International Journal of Multiphase Flow 95 (2017) 199–219

Contents lists available at ScienceDirect

International Journal of Multiphase Flow

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

Homogeneous two-phase flow models and accurate steam-water table

look-up method for fast transient simulations
M. De Lorenzo a,∗, Ph. Lafon a, M. Di Matteo a,b, M. Pelanti a, J.-M. Seynhaeve c,
Y. Bartosiewicz c
a
IMSIA UMR EDF-CNRS-CEA-ENSTA, Palaiseau, 91120, France
b
Politecnico di Torino, Torino, 10129, Italy
c
Université Catholique de Louvain (UCL), Louvain la Neuve, 1348, Belgium

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

Article history: The accurate simulation of fast steam-water transients requires precise algorithms for calculating fluid
Received 28 November 2016 properties. The system of the governing flow equations must be closed with an Equation of State (EoS)
Revised 1 June 2017
to calculate the pressure as a function of the system conservative variables. For water, accurate analytical
Accepted 2 June 2017
EoS for this purpose are not available yet. The aim of this paper is to show an efficient and very accurate
Available online 7 June 2017
algorithm to calculate water properties when the independent variables of the EoS are the density and
Keywords: the specific internal energy. Our algorithm uses a new table look-up method with bicubic interpolation
Two-phase flow based on the IAPWS-IF97 EoS formulation, and it is able to account for metastable states. The liquid
Homogeneous equilibrium model metastability domain is extended until the spinodal curve, here determined and compared with other
Homogeneous relaxation model formulations.
Steam-water tables
Properties calculation The EoS algorithm is coupled to two classical homogeneous two-phase flow models, namely the Homo-
Metastable states geneous Equilibrium Model (HEM) and the Homogeneous Relaxation Model (HRM). HEM and HRM are
Spinodal curve used to simulate fast depressurization, waterhammer and steam explosion problems. Comparison of the
Fast transients numerical results with available experimental data shows the good performance of the proposed algo-
Depressurization rithms.
Waterhammer
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction state representing the thermodynamic conditions of the numerical

simulation. If the thermodynamic conditions of the simulated flow
Computational Fluid Dynamics simulations require accurate and remain close to the reference state, the use of SG EoS can give
fast algorithms for calculating fluid properties. In industrial appli- fairly good results as shown in Daude et al. (2014), Lochon et al.
cations, simulations are performed for a wide range of regimes, (2016) and Saurel et al. (2008). However, for sufficiently large tem-
possibly involving shock wave formation. In such a case, compu- perature deviations from the reference linearization value, the SG
tational algorithms must employ the conservative form of the gov- EoS can provide inaccurate results (see the results in Le Metayer
erning fluid equations. For compressible flow models in conserva- et al., 2004) or even unphysical states (Saurel and Abgrall, 1999).
tive form, the appropriate thermodynamic independent variables For industrial purposes, especially when nuclear safety is in-
to be chosen for the pressure law closure are the density ρ (or the volved, fast high-fidelity CFD simulations are needed. The accuracy
specific volume v = 1/ρ ) and the specific internal energy e. of a CFD simulation depends on many features often intercorre-
For liquid and vapor water, accurate Equations of State (EoS), lated, however, the precision of the thermodynamic properties cal-
having (ρ , e) as independent variable pair, are not available so far. culation is independent from the other numerical aspects, there-
In the last decade, simple EoS such as the Stiffened Gas (SG) one fore, it should be the first feature to deal with. To obtain a realistic
have been adopted for water and other complex fluids. This SG evaluation of the thermodynamic properties, simple EoS are not
EoS is based on a linearization around a reference thermodynamic sufficient and experimental EoS have to be employed. However the
direct use of such EoS is time consuming.
Furthermore, CFD codes are frequently used, hence calculations
∗
Corresponding author. must be performed in an efficient way. For the EoS, the require-
E-mail address: [email protected] (M. De Lorenzo). ment of both accuracy and efficiency can be fulfilled by the tabu-

http://dx.doi.org/10.1016/j.ijmultiphaseflow.2017.06.001
0301-9322/© 2017 Elsevier Ltd. All rights reserved.
200 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

lation of an accurate EoS at the beginning of the simulation and a The most complete model of this type is the seven equa-
look-up table interpolation at each time step, for each cell of the tion model (Baer–Nunziato 1986), which allows pressure, temper-
simulation domain. ature and velocity non-equilibrium between the phases. The sim-
In one-component two-phase transients, metastable phases of- plest one is the Homogeneous Equilibrium Model (HEM) which as-
ten appear, thus, the EoS to be tabulated should be valid also in the sumes pressure, temperature and velocity equilibrium. For a brief
metastable domain. For water, an EoS that satisfies all the features but comprehensive discussion about various models for two-phase
discussed so far is the Industrial Formulation release (1997) of the flows, we refer to Saurel et al. (2008).
International Association for the Properties of Water and Steam One of the main goals of this article is to illustrate the cou-
(IAPWS), commonly called IAPWS-IF97 (Wagner et al., 20 0 0). In- pling between two-phase flow models and accurate EoS, both in
deed it is a simplified form of the most accurate EoS today avail- the equilibrium and metastable domains. For this purpose here
able, that is the IAPWS-95 (Wagner and Pruß, 2002a), but contrar- the simplest two-phase flow models shall be used. In particular,
ily to this release, the IAPWS-IF97 can be easily extrapolated into we will employ the HEM model for the simulation of thermody-
the metastable domain. namically stable mixtures. In water fast transients and in steady-
The aim of this paper is to show an efficient and very accu- state flashing mixtures, the liquid phase attains the metastable do-
rate algorithm to calculate water properties when the independent main evolving beyond its saturated conditions. To extend our anal-
variables are the density, ρ and the specific internal energy e. For yses to those cases, the Homogeneous Relaxation Model (HRM)
a given thermodynamic pair (ρ , e), we wish to obtain the corre- can be employed. This model represents the simplest two-phase
sponding pressure, temperature and speed of sound. The main pur- flow model able to simulate flows of mixtures containing liquid in
pose is to couple this algorithm to two-phase flow models in their metastable conditions.
conservative form. To reach this goal, thermodynamic properties Both the HEM and HRM can be written in a fully conservative
shall be tabulated on the e-v thermodynamic diagram. The best form
attempt presented in literature so far is due to Kunick et al. (IAP,
2015; Kunick, 2015), which is a Guideline from the IAPWS and is ∂t U + ∂z F(U ) = b, (1)
now implemented in RELAP-7 (REL,Berry et al. (2016)). Here we
where U is the vector of the conservative variables, F represents
present a new bicubic look-up table algorithm, which has higher
the flux vector and b accounts for the source terms eventually
accuracy and, like Berry et al. (2016), extends the validity domain
present. To highlight some thermodynamical features of the mod-
to the metastable states of water. Here, this extension is done
els, we also write here these models in their quasi-linear form
up to the assumed spinodal curve, that is, the line that sepa-
rates the unstable domain from the stable/metastable one. More- ∂t U + A(U )∂z U = b, (2)
over, the new algorithm for fluid properties calculation is coupled
to two common two-phase flow models, the Homogeneous Equi- where A(U) is the Jacobian matrix, defined as
librium Model (HEM) (Clerc, 20 0 0) and the Homogeneous Relax-
∂F
ation Model (HRM) (Bilicki and Kestin, 1990), both within the EU- A (U ) = . (3)
ROPLEXUS1 software. Results of simulations are presented for fast ∂U
depressurization, waterhammer and energy injection problems. For In the next two subsections, both the HEM and the HRM are
the fast depressurization test, a novel correlation for the relaxation analyzed to underline their mathematical properties. Both are hy-
time to return to equilibrium conditions is also proposed. perbolic models when appropriate thermodynamical constraints
The paper is organized as follows. In Section 2 we describe the are satisfied. The hyperbolicity is discussed in Appendix A and
homogeneous two-phase flow models that we consider and the Appendix B, respectively.
type of EoS that is needed for their closure. Section 3 is devoted to
the possible strategies to be adopted when an EoS as p = p(ρ , e ) is
required. In Section 4, we detail the EoS today available for water 2.1. Homogeneous equilibrium model
focusing on their capability to deal with metastable fluids. We also
discuss the need of an equation for the spinodal line in order to The Homogeneous Equilibrium Model (HEM) is a set of partial
detect, during the simulation, if one of the two phases has attained differential equations that governs the motion of the total mass,
an unstable state ((∂ p/∂ v)T > 0). Section 5 illustrates the construc- the global momentum and the total energy of a mixture in full
tion of the tabulated domain on the e-v thermodynamic diagram, equilibrium. In spite of its simplicity, this model is extensively used
the bilinear mapping of the diagram and the bicubic interpolation for numerous industrial applications. For instance, to simulate heat
technique. This section ends with the comparisons between the in- exchangers (e.g., Porsching, 1977), for calculating the mass flow
terpolated values and the exact solution using the IAPWS-IF97. Re- rate of critical two-phase flows (Schrock and Amos, 1984) and even
sults of numerical simulations performed by EUROPLEXUS with the for the analysis of a hypothetical Loss Of Coolant Accident (LOCA)
new look-up table algorithm are presented and discussed in the in nuclear power plants (Faucher et al., 2014). Generally, the HEM
Section 6. Some conclusions are written in Section 7. is particularly well adapted for the simulation of dispersed bubbly
flow (Clerc, 20 0 0). The drawbacks of this model are well known: it
2. Homogeneous two-phase flow models cannot reproduce mechanical or thermodynamic non-equilibrium,
typical of annular flows and fast depressurizations, respectively. If
Different models describing two-phase flows are available in the deviation from equilibrium is moderate, correction terms can
the literature. Here we consider compressible flow models that be- be used (drift flux velocity, subcooled boiling models). However
long to the class stemming from the model of Baer and Nunzi- when non-equilibrium effects are more important, additional equa-
ato, 1986. Models in this class differ primarily in the level of non- tions are needed for an accurate prediction of the flow physics.
equilibrium between the two phases that they are able to take into Notwithstanding, HEM has good mathematical properties which
account. make it suitable for the simulation of wave motions, hence, fast
transients.
1
EUROPLEXUS code is owned by the Commissariat à l’Energie Atomique et aux
We shall consider a mixture made of the liquid and vapor
Energies Alternatives (CEA) and the Joint Research Centre (JRC) of the European phases of one-component in full thermodynamic equilibrium. The
Union. HEM system has formally the structure of the single-phase Euler
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 201

