You are on page 1of 17

International Journal of Multiphase Flow 114 (2019) 98–114

Contents lists available at ScienceDirect

International Journal of Multiphase Flow


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

Improved hybrid model applied to liquid jet in crossflow


Douglas Hector Fontes∗, Vitor Vilela, Lucas de Souza Meira, Francisco José de Souza
School of Mechanical Engineering, Federal University of Uberlândia, 2121 João Naves de Ávila Avenue, Uberlândia, Minas Gerais, Brazil

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

Article history: Spray formation in a liquid jet in air crossflow is investigated numerically using a hybrid approach for two
Received 9 October 2018 cases: We = 11 and We = 53, where We is the air Weber number. The VOF method was used to solve the
Revised 27 February 2019
interaction between the liquid jet column and the air flow up to the primary breakup. The Eulerian liquid
Accepted 28 February 2019
portion is converted to discrete Lagrangian drops by the use of source terms of mass and momentum and
Available online 11 March 2019
according to empirical correlations. The effects of two secondary breakup models, as well as two-way
Keywords: coupling with droplet-to-droplet collisions were evaluated in this work, in order to establish an improved
Liquid jet in crossflow hybrid model suitable to solve liquid jet in crossflow at low computational cost. The results were very
VOF Method satisfactory relating droplet velocity and mass fraction distribution for the lower Weber number case. For
Hybrid approach the higher Weber number case, the deviation of the numerical mass fraction distribution in comparison
AB-TAB model with the experimental one was less than 14 × 10−2 . Improvement to the droplet size distribution was
secondary breakup models
observed by using a new secondary breakup model, specially for the higher Weber number case.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction sequent breakup that the detached droplets experience along the
flow (Reitz, 1987; Lin and Reitz, 1998).
Liquid jet in crossflow (LJIC) is an effective configuration for Numerical simulations of LJIC are essential for the design and
spray formation in several combustion systems, such as carbure- evaluation of many combustion systems. These simulations in-
tors, afterburners of a jet engine, gas turbine combustors, ramjet clude the primary and secondary breakups, when drops separate
and scramjet combustors (Hojnacki, 1972; Inamura, 1997; Mad- from the liquid column and from other droplets, respectively. Sim-
abhushi, 2003). The perpendicular interaction of the two phase ulations have been performed based on the Euler-Lagrange ap-
flows, whose density ratio is usually of the order of 103 and proach, Euler-Euler approach or hybrid approach. The choice of
liquids with low viscosity, generates complete disintegration of each method is usually related to the balance of the accuracy re-
the liquid jet in a short space (i.e. less than 10 times the jet quirement and the computational resources.
diameter), for viscosity is a kind of restraining force to atom- The Euler-Lagrange approach applied to LJIC consists of trans-
ization so that lower viscosity may produce smaller droplets porting the gaseous stream based on the Eulerian reference frame
(Strasser, 2019). The aerodynamic, viscous and surface tension and the liquid droplets based on the Lagrangian referential. At its
forces on LJIC influence spray formation in many aspects: liquid early stages, this approach considered the liquid jet as Lagrangian
jet breakup height and mode, droplets breakup mechanism and blobs Reitz (1987). In the blob method, a droplet with diameter
velocity. Kelvin-Helmholtz, Rayleigh-Taylor instabilities, turbulent of the order of the nozzle diameter is released into the gas cross-
eddies and cavitation are the main causes of primary breakup in flow with a frequency that maintains the jet mass flow rate. The
sprays Bravo et al. (2014). LJIC studies have related the primary primary breakup of the blobs and the secondary breakup of the
breakup instabilities and the liquid jet breakup height to the We- subsequent droplets can be modeled based on the growth of insta-
ber number, We, and the momentum flux ratio, J, respectively (Wu bilities, such as Rayleigh-Taylor and Kelvin-Helmholtz (Bravo et al.,
et al., 1997; Mazallon et al., 1998; Becker and Hassa, 2002; Sallam 2014). Reitz (1987) evaluated the LJIC using the Euler-Lagrange ap-
et al., 2003). Additionally, the Weber number also affects the sub- proach, considering droplets collision and coalescence. Different di-
ameters of the generated droplets were calculated, relating them
to the wavelengths on the blob surface. The jet penetration and
droplet sizes were in accordance with the experimental measure-

Corresponding author. ments of Hiroyasu and Kadota (1974). In a recent work (Li et al.,
E-mail addresses: douglas.fontes@ufu.br (D.H. Fontes), vitor.vilela@ufu.br (V. 2017), LJIC in supersonic air flow (Ma = 1.94) based on the Euler-
Vilela), lucas.meira@ufu.br (L.d. Souza Meira), francisco.souza@ufu.br (F. José de Lagrange approach was simulated. The Kelvin-Helmholtz breakup
Souza).

https://doi.org/10.1016/j.ijmultiphaseflow.2019.02.009
0301-9322/© 2019 Elsevier Ltd. All rights reserved.
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 99

