Atomization and Sprays, 27(4): 345–365 (2017)

INVESTIGATION OF HOMOGENEOUS RELAXATION

MODEL PARAMETERS AND THEIR IMPLICATIONS
FOR GASOLINE INJECTORS
K. Saha,1,∗ S. Som,1 & M. Battistoni2
1
Argonne National Laboratory, Argonne, Illinois 60439, USA
2
University of Perugia, Italy
*Address all correspondence to: K. Saha, Argonne National Laboratory, Argonne, Illinois 60439, USA,
E-mail: ksaha@anl.gov

Original Manuscript Submitted: 1/22/2016; Final Draft Received: 8/14/2016

Flash boiling is known to be a common phenomenon for gasoline direct injection (GDI) engine sprays. The Homoge-
neous Relaxation Model has been adopted in many recent numerical studies for predicting cavitation and flash boiling.
The Homogeneous Relaxation Model is assessed in this study. Sensitivity analysis of the model parameters has been
documented to infer the driving factors for the flash-boiling predictions. The model parameters have been varied over
a range and the differences in predictions of the extent of flashing have been studied. Apart from flashing in the near-
nozzle regions, mild cavitation is also predicted inside the gasoline injectors. The variation in the predicted time scales
through the model parameters for predicting these two different thermodynamic phenomena (cavitation, flash) have
been elaborated in this study. Turbulence model effects have also been investigated by comparing predictions from the
standard and Re-Normalization Group (RNG) k-ǫ turbulence models.

KEY WORDS: spray structure, flash boiling, cavitation, degree of superheat, gasoline direct injection

1. INTRODUCTION
A comprehensive understanding of fuel spray formation is needed for optimization of existing technologies, as well as
future developments, related to diesel or gasoline direct injection (GDI) combustion engines. GDI engines have been
preferred to port fuel injection (PFI) engines for almost a decade. Experimental studies have shown that GDI offers
better fuel economy and superior performance when compared to PFI (Harada et al., 1997; Iwamoto et al., 1997). The
GDI engines have become advantageous owing to elegant technologies, such as intelligent piston design, swirl flows,
high-pressure fuel injection, advanced injection strategies, exhaust gas recirculation (EGR), closed-coupled catalysts,
etc.
Flash boiling occurs in modern GDI engines at throttled conditions and it significantly affects spray formation
(Weber and Leick, 2014). Throttled operating conditions create low in-cylinder chamber pressures (≈30 kPa). There-
fore, when heated gasoline-type fuel (at around 363 K) is injected into such in-cylinder environments, liquid fuel is
subjected to a superheated condition. Based on the thermodynamic definition, superheated condition arises when any
liquid substance experiences ambient pressure lower than the saturation pressure corresponding to its temperature.
Flash boiling or flashing causes rapid bulk conversion of liquid fuel to gaseous vapor, leading to volume expansion
and thus affecting spray width and penetration. Consequently, flashing is either advantageous or disadvantageous, de-
pending on chamber design and injection timing as well as injector orientation. As a result, flashing for GDI engines
is becoming a relevant topic of research.
Flashing is typically a near-nozzle-exit phenomenon. Apart from flashing, there are two more phenomena in-
volving a phase change of liquid fuel to gaseous vapor: vaporization and cavitation. A high-temperature environment
in combustion chamber can provide convective heating of liquid fuel, causing “vaporization.” The liquid fuel enters

1044–5110/17/$35.00 © 2017 by Begell House, Inc. 345

346 Saha, Som, & Battistoni

NOMENCLATURE

CFL Courant Friedrichs Lewy number Tsat saturation temperature (K)

D mass diffusivity (m2 /s) u local cell velocity (m/s)
e specific internal energy (J/kg) ui advecting mean velocity (m/s)
hlv latent heat of vaporization (kJ/kg) uj advected mean velocity (m/s)
hm specific enthalpy of mth species V volume of a cell (m3 )
(kJ/kg) x local cell vapor quality
Ja Jakob number x̄ local cell equilibrium quality
k turbulent kinetic energy (m2 /s2 ) Ym mass fraction of mth species
mvap , mliq mass of vapor and liquid phases
in a cell (kg) Greek Symbols
p local pressure (Pa) α void fraction (vapor and
Pch , Pchamber chamber pressure (kPa) non-condensable gases)
pcrit critical pressure (Pa) ǫ turbulence dissipation rate
Pinj injection pressure (MPa) (m2 /s3 )
Psat , psat saturation pressure (Pa) θ,θ0 equilibrium time-scale and
RP pressure ratio empirical time-constant(s)
Sm source term in species transport µ,µt dynamic and turbulent viscosity
equations coefficient (kg/m.s)
Sij strain rate (1/s) ρ, ρv , ρl , ρg density of mixture, vapor, liquid
t time (s) and gas (kg/m3 )
T local cell temperature (K) σij strain rate tensor (1/s)
Tfuel fuel temperature (K) τij Reynold’s stress (Pa)

the injector holes through the space between the moving needle and the needle sac. This behavior resembles the flow
from a large cross-sectional area to a much smaller cross-sectional area, where flow separation occurs near the entry
of the smaller cross-sectional portion. Inside the fuel injector holes, the local pressure in this flow separation region
may drop below the saturation pressure of the liquid fuel, depending on local temperature, injection pressure and
injector geometry. This phenomenon is termed as “cavitation.” Cavitation occurs almost instantaneously, since it is
typically a near-equilibrium process involving negligible heat transfer (Brennen, 1995). Cavitation is usually promi-
nent in diesel injector holes, because the injection pressure in diesel injector (up to 300 MPa) can be at least one order
of magnitude higher than that of a gasoline injector. Flashing is usually perceived as a non-equilibrium phenomenon,
since heat transfer is non-negligible and requires a finite time scale (relative to cavitation) for the phase change to
occur (Neroorkar, 2011).
Several experimental studies have been undertaken to investigate flash boiling effects on GDI spray patterns. Van-
derwege and Hochgreb (1998), in planar laser-induced fluorescence (PLIF) and Mie-scattering studies of a pressure-
swirl injection system, observed the transition of the spray shape from hollow-cone to solid-cone, with an increase
in the degree of superheat of the liquid fuel. The authors concluded from the image intensity comparisons that the
droplet diameter reduction due to flash boiling was on the order of 40%. On the basis of the bubble-point calculation,
they also suggested that superheating by roughly 20 K is required for the flash boiling to generate noticeable effects.
Their experimental results demonstrated the potential advantages of using flash boiling to enhance spray atomiza-
tion. Xu et al. (2013) and Zeng et al. (2012) studied the spray from a multi-hole injector using Mie scattering and
laser-induced exciplex fluorescence (LIEF). The authors quantified the macroscopic spray structure with respect to
penetration length, plume width, and normalized plume distance. They also characterized the different macroscopic