Fig. 1. Isobaric curve on the T-v diagram. The plot has been realized using the
IAPWS-IF97. The point (a) lies on the liquid saturation curve and represents the Fig. 2. Mixture speed of sound for water for HEM, HRM and Wood’s relation
thermodynamic state of the liquid phase for the HEM. Point (b) is a metastable (Wood, 1930). The plot on the left is at 10 MPa whereas the one on the right
state which can represent the state of the liquid phase using HRM. In HRM: Tl (p) is at 0.1 MPa (magnified view). It has to be noted that the speed of sound of the
> Tv (p). Point (c) is the spinodal limit for the isobaric p = 12.7 MPa. In the HRM, HEM has two discontinuities for αv = 0 and αv = 1, as discussed in Appendix A.
the liquid phase cannot overstep this point since the tabulation of the properties is
defined until the spinodal line. Therefore, unstable states are not allowed. Point (d)
lies on the vapor saturation curve and represents the state of the vapor phase for
both HEM and HRM.
In HEM, the flow quality x corresponds to the thermodynamic
quality:
system. In one-dimensional geometry, the HEM equations are:
⎧ mass o f vapor h − hf
x= = ,
⎨∂t ρ + ∂z (ρ u ) = 0, total mass hg − h f
(8)
∂t (ρ u ) + ∂z (ρ u2 + p) = 0, (4)
⎩∂ (ρ E ) + ∂ [(ρ E + p)u] = 0. where the enthalpy is h = e + pv.
t z
The system of Eq. (4) is not closed, there are three equations
The notation is the classical one: ρ is the mixture density, u de- and four unknowns. To close the system, an additional equation
notes the velocity of the mixture, p is the pressure, and E is the must be supplied that links the pressure to the other thermody-
specific total energy, i.e., E = e + u2 /2, where e is the specific in- namic variables. Let us consider the caloric equation of state with
ternal energy of the mixture. the pair (p, ρ ) as independent variables:
This is a system of non-linear hyperbolic conservation laws that
governs the dynamics of a compressible fluid for which the effects e = e( p, ρ ). (9)
of body forces, viscous stresses and heat flux are neglected. How-
ever, for more realistic simulations these phenomena can be taken We are looking for its inverse:
into account as source terms, i.e. as contributions to the vector b
of Eq. (1). p = p( ρ , e ) . (10)
The thermodynamic equilibrium assumptions of the HEM are:
As we stressed before for many fluids like water, an accurate EoS
pl = pv = p, expressed in the form of Eq. (10) is not available. In the next sec-
Tl = Tv = Tsat ( p), tion we shall discuss various approaches to evaluate the pressure
(5) for a given pair of values of specific volume and specific internal
ul = uv = u,
energy.
gl = gv = g, Let us note that a complete description of the thermodynamic
where subscripts l and v indicate, respectively, the liquid phase and conditions of the flow requires an additional relation, the thermal
the vapor one, and where g denotes the Gibbs free enthalpy. The equation of state:
closure laws are:
T = T ( p, ρ ). (11)
e = x ev ( p ) + ( 1 − x ) el ( p ),
v = x vv ( p ) + ( 1 − x ) vl ( p ), (6) However, only the caloric EoS is needed to solve the system of Eq.
(4). Choosing a thermal EoS does restrict the choice of a caloric
where the specific volume is v = 1/ρ and the quality is x = EoS but does not determine it.
mass of vapor
total mass
. With the constraints expressed in Eq. (5), the internal The behavior of the speed of sound that characterizes the HEM
energies and the specific volumes of vapor and liquid are those of with respect to the vapor volume fraction α v is depicted in Fig. 2.
the saturation curve. That is: The expression of the speed of sound is reported in Appendix A. It
ev = eg ( psat ), has to be noted that the speed of sound of the HEM has two dis-
el = e f ( psat ), continuities for αv = 0 and αv = 1. In fact, one of the main charac-
(7) teristics of the equilibrium equations of state is the presence of a
vv = vg ( psat ), “kink” (discontinuity in the derivatives) along the saturation curve.
vl = v f ( psat ), The discontinuities in the speed of sound have been referred to
as unphysical by several authors (Flȧtten and Lund, 2012; Kieffer,
where subscripts f and g indicate, respectively, the saturated liquid
1977; Nakagawa et al., 2009; Stadtke, 2006).
and the saturated vapor. Refer to Fig. 1 for a graphic illustration.
202 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

2.2. Homogeneous relaxation model the sound speed that characterizes the Kapila 5-equation model).
Such a well-known non-monotonic behavior for the mixture sound
At the beginning of Section 2 we pointed out the existence of speed has been experimentally verified by several authors in the
a hierarchy of two-phase flow models based on the level of non- literature (Gouse and Brown, 1964; Karplus, 1958). Let us note
equilibrium that they are able to take into account. In certain fast that the curve of the sound speed of the HRM always lays be-
transients of one-component mixtures it is crucial to take into ac- low the curve of the Wood’s sound speed, and above the curve
count thermal and chemical non-equilibrium, that is the tempera- of the sound speed of the HEM. This is consistent with the well-
ture and Gibbs free enthalpy difference between vapor and liquid. known sub-characteristic condition (Liu, 1987) that requires that
For instance, in depressurizations of subcooled liquids, the pres- the sound speed associated to a two-phase flow model is reduced
sure evolves very quickly and the liquid becomes metastable due whenever an additional equilibrium is assumed (Flȧtten and Lund,
to the finite vaporization rate. Experimental evidences of this phe- 2012). We observe that the magnitude of the speed of sound of the
nomenon are reported in Bartak (1990) and Riegel (1978). HRM can be remarkably different with respect to Wood’s speed of
The simplest conservative two-phase flow model that accounts sound. This difference is smaller at low pressures.
for thermal and chemical non-equilibrium is the Homogeneous Re- The HRM of Bilicki & Kestin (Bilicki and Kestin, 1990; Bilicki
laxation Model (HRM). The HRM system consists of the mixture et al., 1990; Downar-Zapolski et al., 1996), conversely to Flȧtten
Euler equations augmented with a mass equation for one phase and Lund, 2012 and other 4-equation models, for one-component
with a relaxation source term for the flow quality (see Eq. (13)). two-phase flows assumes:
An important feature of HRM is the assumption that the va-
pl = pv = p,
por phase is always at saturation conditions. Note that this model
is different from the 4-equation relaxation model of Flȧtten and Tl = Tv = Tsat ( p), (14)
Lund, 2012, which accounts for chemical non-equilibrium but as- ul = uv = u.
sumes thermal equilibrium. These different assumptions are re-
flected in the definition of the relaxation source term and closure With the relaxations expressed in Eq. (14), the internal energies
relations. Let us also remark that other models such the 5-equation and the specific volumes of vapor are the ones corresponding to
model of Kapila et al., 2001 or the 6-equation model (Pelanti and the saturation curve. Whereas, for the liquid phase, they are in the
Keh-Ming Shyue,Saurel et al. (2009)) allow the description of non- metastable region of the superheated liquid, namely, along the iso-
equilibrium phenomena, however these models are more difficult baric line crossing the vapor saturation curve in (vg , eg ). Further
to handle numerically due to the presence of non-conservative explanations are reported on Fig. 1. Summarizing, we have:
terms in the hyperbolic portion. To better enlighten the novel con- ev = eg ( p),
tribution of this article, we preferred here to adopt the simpler
el = e f ( p),
HRM, which can describe metastable liquids without the need to (15)
address numerical difficulties related to non-conservative products. vv = vg ( p),
This is a coherent choice since in the fast transient flow problems vl = v f ( p).
here considered the processes related to the metastable states of
liquid are crucial. Moreover, using the IAPWS EoS, the description Since in general thermal equilibrium does not hold, the flow qual-
of the properties of the subcooled vapor (metastable with respect ity, x, does not correspond to the thermodynamic quality, i.e.
to liquid) is complicated since there are no experimental data for mass o f vapor h − hf
its thermodynamic properties (Wagner and Pruß, 2002b). x= = . (16)
total mass hg − h f
The governing equations of the HRM, in one dimension, are:
⎧ Another useful relation, obtained from the first two equations
⎪
⎪∂t (αv ρv ) + ∂z (αv ρv u ) = l→v , of Eq. (12), gives the flow quality, or the mass fraction, x:
⎨∂ ρ + ∂ (ρ u ) = 0,
t z
(12) αv ρv
∂t ( ρ u ) + ∂z ( ρ u + p ) = 0,
x= . (17)
⎪
⎪
2
ρ
⎩∂ (ρ E ) + ∂ [(ρ E + p)u] = 0.
t z
This is a relation between the volumetric and mass fractions occu-
To be able to relax the thermal equilibrium constraint, the mass pied by the vapor phase.
transfer between the two phases is different with respect to the The closure laws, for Eq. (12), are
one of a mixture at saturation conditions. The mass transfer is e = x ev ( p) + (1 − x ) el ( p, Tl ),
modeled by means of a source term,  l → v , which appears on the
right hand side in the equation of the mass balance of the vapor v = x vv ( p) + (1 − x ) vl ( p, Tl ). (18)
phase of Eq. (12). This source term tends to relax the current va- Similarly to the HEM, an equation for the pressure is needed.
por mass fraction to the equilibrium mass fraction introducing a With respect to the HEM, here, the flow quality is a variable
certain delay for the liquid vaporization: needed for the evaluation of the mixture pressure. Then, the pres-
x − xeq sure law for the HRM has the form:
l→v = −ρ , (13)
 p = p(x, v, e ). (19)
where xeq is the equilibrium quality.
HRM has good mathematical properties for numerical approxi- 3. Steam-water EoS for two-phase models
mation purposes, due to the absence of non-conservative terms. In
fact, excluding source terms, this model is under full conservative As seen in the previous section, an EoS to determine the pres-
form. Details about the mathematical properties of the HRM are sure is needed to close the systems of Eqs. (4) and (12). There are
discussed in Ambroso et al. (2009), Bilicki and Kestin (1990) and three ways to achieve this goal: to use a simple equation of state,
Faucher et al. (20 0 0). Moreover, HRM also has consistent phys- to employ an iterative algorithm adopting an accurate EoS or to
ical properties in terms of wave propagation. As we can see in use tabulated values.
Fig. 2, in the two-phase domain, the speed of sound for the HRM Generally, simple EoS allow fast properties calculations but they
is continuous and has the same non-monotonic behavior of the are not accurate and sometimes provide unrealistic results. Hence
well known Wood’s speed of sound (Wood, 1930) (which is also accurate EoS should be used for reliable simulations, as especially
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 203