model was used to calculate the droplet stripping and secondary brid approach properly solves the liquid jet and gas crossflow in-
breakup was simulated using Taylor Analogy Breakup (TAB) and teractions, just as it is done in the Euler-Euler, while avoiding
Rayleigh-Taylor (RT) models concomitantly; Large Eddy Simulation the high computational cost in dilute regions, similarly to the
was applied with modified drag coefficient and breakup models, Euler-Lagrange. This approach also requires empirical correlations
depending on compressible effects and droplet deformation. The to the primary breakup, whenever the Eulerian representation is
authors obtained good agreement with experimental data on jet replaced by the Lagrangian one. Arienti et al. (2006) presented a
penetration and Sauter Mean Diameter along the streamwise di- hybrid algorithm to solve LJIC. The VOF method with the algebraic
rection. scheme HRIC was used. The authors used empirical correlations
The main advantage of the Euler-Lagrange approach lies in its from Mazallon et al. (1999) and Sallam et al. (20 03, 20 06) to solve
low computational cost, since the liquid phase is solved as parcels. the primary breakup of the liquid jet. The volume flux, droplets
On the other hand, its main disadvantage is the poor evaluation of velocities and spatial distribution of the droplets diameters were
the interaction between liquid jet and gas crossflow, since the liq- compared to experimental measurements, showing good agree-
uid jet is usually modeled as unrealistic blobs (Reitz, 1987; Bravo ment. An important advantage of the hybrid approach used by
et al., 2014). Other numerical results using the Euler-Lagrange ap- Arienti et al. (2006) was to suitably capture the liquid jet mor-
proach (Apte et al., 2003; Balasubramanyam and Chen, 2008) dis- phology during its interaction with the air crossflow at a lower
play good agreement regarding droplet size and velocity with ex- computational cost than that of the Euler-Euler approach. The au-
periments. In general, the Euler-Lagrange approach is highly de- thors used source terms and empirical correlations in the VOF and
pendent on empirical correlations, therefore, it is more suitable for momentum equations to switch between Eulerian and Lagrangian
use in cases with great complexity and/or when the interaction of frameworks. The secondary breakup of the droplets generated from
phases is not of main importance. the liquid jet was simulated using the Taylor Analogy Breakup
A second formulation used more recently for LJIC is the (TAB) method (O’Rourke and Amsden, 1987). Droplet velocities and
Euler-Euler approach. It may describe the two phases using sev- diameters presented good agreement with the experimental data
eral methods, such as Front-Tracking (Tryggvason et al., 2001), (Shedd, 2009).
VOF (Volume of Fluid) (Hirt and Nichols, 1981) and Level Set In the view of numerically solving the primary breakup with-
(Sussman et al., 1994). Since the secondary breakup usually gen- out any empirical correlations, some hybrid simulations have been
erates droplets of the order of 10−6 m for low liquid viscosity and made (Li et al., 2010; Herrmann, 2010; Li and Soteriou, 2016; Ling
high density ratios as found in water-air flows, this kind of numer- et al., 2017). However, in this high-fidelity simulations, the mesh
ical simulation incurs in a high computational cost, due to the high resolution required to properly solve the primary and secondary
mesh refinement level required. Despite all improvements on par- breakups is in the order of hundreds of millions elements, as the
allel computation and computational methods, there are relatively liquid structure is converted to a droplet only after the secondary
few works in the literature of sprays using the Euler-Euler ap- breakup.
proach. Li and Soteriou (2016) simulated LJIC using the Euler-Euler The purpose of the present work is to bring some advance-
approach with the Coupled Level Set Volume of Fluid (CLSVOF) ments to the hybrid approach which uses empirical correlations. In
and Sharp Interface Ghost Fluid methods to capture the spatial and the previous work based on a hybrid approach using empirical cor-
temporal evolution of the liquid-gas interface with high density relations (Arienti et al., 2006; Arienti and Soteriou, 2007; Arienti
ratio. Li and Soteriou (2016) used two meshing strategies: a uni- et al., 2011), the original TAB model (O’Rourke and Amsden, 1987)
form grid of 39 μm, corresponding to approximately 500 million was frequently used, droplets were assumed to be perfectly spher-
elements, and a dynamic adaptive mesh (AMR) with three levels ical, no analysis were made on two-way or four way coupling
of refinement, resulting in approximately 7 million elements, re- and the primary breakup coefficients were not analyzed. The orig-
fined only at the interface. In order to detect the gas-liquid inter- inal TAB model can be proven not to conserve droplet momentum
face at each timestep and then perform the dynamic mesh adap- and kinetic energy. In this sense, the newly proposed secondary
tation, a numerical approach was developed to change the way breakup model developed by Dahms and Oefelein (2016), herein
the liquid phase is described in dilute regions from Eulerian level- named Accurate Balance TAB method (AB-TAB), is used and com-
set/volume-of-fluid. Domain regions that do not meet the crite- pared with the TAB model; droplet distortion is then accounted
ria for the existence of a liquid-gas interface are then coarsened, for during the secondary breakup; two primary breakup coeffi-
considerably reducing the mesh count. 50 0 0 and 24 cores of a cients were investigated, considering the prediction of mass frac-
supercomputer were used for the uniform grid and AMR simula- tion distribution and velocity; and two-way coupling between dis-
tions, respectively. The authors recognized that uniform grid sim- crete and continuous phases and droplet-to-droplet collisions are
ulations not only present high processing cost, but also pose sig- also accounted for in the present work. The comparisons with ex-
nificant challenges in storing and post-processing the large set of periments demonstrate that the improved hybrid model is a feasi-
data. Spray simulations driven by mechanisms rather other than ble, low-cost approach for LJIC.
LJIC can be found using Euler-Euler Lagrange methods (Shinjo and
Umemura, 2011; Chen et al., 2013; Arienti and Sussman, 2015). In
such works, the simulated region is usually in a denser region at 2. Setup description
the proximity of the jet.
Recently, statistical learning methods have been used as an The LJIC cases analyzed in this work consist of a water jet per-
alternative to computationally intensive methods based on the pendicularly interacting with air crossflow, as evaluated experi-
highly accurate solution of conservation equations. Such methods mentally by Deepe (2006). This interaction breaks the liquid jet
are validated with the aid of high-fidelity simulation/experimental into drops or ligaments, which in turn break into smaller droplets,
database, which results in considerable reduction in computational creating a liquid spray. The velocity and occurrence (which means
time (Ma et al., 2015, 2016; Ling et al., 2018; Tryggvason et al., the number of droplets within a diameter range crossing a sam-
2016). pling plane) of the droplets in a specific distance from the liquid
Alternatively, the hybrid approach uses both Euler-Lagrange and jet are important information for spray designs. Deepe (2006) ob-
Euler-Euler concepts. In LJIC simulations, every segregated parcel tained the droplet velocity and their occurrence in specific trans-
of liquid is modeled as a Lagrangian particle, and both gas and verse planes using Mie scattering and PDPA (Phase Doppler Par-
liquid flows are simulated with the Euler-Euler approach. The hy- ticle Analysis) measurements for two LJIC conditions. Table 1
100 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Fig. 1. Physical domain for LJIC cases.

Table 1 3. Mathematical description


Physical parameters for LJIC cases.