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 347

spray behaviors on the basis of the ratio of ambient pressure to saturation pressure. A set of empirical expressions
was proposed to predict the macroscopic structure of the spray under the effects of flash boiling. In the analysis,
the authors focused more on the macroscopic spray structure rather than on fundamental bubble and droplet dynam-
ics. Weber and Leick (2014) used high-speed Mie scattering for global spray characterization and investigation of
near-nozzle spray structures using high-speed shadowgraphy. Additionally, droplet velocities were estimated by a
shadow particle image velocimetry technique. They observed that individual spray plume length decreases and the
initial cone angle increases as a result of flash boiling. They noticed that during intense flash-boiling situations two
adjacent spray plumes might interact with each other. In a recent study with a GDI injector similar to Spray G (Spray
G ECN Gasoline Spray Combustion, 2014), a comprehensive experimental analysis was documented (Montanaro and
Allocca, 2015). The authors used a Delphi Valve Covered Orifice (VCO) injector with iso-octane fuel. The injection
pressure, the fuel temperature, and the chamber pressure were varied, with the chamber temperature remaining at
296 K. They observed that sub-atmospheric chamber pressure and elevated fuel temperatures led to a large atomized
cloud surrounding the fuel jets.
Significant effort has been made already in terms of numerical studies on the flash-boiling phenomenon. They
can be broadly classified into two types: (a) bubble-dynamics-based models, (b) empirical coefficients-based thermo-
dynamic rate equation models. Blinkov et al. (1993) assumed a mono-disperse bubble population, and their model
consisted of five transport equations addressing vapor formation, interface heat transfer, and nucleation. Chang and
Lee (2002) modified Blinkov’s approach by considering a poly-disperse bubble population in a converging water
nozzle and concluded from their primary analysis that capturing the bubble size distribution is important. Kawano
et al. (2004) incorporated bubble growth, bubble perturbation, and bubble breakup to model flashing and non-flashing
sprays. A similar approach was adopted by Zeng and Lee (2001) to study n-pentane sprays subjected to different
degrees of superheat. They concluded that more flashing can be correlated with an increment in superheat, smaller
drop sizes (∼50% reduction in Sauter mean diameter (SMD), when flashing is considered in the model), wider sprays,
and even plume merging. They also inferred that for lower degrees of superheat aerodynamic forces dominate, while
bubble growth takes over at higher degrees of superheat. Another type of flash-boiling model based on empirical
coefficients is the Homogeneous Relaxation Model (HRM) (Bianchi et al., 2009; Bilicki and Kestin, 1990; Downar-
Zapolski et al., 1996; Schmidt et al., 2010). HRM represents the phase transition by means of one empirical equation,
which estimates the time scale of the phase change. The time scale evaluates the temporal extent of the deviation of
the local condition from thermal equilibrium.
HRM lies in between the two extremes of thermodynamic two-phase models represented by the homogeneous
equilibrium model (HEM) and the homogeneous frozen model (HFM) (Brennen, 1995; Downar-Zapolski et al., 1996).
In the case of HEM, the two phases are assumed to be mixed homogeneously with the heat transfer occurring sponta-
neously. In a real-world scenario of two-phase flows, such as bubbly flows, instantaneous heat transfer is not feasible.
The other extreme, HFM, assumes zero heat transfer, i.e., an infinitely long heat transfer time scale. HRM captures
the in-between practical two-phase flow scenarios. The vapor–liquid equilibrium properties are necessary for HRM
implementation and have been estimated for pure as well as blended fuels, for GDI applications (Gopalakrishnan and
Schmidt, 2008; Neroorkar, 2011). Recently, Moulai et al. (2015) and Saha et al. (2016b) carried out numerical studies
of iso-octane flash boiling in a multi-hole GDI fuel injector (Spray G) under conditions specified in the Engine Com-
bustion Network (ECN) (Spray G ECN Gasoline Spray Combustion, 2014). These studies examined cases where the
outlet chamber is filled with only iso-octane, as well as cases where the chamber is occupied solely by nitrogen gas.
They explored different flashing and non-flashing conditions in their work.
HRM has been successfully used by many researchers for modeling cavitation in diesel injectors in the recent past
(Battistoni et al., 2014a,b, 2015; Morena et al., 2013; Neroorkar et al., 2012; Xue et al., 2015; Zhao et al., 2014). HRM
is a useful model for the industry because of its utility in both diesel and gasoline direct injection sprays. The model
constants of HRM were obtained two decades ago (Downar-Zapolski et al., 1996), on the basis of experiments done in
1974 with a large-scale nozzle (on the order of meters). Recently, Bliss et al. (2012) carried out detailed optimization
study, by comparing HRM predicted mass flow rates with Reitz’s flash-boiling experiments [Reitz (1990); the nozzle
diameter was 340 µm] and came up with slightly different model constants. The new set of constants provided only
marginal improvement with respect to the original model constants. Both sets of constants provided good agreement
with measured mass flow rates (Reitz, 1990). The same model rate equation and same values of model constants

Volume 27, Issue 4, 2017

348 Saha, Som, & Battistoni

in HRM have been used for solving flashing as well as cavitating problems, because HRM provided reasonable
performance for both cases in terms of mass flow rates or discharge coefficients. However, only good agreement
in terms of mass flow rate does not indicate the validity of a model. The prediction of the equilibrium time scales
for the two different processes should be compliant with the thermodynamic characteristics. Ideally, different sets of
equations or different sets of model constants should be used for solving the two different processes. Researchers have
used the HRM widely without providing sufficient insight on how HRM can model two different phenomena with
characteristic time scales varying by orders of magnitude. This fact raises a fundamental question about the validity
of HRM for modeling two thermodynamically different phenomena such as cavitation and flashing. Moreover, the
HRM implementation in OpenFOAM is different from the way it is implemented in CONVERGE [for details, please
see Neroorkar (2011) and Zhao et al. (2014)]. Therefore, the objective of the current study is to provide an in-depth
understanding of HRM, examine the pros and cons of the model setup, and explore fundamental justifications for
using HRM under both flashing and cavitating conditions.
The operating conditions and fuel injector considered in this study are based on the baseline Spray G condition
(Spray G ECN Gasoline Spray Combustion, 2014) and some parametric variations of that baseline condition. The pa-
per is organized in the following way: First, the numerical setup and the governing equations will be described. Next,
validation result will be presented by comparing model predictions with rate-of-injection measurements from the
ECN. This comparison will be followed by simulation results indicating the effect of different boundary conditions
on spray patterns and their thermodynamic implications and classifications of non-flashing, moderate flashing, and
intense flashing. Finally, results from parametric study using HRM and two different eddy-viscosity turbulence mod-
els, and sensitivities of the model predictions with respect to the different model parameters and turbulence models,
will be discussed.

2. MODEL FORMULATION
The problem considered in this study is iso-octane fuel flowing through an eight-hole counter-bored GDI fuel injector,
which is denoted as the Spray G nozzle in the ECN (Spray G ECN Gasoline Spray Combustion, 2014). Details of the
cases studied in this work are summarized in Table 1. From prior work on Spray G cases by our group (Saha et al.,
2016b) and other research groups (Moulai et al., 2015), it has been evident that the Spray G condition is supposed
to be non-flashing, since the chamber pressure is higher than the saturation pressure corresponding to the iso-octane
(liquid) fuel temperature. However, because the chamber temperature is higher than the fuel temperature, vaporization
of the liquid jets is expected owing to convective heating from the hot environment. In the case of Spray G2, flashing
is expected because the chamber pressure is lower than the corresponding saturation pressure of the fuel. For Spray
G2, the chamber is considered to be at room temperature, therefore, convective vaporization due to the ambience
should be very low. The Spray G3 condition is introduced for the first time in this study. Spray G3 has a higher fuel
temperature and is subjected to a chamber pressure of 1 atmosphere. Thus, the degree of superheat is higher than that
of Spray G2, indicating possibility of more than mild/moderate flashing. The chamber temperature under Spray G3
condition is 293 K; hence, convective vaporization of liquid jets due to the ambience should be very low.

2.1 Nozzle Geometry and Computational Domain

The internal geometry of the Spray G nozzle is not symmetric, since it has eight stepped holes. The nominal diameters
of the holes and the counter-bores are 165 µm and 388 µm, respectively. The length-to-diameter ratios of the nozzle
holes and counter-bores are around 1 and 1.2, respectively. There are five dimples (reducing the flow passage) on the
injector wall, resulting in uneven upstream flow passages leading to the counter-bored, i.e., stepped holes. Therefore,
for realistic predictions, full nozzle geometry has been simulated in this study. The nominal geometry of the Spray
G nozzle has been obtained through the ECN (Spray G ECN Gasoline Spray Combustion, 2014). Figure 1 shows the
location of the eight stepped holes with respect to the five dimples. The upstream asymmetry in the nozzle geometry
is evident from this figure. The information about the orientation of the holes is vital for cases where hole-to-hole
variations become conspicuous inside the nozzle and/or in the near-nozzle regions. A hemispherical outlet domain is
needed to ensure that the near-nozzle flow features are modeled accurately. It is difficult to make a correct a priori

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 349

TABLE 1: Cases studied

Parameters Spray G Spray G2 Spray G3
Inj. press. (MPa) 20 20 20
Chamber press. (kPa) 600 53 100
Chamber temp. (K) 573 293 293
Fuel temp. (K) 363 363 413
Chamber fluid N2 N2 N2
Degree of superheat, ∆T † (K) N/A 12.34 40.68
Pressure ratio, RP ‡ 0.13 1.48 2.83
Thermodynamic state Non-flashing, vaporizing Moderate flashing Intense flashing
†
∆T = TFuel − Tsat (Pchamber ).
‡
RP = Psat /Pchamber .