demanded in an industrial context. Iterative algorithms for this 2004). For instance, such an EoS is not suitable for the simulation
type of EoS are time consuming, therefore, these EoS are often of energy injection transients, whereas, the tabulation method pro-
used in tabulated forms. posed here is valid for any test.
Let us detail further these three options in the next paragraphs.
3.2. Use of an iterative algorithm for an accurate EoS
3.1. Use of a simple caloric EoS
A way to accurately evaluate water properties is to use exper-
The behavior of a single-phase fluid or a mixture could be de- imental EoS. In the case of water, the large availability of experi-
scribed by a simple analytical law. Historically, EoS have always mental data allowed the formulation of a certain number of equa-
been formulated in a thermal form, but for very simple thermal tions which have been adopted as International Standards. EoS of
EoS, it is straightforward to obtain a relation as p = p(ρ , e ). How- this type have been adopted for the present work and they will be
ever, as the complexity of the thermal form increases, it becomes discussed more in detail in Section 4.
practically unfeasible to get the corresponding p = p(ρ , e ). The Generally, accurate EoS are expressed as:
simplest modelling approach is to assume that a fluid behaves as f = f (ρ , T ) or g = g( ρ , T ) , (24)
a perfect gas:
where: f = u − T s and g = h − T s are, respectively, the Helmholtz
p = (γ − 1 )ρ e, (20) free energy and the Gibbs free enthalpy. Such EoS are very accu-
where γ = c p /cv is the heat capacity ratio. However this equation rate but extremely costly from a computational point of view. For
is not adapted for stiff fluids such as water. industrial purposes, to reduce the computational time, we have:
A very simple generalization of the ideal-gas thermal EoS is the f = f ( p, T ) or g = g( p, T ). (25)
covolume EoS, also called Clausius-Noble-Able EoS:
These EoS are also called Industrial Formulation EoS. Such accurate
( γ − 1 )ρ e EoS are composed by many terms, even few hundreds. Generally,
p= , (21)
1 − bρ their terms come from a best fit of a huge amount of experimental
where b is the covolume and tries to take into account the volume data, as we shall discuss in Section 4. From such a cumbersome
occupied by the molecules themselves. EoS, it is not feasible to get an equation in the form: p = p(ρ , e ).
Historically, one of the most important EoS, in physics, has been However, those EoS can be coupled to the Euler system of equa-
the Van der Waals EoS: tions by using an iterative algorithm. More specifically, the EoS is a
non-linear equation to be solved. Details are given in Appendix C.
RT a
p= − . (22) Kunick et al. (IAP, 2015; Kunick, 2015) made a comparison be-
v−b v2 tween the iterative inversion of the IAPWS-IF97 and the look-up
The last term accounts for the forces of attraction between table interpolation. The iterative procedure is hundreds of times
molecules. Using a cubic EoS, the hyperbolicity of the Euler sys- slower than the interpolation.
tem of equations cannot be ensured because of the well-known
drawback of the positive isothermal compressibility, i.e. (∂ p/∂ v)T > 3.3. Look-up table interpolation
0, in the two-phase domain. Indeed, in the unstable domain en-
closed by the spinodal curve, the square of the speed of sound In order to have accurate fluid properties evaluations, the time
may become negative and wave propagation has no physical sense consuming iterative procedure for the EoS could be employed at
(Menikoff and Plohr, 1989). the beginning of the computation to create a table to be used
In other words, using a cubic EoS, in the metastable domain, the during the CFD simulation. This method is generally called look-up
return to equilibrium corresponds to a thermodynamic transforma- table interpolation of the fluid properties. Even though such a
tion. However, to preserve the hyperbolicity of the Euler equations, strategy is widely used in many CFD codes (Carpenter et al., 2013,
we are interested in a kinetic process, as explained in Saurel et al. RELAP, 2001; Robbe and Lepareux, 1996; Thomas, 1998; Tiselj
(2008), which does not lead to an imaginary speed of sound. To et al., 2008), most of these codes still use iterative procedures for
follow a kinetic process, a widely used EoS is the stiffened gas EoS the pressure calculation because the tabulation of water properties
(SG-EoS). It is suitable for stiff fluids such as water, since it is the is done with respect to the variable pairs: (p, T), (p, h), (p, s), (p, e)
linearization of the Mie-Grüneisen EoS which was initially devel- or (T, v).
oped for solids. It reads: In order to avoid costly iterative procedures, at least into the
single-phase domain, iterative algorithms (see Appendix C) could
p = (γ − 1 )(e − q )ρ − γ π , (23)
be employed at the beginning of the calculation on a fairly wide
where γ , π , q are constants to be determined. number of couples (v, e), so that, during the CFD calculation, pres-
The SG-EoS cannot be considered a globally-valid EoS. First of sure, temperature and speed of sound may be evaluated just by
all, it is based on a linearization of fluid properties, therefore, it interpolation. Therefore, the requirements of both accuracy and ef-
does not hold in the highly non-linear regions of the thermody- ficiency are fulfilled. For such strategy, the best attempt presented
namic diagram as in the neighborhood of the critical point or close in literature belongs to Kunick et al. (IAP, 2015; Kunick, 2015). In
to the spinodal curve. Furthermore, the linear nature of this EoS the two-phase domain, the calculation may be still iterative, but
does not allow one to accurately calculate first-order derivatives of the saturation curve can be discretized as a conjunction of splines,
thermodynamical properties, such as the speed of sound, c. easy to handle and at the same time sufficiently accurate and
Secondly, depending on choices made to determine the con- rapid.
stant q, this EoS may not exhibit a spinodal curve. This feature is A further beneficial feature of the tabulation strategy on the e-v
in disagreement with both theory and experiments (Debenedetti, diagram is that we know in advance if a pair (v, e) belongs to the
1996; Eberhart, 2009; Lienhard et al., 1986; Skripov, 1972). single phase or to the two-phase domain.
Several authors in the literature have shown that this EoS is ap- In Section 5.1 we show how to increase the accuracy, with re-
propriate for some test problems, including some industrial test spect to IAP (2015) and Kunick (2015), by using bicubic interpo-
cases (Daude et al., 2014; Lochon et al., 2016). However, since its lations. Moreover, with the goal of simulating fast transients, the
parameters are defined around a reference state, a sufficiently large tabulation has to be extended into the metastable domain of wa-
temperature change could lead to an erroneous evaluation of the ter, as has been done in RELAP-7 (Berry et al., 2016). Here, this
thermodynamic properties (see the results in Le Metayer et al., extension is done up to the assumed spinodal curve.
204 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

4. IAPWS-IF97 and the representation of the liquid spinodal the criteria for equilibrium and stability can be expressed as:
curve δ S = 0, criterion of equilibrium,
Nowadays the most reliable EoS for water is the IAPWS- δ n S < 0, for the smallest n at which δ n S = 0, (27)
95 (Wagner and Pruß, 2002a). It replaced the formulation of criterion of stability.
Haar et al. (1984), which was the adopted international standard
Further developments of the last criterion provide two conditions
since 1984 and still widely used within many codes for nuclear
which ensure the stability of a system:
safety. The IAPWS-95 is an analytical equation based on a multi-
parameter fit of all the experimental data, used for general and cv > 0 cr iter ion o f thermal stability,
scientific purposes. The formulation is a fundamental equation for  
∂p (28)
the specific Helmholtz free energy, f, expressed in dimensionless <0 cr iter ion o f mechanical stability.
form. Appropriate combination of derivatives can produce any de-
∂v T
sired thermodynamic property. If both are satisfied, the system is intrinsically stable. The crite-
This formulation behaves reasonably also when extrapolated rion of thermal stability is satisfied for all the EoS developed so
into the metastable domain (Wagner and Pruß, 2002b). However, far, whereas the fulfillment of the mechanical stability criterion
it is rather difficult to find a formulation indicating the spinodal is not always ensured. As discussed in Section 3.1, cubic EoS sat-
curve since it exhibits several stiff minima (also called Marianna isfy the mechanical stability condition until the so-called spinodal
Trench) and stiff maxima (said Himalaya) along an isotherm (Imre, line. The spinodal line separates the unstable domain from the sta-
2013). ble/metastable one. In classical thermodynamics phase transition
Because the IAPWS-95 formulation is quite cumbersome and occurs at saturation conditions, on the contrary, in real systems an
time consuming, IAPWS also maintains a separate formulation rec- isothermal pressure change would very likely carry the state point
ommended for industrial use. It is the IAPWS-IF97 (Wagner et al., into the metastable domain. The deeper the metastability reached,
20 0 0), which replaced the IFC (1967). The IAPWS Industrial Formu- the higher is the probability for the phase change to occur. The
lation 1997 consists of a set of equations for five different regions, spinodal curve is the limit beyond which phase change has already
each of them is covered by a simple basic equation fitting the val- occurred.
ues resulting from the IAPWS-95. The liquid and vapor spinodal lines connect all the minima of
In this work we have adopted the IAPWS-IF97 for three reasons: the isotherms (liquid spinodal) and all the maxima (vapor spinodal)
• IAPWS-IF97 is more suitable for the calculation of the spin- closest to the associated side of the saturation curve within the
odal curve, since there are no stiff minima or maxima along saturation dome. In these minima and maxima points isotherms
isotherms, which instead characterize the IAPWS-95, as dis- have zero slope (see Figs. 1 and 6), hence, spinodal lines mark the
cussed above; location from where the mechanical stability condition is violated.
• IAPWS-IF97 has an ad hoc formulation for the metastable vapor In the present work, the spinodal curve has been determined by
region which involves a low-density gas equation that is con- setting to zero (∂ p/∂ v)T to detect the limit of the mechanical sta-
sidered to be more adequate than IAPWS-95 (see IAP, 2007). bility. The resulting spinodal curve agrees with other curves pre-
This choice has been made in view of future work based on sented in the literature as shown in Fig. 5. The curve (∂ p/∂ v )T = 0
the hyperbolic 6-equation model of Saurel et al. (2009) and has been defined using an optimized Newton–Raphson algorithm
Pelanti and Keh-Ming Shyue which would use both metastable (Press et al., 1997). To this end, the basic equation of Region 3
liquid and vapor. In the current work, however, metastable va- (Fig. 3) was used, even slightly below the temperature of 623.15 K
por states are not considered because HEM and HRM do not which indicates the boundary between Region 1 and 3. As shown
need them; in Fig. 6, this choice provides accurate results even when the ex-
• The discrepancy between the IAPWS-IF97 and the IAPWS-95 is trapolation is done below the limit temperature.
very small (Wagner et al., 20 0 0). Such strategy revealed that the spinodal corresponds to positive
values of the pressure until T ≈ 593.50 K. Below this temperature,
The region subdivision adopted for IAPWS-IF97 is reported in the limit of mechanical stability corresponds to negative pressures.
Fig. 3. The constraint of a positive pressure holds for a gas or vapor, but
In addition, IAPWS also provided some backward equations to not for a liquid. Negative pressures can be reached by depressuriz-
make the calculation of properties faster. However, they are not ing a liquid isothermally below its saturation pressure. The possi-
used in the work presented in this paper, because they are not as bility for a liquid to be at negative pressure means that it possesses
accurate as the basic equations. a tensile strength. At equilibrium, water and other liquids do not
The entire set of equations originally covered the following support an externally applied mechanical tension. They can resist
range of validity: to it in a thermodynamically metastable state, resulting in a neg-
273.15 K ≤ T ≤ 1073.15 K p ≤ 100 MPa, ative pressure. The laboratory study of liquids under tension has a
(26) long history and the literature abounds with discussions on this
1073.15 K ≤ T ≤ 2273.15 K p ≤ 10 MPa.
subject (e.g. Caupin and Stroock, 2013; Debenedetti, 1996; Imre,
The subdivision of the entire domain in different regions allows an 2013; Kiselev and Ely, 2001; Lienhard et al., 1986; Skripov, 1972;
efficient property computation, even if still too slow for CFD cal- Zheng et al., 1991).
culations. The IAPWS-IF97 also deals with metastable states, pro- However, in this work we limited the tabulation to positive val-
viding reasonable values when extrapolated. Last but not least, the ues of the pressure, choosing the isobaric at p = 0 MPa as border of
basic equation of Region 3 allows one to easily define a spinodal the metastable domain for temperatures lower than T ≈ 593.50 K.
curve (see Figs. 4 and 6). See Fig. 4 for a more effective understanding.
All the features discussed here motivated the authors to adopt Regarding the saturation curve, the equilibrium criterion is en-
the IAPWS-IF97 for the tabulation on the e-v diagram. sured by the equality of Gibbs free enthalpy for the two phases:
gl = gv . Along an isotherm, the Gibbs–Duhem equation between
4.1. Spinodal and saturation curve for water
the two saturated states reduces to:
 v
From thermodynamics, it is well known that a closed system
gv − gl = vdp. (29)
attains a maximum value of entropy at equilibrium (Carey). Hence, l
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 205

Fig. 3. Region subdivision of the IAPWS-IF97 on the p-T diagram. Region 4 represents the saturation curve.

Fig. 4. Metastable domain considered in this work. In both the e-v and p-T thermodynamic planes, the saturated liquid curve, the liquid spinodal and the isobaric of 0 MPa
are represented. The unstable domain represents the states where the mechanical stability condition is violated (Eq. (28)). Below the intersection between the spinodal
and the isobaric at p = 0 MPa, the metastable domain refers to negative pressures. Authors decided to restrict the tabulation of the thermodynamic properties just on the
metastable states at positive pressure.