Case vjet vg Wejet Weg Rejet Reg J The mathematical description is composed of the following
[m/s] [m/s] [−] [−] [−] [−] [−] parts: the Eulerian and Lagrangian frameworks, in which the so-
C1 4.26 50 66 11 1 × 103 8.6 × 104 6 lution of the fluid flow PDEs (Partial Differential Equations) and
C2 13.3 110 641 53 3.4 × 103 1.9 × 105 12 the droplets transport ODE (Ordinary Differential Equations) are re-
spectively explained. The empirical models required to convert the
fluid representation from one framework to another are also pre-
sented.
Table 2
Fluids physical properties.

Property Air Water 3.1. Eulerian framework


ρ [kg/m ]
3
1.2 998
μ [Pa · s] 1.8 × 10−5 1.0 × 10−3 This section presents all Eulerian transport equations in their
σ [N/m] 0.0713 integral form using index notation. These equations were dis-
cretized by the Finite Volume Method (FVM) (Ferziger and Perić,
2002).
The time-averaged mass conservation and momentum equa-
tions are expressed by Eqs. (1) and (2), respectively. This set of
shows the main features of the two cases, C1 and C2, studied by equations is known as the Unsteady Reynolds Averaged Navier-
Deepe (2006) that are analyzed numerically in this work. Stokes (URANS), Reynolds (1895),
In this table, vjet is the liquid jet area-averaged velocity entering 
the air duct; vg is the air area-averaged velocity at the air duct en- ui ni · ds = 0, (1)
ρl v2j d j ρg v2g hduct
sc

trance; W e jet = and W eg = represent the Weber


σ σ
numbers based on liquid jet flow and the air flow, respectively;    

ρl v j d j ρg vg hduct ρ ui d v + ρ ui u j · n j ds = − pδi j · n j ds + ρ gi dv
Re jet = and Reg = represent the Reynolds num- ∂t cv cv
μl μg   
cs
cs 
bers based on the liquid jet flow and the air flow, respectively; and ∂ ui ∂ u
μ − ρ ui uj · n j ds
j
+ +
ρl v2j cs ∂ x j ∂ xi
J= represents the momentum flux ratio.  
ρg v2g
The physical domain used for numerical analyses, Fig. 1, con-
+ ρ fα dv + ρ sui dv + fst , (2)
cv cv
sists of a constrained region of the experimental setup. The noz-
zle center, d j = 0.259 mm, is located at 9.52 mm from the air in- where: cs and cv stand for the control surface and control volume;
flow and at 14.2 mm from the lateral walls of the duct. The nozzle δ ij is the Kronecker delta; ui are the velocity vector components;
length is 2.38 mm, which is short to prevent any turbulence on the gi are the gravity vector components; μ is the fluid dynamic vis-
flow, according to the experimental setup (Deepe, 2006). Because cosity; ρ is the fluid density; ui uj is the tensor of the fluctuating
the liquid turbulence is known to affect the atomization process, velocities, which is modeled in this work with the two-layer k − 
the jet flow was kept laminar to allow the investigation of effects turbulence closure model (Launder and Spalding, 1974); fst is the
n
of flux momentum ratio and Weber number. The duct height and surface tension source term; sui = − mdp (Fdi ) is the source term that
length are 25.4 mm and 54.5 mm, respectively. represents the exchange of momentum, related to the drag force
Table 2 depicts the physical properties of water and air: den- (described bellow), Fdi , on droplets and the number density of a
sity, ρ ; dynamic viscosity, μ; and the surface tension coefficient, drop parcel, nd ; and fα is the source term related to the change in
σ . These physical properties were used to obtain the physical con- the momentum when an Eulerian liquid portion is converted to a
ditions presented in Table 1. Lagrangian particle.
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 101

Different RANS turbulence models may affect features of fluid only at the liquid-gas interface. However, from a numerical point
flow in some configurations as shown by Dhakal et al. (2014) that of view, this effect must be mathematically modeled as continu-
evaluated turbulence modeling of a single-phase flow cyclone gasi- ous. The Continuum Surface Force model (Brackbill et al., 1992)
fier. For jet in crossflow there exist many RANS turbulence models was used to describe the source term due to the surface tension
(Karvinen and Ahlstedt, 2005; Lefantzi et al., 2013; Arunajatesan force as a continuous force in the momentum equation, Eq. (2).
and Laboratories, 2012), however, achieving convergence at high- Thus, in the VOF method, this source term is related to the vol-
density ratio revealed to be a formidable task for k −  alternatives, ume fraction gradient, the interface curvature, κ , and the surface
like Reynolds Stress Model (RSM). The Shear Stress Transport (SST) tension coefficient of the phases pair, σ , according to the Eq. (8) in
was also experimented with, but the results for size distribution the integral form,
and droplet velocity were indistinguishable from those obtained 
with the k −  model. Therefore, the k −  model has been chosen fst = σ κ α n j ds. (8)
sc
as a compromise between accuracy and robustness.
Furthermore, the work of Karvinen and Ahlstedt (2005) showed The interface curvature can be obtained from the divergent of
that the results of standard k −  model were the closest to the the unit vector of the interface (Ubbink, 1997), which is based on
measured values, whereas results of the SST were also quite satis- the gradient of the volume fraction,
factory. The RNG k −  model and the standard k − ω model gave  
∇α
nonphysical velocity profiles. κ = −∇ · . (9)
Turbulence source/sink terms for two-way coupling were not |∇α|
considered in the present work. Strasser (2007) performed prelim-
Computing curvature accurately still remains a challenge in un-
inary test using a two-equation turbulence model to estimate the
structured meshes. In this work, we compute the volume fraction
influence of these terms for the current separator flow scenario.
derivatives, which are necessary for calculating the interface cur-
The results with and without the source terms were indistinguish-
vature, based on the second-order node-averaged Gauss method
able.
described in Maric et al. (2013). Since the volume fraction is dis-
The two-phase flow is solved as one fluid, whose physical prop-
continuous across the interface, its derivatives will be noisy. In
erties depend on the local volume fraction, following the concept
order to avoid this effect, the volume fraction is smoothed by a
of the VOF method (Ubbink, 1997). The volume fraction is de-
nodal smoother prior to the derivatives calculations. This ensures
fined in the range 0 ≤ α ≤ 1 and expresses the ratio of liquid vol-
Vl smooth derivatives. The method can be proven to be second-order
ume to the total volume, α = . This variable indicates which in the classical static droplet problem. Parasitic currents inevitably
Vl + Vg
phase is present in each numerical cell: fully liquid phase, α = 1, appear, but their magnitude is not relevant to the velocity scales
fully gas phase, α = 0. Fluid density, Eq. (3), and dynamic viscos- involved in this work.
ity, Eq. (4), in the mass conservation and momentum equations are Some assumptions were made regarding the numerical model-
weighted according to the volume fraction, ing of the flow on the Eulerian framework:

ρ = (1 − α )ρg + αρl , (3) • Both fluids are treated under the continuum hypothesis;
• The flow equations are averaged in time - RANS approach;
• Fluid phases are considered immiscible;
μ = (1 − α )μg + αμl , (4) • No thermal effects are accounted;
where the subscripts g and l indicate gas and liquid phases, respec- • Physical properties are considered constant.
tively.
In an LES context, Liovic and Lakehal (2007) showed that by fil-
The volume fraction transport equation, Eq. (5), is obtained
tering the single-fluid momentum equations, not only the subgrid
from the continuity equation for an individual phase.
tensor appears. Because of the surface tension force, two additional
  
∂ terms, the interphase net force between the two fluids at the sub-
α dv + α u j · n j ds = sα d v, (5)
∂t cv cs cv grid level and a non-linearity error, are produced. The latter repre-
sents the non-resolved counterpart of surface tension. Such terms
where sα is the source/sink term related to the Euler/Lagrange
are difficult to characterize, not only because of the lack of a crit-
frameworks exchange. It is calculated as the ratio of the liquid vol-
ical scale, but also because the resolved surface tension force it-
ume to the cell volume multiplied by the time step (Arienti et al.,
self is hardly computed with sufficient accuracy. In the context of
2011)
RANS, analogous terms arise in the averaging procedure. As in the
Vl α work of Liovic and Lakehal (2007), such terms are not modeled
sα = − =− , (6)
V t t for the same reasons. Their effect is undesirably embedded in the
and it assumes a positive sign if a Lagrangian droplet is converted modeling, but fortunately limited to the liquid-gas interface.
to the Eulerian framework. The volume fraction is converted into The numerical mesh created to simulate the LJIC cases was
Lagrangian particles from a specific column height according to the composed of hexahedra, with near wall refinement in order to
empirical models (Wu et al., 1997; Sallam et al., 2003), so that no match the dimensionless wall distance y+ ≈ 1. Since the liquid-
marking cells algorithm is needed. gas interface must be well captured, high refinement was applied
The source/sink term related to the Euler/Lagrange frameworks also in the vicinity of the liquid jet. In order to obtain grid inde-
exchange in the time-averaged momentum equation, Eq. (2), is pendence, several mesh tests were performed, mainly monitoring
droplet velocity. Additionally, all meshes displayed orthogonality
Vlvd αv
fα = − = − d, (7) above 0.5 for all elements.
V t t Regarding the boundary conditions, symmetry was applied to
where vd is the drop velocity vector. It considers that an amount both lateral surfaces of the simulated domain; the no-slip condi-
of momentum is lost by the continuous phase at the Euler to La- tion was considered for the liquid and gas phases at the bottom,
grange conversion. top and tube walls; velocity profiles were imposed at the air and
The interactions at the interface are modeled through the sur- jet inlets; the static pressure was prescribed at the duct exit. Be-
face tension force, which is physically discontinuous and defined cause of the long section upstream of the test section used in the
102 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Fig. 2. Two possible kinds of droplets interaction.

experiments, the air inlet is fully turbulent. For the air inlet, a tur- Regarding boundary condition, droplets are allowed to escape
bulent velocity profile with turbulent length of 2.54 mm and tur- when hitting the walls, since spray-wall interaction and liquid film
bulence intensity of 3% was imposed, according to the power law, transport model are not considered in this work. Actually, the
Eq. (10), choice of droplet boundary condition was seen to be irrelevant in
 1/7 this work, since there is no film formation in the experiments con-
dw sidered.
δT > δw → u = u∞ ,
δT Droplet velocity and position are obtained from the solution of
δT < δw → u = u∞ , (10) the ordinary differential equations, Eqs. (12) and (13),
du pi
where dw is the distance to the nearest wall and δ T is the turbulent mp = Fdi + Fw,bi , (12)
boundary layer thickness at the air inlet. The turbulent boundary
dt
layer thickness is estimated following a fully developed boundary dx pi
layer on a plate, = u pi , (13)
dt
δT = 0.37(ν /u∞ )0.2 L0T .8 , (11) where: the subscript p refers to discrete particle; u and x are,
respectively, the particle velocity and position; m is the droplet
assuming LT = 200 mm Arienti and Soteriou (2007). In Arienti and
mass; and subscript i indicates the three components of a vector.
Soteriou (2007), both steady and pulsed jet flow were investigated.
Only the drag and the combined buoyancy-weight forces, respec-
In the present work, only the steady case was simulated. For the
tively Eqs. (14) and (15),
water inlet, a uniform velocity profile with turbulent length of
0.032 mm and turbulence intensity of 3% was set at the jet noz- 3ρCD
Fdi = m p (ui,t − u pi ), (14)
zle entrance. 4ρ p d p
 