FIG. 1: (Left) Layout of the eight stepped holes of the Spray G injector (bottom view), and (Right) the computational domain
considered in this study

estimation of the extent of the outlet domain. Results using an outlet domain of 9 mm diameter are reported in this
study. A previous study (Saha et al., 2016b) indicated that results using a 9 mm outlet domain could be reasonable for
internal and near-nozzle flow analysis at a physical time-stamp of 0.1 ms. Past studies with this injector showed the
possibility of back-flow of chamber gas into the counter-bore (Moulai et al., 2015; Saha et al., 2016a). The stepped
or counter-bored hole with the back-flow velocity vectors is shown in Fig. 2.
Simulations are performed using the CONVERGE code (Richards et al., 2015), which uses a cut-cell technique
to generate the mesh automatically during the run time. A vertical cut-plane showing the mesh with 17.5 µm as the
minimum grid size is shown in Fig. 3. Adaptive Mesh Refinement (AMR) has not been utilized, to ensure efficient
load-balancing while executing parallel computing with millions of cells in the computational domain. Fixed embed-
ding has been used near the walls, inside the holes and in the near-nozzle regions, and inside the chamber to capture
the sharp gradients in species, velocity, temperature, etc. Three levels of embedding have been used, meaning that
the smallest cell has a dimension 23 = 8 times smaller than the largest cell size (base grid size) in the domain. In
this work, results using 140 µm as the base grid, i.e., 17.5 µm minimum grid size, have been presented, because this
grid resolution has been shown to provide reasonable estimates with affordable wall-clock times (Saha et al., 2016a).
There were approximately 2.8 million cells in the domain. On 60 processors, it took approximately 10 days to reach
the time-stamp of 0.1 ms for the Spray G case.

Volume 27, Issue 4, 2017

350 Saha, Som, & Battistoni

FIG. 2: Back-flow velocity vectors in the counter-bore region for the Spray G case

FIG. 3: Vertical cut-plane showing the mesh with 17.5 µm as minimum grid size and 9 mm outlet domain

2.2 Governing Equations

The mixture multiphase model in the Eulerian framework has been used for this study. The numerical setup for flash
boiling requires solving the set of governing equations of mass, momentum, species and energy conservation, addi-
tional scalar equations for turbulence, and constitutive equations for liquid and vapor phases to account for variable

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 351

density due to pressure changes. The governing equations adopted in this study are available in the literature (Bat-
tistoni et al., 2014b; Richards et al., 2015; Xue et al., 2014, 2015; Zhao et al., 2014) and are provided here as well
for the convenience of the readers. The CONVERGE code is used for the simulations, and the mass, momentum, and
energy conservation equations are as follows:

∂ρ ∂uj ρ
+ =0 (1)
∂t ∂xj

∂ρui ∂ρui uj ∂p ∂σij

+ =− + (2)
∂t ∂xj ∂xi ∂xj
!  
∂ρeui ∂ρeuj ∂uj ∂uj ∂ X ∂Ym ∂ ∂T
+ = −p + σij + ρ Dhm + k (3)
∂t ∂xj ∂xj ∂xj ∂xj m
∂xj ∂xj ∂xj
where   
∂ ∂ui ∂uj 2 ∂ui
σij = (µ + µt ) + − (4)
∂xj ∂xj ∂xi 3 ∂xi
and µ is the molecular viscosity and µt is the turbulent viscosity, which is modeled as

k2
µt = Cµ ρ (5)
ǫ
using the standard k-ǫ model. The mixture density is calculated from
X
ρ= ρ m αm (6)

and
ρ
αm = Ym (7)
ρm
where ρ is the mixture density and ρm stands for the density of any constituent species of the mixture. The species
considered are liquid fuel, vapor fuel, N2 , and O2 . The densities of all the gaseous species have been estimated
using the Ideal Gas Equation. αm and Ym represent the volume fraction and mass fraction of the individual species,
respectively. The individual mass fractions are determined from the species conservation equations:
 
∂ρYm ∂uj ρYm ∂ ∂Ym
+ = ρD + Sm (8)
∂t ∂xj ∂xj ∂xj

where Sm represents the source or sink term. There are no source or sink terms for non-condensable gases (N2 and
O2 ). Sm in the vapor species transport equation provides the rate of phase change for cavitating, flashing and vaporiz-
ing conditions. The relevant details for estimating Sm in phase change scenarios are provided in the “Homogeneous
Relaxation Model” subsection.

2.3 Turbulence Model

Reynolds-averaged Navier–Stokes (RANS)-based turbulence modeling approaches (eddy-viscosity) have been adop-
ted in this work. Such approaches typically constitute two transport equations for solving The turbulent kinetic energy
k and the turbulent dissipation rate ǫ are determined from the turbulence model. The standard and Re-Normalization
Group (RNG) k-ǫ turbulence models are the only two options available in CONVERGE v2.2 that can be applied for
nozzle flow calculations. Both the models have been tested here. The RNG k-ǫ model was developed to consider
the effects of smaller length scales. In the standard k-ǫ model, the turbulent viscosity is estimated from a single
turbulence length scale. Hence, the calculated turbulent diffusion occurs only at that specified scale, whereas in
reality all scales of motion should be taken into account. The RNG k-ǫ approach is a mathematical technique, used

Volume 27, Issue 4, 2017

352 Saha, Som, & Battistoni

to derive a turbulence model similar to the standard k-ǫ model, such that the effects of different length scales are
incorporated through the formulation of the production terms in the dissipation transport equation. For the standard
k-ǫ turbulence model, the conservation equations are
∂ρk ∂ρuj k ∂ui ∂ µ ∂k
+ = τij + − ρǫ (9)
∂t ∂xj ∂xj ∂xj Prk ∂xj
   
∂ρǫ ∂ρuj ǫ ∂ µ ∂ǫ ∂uj ∂ui ǫ
+ = + Cǫ3 ρǫ + Cǫ1 τij − Cǫ2 ρǫ (10)
∂t ∂xj ∂xj Prǫ ∂xj ∂xj ∂xj k
where Reynolds stress is  
′ u′ = 2µ S − 2 δ ∂ uej
τij = −ρug
i j t ij ij ρk + µt
3 ∂xj
The model constants are as follows:
Cµ = 0.09; Cǫ1 = 1.44; Cǫ2 = 1.9; Cǫ3 = −1.0; 1/Prk = 1.0; 1/Prǫ = 0.77
The RNG k-ǫ model equations are:
∂ρk ∂ρuj k ∂ui ∂ µ ∂k
+ = τij + − ρǫ (11)
∂t ∂xj ∂xj ∂xj Prk ∂xj
   
∂ρǫ ∂ρuj ǫ ∂ µ ∂ǫ ∂uj ∂ui ǫ
+ = + Cǫ3 ρǫ + Cǫ1 τij − Cǫ2 ρǫ − ρR (12)
∂t ∂xj ∂xj Prǫ ∂xj ∂xj ∂xj k


where R = {[Cµ η3 (1 − η/η0 )]/ 1 + βη3 }[ǫ2 /k] and η = (k/ǫ)|Sij |. The model constants for the RNG k-ǫ model
are as follows:
Cµ = 0.0845; Cǫ1 = 1.42; Cǫ2 = 1.68; Cǫ3 = −1.0; β = 0.012