Then, the saturation curve, psat (T), can be constructed by requiring best approach is to tabulate the required properties on a e-v di-
the vanishing of Eq. (29). The IAPWS-IF97 provides a suitable ba- agram. Such a diagram is quite uncommon in the literature, hence
sic equation that relates saturation pressure and temperature. This Fig. 7 should make the reader more familiar with the behavior of
simple relation to define the saturation states has been adopted properties in the considered thermodynamic plane.
herein. In this section we present the property evaluation method that
has been coupled to the HEM and HRM in the EUROPLEXUS code.
To the best of authors’ knowledge, the method proposed in the fol-
5. Look-up table method for thermodynamic properties
lowing section is the most accurate among those presented in the
evaluation on the e-v diagram
literature.
Fluid properties calculation during CFD simulations should be
as fast as possible, and, at the same time, highly accurate. A look- 5.1. Grid construction on e-v diagram and bicubic interpolation
up table method can meet these requirements through the use
of simple interpolation techniques to determine the thermody- The physical domain can be discretized by using a sufficiently
namic properties. The Euler equations and mixture model systems high number of nodes. For easiness, the irregular physical domain,
for compressible flows such as the ones considered in this paper e-v, has been transposed in Cartesian transformed domain, Y-X.
require a closure law of the form p = p(ρ , e ), consequently the This feature is depicted in Fig. 8.
206 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

Fig. 5. Trend of spinodal curves on the p-T diagram from different formulations. The one determined in this work (labeled as IAPWS-IF97) agrees with the other ones
proposed in literature (Eberhart, 2009; Imre, 2013).

the grid. Then, constraints were imposed to ensure the continuity

of first derivatives throughout the boundaries of the cells of the e-v
plane.
Differently from IAP (2015), here a bicubic interpolation is per-
formed. Thus, the values of the derivatives are provided in each
node of the grid. The bicubic interpolation for the pressure calcu-
lation is the object of the following section.

5.2. Bicubic interpolation

To perform a bicubic interpolation, the cell must be square. For

a given non-uniform quadrilateral on the e-v diagram, its transfor-
mation onto a regular cell of an arbitrary Y-X diagram is required.
Let us rescale the coordinates Y-X in order to have a square cell of
unit length.
The goal is to construct a function such as (v, e ) = (X, Y ),
where X ∈ [0, 1] and Y ∈ [0, 1], which describes the entire point
Fig. 6. Isotherms on the p-v diagram calculated at three different temperatures ac- space enclosed by the quadrilateral cell. This function, yet to be
cording to three basic equations of state from IAPWS-IF97. The basic equation of determined, forms a map that allows us to transform the quadri-
Region 3 provides accurate results even when extrapolated in metastable region at
lateral cell, that is:
T < 623.15 K.

: R2 → R2 such that : ∀(v, e ) ∈ D ⇒ (X, Y ) ∈ D ,

The physical domain has been mapped as follows. In the liq- (31)
uid and supercritical domain we imposed an equidistant pattern of
nodes, hence the distribution is linear, whereas, in the vapor do- where D is the physical domain and D the transformed one. The
main the node distribution is logarithmic. Handling such a regu- chosen mapping function, fulfilling the requirements of the expres-
lar distribution of nodes ensures that, the cell (i, j) containing the sion in (31), is a bilinear mapping (Hughes, 20 0 0; Ruas, 2016), that
point can be immediately found: is:
Y0 − Ymin X0 − Xmin
i = int , j = int , (30) v = α1 + α2 X + α3Y + α4 X Y, (32)
Y X
where X and Y are the coordinates on the transformed space. i and and
j are the indices of the cells along the Y and X directions, respec-
tively. Such a feature is very beneficial in terms of computational e = β1 + β2 X + β3Y + β4 X Y. (33)
time. Then, a bicubic interpolation is performed on the Cartesian
diagram Y-X, using stocked values on the e-v diagram. Setting the vertices of the irregular quadrilateral cell on the e-v
In IAP (2015), the interpolation was a spline based one, hence plane to be the vertices of the unit length square of the trans-
the coefficients of spline polynomials were determined by provid- formed space, the α i , β i coefficients can be obtained. The indicators
ing values of the exact derivatives just at the outer boundary of i are distributed in the clockwise direction starting from the node
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 207

Fig. 7. Thermodynamic diagram e-v. (a) reports the region subdivision of the IAPWS-IF97 already illustrated in Fig. 3. In (b) we show the trend of the isoquality curves in
the two-phase domain. In (c) the reader can see the isobaric curves in the liquid domain (bottom left corner), two-phase domain and vapor region until T = 1073.15 K. (d)
depicts the isothermal curves. The thermodynamic domain on the e-v plane is highly irregular, hence, particular attention is demanded for the node distribution.

in the bottom left corner. The coefficients are calculated solving simple enough to be solved analytically. In fact, Y0 is the solution
⎡ ⎤⎡ ⎤ ⎡ ⎤ of the following quadratic equation:
1 0 0 0 α1 v1
⎢1 0 1 0⎥⎢α2 ⎥ ⎢v2 ⎥ aY02 + bY0 + c = 0, (37)
⎣1 1 1 1⎦⎣α3 ⎦ ⎣v3 ⎦
= . (34)
for β2 , β4 = 0, the coefficients are
1 1 0 0 α4 v4
a= α4 β3 ,
A similar expression can be written for the relation between ei and
β i . Finally, in a general framework: b= α4 (β1 − e0 ) + α2 β3 ,
c = α2 (β1 − e0 ). (38)
α1 = v1 ,
The physical solution is:
α2 = −v1 + v4 , √
(35) −b + b2 − 4ac
α3 = −v1 + v2 , Y0 = . (39)
2a
α4 = v1 − v2 + v3 − v4 , Once Y0 is known, it is straightforward to find X0 :
v0 − α1 − α3Y0
β1 = e1 , X0 = . (40)
α2 + α4Y0
β2 = −e1 + e4 , Bicubic interpolation needs first and second order cross deriva-
(36)
β3 = −e1 + e2 , tives to be known into the transformed space. Let us call F(v, e)
β4 = e1 − e2 + e3 − e4 . the function into the physical space and F˜ (X, Y ) the correspond-
ing function into the transformed space. In our case, F corresponds
Given that the grid is formed by horizontal iso-e lines: β2 , β4 = 0. to pressure, temperature or speed of sound. Using both first and
During a numerical simulation, the couple (v0 , e0 ) must be second order chain rules:
transformed into the point (X0 , Y0 ). That is, the system formed by
∂ F˜ (X, Y ) ∂ F ∂v ∂ F ∂ e
Eqs. (32) and (33) should be solved. This system is non-linear but = + , (41)
∂X ∂v ∂ X ∂ e ∂ X
208 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

Fig. 8. Grid of nodes in the physical domain and in the transformed one.

interested in the pressure calculation, thus:

∂ F˜ (X, Y ) ∂ F ∂v ∂ F ∂ e
= + , (42) p ( v, e ) = 
p ( X, Y )
∂Y ∂v ∂ Y ∂ e ∂ Y
   T (46)
= 1 X X2 X3 A 1 Y Y2 Y3
∂ F˜ (X, Y )
2
∂F ∂ v 2
∂F ∂ e 2
= +
∂ X ∂Y ∂v ∂ X ∂ Y ∂ e ∂ X ∂ Y where A is the coefficient matrix of Eq. (45).
∂ 2 F ∂v ∂v ∂ 2 F ∂ e ∂ e The function p˜ (X, Y ) refers to a Cartesian square grid and, by
+ 2 + (43)
∂v ∂ X ∂Y ∂ e2 ∂ X ∂ Y construction (Press et al., 1997), has the following properties: i) the
∂ 2 F ∂v ∂ e ∂v ∂ e values of the function and the specified derivatives are reproduced
+ + ,
∂ v∂ e ∂ X ∂ Y ∂ Y ∂ X exactly at the nodes, and, ii) the function and the specified deriva-
where the first order partial derivatives of the physical coordinates tives change continuously at the edges of the square cells. Further-
with respect to the transformed ones (i.e. ∂∂vX , ∂∂Xe , ∂∂vY , ∂∂Ye ) are eval- more, these smoothness properties do not depend on the accuracy
uated by using Eqs. (32) and (33). Whereas, first order derivatives of the specified derivatives (Press et al., 1997).
of the function F (i.e. ∂∂vF , ∂∂Fe ) have been analytically calculated by Due to the choice of the bilinear transformation, p(v, e) is still
using the equations of state provided by IAPWS-IF97. Second order a continuous function across the cell edges in the e − v diagram.
derivatives are approximated by difference quotients of first deriva- However, not all the derivatives are continuous across the edges.
tives instead. More details about derivative calculations are given Further details about derivatives calculation are given in the next
in Appendix D. section (cf. Section 5.3).
The interpolating surface is a polynomial composed of 16 ad- This procedure for calculating thermodynamic properties has a
dends, and it describes the function F on the transformed space: high level of accuracy. As we can see in Fig. 9, the fidelity in the
reproduction of the IAPWS-IF97 is very well ensured. Fig. 9 reports

3 
3 the relative error between the values calculated using the IAPWS-
F˜ (X, Y ) = ai j X i Y j . (44) IF97 and the look-up table method here proposed. Pressure is the
i=0 j=0 physical magnitude which has the strongest gradients on the e-v
The 16 coefficients aij are the elements of the matrix A: diagram, thus, the most difficult field to be represented. The speed
⎡ ⎤ of sound and the temperature have a much smoother trend.
1 0 0 0 It is worth reminding that the tabulation has been carried out
⎢ 0 ⎥
⎢ 0 1 0 ⎥ only in the single-phase domain and in the metastable one. Prop-
A=⎢ ⎥ erties calculation into the two-phase domain relies on iterative al-
⎣ −3 3 −2 −1 ⎦
gorithms which use a spline reconstruction of the saturation curve
⎡2 ˜ −2 ˜ 1 1
⎤ (Appendix C). A comparison with the results obtained by Kunick
F1 F2 ∂Y F˜1 ∂Y F˜2 et al. (Kunick, 2015) shows that the method here proposed is more
⎢ F˜ F˜3 ∂Y F˜4 ∂Y F˜3 ⎥ accurate. In spite of a lower number of nodes with respect to
⎢ 4 ⎥
×⎢ ⎥ (45) Kunick (2015) (160,0 0 0 nodes here vs. 230,0 0 0 nodes in Kunick
⎣ ∂X F˜1 ∂X F˜2 ∂X,Y F˜1 ∂X,Y F2 ⎦
˜ (2015)), relative errors are around 2 orders of magnitude less on
the whole tabulated domain. Furthermore, in the present work, the
⎡ ∂X F4 ∂X F˜3 ∂X,Y F˜⎤ ∂X,Y F˜3
˜ 4
1 0 −3 2 domain has been extended to the high temperature region and to
⎢ 0 ⎥ metastable liquids.
⎢ 0 3 −2 ⎥
×⎢ ⎥,
⎣ 0 1 −2 1 ⎦ 5.3. Derivatives calculation from bicubic splines
0 0 −1 1
where, for instance, ∂X F˜1 indicates the derivative of the function F For a general function F(v, e), the derivatives are calculated us-
with respect to the coordinate X at node 1, that is, the node in the ing the chain rule. Let us rewrite Eqs. (41) and (42) in matrix form:
bottom left corner.    ∂F 
Once the coefficients of the bicubic interpolation are deter- ∂ F˜
∂X
mined, the function is calculated as follows. Let us suppose to be = J ∂v . (47)
∂ F˜ ∂F
∂Y ∂e
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 209