ρ
3.2. Lagrangian framework Fw,bi = 1− m p gi , (15)
ρp
Droplets created in the Eulerian to Lagrangian conversion are were considered in this work. Other forces such as lift, virtual mass
modeled as discrete parcels. The physical modeling of the droplet and Basset forces, are normally not relevant because of the high
motion follows the Newton’s second law. Thus, flow forces on liquid/air density ratio.
droplets are accounted by adding source terms to the particle In Eq. (14), ui,t = ui + ui is the instantaneous fluid velocity, com-
equation motion, Eq. (12). Droplets also may affect the air flow and posed by the averaged and fluctuating velocities. The averaged
interact with other droplets. component is interpolated from the Eulerian mesh to the par-
Two-way particle coupling considers that droplets affect the air ticle position, and the fluctuating component is obtained from
flow, besides the effect of the air flow on droplets. It is modeled by the Langevin dispersion model proposed by Sommerfeld (2001).
the source term sui on the momentum equation, Eq. (2). The inter- The drag coefficient, CD , is based on the correlation of Feng and
action between droplets is usually relevant in dense spray regions, Michaelides (2001), which gives, Eq. (16),
i.e. high droplet concentration, whereas in dilute spray regions, i.e.  4
low droplet concentration, this effect can be neglected. 10 0 0 ≥ Re p > 5 → CD = 17.0Re−2 /3
Considering binary collision, droplets interaction can be of two
p
λ+2
 
kinds: grazing collision or coalescence, Fig. 2. In grazing collision, 24
0.687
λ − 2
+ 1 + 0.15Re p
droplets keep their size but change their velocity. In coalescence, Re p λ+2
two droplets become a bigger droplet with modified velocity. Coa-     
lescence happens whenever surface tension dominates over liquid 8.0 3λ + 2 3λ + 2
0 ≤ Re p ≤ 5 → CD = 1 + 0.05 Re p
flow inertia (Reitz, 2006). The droplet Weber number represents Re p λ + 1 λ+1
the ratio of the liquid inertia and the surface tension forces. Thus,  
3λ + 2
the interaction of droplets with high Weber number generates − 0.01 Re p ln(Re p ). (16)
grazing collision instead of coalescence. The outcome of droplet λ+1
collision was modeled by the stochastic model of O’Rourke (1981).
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 103

The present work used this correlation aiming at verifying the


influence of the viscous flow inside the droplets and their distor-
tion on spray formation in LJIC. Strictly speaking, this internal flow
is important, although the use of simpler correlations produced
similar results for the droplet velocity. The droplet Reynolds num-
ber is calculated according to
ρ|ui − u pi |d p
Re p = . (17)
μ
The interactions between droplets were evaluated considering
grazing collision and coalescence. The collision occurrence and its
outcome were modeled by the stochastic method, O’Rourke (1981).
This method considers that a collision can occur only with pairs
of parcels in the same cell. A parcel represents a number of real
droplets with the same volume and velocity. In the current hybrid
method, parcels are initially generated by primary breakup at each
timestep. Since the timestep is rather low, the liquid volume con-
verted to parcels is also very small. In most simulations, this pri- Fig. 3. Schematic representation of the Taylor Analogy Breakup.
mary parcel was made up of less than one droplet. When collision
is detected, its type is determined comparing the collision impact Table 3
The analogy between a forced mass-spring-damper system and an oscillating
parameter,
√ and distorting droplet.
b = ( r1 + r2 ) Y , (18)
Forced mass-spring-damper Oscillating and distorting Mathematical
with the critical collision impact parameter, system droplet equivalence


 Spring force Surface tension force k
= Ck ρσr3
2.4 rr3 − 2.4rr2 + 2.7rr
m l p

= Cd ρμrl2
= (r1 + r2 )min 1.0,
c
Damping force Viscous force
bcrit . (19) m l p
ρ u2
W ec External force Drag force F
m
= CF ρg r p
l

Where Y is a random number between 0 and 1, r1 and r2 are


respectively the radius of the biggest and smallest parcel, and Wec The diameter of the droplets generated at the column breakup
is the collisional Weber number, is estimated by the Eq. (22),
ρUrel
2

(20) dp 1/3
W ec = . = (1.5λb ) , (22)
σ d jet
Urel is the relative velocity between the parcels and D̄ is their
where λb = 16.3W e−0.79 is the dimensionless wavelength
mean diameter. Film draining between two coalescing particles is
(Sallam et al., 2003).
neglected in this work.
The shear breakup follows the model described by
3.3. Euler-Lagrange conversion Arienti et al. (2011). In this mechanism, some droplets are stripped
from the liquid column surface every time the aerodynamic forces
Primary breakup of the simulated LJIC cases consists of two exceed the cohesion forces; the criterion is
mechanisms: column and shear breakup. In both mechanisms, nu- σ
ρg |Ug − Ul |κ > Cσ , (23)
merical criteria are monitored in the Eulerian framework in or- dp
der to update the mass and momentum source terms of transport  g and U
 l are the spatial averaged velocity around the con-
where U
equations. From the Lagrangian perspective, new droplets are cre-
trol volume where the droplet is released, when the above crite-
ated whenever the breakup criteria are met. Droplet position and
rion is met, respectively for gas and liquid phases; κ is the inter-
velocity are calculated following empirical correlations and per-
face curvature; Cσ = 2.0 is a constant accounting for the nonideal-
forming mass and momentum balance. It is emphasized that, un-
ity of this process; σ is the surface tension coefficient; and dp is
like high-fidelity simulations, correlations are used instead to esti-
the droplet diameter,
mate the size of a droplet breaking off of the liquid jet.
 1/2
Wu et al. (1997) and Sallam et al. (2003) experimentally stud- dp νl y
ied the column breakup height, yb , and found it is a function of = 3.36 , y/yc ≤ 1, (24)
d jet v jet d2jet
momentum ratio, J, and jet diameter, dj ,
yb  dp
= Cb J. (21) = 0.132, y/yc > 1. (25)
dj d jet
Wu et al. (1997) and Sallam et al. (2003) obtained different values ν l is the liquid kinematic viscosity, vjet is the liquid velocity
for the empirical coefficient, respectively Cb = 3.44 and Cb = 2.6. at the nozzle inlet and yc /d jet = (0.001v jet d jet )/(νl ). The values of
According to Sallam et al. (2003), their smallest coefficient was these constants were obtained from Sallam et al. (2003).
probably due to improvements in the ability to locate the end of In particular, no significant shear breakup was observed in all
the liquid column than that available for Wu et al. (1997). As can the simulations carried out in this work. This is not surprising as
be concluded from the above correlation, the location for the re- both cases are dominated by column breakup.
lease of droplets due to column breakup changes with the feed
conditions, specifically with the momentum flux ratio. From an im- 3.4. Secondary breakup
plementation point of view, the domain is searched along the ver-
tical direction so that when the cell at yb is found, a droplet is The secondary breakup was modeled with both the original
produced. This procedure is repeated every timestep. Taylor Analogy Breakup model (TAB) (O’Rourke and Amsden, 1987)
104 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