2.4 Homogeneous Relaxation Model

The source/sink terms Sm in the fuel species conservation equations are evaluated through the HRM formulation.
HRM estimates the rate at which the local vapor quality (x) approaches the equilibrium vapor quality. The vapor
quality is mathematically represented as x = Yv /(Yv + Yl ) = mvap /(mvap + mliq ). The rate of change of local
vapor quality, (Dx/Dt), provides the estimate of Sm . HRM enables one to express Dx/Dt as

Dx x−x
= (13)
Dt θ
The time scale θ is calculated as
θ = θ0 α−0.54 ψ−1.76 (14)
where
|psat − p|
θ0 = 3.84 × 10−7 ; ψ = ; α = αv + αN2 + αO2
pcrit − psat
These specific values of the parameters have been known to be effective from engineering perspective for flow prob-
lems with injection pressure higher than 10 bar (Downar-Zapolski et al., 1996; Neroorkar, 2011). The injection pres-
sure for the GDI cases investigated in this study is 20 MPa (200 bar). HRM is capable of predicting phase changes
from liquid to vapor and vice versa, whenever the condition is thermodynamically feasible. Therefore, HRM is used
to capture both cavitation and flash boiling. Naturally, in a high-temperature environment, such as the Spray G case,
HRM can predict vapor formation due to convective heating of the liquid jets. Since HRM was developed/verified
only for flashing and cavitation problems, vaporization predictions by HRM due to convective heating may not be
reasonable.

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 353

Non-condensable gases have been included in α, despite the fact that gases were not considered in the original
HRM formulation. It is our understanding that in a given cell, when the liquid volume fraction is very low compared to
the gaseous species, considering only the fuel vapor volume fraction may not provide a proper physical representation
of the thermodynamic condition in that cell. Moreover, the work done by Downar-Zapolski et al. (1996) indicated
that the relaxation time exponentially decreases to very small values with increments in void fraction. Once the void
fraction exceeds 10%, the relaxation time scale continues to decrease asymptotically (cf. Fig. 3). This indicates
that mathematically, inclusion of gases in the void fraction should not hamper prediction of the relaxation time
scale. Additionally, under flash-boiling conditions the gases do not need to provide energy for phase change. The
superheated liquid provides the necessary enthalpy for phase change under flash-boiling conditions. For the flashing
cases Spray G2 and Spray G3 the chamber temperatures were specifically kept low to ensure that there is negligible
phase change due to convective heating by the ambience. It is also vital to understand the relation between Sm and
(Dx)/(Dt). Battistoni et al. (2015) elaborated this relationship in their work on the effect of non-condensable gases
on cavitation. If x0 is the computed current local vapor quality of a computational cell and x1 is the corresponding


value for the following time step, with a time-step size of ∆t, then x1 can be computed as x1 = x̄ − x̄ − x0 e−∆t/θ .
Therefore,
(x1 − x0 )(mvap + mliq ) (Yv + Yl )
Sm = = (x1 − x0 )ρ (15)
V ∆t ∆t
where V is the volume of the computational cell and mvap and mliq are the masses of the vapor and liquid phases,
respectively, in that computational cell.

2.5 Boundary and Initial Conditions

Pressure is identified at the inlet and outlet of the nozzle, and a no-slip boundary condition is considered at the
walls. The no-slip condition is also assumed at the interface of liquid and vapor for the mixture multiphase approach.
Turbulence boundary conditions are specified by the turbulent kinetic energy and turbulence dissipation rate at the
inlet and the outlet. The standard law-of-the-wall for turbulence is used to model the boundary layer.
A liquid fuel cannot be 100% pure; therefore, it is reasonable to consider the presence of a very small amount
(≈10−5 mass fraction) of non-condensable gas (Battistoni et al., 2014a; Saha, 2014), which acts like nucleation sites
for cavitation initiation. The volume fraction of the gaseous phase obtained from the solution includes both the vapor
and non-condensable gases. Initially, no vapor is present in the computational domain. However, non-condensable
gases are present in the liquid fuel in very small mass fractions (N2 = 10−5 ; O2 = 10−5 ). The chamber is initialized with
chamber pressure, chamber temperature, and only nitrogen. The eight holes, along with counter-bores, are initialized
with chamber pressure, fuel temperature, and 99.998% liquid fuel and trace amounts of non-condensable gases, as
mentioned before. The needle sac and the upstream of sac region up to the inlet are initialized with fuel injection
pressure, fuel temperature and 99.998% liquid fuel and trace amounts of non-condensable gases. Simulations are
transient, and all calculations are run until both the inflow and the outflow mass flow rates are stabilized.

2.6 Solution Procedure

The commercial software CONVERGE v2.2 is used in the current study to solve the governing equations using the
finite volume approach. The Rhie–Chow algorithm has been adopted to have all the transport variables collocated
at the cell center. The pressure implicit with splitting of operator (PISO) algorithm is used for pressure–velocity
coupling, since it is expected to work well for unsteady compressible flow problems (Richards et al., 2015). For the
momentum equation, second-order central differencing is employed, while for the other transport equations, first-
order upwind scheme has been applied. Successive Over-Relaxation (SOR) algorithm has been observed to offer
better numerical stability. The flash-boiling problem studied in this work is a high-speed, two-phase compressible
flow where sharp or nearly sharp interface between liquid and vapor phases is not expected. Hence, no interface-
capturing scheme has been used. The mixture multiphase approach is adopted for the simulations and not the Volume
of Fluid (VOF) method coupled with piecewise-linear interface calculation (PLIC) scheme. VOF/PLIC could be
anyway tricky to implement in high-speed compressible flows. A special smoothing operation and consideration of

Volume 27, Issue 4, 2017

354 Saha, Som, & Battistoni

surface tension effects are needed to capture the interface characteristics of complex flows with steep density gradients
(Pianet et al., 2010). However, such modified approaches are not needed in the current study.

2.7 Material Properties

The majority of the material properties utilized in this study are obtained from the material database of the code
(Richards et al., 2015). Modifications were required for capturing liquid density changes with changes in pressure
and temperature. For implementing liquid compressibility, reference pressure, reference density, and bulk modulus
information are incorporated using the data available in the literature (Dymond et al., 1985).

3. RESULTS AND DISCUSSION

In this section, first comparison with experiment, and then the effect of thermo-dynamically different boundary condi-
tions will be demonstrated through HRM predictions. After that, results from the parametric study will be presented,
followed by turbulence model variation effects.

3.1 Model Assessment

Experimentally measured mass flow-rate values are available through ECN (Spray G ECN Gasoline Spray Combus-
tion, 2014) for the Spray G Submerged condition. In this condition, the average injection pressure (in the quasi-steady
region) is slightly low, i.e., 19.5 MPa. The chamber is filled with liquid iso-octane. The real injector geometry infor-
mation and needle lift data were not available at the time of this simulation. Hence, a fixed needle lift simulation with
the ideal geometry has been carried out. This result is also available in our prior publication (Saha et al., 2016a) for a
shorter time duration. In the real geometry, the dimensions of the holes are different, which is not the case for the ideal
geometry considered in this study. As a result, some discrepancies can be expected between the simulation predictions
and the measured values. The mass flow-rate predictions for Spray G Submerged are shown, along with experimental
values, in Fig. 4. Overall, we see some over-prediction after the solution stabilizes. It is seen that the peak value of
the measured ROI is around 14.5 mg/ms and in the stabilized region the prediction fluctuates mainly between 14.5
and 15.5 mg/ms. If these values are averaged over the stable region, there is an over-prediction of approximately 4%.
These predictions are reasonable considering that this simulation is done with fixed needle lift and ideal geometry.
These mass flow-rate predictions are better than the ones presented in our previous work (Saha et al., 2016a). In our
previous work the cell count was 2.8 million with the minimum cell size of 17.5 µm. Rigorous investigation led to the
finding that having larger domains of fine mesh in the sac region helped in mass flow-rate predictions. In the current
validation result the cell count was approximately 3.9 million with the minimum cell size as 17.5 µm. The following
results in this study were carried out with the cell count of 2.8 million. This way computational cost was reduced as
well, as it was ensured that the qualitative findings from the parametric study will be still valid.