Fig. 9. Distribution of the relative error for pressure between values calculated using the IAPWS-IF97 and the look-up table method here discussed. The error refers to the
calculation of p = p(ρ , e ) for the Euler system. The first plot reports the error on the whole domain of the IAPWS-IF97 on the e-v diagram. The second plot is a magnified
view for the liquid region. Here we can observe the trend of the saturation curve, the liquid spinodal and the isobaric of p = 0 MPa. The grid of nodes adopted to map the
physical domain consists of 160,0 0 0 nodes and roughly 3 million points have been tested to produce this plot.

where the Jacobian of the transformation is and its inverse is

 ∂v ∂e   
1
∂X ∂X α2 +α4Y 0
J= , (48) J −1
= . (51)
∂v ∂e
∂Y ∂Y − β α(α3 ++αα4 XY ) 1
β3
3 2 4
Thus,
 ∂F    Given that the interpolating function is the bicubic spline in (44),
∂ F˜
∂v −1 ∂X the derivatives in the Y − X domain are
=J . (49)
∂F ∂ F˜
∂e ∂Y ∂ F˜  3  3
= ai j iX i−1Y j ,
The Jacobian of the transformation must be nonsingular, in or- ∂X i=0 j=0
der to have an invertible transformation. This requirements ensures
that each point within the cell in e − v corresponds to one and only ∂ F˜  3  3

one point in the canonical cell in Y − X. = ai j jX iY j−1 . (52)

∂Y i=0 j=0
For the bilinear transformation expressed in Eqs. (32) and (33),
with β2 , β4 = 0, the Jacobian of the transformation is Therefore, once the point (X0 , Y0 ) is known, the first order partial
 
α2 + α4Y 0 derivatives can be evaluated by using Eqs. (49), (51) and (52).
J= . (50)
α3 + α4 X β3
210 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

Note that first order partial derivatives in the Y − X diagram As described above, Eq. (56) is solved in two steps: convec-
are continuous at the inter-cell edges, however, this property is tion and relaxation. The convection step over the time interval
not generally verified in the e − v diagram. Analyzing Eq. (51), one [t n , t n + t] provides the evolution of the quality from xn to x˜n .
can observe that continuity is verified only for ∂∂vF across the hor- The relaxation equation becomes
izontal edges thanks to the regular pattern on nodes along the e- dx x − xeq
axis. The other derivatives are not continuous but numerical tests =− . (57)
dt 
have shown that, at the edges, the discrepancy is small. Nonethe-
less, partial derivatives are not often employed in numerical simu- This equation is integrated over the time interval [t n , t n + t] us-
lations. For instance, HEM and HRM here tested do not need partial ing as initial value x˜n . Due to the simple form of Eq. (57), the inte-
derivatives evaluations. gration is analytical. If  = (x, t):
t
xn+1 = xeq − (xeq − x˜n )e−  . (58)
6. Numerical simulations of steam-water transients
By definition, xeq > x˜n ,
therefore xn+1 < xeq .
The method for properties calculations discussed in the previ- Consistently with its construction, the HRM tends to the HEM
ous sections has been implemented in the EUROPLEXUS fast tran- if  → 0+ , in fact
 
sient dynamics software, a simulation tool employed for instance t
lim xeq − (xeq − x˜n )e−  = xeq . (59)
in the nuclear safety field. →0+
The numerical method used for this work relies on a Finite Vol- The mass transfer is activated just if metastability is detected. For
ume discretization for the solution of the PDEs reported in Eqs. instance, when a liquid undergoes a depressurization, it becomes
(4) and (12). The numerical scheme is the HLLC-type method pre- metastable. In the case of the HRM, this condition is detected by
sented and assessed in Daude et al. (2014). As mentioned above, using the steam-water tables here proposed. A liquid is metastable
HEM is a homogeneous set of partial differential equations, thus, it if the couple (e, v) belongs to the region bounded by the saturation
does not need additional numerical techniques for its solution. curve and the liquid spinodal. Refer to Figs. 4 and 9b for a graphic
For the HRM, a fractional step technique is used to solve the illustration. For a two-phase mixture, the metastability is detected
system of equations. It means that the numerical method proceeds when x˜n < xeq .
in two steps. At each time step, first the homogeneous hyperbolic
portion of the system with no mass transfer is solved. This step 6.2. Steam-water transients
provides the non-equilibrium hydrodynamic field. Then, the mass
transfer term is activated where metastable liquid is detected (re- Now that the numerical strategies have been discussed, we can
laxation step). assess them on three challenging fast transients tests: fast depres-
The first step represents the convection: surization, waterhammer and steam explosion problems.
∂t U + ∂z F(U ) = 0. (53)
6.2.1. Depressurization transients
For the HEM the vector of the conservative variable is UHEM = In the context of nuclear safety, one of the Design Basis Acci-
(ρ , ρ u, ρ E )T , whereas, for the HRM UHRM = (αv ρv , ρ , ρ u, ρ E )T . dents (DBA) is the Loss of Coolant Accident (LOCA), which would
By using the HLLC-type Riemann solver, we compute the ap- occur in the case of a sudden rupture of a primary system pipe.
proximate solution over the time interval [t n , t n + t] with given In the ’70s–’80s, few experimental facilities have studied the rapid
initial value Un . The approximate solution of Eq. (53), U ˜ n , will be depressurization of vessels or pipes containing water at subcooled
the initial value for the second step, i.e., the relaxation step. By conditions (Bartak, 1990; Edwards and O’Brien, 1970; Riegel, 1978).
solving the following ODE: In this section we try to validate the model discussed above
dU against these physically complex and challenging experiments. The
= b, (54) benchmark is performed using the experimental data of Bartak
dt
(1990) and Super Canon test rig (Riegel, 1978). The initial pres-
we finally obtain the approximate solution Un+1 corresponding to sure and temperature conditions for the first test are 12.5 MPa and
the time level t n+1 . 563.15 K, and for the second test 15 MPa and 573.15 K. The cor-
The source term b could contain dissipative terms such as wall responding subcooling degrees, i.e. Tsat ( pin ) − Tin , are respectively
friction, body forces or injected energy. In the case of the HRM, the 38 K and 42 K.
vector b contains the mass transfer term,  l → v . The arrangement of the Super Canon facility and the measure-
ment devices are shown in Fig. 11. Both experiments are triggered
6.1. Mass transfer for the HRM by the sudden opening of a rupture disc located at one extrem-
ity of the pipe. At that moment, a rarefaction wave originates at
The Homogeneous Relaxation Model is handled by adding one the break and moves backward. The very first milliseconds follow-
more differential equation (i.e. the first equation of system (4)) to ing the rupture are well represented by the Bartak data, whereas,
the HEM. This supplementary equation drives the evolution of the the whole transient is better described by the Super Canon exper-
quality x. The source term  l → v describes how the quality x tends iments.
towards its local unconstrained-equilibrium value xeq . In fact the The comparisons of HEM and HRM simulations against the ex-
quality is no longer assumed to instantaneously reach the equilib- perimental data are reported in Figs. 10 and 12. From the exper-
rium value of xeq , hence a relaxation time,  > 0, is introduced. iments, we know that the measured velocity of the rarefaction
The vapor rate equation is wave compares well with the isentropic speed of sound of single-
x − xeq phase water. This feature is well reproduced by both models since
∂t (αv ρv ) + ∂z (αv ρv u ) = −ρ . (55)
 the rarefaction wave propagates in the single-phase liquid.
By using the continuity equation, this gives The depressurization is initially abrupt and fast. The pressure
reaches values lower than the saturation pressure corresponding to
Dx x − xeq the initial stagnation temperature (7.4 MPa in Bartak experiment,
=− , (56)
Dt  8.6 MPa for Super Canon). The difference between the local pres-
where D/Dt ≡ ∂t + u∂z . sure and the saturation pressure is as high as 5 MPa, hence, the
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 211

oped on the (p, T) diagram. Now, however, the code implements

the bicubic look-up table algorithm presented in Section 5. The di-
rect calculation of properties allows us to save about the 60–65% of
the total computational time in the case of a Super Canon experi-
ment. Hence, the calculation lasts 1/3 of the time needed before.

6.2.2. Relaxation time correlations

The relaxation time  accounts for the delay necessary to reach
equilibrium conditions. The relaxation term is of paramount impor-
tance in the HRM.
Downar-Zapolski et al. (1996) proposed two correlations based
on the experimental data of the Moby Dick (Reocreux, 1974) facil-
ity, which was a test rig for the stationary critical flow in flashing
steam-water (De Lorenzo et al., 2017). The first correlation, here
indicated as DZ1, is:
−2.24
psat − p
 = 6.51 · 10−4 αv−0.257 , (60)
psat
the second one, indicated as DZ2, reads:
Fig. 10. Pressure evolution in the first 16 ms of the Bartak’s experiment (Bartak,
−1.76
psat − p
1990) (12.5 MPa, 563.15 K). Assessment of HEM and HRM using the relaxation time
 = 3.84 · 10−7 αv−0.54 , (61)
correlation proposed in Eq. (62). pcrit − psat
where pcrit is the pressure of the critical point (22.064 MPa). These
liquid water tends to penetrate in depth into the metastable do- correlations express the relaxation time  in seconds. Gale et al.
main. The undershoot is well reproduced by the HRM, whereas, (2008) proved that these correlations well represent the physics
for the HEM, the mixture remains at saturated condition causing of vaporization even in unsteady conditions. However they fail in
the pressure to stay constant for few hundreds milliseconds. simulating the Super Canon experiments. Here we propose a cor-
Fig. 10 shows that the depressurization is stopped by bubble relation that better fits the pressure and void fraction evolution at
nucleation, which leads to a small but extremely quick increase of high pressures:
pressure. Skripov et al. (1980), Alamgir and Lienhard (1981) and −1.8
psat − p
Bartak (1990) agree that bubble nucleation occurs too quickly to be  = 4 · 10−5 αv−0.25 . (62)
the classical heterogeneous nucleation at the wall cavities. Hence, pcrit − psat
it could be the homogeneous nucleation (Carey), provoked by the The comparison of the three correlations is reported in Figs. 13–
high metastability reached by the liquid. The HRM can reproduce 15. Globally, the correlation here proposed represents better the
this phenomenon thanks to the exponential rate of return to equi- whole transient of Super Canon experiments. Additionally, it well
librium conditions. The mass transfer from liquid to vapor is acti- captures the pressure undershoot following the rarefaction wave
vated only if metastable conditions are detected. (Fig. 10). In fluid-structure interaction problems, this feature is cru-
After the explosion-like nucleation, the pressure remains con- cial because the amplitude of the rarefaction wave is an important
stant for a while, but at values lower than the saturation one. After factor in calculating the mechanical stress on the structures. For
100 ms, the wave propagation phenomena are no longer relevant instance, in the event of a full-break LOCA mentioned above, the
and the pressure decreases due to the emptying of the capacity. correct simulation of the rarefaction wave may provide informa-
In fact, the closed-end system empties at the rate imposed by the tion on the possible damage to the nuclear core.
two-phase critical flow that sets at break (De Lorenzo et al., 2017).
Finally, the simulation performed with the HRM agrees with 6.2.3. Waterhammer
most of the features experimentally observed. In this test, the HEM Simpson’s column-separation-induced waterhammer experi-
seems to not be sufficient for modelling the acoustic phase of this ment (Simpson, 1989) is a benchmark of many two-phase com-
transient flow due to the impossibility of dealing with metastable puter codes because of its simple geometry and the richness of
liquids. phenomena involved. A pipe is filled by a slow steady state liquid
The HEM was already implemented in EUROPLEXUS, but cou- water flow which is interrupted by a sudden closure of the valve
pled to an iterative algorithm for the thermodynamic properties at the end of the pipe. The pipe is 36 m long, has constant inter-
calculation. The iterative procedure used steam-water tables devel- nal diameter of 0.01905 m and the wall thickness is 0.001588 m.

