and the common log of the phase change timescale H for the 4 MPa saturated water experiment of Tikhonenko.

The second test case was chosen to emphasize two-dimensional iments. Tikhonenko’s experiment was at an upstream pressure
effects. In these experiments Fauske (1965) studied saturated very close to the middle of the range of Fig. 6 and both were satu-
water discharge through short tubes. He noted the maximum dis- rated. The L/D ratio for the data in Fig. 6 are for L/D of 4, compared
charge rates as a function of L/D and upstream stagnation pressure. to Tikhonenko’s L/D of 6. The diameter of Tikhonenko’s nozzle was
Of the various nozzles that Fauske tested, a relatively short nozzle about four times larger than Fauske’s, which could be a factor.
was chosen for validation, with L=D ¼ 4, in the next test case. It is Next, the internal flowfield details were observed in order to
expected that the inlet corners will cause large variations of void understand how the two-dimensional effects were manifesting
fraction and velocity in the radial direction. These two-dimensional themselves in the flowfield. The first figure illustrates a simulation
effects are likely to be more pronounced than in the longer nozzle of Fauske’s experiment with an upstream pressure of 1.38 MPa
discussed above. (200 PSIA). Fig. 7 shows the volume fraction of vapor in the upper
First, the mass flow rates were compared for 6.35 mm diameter half of the figure and approximate stream lines in the bottom half.
tubes at stagnation pressures of 1.37 MPa, 4.13 MPa, and 6.89 MPa. The streamlines are not from a solution of a stream function, since
The calculated mass flow rates using the low-pressure correlation, the velocity field is not divergence-free, but are rather calculated
Eq. (8) are compared to Fauske’s measurements in Fig. 6. The from Runge–Kutta integration of particle trajectories using the fro-
agreement of the data is excellent, with the computed results lying zen, discrete velocity field, incurring a discretization error compa-
within the scatter of the experimental data. The good agreement rable with the CFD computations.
produced by the low-pressure correlation is somewhat surprising The streamlines in Fig. 7 show the separation and formation of a
and much better than the high-pressure correlation, Eq. (10), vena contracta just downstream of the nozzle inlet. The outer flow
which under-predicted mass flow rate by a factor of two. recirculates downstream of this corner, forming an area of high va-
In the calculations shown back in Fig. 2 the high pressure corre- por concentration. This outer fluid likely has a long residence time
lation performed better, as one might expect given the 4 MPa up- in the nozzle due to the recirculation, which may explain why Fau-
stream pressure. In the simulations with Fauske’s experiments, ske did not observe any sensitivity to nucleation. The recirculating
the low-pressure correlation was clearly better. This observation fluid has a relatively long opportunity to change phase, compared
is especially curious given the similarities between the two exper- to the central flow. As the vapor fraction shows, the core begins to

4
5×10
Fauske Expt.
CFD Simulation
Mass Velocity [kg/s/m ]

4
2

4×10

4
3×10

4
2×10

4
1×10

0
0 1 2 3 4 5 6 7
Upstream Pressure [MPa]
Fig. 6. Measured (Fauske, 1965) mass flow rates for a nozzle with L/D = 4 compared with the present calculations.
 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292 291

Fig. 7. Simulation of Fauske’s experiment with 1.38 MPa saturated liquid discharge. This figure shows volume fraction of vapor and approximate streamlines.