3.2 Boundary and Operating Conditions

In the current study, we do not show any more validation of the simulation setup. Our previous studies (Saha et al.,
2016a,b) assessed the predictions qualitatively as well as quantitatively and proved the credibility of the numerical
predictions using the current setup. Hence those findings are not shown here for the sake of brevity. The conditions
studied are Spray G, Spray G2, and Spray G3 as shown in Table 1. The expected outcomes for these three conditions
have already been stated earlier through thermodynamic assessment. Figure 5 presents the vapor mass fraction con-
tours at 0.1 ms. The predicted results are in accordance with prior estimation based on thermodynamic analysis. The
Spray G calculation indicates that the peripheries of the liquid jets are vaporizing because of convective heating in
the high-temperature chamber. Spray G2 is moderately flashing, since the degree of superheat is not very high. The
degree of superheat is very high for Spray G3 and hence there is abundant vapor formation, i.e., intense flashing in
the chamber and significant plume-to-plume interaction, as shown in the simulation result. HRM is used to model
any sort of phase change in the simulations. However, HRM has not been proven in the literature to be effective for

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 355

FIG. 4: Mass flow-rate predictions compared with measured ROI for Spray G submerged condition

vaporization caused by convective heating due to high ambient temperature. Therefore, vapor formation prediction
for Spray G may not necessarily be an accurate representation of the actual non-flashing but vaporizing scenario.
Detailed experimental findings with quantified information on vapor concentration are still not available in literature.
At this stage, it is not feasible to provide a comparison of predictions of vapor concentration with experimental data.
However, the prior thermodynamic calculations, along with observations by Weber and Leick (2014), point to the
fact that when the pressure ratio (RP = Psat /Pchamber ) exceeds 2, there is the potential of aggressive flash-boiling.
Weber and Leick also observed that under intense flash-boiling conditions it is feasible for two opposing spray plumes
to interact with each other (cf. Fig. 6). The vapor fuel mass fraction contour for Spray G3 (in Fig. 5) shows that the
HRM captures the essence of aggressive flash boiling and plume merging.
The above result and corresponding discussion only depict how the mass fraction contours vary when the system
is subjected to different thermodynamic conditions. The density of vapor fuel is a few orders of magnitude lower
than that of liquid fuel. Therefore, the volume fraction of vapor fuel is going to be significantly higher than the mass
fraction values. Figure 7 presents the mass fraction and volume fraction contours of vapor iso-octane for the Spray

FIG. 5: Effect of operating and boundary conditions on two-phase flow through Spray G injector under G, G2, and G3 conditions.
Vapor mass fraction contours are shown through a cut-plane showing two of the eight holes

Volume 27, Issue 4, 2017

356 Saha, Som, & Battistoni

FIG. 6: Effect of changing the exponent of α on the fuel vapor mass fraction contour for the Spray G injector under the G2
condition from Table 1

G2 case. As a result, it is clear that, even though a considerable mass of vapor is not formed during moderate flashing,
the vapor phase occupies a substantial volume in the counter-bores and farther downstream along the spray plumes.
This information is useful for detailed spray calculations for different chamber conditions, which will be undertaken
in the future. The Spray G2 mass fraction contours in Figs. 5 and 7 look different because of the differences in the
scales. Figure 5 was prepared to emphasize the effects of different thermodynamic conditions.

3.3 HRM Parametric Study

The constants of the HRM have been optimized for cases where water flowing through a nozzle is flashing (Bilicki and
Kestin, 1990; Downar-Zapolski et al., 1996). In the current study, the fluid, operating, and boundary conditions are
different from those used by Downar-Zapolski et al. (1996), wherein these model constants were optimized through
curve-fitting with measured data for water. Hence, analyzing the sensitivity of the model predictions with respect
to the model constants is warranted in studies involving different fluids with HRM. The time-scale constant θ0 and
exponents of α and ψ from Eq. (14) are varied for this sensitivity study. The ranges of uncertainty of these model
constants are not documented in the literature, to the best of our knowledge. In the optimization study by Bliss et al.
(2012), the new proposed value of θ0 was of the same order as in the original HRM. Despite that change, only marginal
improvement was seen in their study. In that study, the other parameters (exponents of α and ψ) were also different
from the ones in the original HRM. It is difficult to assess the role of θ0 with simultaneous variations of all the model
constants. The time scale of flashing can vary by 1 or 2 orders of magnitude, depending on local thermodynamic
conditions, as seen in the work by Downar-Zapolski et al. (1996). As a result, the order of θ0 is varied by an order
of magnitude. For exponents of α and ψ, positive values would not be correct. HRM is a phenomenological model
and changing the sign of the exponent would change the physical dependency of θ on the two variables. Because of
this physical constraint, it was decided that a ±50% variation about the values in the original HRM would be a wide
enough range to carry out the sensitivity study.

FIG. 7: Mass fraction and volume fraction contours of vapor iso-octane under the G2 condition from Table 1

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 357

The parameter θ0 has been varied from the order of 10−6 to 10−8 . Figure 8 presents the effect of changing θ0 on
vapor formation for the Spray G2 condition. A lower value of θ0 seems to trigger more flashing, which is expected
since θ0 considerably lowers the equilibrium time scale, leading to a significant increment of (Dx)/(Dt) throughout
the run time. Therefore, it is evident that θ0 is a very influential factor, at least for the case of moderate flashing.
Intense flashing is expected when the pressure ratio (RP ) is greater than 2 or 3 (Weber and Leick, 2014), and for
Spray G2, RP < 2. Therefore, Spray G2 should be moderately flashing, and θ0 ≈ 10−7 would still seem to be
reasonable for the Spray G2 condition.
The exponent of α has been varied from −0.27 to −0.81, with the default value being −0.54. The effect of this
variation is shown in Fig. 6. There is hardly any change in the prediction, even with ±50% variation about the default
value of −0.54. For this Spray G2 condition, the chamber is filled with non-condensable gases and α comprises both
vapor fuel and non-condensable gases. Unlike θ0 , α is a variable that approaches a constant value (≤1) as the fluid
travels through the nozzle and is convected downstream. Hence, along the fluid path line, the influence of α on θ
diminishes. Therefore, α is not going to be a dominant factor for the cases of interest in this study.
The parameter ψ, in Eq. (14), incorporates the effect of fuel physical properties in the time-scale estimation, as
well as the essence of the local thermodynamic conditions. When the fuel temperature is raised, the corresponding
saturation pressure increases and so does the value of ψ. This is because that with the increase in psat , the denom-
inator of ψ decreases and the numerator increases, provided the local pressure remains the same. The exponent of
ψ has been varied from −0.88 to −2.64, with the default value being −1.76. ψ significantly affects the vapor for-
mation, as depicted in Fig. 9. ψ varies by orders of magnitude depending on local pressure values. Consequently, in
flashing-prone zones, ψ has significantly high values and it considerably influences the vapor formation. As previ-
ously mentioned, the Spray G2 condition is supposed to moderately flash and hence, the default value of −1.76 as
the exponent of ψ should be reasonable for the Spray G2 condition.
In the code, there is another crucial parameter called ψmin . For cases where local pressure is equal to the satura-
tion pressure, the value of (Dx)/(Dt) in Eq. (13) will blow to infinity, because θ in the denominator becomes zero.