Fig. 11. Schematic of Super Canon experimental facility and location of the measurement devices.
212 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

Fig. 12. Assessment of HEM and HRM on a Super Canon experiment. The initial stagnation conditions are 15 MPa and 573.15 K. For this calculation, the 4.389 m long tube
has been discretized into a 10 0 0 cells mesh. In the first milliseconds, the difference between local pressure and saturation pressure is as high as 5 MPa. The relaxation time
correlation used for this simulation is the one proposed in Eq. (62).

The initial thermodynamic state in the pipe in terms of pressure reflected at both pipe ends. According to the initial velocity, pres-
and temperature is 0.3419 MPa and 297.05 K, hence, water is single- sure may decrease enough to attain saturation conditions, thus,
phase liquid. Two experiments are analyzed here, in the first the generating a vapor pocket at the valve.
liquid initially flows at 0.239 m/s, in the second the initial velocity For this experiment, the wave propagation velocity is affected
is 0.401 m/s. The wave propagation in the tube is triggered by the by the elasticity of the pipe. In order to obtain a fair agreement be-
sudden closure of the valve, waves propagate in the pipe and are tween the simulations and experimental data, the speed of sound
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 213

Fig. 13. Comparison between HEM and HRM using different correlations for  on Fig. 14. Comparison between HEM and HRM using different correlations for  on
pressure and void fraction evolution in the Super Canon experiment with initial pressure and void fraction evolution in the Super Canon experiment with initial
data: 15 MPa, 553.15 K. In the legend, DZ1 and DZ2 indicate the correlations pro- data: 15 MPa, 573.15 K. In the legend, DZ1 and DZ2 indicate the correlations pro-
posed by Downar-Zapolski et al. (1996). posed by Downar-Zapolski et al. (1996).

is modified according to Allievi’s correlation (Allievi, 1913). Here

Similar considerations about the computational cost as those
we used the optimal Young’s elasticity modulus of the pipe sug-
made in Section 6.2.1 can be made also for the Simpson’s exper-
gested in Gale et al. (2008), i.e. 75 GPa, however the real one is
iment. A comparison between the calculation performed by using
around 119 GPa.
the numerical implementation of the HEM previously present in
Fig. 16 shows the assessment of the HEM and the HRM on the
EUROPLEXUS, and the calculation carried out by using the method
experimental data of Simpson (1989). The top graph shows the
presented in this work, suggests that the new algorithm that we
pressure history near the valve for the test with initial velocity
have developed is less time consuming. The use of an interpolation
equal to 0.239 m/s. This value is low enough to not provoke va-
technique for water properties calculation rather than an iterative
por formation. The comparison with experimental data shows very
one allows us to save the 20–25% of the total computational time
good agreement. Pressure peaks are well described by the simula-
that was required before.
tion except for the time-decreasing amplitude of pressure waves.
This feature is not well represented by simulations due to the ab-
sence of pressure losses in our numerical method. Of course the 6.2.4. Steam explosion
HEM and the HRM exactly agree since the fluid is always single- A steam explosion is an interaction process between two fluids
phase liquid. at very different temperatures. Namely, the temperature of the hot
The bottom graph of Fig. 16 shows the pressure history near liquid is higher than the saturation temperature of the cold one.
the valve for the test with initial velocity equal to 0.401 m/s. Con- Such phenomena are typical of iron foundries due to the interac-
versely to the previous test, cavitation occurs and a vapor pocket is tion between melting metal and coolant water or in petroleum en-
created close the valve. Moreover, the disappearance of this pocket gineering because of the contact of water and liquefied natural gas.
generates a new pressure wave, thus a combination of several In the nuclear industry, steam explosion may occur during a severe
waves occurs as it is visible at t ≈ 175 ms. Again, the lack of dissi- reactor accident when the molten fuel comes into contact with the
pative phenomena in our simulations leads to a slight overestima- coolant water.
tion of pressure peaks. In this simulation, the HEM and the HRM During a steam explosion the energy of the hot liquid is trans-
are coincident. Few discrepancies between the two models are de- ferred to the coolant in a time scale smaller than the time scale for
tected but irrelevant for the global description of the experiment. system pressure relief. Therefore the vaporization leads to the for-
214 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

Fig. 16. Pressure history near the valve of Simpson’s waterhammer experiment
Fig. 15. Comparison between HEM and HRM using different correlations for  on
(Simpson, 1989). The top graph shows results for the experiment with initial veloc-
pressure and void fraction evolution in the Super Canon experiment with initial
ity 0.239 m/s, the bottom one for the experiment with initial velocity 0.401 m/s. The
data: 15 MPa, 593.15 K. In the legend, DZ1 and DZ2 indicate the correlations pro-
initial pressure and temperature conditions for both experiments are 0.3419 MPa
posed by Downar-Zapolski et al. (1996).
and 297.05 K. For this calculation, the 36 m long tube has been discretized into a
10 0 0 cells mesh.

mation of shock waves that might endanger the surrounding struc-

tures. cation of the first 55 ms of the transient is reported in Fig. 19. The
In the context of nuclear safety, Board et al. (1975) recognized results are obtained by using the homogeneous equilibrium model.
that a steam explosion is very similar to a chemical detonation. We can see a pressure peak of 16 MPa during the first 15 ms
Similarly to a detonation, the propagation of a shock wave through due to the initial confinement of water. As the velocity increases,
a pre-mixture of fuel and coolant leads to a quick and fine frag- the pressure decreases rapidly, then increases again smoothly (see
mentation of the fuel, which releases a huge amount of its inter- Fig. 19). Therefore a pressure wave originates and moves right-
nal heat. The rapid heating of the coolant produces high pressures ward. At 80 ms we notice a pressure peak of 85 MPa due to the
which drive the shock forward (Berthoud, 20 0 0). quick condensation of the vapor at the end of the tube. From now
If the steam explosion can be assumed as a chemical detonation the transient becomes similar to a waterhammer as in Section 6.3.
(Board et al., 1975), a simple numerical modelling might be a quick Waves propagate in the tube reflecting at the extremities and/or at
energy injection. Here we performed a test proposed in Lepareux the location of a sudden change of the acoustic impedance Z = ρ c.
(1994). A schematic of the injection test is depicted in Fig. 17. This occurs, for instance, when the wave is propagating into a liq-
A 12 m long tube is divided into three parts, the leftmost one uid and encounters a two-phase mixture.
and the central one contain liquid water, the rightmost part con- At the end of the transient, the stagnation pressure is increased
tains vapor. The pressure field is uniform and equal to 3.517 MPa and has reached the value of 12 MPa. In fact the pipe can be seen
(Tsat = 516 K), the liquid is at a temperature of 515.15 K whereas as a closed system without mass and energy transfers with the en-
the vapor is at 516.15 K. At t = 0, the energy injection starts: vironment. The energy injection represents the heat exchange with
10MW/kgH20 are injected in the first portion of the tube for 50 ms. the molten fuel, then, at the end of the transient the injected en-
The volume of this portion is roughly 2.36 m3 , containing 1909 kg ergy is completely transformed in internal energy.
of liquid water. Hence the injected energy throughout the transient This kind of transients are really fast and lead to huge pres-
is 0.95 GJ. sure peaks. Experimental tests of steam explosion are KROTOS
The temporal evolutions at points A and B of pressure, velocity, (Huhtiniemi et al., 1999, pressure peak of 100 MPa in test K49) or
temperature and void fraction are illustrated in Fig. 18. A magnifi- SANDIA (Evans et al., 1982). Those experiments were not repro-
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 215

Fig. 17. Schematic of the injection test. The evolution of pressure, velocity, temperature and void fraction at points A and B are illustrated in Fig. 18 A and B.

Fig. 18. Temporal evolutions of pressure, velocity, temperature and void fraction at points A and B (Fig. 17). Test performed with 400 cells.

Fig. 19. Magnification of the temporal evolutions of pressure and velocity during the first 55 ms.
216 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

duced here since the modelling of the heat transfer in such situa- bolic 6-equation model presented in Pelanti and Keh-Ming Shyue
tions is complicated (Berthoud, 20 0 0) and is not in the objectives 2014.
of this work. The purpose of this test was to assess the robustness
of the numerical method and its capability of dealing with very Acknowledgments
high pressures and multiple phase changes.
The authors would like to gratefully thank Pascal Galon (CEA)
and Frédéric Daude (EDF) for helpful discussions and advices for
7. Conclusions the implementation of the flow models in EUROPLEXUS.

In this article we have proposed a new efficient EoS look-up ta- Appendix A. HEM - Speed of sound using IAPWS-IF97
ble method for steam-water flow simulations. Our algorithm uses
the IAPWS-IF97 EoS formulation and evaluates water properties by The speed of sound in two-phase flows does not correspond
interpolation of values tabulated on the e-v thermodynamic dia- to an intrinsic thermodynamic property of the substance, and it
gram. The main features of this algorithm consist in transforming depends both on the mixture composition and on the degree of
the irregular shaped e-v physical domain into a regular one and inter-phase equilibrium attained between the phases. For models
in performing a bicubic interpolation in the transformed domain. describing two-phase flows the expression of the speed of sound
Such a technique makes the procedure very efficient and at the stems mathematically from the governing systems of equations,
same time it provides remarkable precision. In particular, in con- and it depends on the choice of the EoS. For the Homogeneous
trast to simple EoS formulations, our method allows an accurate Equilibrium Model the corresponding speed of sound can be ob-
description of regimes where water has a significant non linear be- tained by using the definition (formally equal to the one of the
havior, i.e. close to the saturation curve, in the metastable region single-phase case)
and close to the critical point.  
In fast transients, especially during depressurizations, water be- 2 ∂p
cHEM = . (A.1)
comes metastable. In order to take into account this phenomenon, ∂ρ s
the tabulation has been performed until the spinodal curve. How-
ever the spinodal line has not been clearly experimentally identi- together with the closure EoS relations for the mixture detailed in
fied yet. Here, it has been determined by extrapolating the basic Section 2.1. Saurel et al. (2008) obtained:
equation provided by the IAPWS-IF97 EoS.   2  2 
The steam-water table look up algorithm has been coupled 1 αv 1 − αv αv ρv dsv αl ρl dsl
= + +T +
to two simple two-phase flow models, the Homogeneous Equilib- ρ cHEM
2 ρv cv2 ρl cl2 c p,v dp c p,l dp
rium Model, and the Homogeneous Relaxation Model. Both mod-
els are numerically solved by Finite Volume discretizations with (A.2)
a HLLC-type scheme. The methods and algorithms presented in Based on the expression above, it is easy to see that the HEM sys-
this work have been implemented in EUROPLEXUS, a fast transient tem is hyperbolic with the natural conditions:
dynamics software adopted for instance in the field of nuclear
safety. ρk > 0 , T > 0, c p,k > 0 (A.3)
The performance of the HEM and HRM coupled with the new where k ∈ {v, l}.
steam-water table look-up method has been assessed on three As well known in literature, the HEM has a discontinuous speed
challenging fast transients problems: fast depressurization, water- of sound (see Fig. 2). This feature is exhaustively explained in
hammer and steam explosion tests. For the first two test problems Flȧtten and Lund (2012).
the comparison of numerical results with experimental data show The employment of the expression of Eq. (A.2) for CFD simu-
the very good performance of the computational techniques that lations is not efficient for the steam-water table look-up method
we have developed. Let us note that for the fast depressurization here presented. Let us derive another formulation more conve-
test, although both the HEM and HRM models prove to be able to nient and less time-consuming. Following Eq. (2), for the HEM, the
describe the main features of the transient flow dynamics, only the quasi-linear form reads:
HRM can take into account the deep metastability reached in the        
very first milliseconds of the experiment. For this HRM model, a ρ u ρ 0 ρ 0
correlation for the relaxation time to return to equilibrium condi- ∂t u + 0 u v ∂z u = 0 , (A.4)
tions has been proposed and compared to other correlations found p 0 γHEM p u p 0
in literature. The correlation proposed here proves to be more ac-
where the dimensionless quantity, γ HEM , is:
curate for the description of the pressure evolution, which is a very
important feature in particular for fluid-structure interaction prob- 1 p
γHEM = − ρ ∂ρ e . (A.5)
lems. p ∂ pe ρ
Our method has also been assessed for its capability of dealing
The eigenvalues of the system in (A.4) are:
with very high pressures and multiple phase changes. These are
the characteristic features of the considered steam explosion test λ1 = u − cHEM , λ2 = u, λ3 = u + cHEM . (A.6)
case.
We define the speed of sound of the HEM, cHEM , as:
Finally, let us stress that the algorithm here proposed is