and the Accurate Balance equation Taylor Analogy Breakup model Breakup is triggered when the dimensionless droplet distortion,
(AB-TAB) proposed by Dahms and Oefelein (2016). The AB-TAB y, exceeds unity. However, in practice a breakup time measurement
model ensures conservation of momentum (linear and angular) is more convenient to control this event. Therefore, an expression
and kinetic energy during breakup; also, it properly accounts for for the breakup time, tB , can be obtained considering y = 1,
droplet non-sphericity effects upon secondary breakup.
1
  1 − We 
−φ ,
crit
In the original TAB model (O’Rourke and Amsden, 1987), the tB = cos (35)
breakup of an oscillating and distorting droplet is associated with ω A
a forced mass-spring-damper system, Fig. 3, where parameters A and φ are
2
d x c dx k F
  2
+ + x= , (26) 1 dy
dt 2 m dt m m A= (y − W ecrit )2 + , (36)
ω dt
where m is the mass of the system, k is the spring stiffness, c is
the damping coefficient, F is the external force applied to the sys- and
tem, and x is system shift correspondent to the droplet pole dis-  
placement. Table 3 shows all correspondences of the forced mass- 1 dy
φ = arctg . (37)
spring-damper variables to the oscillating and distorting droplet. In ω (y − W ecrit ) dt
Table 3, rp is the droplet radius; u is the relative velocity between
the droplet and the surrounding gas; CF , Ck and Cd are empirical The diameter and velocity of the droplets created at the sec-
dimensionless constants, whose values are 1/3, 8 and 5, respec- ondary breakup are calculated considering the energy conservation
tively. during breakup, as presented in the work of O’Rourke and Ams-
The dimensionless, time-dependent distortion equation can be den (1987).
written as The AB-TAB model, developed by Dahms and Oefelein (2016),
  follows the main concepts of the original TAB model. However,
t a refined balance of mass, momentum, energy and surface en-
y(t ) = W ecrit + exp − [(y0 − W ecrit ) cos(ω t )
tD ergy was assured in the calculation of secondary droplets diam-
  eter and velocity, in addition to the energy balance performed by
1 dy0 y0 − W ecrit
+ + sin(ω t )], (27) O’Rourke and Amsden (1987). Therefore, the breakup time is more
ω dt tD realistically estimated, since the oscillation energy is properly ac-
x counted, not requiring the use of the empirical value of the crit-
using y = as dimensionless distortion and Cb = 0.5.
Cb r p ical Weber number (Dahms and Oefelein, 2016). Other secondary
The critical Weber number, Eq. (28), is proportional to the breakup models, for instance the Enhanced Taylor Analogy Breakup
droplet effective Weber number, We∗ , which in turn is calculated (ETAB) (Tanner, 1997), had already been developed aiming at solv-
based on the Weber, We, and Ohnesorge, Oh, numbers of the ing some under-predictions of droplet diameter and some incon-
droplet, sistencies on the spray angle prediction of the original TAB model.
Nevertheless, empirical constants were also considered without a
CF
W ecrit = W e, (28) rigorous physical derivation. Therefore, the AB-TAB model appears
CkCb to be physically more consistent to describe the secondary breakup
than the original TAB model and similar models that make use of
We
W e∗ = , (29) empirical constants.
1 + 1.077Oh1.6

ρg u 2 r p (30)
4. Numerical methods
We = ,
σ
The in-house numerical code UNSCYFL3D was used in this
μ work. It has been extensively validated for turbulent, single-phase
Oh =  l . (31)
ρl d p σ and particle flow simulations (De Souza et al., 2012, 2014; Duarte
et al., 2015, 2017), and more recently for interfacial flow simula-
The oscillating damping time, tD , and the droplet oscillating fre- tions using the VOF method (Fontes et al., 2018). In particular, be-
quency, ω, are given by cause of number of submodels involved in the simulation of the
2ρl rd2 jet in crossflow, validations/verifications of individual elements has
tD = , (32) been carried in order to ensure that results are reproducible with
Cd μl
other codes using the same models. The implementations of cor-
and relations for primary breakup (droplet diameter and velocity) have
 been tested, as well as the liquid mass conservation during pri-
σ 1
ω= Ck − . (33) mary breakup. Furthermore, the conservation of mass, momentum
ρl r3p tD2 and kinetic energy during the secondary breakup have also been
The distortion velocity (O’Rourke and Amsden, 1987) is ex- confirmed for the AB-TAB model by simple simulations involving a
pressed as prescribed base flow and a number of parcels.
Fig. 4 shows an overview of the numerical methods applied to
dy W ecrit − y solve liquid jet in crossflow based on the hybrid approach devel-
=
dt tD oped in UNSCYFL3D.
   
t 1 dy0 y0 − W ecrit As shown in the above flowchart, even if no primary breakup
+ exp − + cos(ω t ) occurs during the current timestep, there may be droplets in-
tD ω dt tD
 side the domain, which must be transported and may undergo
secondary breakup. Regarding the numerics, the Finite Volume
− (y0 − W ecrit ) sin(ω t ) . (34)
method (Ferziger and Perić, 2002) is used to discretize the par-
tial differential equations on unstructured grids. The second-order
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 105

Fig. 4. Flow chart of numerical methods.

upwind scheme is used for the advective term in the momen-


tum equations, while for temporal discretization, the three-time-
level is employed. The SIMPLE method was applied in the velocity
and pressure coupling. For turbulence closure, the two-layer k − 
model (Ferziger and Perić, 2002) was used, so that the viscous sub-
layer is resolved. In order to improve the predictions in the near-
wall region, the standard law of wall is used. No turbulence source
terms due to the droplets are considered as preliminary simula-
tions showed them to be negligible.
The discretization of the advection term of the VOF equation,
Eq. (5), was performed by the modified High Resolution Inter-
face Capturing scheme (HRIC), which is a TVD scheme suitable for
sharp interfaces. The Momentum Interpolation Method, required
to avoid unphysical pressure fluctuations on collocated grid, was
implemented based on the works of Denner (2013) and Rhie and
Chow (1983).
The Particle Tracking Method proposed by Haselbacher et al.
(2007) has proven to be suitable for tracking particles on unstruc-
tured grids, and was applied in this work.
The timesteps were 1 and 0.1 microseconds for C1 and C2, re-
spectively. The maximum Courant numbers were lower than 2 for
both cases, also in order to maintain stability. Convergence within
each timestep is declared when the normalized residuals for all
linear systems, i.e., velocity components, volume fraction and tur-
bulence variables, reaches 0.0 0 01. For C1, typically 1 or 2 inner it-
erations were needed for convergence within each timestep. For
C2, more iterations were necessary, 7–10, mostly.
In this work, the most LJIC simulations took about ten days us-
ing a serial Intel® i7-4790k processor with 8 GB of RAM.