FIG. 8: Effect of changing θ0 on fuel vapor mass fraction contour for the Spray G injector under the G2 condition from Table 1

FIG. 9: Effect of changing the exponent of ψ on fuel vapor mass fraction contours for Spray G injector under G2 condition from
Table 1

Volume 27, Issue 4, 2017

358 Saha, Som, & Battistoni

To avoid this problem, ψmin has been used as a threshold parameter, such that when the value of ψ is below ψmin ,
the value of ψmin will be used instead of ψ. Typically, the value of this parameter is set as 10−5 . The values have
been varied by a few orders, from 10−3 to 10−6 . Like α, ψ varies during the run time and ψmin becomes useful only
when local pressure is very close to saturation pressure. However, it is clear from Fig. 10 that prediction of vapor
formation is not very sensitive to the ψmin values considered here. Overall, it seems that the default values of the
HRM constants are providing reasonable predictions. Model constants optimized for a different liquid species and
under different operating conditions perform very decently for GDI systems.
On the basis of results presented at this point, it can be inferred that time scale θ is strongly influenced by θ0 and
ψ as well as its exponent. Nevertheless, it is not very clear how θ affects the vapor production in order to capture the
phase changes under various thermodynamic conditions. A simple mathematical exercise can help in visualizing the
qualitative trend of θ. It is known that local pressure varies from tens of Pa inside the injector holes to a few hundred
kPa in the near-nozzle region in GDI engines, where flashing usually occurs. Using the mathematical expression θ in
Eq. (14), the pattern of variation of θ can be investigated for a range of α and ψ, i.e., local pressure, for a given fuel
temperature. For fuel temperatures of 300 K and 360 K, Fig. 11 presents the variation of θ with α and local pressure.
For a fixed fuel temperature, the saturation pressure is fixed, and therefore, the variation of local pressure over a wide
range would encompass the possible range of variation of ψ. The local pressures have been selected such that the
qualitative pattern of variation of θ becomes apparent. The peak values are typically occurring around the saturation
pressure for both 300 K (psat = 7320 Pa) and 360 K (psat = 70,900 Pa). The peak value of θ, however, will definitely
change depending on the selection of local pressure, close to the saturation pressure. The variations of θ indicate,
again, that the local pressure or, to be precise, the ψ parameter is a more dominant factor than α for influencing
θ. For both the cases, it is seen that θ decreases when α approaches unity; this behavior is more evident from the
magnified view of the 360 K case. Under typical phase change conditions, the tendency to produce vapor is supposed
to decrease as the gas phase approaches its peak concentration. HRM was originally developed with measured data
for a mixture of liquid and vapor phases only. Moreover, in computational cells where the gaseous phase (vapor fuel
+ non-condensable gases) is predominant, the chance that local vapor quality will be close to equilibrium is high. As
a result, the behavior of θ with the change in α appears to be logical.

FIG. 10: Effect of changing the ψmin on the fuel vapor mass fraction contour for the Spray G injector under the G2 condition
from Table 1

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 359

FIG. 11: Variation of θ with α and local pressure (kPa) for fuel temperatures of 300 K and 360 K. The enlarged view highlights
the extent of variations when θ is approaching the peak values for fuel at 360 K

Since ψ is the more dominant factor, it is important to focus on its effect on θ. Figure 12 helps in clarifying
the role of ψ in influencing θ. Close to saturation pressure for a given fuel temperature, ψ decreases drastically,
leading to an abrupt rise in θ, as seen in Fig. 11. Another noteworthy aspect is that the variation of ψ (and as a
consequence the variation of θ) in Fig. 11 about the saturation pressure is symmetric. In this regard, HRM suffers
from the same fundamental inadequacy as the Linear Rayleigh Equation-based cavitation models used in single-fluid
(mixture or VOF) and two-fluid (Eulerian–Eulerian) models for studying in-nozzle analyses of diesel injectors (Saha,
2014). Fundamentally, the processes of bubble growth and collapse are not the same. Bubble collapse occurs at a
much faster rate compared to bubble growth (Brennen, 1995). The time scale of complex bubble collapse is too small
(≈nanoseconds) to be resolved in a multi-dimensional simulation involving clusters of bubbles or cavities (Saha,
2014). As a result, the effectiveness of HRM, in terms of vapor condensation, for engineering calculations is not
inferior compared to the commonly adopted, simple bubble-based approaches.
The next aspect to observe is that the peak values of θ are similar for both the fuel temperatures. However,
for chamber pressures on the order of 50 kPa and higher, θ values are larger for 360 K compared to 300 K. As
explained previously, flashing is a non-equilibrium phenomenon, which tends to occur at elevated fuel temperatures.
Additionally, θ for 360 K drops down noticeably with a further decrease in local pressure. The local pressures inside
the GDI holes are on the order of a few hundred Pascals. Consequently, the equilibrium time scales inside the holes are
smaller than those outside the injector, in the chamber. As a result, phase changes inside the holes can be considered
cavitation, as the non-equilibrium effects are small because of small values of θ. Keeping in mind that the chamber
pressure of the Spray G2 condition is 53 kPa, the predictions of HRM are very realistic in a physical sense for
GDI applications. However, for the 300 K case, θ peaks at a very low local pressure, since at low fuel temperature
the saturation pressure will also be low. Cavitation, being a near-equilibrium phenomenon, should be represented by
small equilibrium time scales; this condition is clearly not being satisfied. In these calculations, negative local pressure

Volume 27, Issue 4, 2017

360 Saha, Som, & Battistoni

FIG. 12: Variation of ψ with local pressure (kPa) for fuel temperatures of 300 K and 360 K

was not considered, primarily because injector flow is compressible and the presence of gas lowers the threshold for
cavitation (Brennen, 1995). Negative pressure is a possibility only under very specific conditions (Caupin and Herbert,
2006) that are not characteristic of fuel injection scenarios. Therefore, for simulation of cavitation in high-pressure
diesel injectors with low fuel temperatures, θ predictions could not be realistic.
Figures 10, 11, and 12 in conjunction with Eqs. (13), (14), and ψmin , provide the justifications for using HRM
in a CFD simulation for flashing as well as cavitating conditions. Figure 10 shows that the extent of vapor prediction
does not change noticeably by changing ψmin from 10−3 to 10−6 . Figure 12 reveals that for small changes in local
pressure (in the vicinity of Psat ) the ψmin parameter changes its values from 10−3 to 10−6 . Therefore, it is clear
that ψmin acts as a clipping parameter that truncates the equilibrium time scale (θ) predictions and ensures that
not unreasonably high values of θ are used in the CFD calculations. Consequently, Fig. 11 shows that for high
fuel temperature, under flashing condition (in the chamber) the predicted values of θ are higher than the values of
under cavitating condition (in holes). This trend is physically reasonable. However, that is not the case with low fuel
temperatures, where the θ values under cavitating conditions tend to be very high especially for local pressure values
in the vicinity of corresponding saturation pressure. At low fuel temperatures, under cavitating conditions θ values
are instead expected to be low. In CFD calculations, due to the usage of ψmin , extremely low values of ψ and hence,
extremely high values of θ could be avoided. Therefore, it is still possible to get the reasonable results through CFD
calculations with unphysical high predicted values for θ through HRM. Thus the current model formulation and the
code implementation provide reliable predictions specifically for high fuel temperatures.
The θ variations for different fuel temperatures with respect to the corresponding saturation pressure are exactly
symmetric about the saturation pressure. The peak simply shifts left with lowering of fuel temperature. This behavior
is not physical and it is due to the model formulation. Depending on fuel temperature the tendency to reach thermody-
namic equilibrium changes and the current patterns of θ variations do not seem to capture that physical characteristic.
HRM is a phenomenological model, because it is not derived from first principles. Therefore, despite having some
physical basis, θ variation might not always be fundamentally correct. Ideally, a phase change sub-model should con-
sider the effect of local pressure, local temperature, and physical properties of the fuel. Inclusion of non-dimensional
numbers like the Jakob number in the model formulation could thus be useful for physically meaningful sub-models.
Such endeavors constitute the scope for future work.
Eventually, it will be interesting to see how the above analysis can be reflected in situation typical of GDI
engines. For Spray G, G2, and G3 conditions, Fig. 13 shows the vapor concentrations, as well as the θ values in
different computational cells, in scatter plots of local pressure versus local temperature. The blue lines correspond