very efficient in terms of time consumption. Indeed the global γHEM p
computational time decreased significantly (even of 65% for the cHEM  . (A.7)
ρ
Super Canon experiment) through the use of a direct proper-
ties calculation rather than an iterative one when the HEM is Using a caloric EoS of the form e = e(ρ , p), for the HEM, the
used. speed of sound reads:
A further evolution of the present work would be the extension ∂e
 
ρ − ρ ∂ρ p
p
of our technique to the case of metastable states of vapor and its 2
cHEM =  ∂e  (A.8)
coupling with a more complex two-phase flow model as the hyper- ρ ∂p ρ
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 217

where: A more convenient form would be:

   
∂e ∂x −v2
= ( ev − el )
2
cHRM =    ∂v   ∂ s   ∂ s −1  . (B.9)
∂ρ ∂ρ p ∂v
p ∂ p T,x − ∂ T p,x ∂ p T,x ∂ T p,x
( ev − el )
=− (A.9)
ρ 2 ( vv − vl ) The calculation of these derivatives with IAPWS-IF97 is straight-
forward if we refer to the methodology shown in Appendix D.
and However, we remind that even for the calculation of the speed of
        −1 sound, the vapor must be set at saturation (see Eq. (14)).
∂e ∂e ∂e ∂ρ ∂ρ
= −
∂p ∂p x ∂x p ∂ p x ∂x p
ρ     (A.10) Appendix C. Iterative algorithms for the IAPWS-IF97
∂e ev − el ∂v
= − .
∂ p x vv − vl ∂ p x Using an Industrial Formulation EoS as the IAPWS-IF97, we can
Finally, cHEM can be determined through the following relation: define a residual function, F = F( p, T ), whose zero gives the ther-
modynamic variable pair (p, T) we are looking for. The residual
function is defined as F :  → R2 , where  ∈ R2 is the validity do-
p + vevv −e l
−v l
2
cHEM =     (A.11) main of the EoS. For instance, in the single-phase domain, for both
ρ2 ∂e − vevv −e l ∂v the HEM and HRM, we need to solve:
∂p −v l ∂ p
x x
 
v ( p) + (1 − x ) del ( p) , v ( p ) + ( 1 − x ) d vl ( p ) .
e − eIAPW S97 ( p, T )
with ( ∂∂ ep ) = x dedp ( ∂∂vp ) = x dvdp F ( p, T ) = . (C.1)
x dp x dp v − vIAPW S97 ( p, T )
In such a form, it is simple to calculate the speed of sound by
using a spline reconstruction of the saturation curve as vv (p), vl (p), In thermodynamics, according to Gibbs phase rule, the degrees of
ev (p), el (p). For this work, the spline coefficients have been evalu- freedom are 2 in the single-phase region, 1 in the stable two-phase
ated using the IAPWS-IF97. region and 0 for both triple and critical points. Therefore, in the
stable two-phase region, the residual function, F, becomes F : D →
R, where D represents the two-phase domain. In practice, for F =
Appendix B. HRM - Speed of sound using IAPWS-IF97
F( p), the domain is D = [ ptriple , pcritic ]. For F = F(T ), we have D =
[Ttriple , Tcritic ].
Similar to the HEM in Appendix A, the quasi-linear form of the
For instance, considering F = F( p), in the stable two-phase do-
HRM reads:
main, the residual function is:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
x u 0 0 0 x 0
⎢ρ ⎥ ⎢0 u ρ 0⎥ ⎢ρ ⎥ ⎢0⎥ F = e − x ev ( p ) − ( 1 − x ) el ( p ) (C.2)
∂t ⎣ ⎦ + ⎣ ∂
v⎦ z ⎣ u ⎦ ⎣0⎦
= , (B.1)
u 0 0 u
where:
p 0 0 γHRM p u p 0
e − el ( p )
where the dimensionless quantity γ HRM , is: x = x ( p) = . (C.3)
ev ( p ) − el ( p )
1 p
γHRM = − ρ ∂ρ e . (B.2) Eq. (C.2) just holds for the stable two-phase domain. Hence, it is
p ∂ pe ρ not adequate for the HRM. In the metastable two-phase domain,
The eigenvalues of the system of equations (B.1) are: for the HRM, we need to solve:
 
λ1 = u − cHRM , λ2,3 = u, λ4 = u + cHRM . (B.3) e − x ev ( p) − (1 − x ) el ( p, T )
F ( p, T ) = , (C.4)
v − x vv ( p) − (1 − x ) vl ( p, T )
We define the speed of sound of the HRM, cHRM , as:
 where the flow quality, x, is obtained from Eq. (17).
γHRM p Eqs. (C.1)and (C.4) are two-dimensional non-linear equations.
cHRM  . (B.4)
ρ Solving systems of more than one nonlinear equation is gener-
ally hard. To get a globally convergent solver, optimization and
Again, the System (12) is hyperbolic if and only if γ HRM > 0.
backtracking methods are often employed. Various methods of op-
Let us derive a formulation of the speed of sound compatible
timization/backtracking exist, authors suggest the ones in Press
with IAPWS-IF97. The caloric EoS is e = e( p, ρ , x ). Therefore, the
et al. (1997). Even though such a procedure allows one to imple-
speed of sound reads:
ment very accurate EoS, it is extremely time consuming (hundreds-

∂ e( p,ρ ,x )

ρ −ρ
p thousands times longer than a look-up table interpolation (Kunick,
∂ρ p,x
2
cHRM =  ∂ e( p,ρ ,x)  . (B.5) 2015)).
ρ ∂p ρ ,x

By using the following thermodynamic relations: Appendix D. Derivatives of IAPWS-IF97 for the Bicubic
Interpolation
de = T ds − pdv, (B.6)
This paragraph outlines how the needed derivatives for the
dh = T ds + vdp, (B.7) evaluation of the bicubic interpolation coefficients have been cal-
culated. Therefore it serves as a support tool for a more effective
Eq. (B.5) becomes: understanding of the procedures presented in Section 5.2.
  Let us consider the pressure p as our function F of interest. If
2 ∂p
cHRM = . (B.8) the point defined by the couple (v, e) belongs to Region 1, 2 or 5
∂ρ s,x
218 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219

of the IAPWS-IF97 domain, first order derivatives can be evaluated    

as: ∂p ∂p  
= −ρ 2 = −ρ 2 RT 2δ δ + δ 2 δδ , (D.12)
   −1        −1 −1 ∂v T
∂ρ T
∂p ∂v ∂v ∂v ∂e ∂e  
= = − , ∂e
∂v ∂p ∂p ∂T ∂p ∂T = −Rτ 2 τ τ , (D.13)
e e T p T p
∂T v
(D.1)
   