vaporize closer to the exit. The predictions of the annular vapor 6. Conclusions
sheath and core vaporization are consistent with Fauske’s observa-
tions and published sketches of the flow. The Homogenous Relaxation Model has been tested in multi-
The phase change process is accompanied by acceleration as a dimensional CFD calculations. The non-equilibrium thermodynam-
consequence of conservation of mass and momentum. This accel- ics in conjunction with the assumption of no sub-cell interphase
eration is evident in the upper half of Fig. 8. The initial contraction slip produced a model that required a special numerical approach.
near the throat of the vena contracta produces an acceleration as A method for connecting the predicted phase change to conserva-
the core flow passes through a reduced cross-sectional area in tion of mass and momentum was developed that used the chain
the nozzle. A second acceleration occurs near the exit, where vapor rule in lieu of a typical equation of state. This new numerical con-
is formed. struction produced a closed set of equations that could be solved
As vapor is formed, more interfacial area is available for heat using the finite volume method. The model was implemented into
transfer, which feeds back into the phase change process by short- a CFD code and demonstrated with two-dimensional calculations.
ening the timescale. The feedback of between interfacial area and Experiments from the open literature were used for validation with
the timescale can be seen in the lower half of Fig. 8, where the low- measurements of wall pressure and mass flow rate. The simula-
est values of the timescale represent regions where the fluid will tions showed reasonably good fidelity in predicting pressures at
move more quickly towards the equilibrium quality. nozzle walls in the case of saturated discharge. Choking was also
The above figures were only considering the lower end of predicted under these conditions.
the range of upstream pressures. However, the general charac- The results indicate that a properly formulated one-dimensional
ter of the nozzle flow in these saturated discharge calculations closure for flashing flow can be extended to multi-dimensional com-
is relatively insensitive to variations in the upstream pressure. putations. The model used in the present work, the Homogenous
Though velocities increase with increasing upstream pressure, Relaxation Model, performed well in various tests of flashing flow
the vena contracta remains a relatively constant feature. The without adjustment to the previously reported empirical parame-
amount of vapor does not change much with a factor of five in- ters, though neither of the two correlation suggested by Downar-
crease in upstream saturation pressure, as shown in Fig. 9. The Zapolski et al. (1996) were sufficient for all cases. Given that the pres-
stability of the vena contracta is well-known from previous ent work is an extension of the Homogenous Relaxation Model be-
studies of cavitating flow and single-phase nozzle flow (Nurick, yond its original intent, some adjustment of these parameters
1976). might be a useful exercise for future work. The assumption of equal

Fig. 8. Simulation of Fauske’s experiment with 1.38 MPa saturated liquid discharge. This figure shows velocity magnitude and the common logarithm of the timescale of
phase change.
292 D.P. Schmidt et al. / International Journal of Multiphase Flow 36 (2010) 284–292

Fig. 9. Comparison of the volumetric vapor fraction with 1.38 MPa saturated liquid discharge (upper half) and 6.89 MPa (lower half).