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 361

FIG. 13: Scatter plots of local pressure and local temperature in the different computational cells in one of the eight holes (includ-
ing the counter-bore) of the Spray G injector under Spray G, G2, and G3 conditions, colored according to θ (left) and vapor mass
fraction (right). The blue line corresponds to the saturation pressure versus saturation temperature curve for iso-octane

to the saturation pressure versus saturation temperature curve for iso-octane. The cells belong to one of the eight
holes and the cells in the corresponding counter-bore are also included. It is evident that cells having local pressure
lower than the saturation pressure comprise the cells occupied mostly by fuel vapor. The cells with local pressure
close to the saturation pressure have the peak θ values. Interestingly, a few cells with local pressure above the sat-
uration pressure also have some fuel vapor, owing to the mathematical nature of θ, which feeds into the source
term of the fuel vapor transport equation. The temperature variations for different conditions are caused by dif-
ferences in the chamber temperatures. Previous studies (Moulai et al., 2015; Saha et al., 2016a,b) have indicated
that there is back-flow of the chamber gas into the counter-bores. Spray G is exposed to high chamber temper-
ature (573 K), while for Spray G2 and G3 conditions the chambers are at room temperature (293 K). Since the
fuel temperature is 363 K for G and G2, and 413 K for G3, the distributions of the cells in the scatter plots make
sense for all the three conditions. In the Spray G condition, the heating of the fuel (in the chamber) occurs en-
tirely at pressure levels above the saturation curve. Cavitation is clearly shown by those cells (in the holes) which
undergo isothermal pressure drop. In the Spray G2 condition, the low chamber temperature induces some fuel cool-
ing, while vapor is simultaneously being formed because of the pressure is below the saturation level. In Spray
G3, the same process occurs, but at higher degree of superheat, leading to a larger amount of fuel vapor forma-
tion.

Volume 27, Issue 4, 2017

362 Saha, Som, & Battistoni

3.4 Turbulence Models

Typically, for all the cases, the standard k-ǫ turbulence model has been used. To understand the role of the turbulence
model, the RNG k-ǫ turbulence model has also been used for the Spray G condition in Table 1. Figure 14 shows the
effect of turbulence model variation. Considerable differences have been observed, especially the lower “feedback”
from the pressure outlet boundary condition for the RNG k-ǫ turbulence model. The excessive “feedback” with the
standard k-ǫ model, as noted by Saha et al. (2016a,b), is reduced by using larger outlet domains. However, with the
RNG k-ǫ turbulence model that specific problem is almost non-existent with 9 mm domain, at least under the Spray
G condition.
In order to understand the reason for the differences between the standard and RNG k-ǫ models, it is worthwhile
to look at the turbulent viscosity distributions in the chamber with two different turbulence modeling approaches. We
have verified that for RANS-based turbulence simulations, at reasonably fine mesh resolutions, turbulent viscosity
is dominant compared to molecular viscosity and numerical viscosity. The differences in the viscosity contours are
noteworthy, as seen in Fig. 15. Higher viscosity predictions with the standard k-ǫ model imply more turbulence-
induced mixing, which can contribute to higher vapor dispersion, as seen in Fig. 15. Unlike the standard k-ǫ model,

FIG. 14: Effect of changing the RANS turbulence model on the fuel vapor distribution for the Spray G injector under the G
condition from Table 1

FIG. 15: Effect of changing the RANS turbulence model on the turbulent viscosity distribution for the Spray G injector under the
G condition from Table 1

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 363

the RNG k-ǫ model accounts for low, medium, and high strain rates in the flows (Saha, 2014). In our prior publication
(Saha et al., 2016a), the effect of increasing the outlet domain was investigated. The issue of vapor buildup near
the outlet boundary could be mitigated to a great extent, even when the standard k-ǫ model was used. However,
the larger domain would mean more cells and hence, the need for more computational resources. Hence, the RNG
k-ǫ turbulence model should be a reasonable choice for these problems with in-nozzle and near-nozzle analyses.
Although most simulations are performed with the standard k-ǫ model, the qualitative findings for effects of different
parameters are still valid.

4. SUMMARY AND CONCLUDING REMARKS

A numerical parametric study has been carried out for a typical GDI fuel injector. The HRM is shown to be capable of
predicting non-flashing yet vaporizing, moderately flashing, and intensely flashing scenarios in accordance with the
calculations of the pressure ratio and the degree of superheat. The HRM constants were originally obtained for water
and under conditions very different from GDI systems. Despite that discrepancy, HRM provides promising results in
the current study. The set of parametric studies leads to the following inferences:

1. ψ varies over several orders of magnitude, both spatially and temporally, in a simulation. Thus, local pressure
and physical properties significantly affect HRM predictions, since ψ turned out to be a dominant factor for
vapor production. For a given local pressure and a given fuel, ψ will change with a change in fuel tempera-
ture and so will θ. Therefore, ψ has the potential of capturing the essence of variation in the thermodynamic
conditions.

2. The time-scale constant (θ0 ) is another dominant factor affecting the vapor distribution considerably. Larger
values of θ0 seem to hinder flashing, because a large relaxation time scale means more time to reach the
equilibrium vapor quality. Hence, at a given time, less vapor will be produced with large values of θ0 .

3. Void fraction, i.e., α, did not appear to be an influential variable in the HRM prediction. Typically, α can only
vary from 0 to 1. For Spray G conditions, the chamber is initially filled with N2 gas and hence α does not vary
much during the run time, especially for flashing conditions. Experimental findings in the literature indicate
that the relaxation time scale is a weak function of void after exceeding 10% void in the mixture.

4. HRM is effective for simulating phase changes for GDI spray applications, despite the fact that it is a phe-
nomenological model with tunable coefficients. This is because, at typical GDI engine chamber pressures,
HRM is capable of predicting larger equilibrium time scales for high fuel temperatures when compared to
those for low fuel temperatures. Thus, HRM provides a realistic depiction of the flashing scenario.

5. For cavitation problems at low fuel temperatures, HRM has a tendency to predict non-physically large relax-
ation time scales. This can lead to erroneous predictions or delayed vapor formation. Therefore, researchers
have to be cautious when using HRM for low-temperature cavitating flows. Typically, the fuel temperatures are
not very low for diesel engines. Hence, for automotive applications, HRM could still be useful.

6. The RNG k-ǫ turbulence model can be suitable for the GDI spray applications, because it can minimize the
physical vapor buildup in small outlet domains. Earlier work indicated that larger domains were needed with
standard k-ǫ model. Therefore, RNG k-ǫ model turns out to be an economic, yet physically correct turbulence
modeling approach for these Eulerian calculations.

Overall, HRM, even with its caveats is a desirable option for internal combustion engines, whether diesel or
gasoline fuel is used. Future modifications in HRM could include effects of the Jakob number to further improve its
fundamental basis.

Volume 27, Issue 4, 2017

364 Saha, Som, & Battistoni

ACKNOWLEDGMENTS

This research was partially funded by the U.S. Department of Energy (DOE) Office of Vehicle Technologies, Office
of Energy Efficiency and Renewable Energy under contract no. DE-AC02-06CH11357. The authors wish to thank
Gurpreet Singh and Leo Breton at DOE, for their support. The authors would like to thank Ronald Grover at General
Motors and Lyle Pickett at Sandia National Laboratory for providing the nominal geometry of the Spray G nozzle. The
authors want to express their gratitude to Yanheng Li and Eric Pomraning at Convergent Sciences Inc. for providing a
better understanding of the volume-averaging method for VOF simulations. The authors would like to thank Prithwish
Kundu at Argonne National Laboratory for his assistance during the preparation of this manuscript. The authors are
grateful to LCRC computing resources at Argonne National Laboratory.