∂e ∂e
= −ρ 2
= −ρδ RT ∗ δτ , (D.14)
 ∂v ∂ρ
   −1       −1 −1 T T
∂p ∂e ∂e ∂e ∂v ∂v    
= = − , where δ = (∂ /∂ δ )τ , δδ = ∂ 2 /∂δ 2 τ , ττ = ∂ 2 /∂τ 2 δ
∂e ∂p ∂p ∂T ∂p ∂T
v v T p T p and δτ = ∂ 2 /(∂ δ∂ τ ).
(D.2) Then, second order derivatives are approximated as difference
quotients of first derivatives:
where v and e are the coordinates of the thermodynamic plane,
  ∂p ∂p
while p and T are pressure and temperature that have been deter- ∂2 p ∂v v+δ,e − ∂v v−δ,e
= , (D.15)
mined by setting to zero Eq. (C.1). ∂v2 2δ
e
In IAPWS-IF97 Wagner and et (20 0 0) the basic equations for
  ∂p ∂p
Region 1, 2 and 5 are expressed in terms of dimensionless specific
∂2 p ∂ e e+δ,v − ∂ e e−δ,v
Gibbs free enthalpy: = , (D.16)
∂ e2 2δ
v
g( p, T )
= γ (π , τ ), (D.3)   ∂p ∂p
∂ e v+δ −
RT ∂2 p ∂ e v −δ
where π = p/p∗ and τ = T ∗ /T with given p∗ and T∗ depending on = . (D.17)
∂ e∂ v 2δ
the region. R represents the specific gas constant of ordinary water.
Therefore, it is possible to evaluate the partial derivatives which References
appear in the square parentheses of Eq. (D.1) and (D.2) as follows:
  RELAP. -7 progress report: FY-2015 optimization activities summary report INL/EXT-
∂v RT 15-36771. Idaho Falls, USA.
= 2 π 2 γππ , (D.4) International formulation committee, 1967. Of the 6th international conference
∂p T
p on the properties of steam. The 1967 Formulation for Industrial Use. Verein
  Deutscher Ingenieure, Dusseldorf
∂v R RELAP, 2001. 5/MOD3.3 code manual, Vol. I. Idaho Falls, USA.
= π (γπ − τ γπτ ), (D.5) International association for the properties, 2007. Of water and steam. Revised Re-
∂T p
p lease on the IAPWS Industrial Formulation 1997 for the Thermodynamic Prop-
erties of Water and Steam, Lucerne, Switzerland.
International association for the properties, 2015. Of water and steam. Guideline on
      the Fast Calculation of Steam and Water Properties with the Spline-Based Table
∂e RT ∗ γπ π ∂v ∂v Look-Up Method (SBTL).
= γπτ − − γππ = −T −p ,
∂p T
p∗ τ τ ∂T p ∂p T Alamgir, M., Lienhard, J.H., 1981. Correlation of pressure undershoot during hot wa-
ter depressurization. J. Heat Transfer 103, 52–55.
Allievi, L., 1913. Teoria del colpo d’ariete. atti collegio ing. arch.(English translation
(D.6)
by Halmos E.E., 1929, The theory of waterhammer, Trans. ASME).
Ambroso, A., Hérard, J.-M., Hurisse, O., 2009. A method to couple HEM and HRM
  two-phase flow models. Comput. Fluids 38, 738–756.
∂e π π
= −Rτ 2 γτ τ + γ − γ , (D.7) Baer, M.R., Nunziato, J.W., 1986. A two-phase mixture theory for the deflagration–
∂T p
τ 2 π τ πτ to-detonation transition (ddt) in reactive granular materials. Int. J. Multiphase
Flow 12, 861–889.
  Bartak, J., 1990. A study of the rapid depressurization of hot water and the dynamics
with γπ = (∂ γ /∂ π )τ , γπ π = ∂ 2 γ /∂π 2 τ and γπ τ = of vapour bubble generation in superheated water. Int. J. Multiphase Flow 16,
∂ γ /(∂ π ∂ τ ).
2 789–798.
Berry, R. A., Zou, L., Zhao, H., Zhang, H., Peterson, J. W., Martineau, R. C., Kadioglu, S.
If the point belongs to Region 3: Y., Andrs, D., 2016. RELAP-7 theory manual, idaho national laboratory technical
         −1 report INL/EXT-14-31366(revision 2).
∂p ∂p ∂p ∂e ∂e Berthoud, G., 20 0 0. Heat transfer modeling during a vapor explosion. Nucl. Technol.
= − , (D.8)
∂v e
∂v T
∂ T v ∂v T ∂ T v 130, 39–58.
Bilicki, Z., Kestin, J., 1990. Physical aspects of the relaxation model in two-phase
    −1 flow. Proc. R. Soc. A, Math. Phy. Sciences 428, 379–397.
∂p ∂p ∂e Bilicki, Z., Kestin, J., Pratt, M.M., 1990. A reinterpretation of the results of the moby
= . (D.9) dick experiments in terms on the nonequilibrium model. ASME J. Fluids Eng.
∂e v
∂T v
∂T v 112, 212–217.
Board, S.J., Hall, R.W., Hall, R.S., 1975. Detonation of fuel coolant explosions. Nature
The basic equation for this region is a fundamental equation for 254, 319.
Carey, V. P.,. Liquid-Vapor Phase-Change Phenomena. Francis & Taylor, New York,
the specific Helmholtz free energy f, expressed in dimensionless
USA.
form: Carpenter, J.H., Belcourt, N., Nourgaliev, R., 2013. General Purpose Steam Table Li-
f (ρ , T ) brary. Sandia National Laboratories, USA. Report SAND2013-7328
= (δ, τ ), (D.10) Caupin, F., Stroock, A.D., 2013. The stability limit and other open questions on water
RT at negative pressure. In: Liquid Polymorphism: Advances in Chemical Physics,
ed Stanley HE, 152. Wiley, New York, pp. 51–80.
where δ = ρ /ρ ∗ , τ = T ∗ /T with ρ ∗ and T∗ equal to the critical val- Clerc, S., 20 0 0. Numerical simulation of the homogeneous equilibrium model for
ues as defined in Wagner and et (20 0 0). Relating the thermody- two-phase flows. J. Comput. Phys. 161, 354–375.
namic properties to the dimensionless Helmholtz free energy and Daude, F., Galon, P., Gao, Z., Blaud, E., 2014. Numerical experiments using a HLLC–
type scheme with ALE formulation for compressible two-phase flows five-equa-
its derivatives when using Eq. (D.10), it is possible to obtain: tion models with phase transition. Comput. Fluids 94, 112–138.
    De Lorenzo, M., Lafon, P., Seynhaeve, J.-M., Bartosiewicz, Y., 2017. Benchmark of
∂p δ
= −Rδ 2 τ ρ ∗ δτ − , (D.11) delayed equilibrium model (DEM) and classic two-phase critical flow models
∂T v
τ against experimental data. Int. J. Multiphase Flow 92, 112–130.
 M. De Lorenzo et al. / International Journal of Multiphase Flow 95 (2017) 199–219 219

Debenedetti, P.G., 1996. Metastable Liquids: Concepts and Principles. Princeton Uni- Nakagawa, M., Harada, A., Berana, M.S., 2009. Analysis of expansion waves appear-
versity Press, Princeton, NJ. ing in the outlets of two-phase flow nozzles. HVAC&R Res. 15, 1065–1079.
Downar-Zapolski, P., Bilicki, Z., Bolle, L., Franco, J., 1996. The non-equilibrium relax- Pelanti, M., Keh-Ming Shyue, K.-M., 2014. A mixture-energy-consistent six-equation
ation model for one-dimensional flashing liquid flow. Int. J. Multiphase Flow 22, two-phase numerical model for fluids with interfaces, cavitation and evapora-
473–483. tion waves. J. Comput. Phys. 259, 331–357.
Eberhart, J.G., 2009. A new four-parameter equation of state and its application in Porsching, T.A., 1977. A finite difference method for thermally expandable fluid tran-
predicting the spinodal temperature of water. Water J. 1, 85–91. sients. Nucl. Sci. Eng. 64, 177.
Edwards, A.R., O’Brien, T.P., 1970. Studies of phenomena connected with the depres- Press, W., Teukolsy, S., Vetterling, W., Flannery, B.P., 1997. Numerical Recipes. Cam-
surization of water reactors. J. British Nuclear Society 9, 125–135. bridge University Press, USA.
Evans, N. A., Mitchell, D. E., Nelson, L. S., Corradini, M. L., 1982. Recent results form Reocreux, M., 1974. Contribution à l’étude des débits critiques en écoulement
the SANDIA steam explosion program. SAND82-2269C. diphasique eau-vapeur. Université Scientifique et Médicale de Grenoble Phd the-
Faucher, E., Hérard, J.M., Barret, M., Toulemonde, C., 20 0 0. Computation of flashing sis.
flows in variable cross section ducts. Int. J. Comp. Fluid Dyn. 13, 365–391. Riegel, B., 1978. Contribution à l’étude de la décompression d’une capacité en
Faucher, V., Crouzet, F., Debaud, F., 2014. Mechanical consequences of LOCA in PWR: régime diphasique. Institut National Polytechnique de Grenoble Phd thesis.
full scale coupled 1d/3dsimulations with fluid-structure interaction. Nucl. En- Robbe, M.F., Lepareux, M., 1996. Programme PLEXUS. matériau EAU. modèle ho-
gng. Des. 270, 359–378. mogène non equilibré. Rapport DMT 96/142, CEA Saclay, France.
Flȧtten, T., Lund, H., 2012. Relaxation two-phase flow models and the subcharacter- Ruas, V., 2016. Numerical Methods for Partial Differential Equations: An Introduc-
istic conditions. Math. Models Methods Appl. Sci. 21, 2374–2407. tion. Wiley, New York.
Gale, J., Tiselj, I., Horvat, A., 2008. Two-fluid model of the WAHA code for simula- Saurel, R., Abgrall, R., 1999. A multiphase godunov method for compressible multi-
tions of water hammer transients. Multiphase Sci. Technol. 20, 291–322. fluid and multiphase flows. J. Comput. Phys. 150, 425–467.
Gouse, S., Brown, G. A., 1964. A survey of the velocity of sound in two-phase mix- Saurel, R., Petitpas, F., Abgrall, R., 2008. Modelling phase transition in metastable
tures. ASME Paper 64-WA/FE-35. liquids. application to cavitating and flashing flows. J. Fluid Mech. 607, 313–350.
Haar, L., Gallagher, J., Kell, G. S., 1984. NBS/NRC steam tables. hemisphere. Washing- Saurel, R., Petitpas, F., Berry, R.A., 2009. Simple and efficient relaxation methods for
ton, DC. interfaces separating compressible fluids, cavitating flows and shocks in multi-
Hughes, T.J.R., 20 0 0. The Finite Element Method. Dover Publications. phase mixture. J. Comput. Phys. 228, 1678–1712.
Huhtiniemi, I., Magallon, D., Hohmann, H., 1999. Results of recent KROTOS FCI tests: Schrock, V.E., Amos, C.N., 1984. Two-Phase Critical Flow, in Two-Phase Flow and
alumina versus corium melts. Nucl. Eng. Des. 189, 379–389. Heat Transfer. China US Progress, Hemisphere, Washington, USA.
Imre, A.L., et al., 2013. Estimation of the thermodynamic limit of overheating for Simpson, A.R., 1989. Large water-hammer pressures for column separation in
bulk water from interfacial properties. Int. J. Thermophys. 34, 2053–2064. pipelines. J. Hydraul. Eng. 117, 1310–1316.
Kapila, A., Menikoff, R., Bdzil, J., Son, S.F., Stewart, D.S., 2001. Two-phase modeling of Skripov, V.P., 1972. Metastable Liquids. Wiley, New York.
deflagration-to-detonation transition in granular materials: reduced equations. Skripov, V.P., Sinicyn, P.A., et, a., 1980. Teplofizicheskyie Svoyistva Zhidkostey v
Phys. Fluids 13. 3002–25. Metastabil’nom Sostoyanyii. Atomizdat, Moskva.
Karplus, H., 1958. The velocity of sound in a liquid containing gas bubbles. Tech. Stadtke, H., 2006. Gasdynamic aspects of the two-phase flow. Wiley-VCH Verlag
Rep. COO-248. Illinois Inst. GmbH & CO. KGaA, Weinheim, Germany.
Kieffer, S.W., 1977. Sound speed in liquid-gas mixtures: water-air and water-steam. Thomas, P., 1998. Extension des table de l’eau utilisées dans le code CATHARE 2.
J. Geophys. Res. 82, 2895–2904. EDF-CEA-FRAMATOME, France.
Kiselev, S., Ely, J.F., 2001. Curvature effect on the physical boundary of metastable Tiselj, I., Horvat, A., Gale, J., 2008. Numerical scheme of the WAHA code. Multiphase
states in liquids. Physica A 299, 357–370. Sci. Technol. 20, 323–354.
Kunick, M., et al., 2015. CFD analysis of steam turbines with the IAPWS standard Wagner, W., et al., 20 0 0. The IAPWS industrial formulation year 1997 for the ther-
on the spline-based table look-up method (SBTL) for the fast calculation of real modynamic properties of water and steam. ASME J. Eng. Gas Turbines Power
fluid properties. In: Turbine Technical Conference and Exposition: Proceedings 122, 150–182.
of ASME Turbo Expo 2015. Wagner, W., Pruß, A., 2002a. The IAPWS formulation year 1995 for the thermody-
Le Metayer, O., Massoni, J., Saurel, R., 2004. Elaborating equations of state of a liquid namic properties of ordinary water substance for general and scientific use. J.
and its vapor for two-phase flow models. Int. J. Thermal Sci. 43, 265–276. Phys. Chem. Ref. Data 31, 387–535.
Lepareux, M., 1994. Materiau EAU, report DMT 94/398, CEA saclay, france. Wagner, W., Pruß, A., 2002b. The IAPWS formulation year 1995 for the thermody-
Lienhard, J.H., Shamsundar, N., Biney, P.O., 1986. Spinodal lines and equations of namic properties of ordinary water substance for general and scientific use. J.
state - a review. Nucl. Eng. Des. 95, 297–313. Phys. Chem. Ref. Data 31.
Liu, T.-P., 1987. Hyperbolic conservation laws with relaxation. Commun. Math. Phys. Wood, A.B., 1930. A Textbook of Sound. G. Bell and Sons Ltd, London, UK.
108, 153–175. Zheng, Q., Durben, D.J., Wolf, G.H., Angell, C.A., 1991. Liquids at large negative pres-
Lochon, H., Daude, F., Galon, P., Hérard, J.M., 2016. Comparison of two-fluid models sures: water at the homogeneous nucleation limit. Science 254 (5033), 829–832.
on steam-water transients. ESAIM, (not printed yet).
Menikoff, R., Plohr, B.J., 1989. The riemann problem for fluid flow of real materials.
Rev. Mod. Phys. 61, 75–130.