velocities and unequal temperatures does appear sufficient for most Issa, R.I., 1986. Solution of the implicitly discretised fluid flow equations by
operator-splitting. J. Comput. Phys. 62, 40–65.
of the channel flow studied here. Certainly, the effects of turbulent
Jasak, H., 1996. Error Analysis and Estimation for the Finite Volume Method with
flow and mixing are not represented in the current framework, Applications to Fluid Flows. PhD thesis, Imperial College.
and are likely candidates for future development. Jasak, H., Weller, H.G., Gosman, A.D., 1999. High resolution NVD differencing
Nucleation effects of dissolved gasses is another extensive area scheme for arbitrarily unstructured meshes. Int. J. Numer. Methods Fluids 31,
431–449.
of study that might impact current model accuracy. In the current Kato, H., Kayano, H., Kageyama, Y., 1994. A consideration of thermal effects on
work, the void fraction is arbitrarily limited to a small positive va- cavitation bubble growth. In: Cavitation and Multiphase Flow, FED, vol. 194.
lue. A more sophisticated approach would be to implement a ASME.
Kim, Y., O’Neal, D.L., 1993. An experimental study of two-phase flow of hfc-134a
nucleation factor in the calculation of minimum void fraction through short tube orifices. Heat Pump Refrig. Syst. Des., Anal., Appl. 29, 1–8.
based on existing nucleation models. Though this is not anticipated Knapp, Robert T., Daily, James W., Hammitt, Fredrick G., 1970. Cavitation. McGraw-
to alter predictions near the inlet corners, where phase change is Hill.
Kubota, A., Kato, H., Yamaguchi, H., 1989. Finite difference analysis of unsteady
geometrically induced, flashing could be promoted near the center cavitation on a two-dimensional hydrofoil. In: Proceedings of the 5th
of the channels. International Conference on Numerical Ship Hydrodynamics. Hiroshima, pp.
667–683.
Lemmon, E.W., Huber, M.L., McLinden, M.O., 2007. NIST Standard Reference
Acknowledgments Database 23: Reference Fluid Thermodynamic and Transport Properties-
REFPROP. National Institute of Standards and Technology, Gaithersburg,
We thank Prof. Gian Marco Bianchi from the University of Bolo- version 8.0 standard reference data program edition.
Minato, Akihiko, Takamori, Kazuhide, Susuki, Akira, 1995. Numerical study of two-
gna, Dr. Ronald Grover from General Motors Research Center, and
dimensional structure in critical steam-water two-phase flow. J. Nucl. Sci.
Dr. Jerry Lee from United Technologies Research Center for their Technol. 35, 464–475.
technical advice. We also acknowledge the financial support pro- Moody, F.J., 1965. Maximum flow rate of a single component, two-phase mixture. J.
vided by General Motors Inc. and United Technologies Research Heat Transfer 87, 134–142.
Nurick, W.H., 1976. Orifice cavitation and its effect on spray mixing. J. Fluids Eng. –
Center. Funding has been provided under NSF Grant IP-0610613 Trans. ASME 98, 681–687.
and NASA Contract 06-SUP-06-0094. This research was supported Reitz, Rolf D., 1998. A photographic study of flash-boiling atomization. Aerosol Sci.
in part by the National Science Foundation through TeraGrid re- Technol. 12, 561–569.
Schmidt, David P., 1997. Cavitation in Diesel Fuel Injector Nozzles. PhD thesis,
sources provided by the National Center for Supercomputing University of Wisconsin Madison, December 1997.
Applications (NCSA), University of Illinois Urbana-Champaign, Schmidt, D.P., Corradini, M.L., Rutland, C.J., 1999a. A two-dimensional, non-
Grant number CTS060044T. equilibrium model of flashing nozzle flow. In: Proceedings of 3rd ASME/JSME
Joint Fluids Engineering Conference, FEDSM.
Schmidt, D.P., Rutland, C.J., Corradini, M.L., 1999b. A fully compressible model of
References cavitating flow. Atomization Sprays 9, 255–276.
Senocak, Inanc, Shyy, Wei, 2002. A pressure-based method for turbulent cavitating
Bilicki, Z., Kestin, J., 1990. Physical aspects of the relaxation model in two-phase flow computations. J. Comp. Phys. 176, 363–383.
flow. Proc. Roy. Soc. Lond. A. 428, 379–397. Sher, E., Bar-Kohany, T., Rashkovan, A., 2008. Flash-boiling atomization. Prog.
Boure, J.A. et al., 1976. Highlights of two-phase flow: on the links between Energy Combust. Sci. 34, 417–439.
maximum flow rates, sonic velocities, propogation and transfer phenomena in Simões Moreira, J.R., Bullard, C.W., 2003. Pressure drop and flashing mechanisms in
single and two-phase flows. Int. J. Multiphase Flow 3, 1–22. refrigerant expansion devices. Int. J. Refrig. 26, 840–848.
Chen, Yongliang, Heister, Stephen D., 1994. A numerical treatment for attached Tikhonenko, L.K., Kevorkov, L.P., Lutovinov, S.Z., 1978. An investigation of the local
cavitation. J. Fluids Eng. 116, 613–618. paramters of critical flow of hot water in straight pipes with a sharp inlet edge.
Downar-Zapolski, P., Bilicki, Z., Bolle, L., Franco, J., 1996. The non-equilibrium Teploenergetika 25, 41–44.
relaxation model for one-dimensional flashing liquid flow. IJMF 22, 473–483. Valero, E., Parra, I.E., 2002. The role of thermal disequilibrium in critical two-phase
Duan, Ri-Qiang, Jiang, Sheng-Yao, Koshizuka, Seiichi, Oka, Yoshiaki, Yamaguchi, flow. Multiphase Flow 28, 21–50.
Akira, 2006. Direct simulation of flashing liquid jets using the mps method. Int. Vortmann, C., Schnerr, G.H., Seelecke, S., 2003. Thermodynamic modeling and
J. Heat Mass Transfer 49, 402–405. simulation of cavitating nozzle flow. Int. J. Heat Fluid Flow 24, 774–783.
Einstein, A., 1920. Über Schallschwingungen in teilweise dissocierten Gasen. Wallis, G.B., 1980. Critical two-phase flow. Multiphase Flow 6, 97–112.
Sitzung Berl. Akad. Physik Chemie, 380–385. Weller, Henry G., Tabor, G., Jasak, Hrvoje, Fureby, C., 1998. A tensorial approach to
Fauske, H.K., 1965. The discharge of saturated water through tubes. Chem. Eng. computational continuum mechanics using object-oriented techniques.
Prog. Symp. Ser. 6, 210–216. Comput. Phys. 12, 620–631.
Ferziger, Joel H., Peric, Milovan, 2002. Computational Methods for Fluid Dynamics, Yu, H., Mironov, V., Razina, N.S., 1987. Supersonic effects with subcooled water
3rd ed. Springer. flowing through cylindrical nozzles with a sharp inlet edge. Therm. Eng. 34,
Henry, R.E., Fauske, H.K., 1971. The two-phase critical flow of one-component 587–595.
mixtures in nozzles, orifices, and short tubes. Heat Transfer 93, 179–187.