REFERENCES
Battistoni, M., Som, S., and Longman, D.E., Comparison of mixture and multifluid models for in-nozzle cavitation prediction,
ASME J. Eng. Gas Turbines Power, vol. 136, no. 6, 061506, 2014a.
Battistoni, M., Xue, Q., Som, S., and Pomraning, E., Effect of off-axis needle motion on internal nozzle and near exit flow in a
multi-hole diesel injector, SAE Int. J. Fuels Lubricants, vol. 7, no. 1, pp. 167–182, 2014b.
Battistoni, M., Duke, D.J., Swantek, A.B., Tilocco, F.Z., Powell, C.F., and Som, S., Effects of non-condensable gas on cavitating
nozzles, Atomization Sprays, vol. 25, no. 6, pp. 453–483, 2015.
Bianchi, G.M., Negro, S., Forte, C., Cazzoli, G., and Pelloni, P., The prediction of flash atomization in GDI multi-hole injectors,
SAE Technical Paper, 2009-01-1501, 2009.
Bilicki, Z. and Kestin, J., Physical aspects of the relaxation model in two-phase flow, Proc. R. Soc. London, Ser. A, vol. 428,
pp. 379–397, 1990.
Blinkov, V.N., Jones, O.C., and Nigmatulin, B.I., Nucleation and flashing in nozzles—2. Comparison with experiments using a
five-equation model for vapor void development, Int. J. Multiphase Flow, vol. 19, no. 6, pp. 965–986, 1993.
Bliss, J., Schmidt, D., Smith, J., and Neroorkar, K., Model constant optimization using an ensemble of experimental data, SAE
Technical Paper, 2012-01-0131, 2012.
Brennen, C.E., Cavitation and Bubble Dynamics, New York: Oxford University Press, 1995.
Caupin, F. and Herbert, E., Cavitation in water: A review, C. R. Physique, vol. 7, pp. 1000–1017, 2006.
Chang, D.L. and Lee, C.F., Preliminary computational studies of flash boiling for fuel injectors in gasoline direct injection auto-
motive engines, Proc. Intersoc. Energy Conversion Eng. Conf., vol. 37, pp. 464–469, 2002.
Downar-Zapolski, P., Bilicki, Z., Bolle, L., and Franco, J., The non-equilibrium relaxation model for one-dimensional flashing
liquid flow, Int. J. Multiphase Flow, vol. 22, no. 3, pp. 473–483, 1996.
Dymond, J.H., Glen, N.F., and Isdale, J.D., Transport properties of nonelectrolyte liquid mixtures VII. Viscosity coefficients for
isooctane and for equimolar mixtures of isooctane + n-octane and isooctane + n-dodecane from 25 to 100 at pressures up to
500 MPa or to the freezing pressure, Int. J. Thermophys., vol. 6, no. 3, pp. 233–250, 1985.
Gopalakrishnan, S. and Schmidt, D., A computational study of flashing flow in fuel injector nozzles, SAE Technical Paper, 2008-
01-0141, 2008.
Harada, J., Tomita, T., Mizuno, H., Zenichiro, M., and Ito, Y., Development of direct injection gasoline engine, SAE Technical
Paper, 970540, 1997.
Iwamoto, Y., Noma, K., Nakayama, O., Yamauchi, T., and Ando, H., Development of gasoline direct injection engine, SAE Tech-
nical Paper, 970541, 1997.
Kawano, D., Goto, Y., Odaka, M., and Senda, J., Modeling atomization and vaporization processes of flash-boiling spray, SAE
Technical Paper, 2004-01-0534, 2004.
Montanaro, A. and Allocca, L., Flash boiling evidences of a multi-hole GDI spray under engine conditions by Mie-scattering
measurements, JSAE, 2015-01-1945, 2015.
Morena, J.D., Neroorkar, K., Plazas, A.H., Peterson, R.C., and Schmidt, D.P., Numerical analysis of the influence of diesel nozzle
design on internal flow characteristics for 2-valve diesel engine application, Atomization Sprays, vol. 23, pp. 97–118, 2013.

Atomization and Sprays

Investigation of Homogeneous Relaxation Model Parameters 365

Moulai, M., Grover, R., Parrish, S., and Schmidt, D., Internal and near-nozzle flow in a multi-hole gasoline injector under flashing
and non-flashing conditions, SAE Technical Paper, 2015-01-0944, 2015.
Neroorkar, K., Shields, B., Grover Jr., R., Plazas Torres, A., and Schmidt, D., Application of the homogeneous relaxation model
to simulating cavitating flow of a diesel fuel, SAE Technical Paper, 2012-01-1269, 2012.
Neroorkar, K.D., Modeling of flash boiling flows in injectors with gasoline-ethanol fuel blends, Ph.D. Thesis, University of
Massachusetts-Amherst, 2011.
Pianet, G., Vincent, S., Leboi, J., Caltagirone, J.P., and Anderhuber, M., Simulating compressible gas bubbles with a smooth
volume tracking 1-fluid method, Int. J. Multiphase Flow, vol. 36, pp. 273–283, 2010.
Reitz, R.D., A photographic study of flash-boiling atomization, Aerosol Sci. Technol., vol. 12, pp. 561–569, 1990.
Richards, K.J., Senecal, P.K., and Pomraning, E., CONVERGE v2.2 Documentation. Convergent Sciences Inc., 2015.
Saha, K., Modelling of cavitation in nozzles for diesel injection applications, Ph.D. Thesis, University of Waterloo, 2014.
Saha, K., Som, S., Battistoni, M., Li, Y., Pomraning, E., and Senecal, P.K., Numerical investigation of two-phase flow evolution
of in- and near-nozzle regions of a gasoline direct injection engine during needle transients, SAE Int. J. Engines, vol. 9, no. 2,
2016a.
Saha, K., Som, S., Battistoni, M., Li, Y., Quan, S., and Senecal, P.K., Modeling of internal and near-nozzle flow for a GDI fuel
injector, ASME J. Energy Resources Technol., vol. 138, no. 5, pp. (052208)1–11, 2016b.
Schmidt, D.P., Gopalakrishnan, S., and Jasak, H., Multi-dimensional simulation of thermal non-equilibrium channel flow, Int. J.
Multiphase Flow, vol. 36, no. 4, pp. 284–292, 2010.
Spray G ECN Gasoline Spray Combustion, Sandia National Laboratory, http://www.sandia.gov/ecn/index.php, December, 2014.
Vanderwege, B.A. and Hochgreb, S., The effect of fuel volatility on sprays from high-pressure swirl injectors, Proc. Combust.
Inst., vol. 27, pp. 1865–1871, 1998.
Weber, D. and Leick, P., Structure and velocity field of individual plumes of flashing gasoline direct injection sprays, In ILASS-
Europe 2014, 26th Annual Conf. on Liquid Atomization and Spray Systems, Bremen, Germany, September 8–10, 2014.
Xu, M., Zhang, Y., Zeng, W., Zhang, G., and Zhang, M., Flash boiling: Easy and better way to generate ideal sprays than the high
injection pressure, SAE Technical Paper, 2013-01-1614, 2013.
Xue, Q., Battistoni, M., Som, S., Quan, S., Senecal, P.K., Pomraning, E., and Schmidt, D., Eulerian CFD modeling of coupled
nozzle flow and spray with validation against x-ray radiography data, SAE Int. J. Fuels Lubricants, vol. 7, no. 2, pp. 1061–1072,
2014.
Xue, Q., Battistoni, M., Powell, C.F., Quan, S., Pomraning, E., Senecal, P.K., Schmidt, D.P., and Som, S., An Eulerian CFD model
and x-ray radiography for coupled nozzle flow and spray in internal combustion engines, Int. J. Multiphase Flows, vol. 70,
pp. 77–88, 2015.
Zeng, W., Xu, M., Zhang, G., Zhang, Y., and Cleary, D.J., Atomization and vaporization for flash-boiling multi-hole sprays with
alcohol fuels, Fuel, vol. 95, pp. 287–297, 2012.
Zeng, Y. and Lee, C.F., An atomization model for flash boiling sprays, Combust. Sci. Technol., vol. 169, pp. 45–67, 2001.
Zhao, H., Quan, S., Dai, M., Pomraning, E., Senecal, P., Xue, Q., Battistoni, M., and Som, S., Validation of a three-dimensional
internal nozzle flow model including automatic mesh generation and cavitation effects, ASME J. Eng. Gas Turbines and Power,
vol. 136, no. 9, pp. (092603)1–10, 2014.

Volume 27, Issue 4, 2017