5. Results and discussions

In this section, the results of LJIC simulations using the im-


proved hybrid approach are presented and discussed. More specif-
ically, the following issues are addressed: strategy for domain re-
duction; LJIC characteristics; evaluation of different methods and
parameters to solve the spray formation in LJIC.

5.1. Strategy for domain reduction

LJIC simulations usually require mesh resolutions of several mil-


lion cells in order to properly represent the liquid jet interface,
hence a strategy to relax such requirement is always useful. In
this work, the symmetry assumption along the tunnel longitudinal
Fig. 5. Grid and refinement zone near jet. plane, x − plane, was evaluated based on C1, Table 1. Fig. 5 shows
the general view and the mesh refinement on the x − plane for two
analyzed domains: half domain and full domain. Cells with high
aspect ratio can be seen, which are known to deteriorate gradient
106 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Fig. 6. Comparison of volume fraction field at x − plane and droplet size distribution and velocity at z − plane for both analyzed domains.

calculations. Thus the maximum aspect ratio, nearly 200, was al- 5.2. LJIC Characteristics using hybrid approach
lowed only in regions where the flow was fully developed. In prin-
ciple, such strategy should not be sensitive to the non-dimensional In the hybrid approach, the interaction between the air cross-
parameters, even at higher Weber numbers, specifically when deal- flow and the liquid jet is solved through the VOF method, while
ing with RANS models. Because of the natural symmetry of the the droplets transport in the air-flow is solved as discrete particles.
configuration, in average terms, no flow asymmetry would be ex- Since the general features of the two-phase flow using the hy-
pected. This does not hold for the for LES/DNS simulations, as brid approach are quite similar for different methods and experi-
the symmetry boundary condition cannot be applied to the in- mental parameters, the flow characteristics for C1 and C2, Table 1,
stantaneous flow. The use of a symmetric plane reduces the num- are only shown for: primary breakup coefficient Cb = 3.44; original
ber of elements approximately by half. For C1, mesh resolutions TAB method for secondary breakup; one-way coupling.
of ≈ 640 0 0 0 elements for full domain and of ≈ 3110 0 0 elements The presence of the liquid jet in crossflow changes the gas ve-
for half domain were enough to describe properly the LJIC flow. In locity field, Fig. 7, creating a downstream recirculation, while the
both domains, the smallest elements have characteristic length of liquid jet is curved by the flow towards its main direction. Fig. 8
about 0.03 mm. shows the streamlines of the air interacting with the liquid jet.
Fig. 6 shows the similarity between the numerical results of full They are colored by the perpendicular component, y, of the gas
and half domains. The volume fraction, α , is insensitive to that vorticity, and the liquid jet column is represented by the isocon-
simplification, mainly because this is a RANS simulation. If a LES tour α = 0.5. The recirculation structures presented in C2 are more
simulation were used instead, the liquid column would be time complex than in C1, because the air velocity and liquid column
oscillating and the symmetry assumption would be inappropriate. height are higher in C2 than C1. In both cases this structure af-
The normalized mass fraction and the droplet velocity were ob- fects the droplet secondary breakup, since the drops released from
tained at the z − plane located at 3.81 cm from the liquid jet exit the top of the column jet may be carried to this recirculation re-
for both domains. Clearly, normalized mass fraction and velocity gion, changing the drop Weber number that is used to calculate
are rather insensitive to the assumption of symmetric flow. There- the criteria for secondary breakup. It is worth mentioning that the
fore, the strategy for domain reduction considering symmetric flow counter-rotating vortex, clearly identified in both cases, cannot be
is validated and will be used throughout this work. Yet, it is impor- represented numerically by the classical Euler-Lagrange approach,
tant to highlight that atomization is always unsteady. More accu- since the liquid column is usually represented by blobs. Some more
rate turbulence models might trigger the natural unsteady behav- recent, specific Lagrangean approaches may produce clusters large
ior. enough to bend the flow.
Following those assessments, a mesh with about 2 million cells The primary breakup represents the conversion of the Eule-
was created based on the half domain. This resolution produced rian approach (VOF method) to the Lagrangian approach (particle
grid-converged solutions for C2, which has higher Weber and mo- tracking). Fig. 9 shows the general view of the first and secondary
mentum ratio numbers. breakups: the droplets, created at the top of the liquid jet, have
large diameter and low velocity; these droplets are broken into
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 107

These two cases were selected because of the availability of a doc-


umented experimental database. Investigations at higher momen-
tum flux ratios, i.e., with relevant surface breakup, will be carried
out in the future to test the model generality. The shear breakup
mechanism was included in all simulations, and the results were
not sensitive to it, as expected. In principle, as long as all the
mechanisms for primary droplet generation are accurately mod-
eled, the model is to be generic.
The droplet spatial distribution in the two cases are shown in
Fig. 9. Since the droplets created at the top of the liquid jet have
approximately the same velocity as the liquid jet, considering the
curvature of the liquid jets in C1 and C2, it was expected that the
release velocity of the droplets in C1 would be more perpendicular
to the air flow than in C2, in which the liquid jet is more aligned
to the air flow. Besides, the air velocity in C2 changes the droplet
direction faster than the air velocity in C1, since the bigger droplets
(higher inertia) in both cases have diameters in the same order
(10−4 m).
More droplets are generated by the secondary breakup in C2
than in C1. Due to the fact of the air velocity in C2 is higher than in
C1, the shear stress on droplets surfaces is likewise higher, surpass-
ing the surface tension forces more times than in C1. From a nu-
merical point of view, the critical Weber number is achieved more
frequently in C2 than in C1, enhancing the secondary breakup.
In both cases, the droplet velocity increases as the droplets
move away from the liquid jet nozzle, as observed in Fig. 9b and
d. As expected, the small droplets tend to achieve the equilibrium
with the air velocity more rapidly than larger droplets, due to their
low relaxation time. This fact can be verified in Fig. 6(d), where
higher velocities can be observed for small droplets at the same
cross-section sample plane.

5.3. Evaluation of primary breakup coefficients and secondary


breakup models

This section examines the effects of case, primary breakup co-


efficient and secondary breakup model over the numerical to ex-
Fig. 7. Streamwise velocity component. perimental agreement. The following combinations are tested for
both cases, C1 and C2 (Factor A): two primary breakup coefficients,
Cb = 2.6 (Sallam et al., 2003) and Cb = 3.44 (Wu et al., 1997) (Fac-
other droplets due to the prevalence of aerodynamics forces over tor B); two secondary breakup models, TAB O’Rourke and Ams-
surface tension forces on their surfaces; the secondary droplets, in den (1987) and AB-TAB (Dahms and Oefelein, 2016) (Factor C).
turn, have small diameter and are accelerated by the gas flow. In the present work, a 2k=3 factorial design with three factors
According to the operating conditions depicted in Table 1 and is performed with no replicate (n = 1) aiming at analyzing their ef-
Fig. 10, C1 and C2 are classified as column breakup regime. There- fects on the l2 norm of numerical and experimental mass fraction
fore, in both cases the primary breakup can be modeled consid- distribution difference,

ering only the column breakup, without breakup along the lat- 
2
eral surface of the jet column, known as surface/shear breakup. l2 = m∗num − m∗exp , (38)

Fig. 8. Streamlines of air flow around the liquid jet.


108 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Fig. 9. General view of the first and secondary breakups.

Fig. 11. Sampling plane at 3.81 cm, red line. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 10. Map of momentum flux ratio versus Weber number for the primary
breakup of the LJIC, adapted from Wu et al. (1997).

where m∗ is the normalized mass fraction that represents the


droplet size pdf.
Factor A is quantitative and related to operation condition; Fig. 12. Geometric view of a 2k=3 factorial design (Montgomery, 2013).
since its levels differ significantly, the simulation responses may be
nonmonotonic functions. Factor B and Factor C are categorical and
related to empirical and mathematical models, respectively; thus, Regarding droplet velocity, it is temporally averaged for each
monotonic functions are expected. droplet diameter, and achieves a statistically steady state. In order
The spatial and temporal integration methods, as well as the to obtain statistically converged results, typically five or six droplet
time step and grid size, were held constant for all simulations to residence time are used. The residence time is calculated based on
isolate the chosen factors. an estimate for the slowest droplet and the domain length. For all
According to the experiment carried out by Deepe (2006), a cases, longer sampling did not lead to changes in the average and
transverse sampling plane 3.81 cm downstream the liquid jet noz- RMS results.
zle is defined to obtain the mass fraction distribution and droplet The applied factorial design leads to 8 simulation combinations,
velocity, Fig. 11. i.e., 8 values of the l2 norm, which are geometrically represented
The normalized mass fraction of the droplet size distribution by the vertices of a cube, Fig. 12. Tables 4 and 5 show these values
on the sampling plane is obtained by the following procedure: the for the right side face (C2) and left side face (C1) of the cube.
droplet diameter is divided into bins of range 3 μm; the mass of In Fig. 12, the symbols (1), a, b, c, ab, ac, bc, abc represent the
the droplets whose diameter lies in a specific bin is accumulated; total of the l2 norm at all n replicates taken at the simulation com-
and the final result is normalized by the total mass. bination; in this work n = 1.
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 109

Table 4 worse at C2. Besides, just for C1, there is a significant interaction
l2 − norm of the mass fraction distribution for C1.
effect, which is represented by the deviation from the dashed, par-
Secondary breakup Cb l2 allel lines.
TAB model 2.6 3.78 × 10−2 The conducted analyses restrict the factors effect on the set of
AB-TAB model 2.6 3.01 × 10−2 models, numerical parameters, and initial and boundary conditions
TAB model 3.44 3.84 × 10−2 applied in the presented simulations. Also, one cannot interpret
AB-TAB model 3.44 3.57 × 10−2 them aside from errors associated with temporal and spatial in-
tegration schemes; grid and time step dependence.
Table 5 The following sections show through tables, graphs, and figures
l2 − norm of droplet size distribution for C2. the data and results cited above, with a focus on mass fraction dis-
Secondary breakup Cb l2 tribution and velocity analyses. Also, apart from the design of ex-
periment, these sections also address the tendency of numerical
TAB model 2.6 13.97 × 10−2
agreement with the experimental data for Cb = 3.44 rather than
AB-TAB model 2.6 10.21 × 10−2
TAB model 3.44 9.80 × 10−2 Cb = 2.6 and AB-TAB rather than TAB on physical and numerical
AB-TAB model 3.44 6.56 × 10−2 perspectives.

5.3.1. Droplet size distribution


Fig. 14 shows the normalized mass fraction distribution for C1
using combinations of the two column breakup coefficients and
the two secondary breakup models. The frequency of droplets
crossing the sampling plane is predominantly influenced by the
secondary breakup model, with little influence of the column
height (Cb ). The AB-TAB model presents an overall good prediction
of the droplet diameters, whereas the peak of the mass fraction
distribution is shifted to the left when the original TAB model is
used. Despite the AB-TAB appears to better predict the peak diam-
Fig. 13. Main and interaction effects of the primary breakup coefficient (blue: Cb = eter in terms of mass distribution, the prediction for intermediate
2.6, red: Cb = 3.44) and secondary breakup model (TAB and AB-TAB) expressed by
diameters (40 μm < Dp < 90 μm) is slightly better when the original
the l2 norm. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.) TAB model is used.
The l2 − norm, represented by Eq. (38), is a way to quantita-
tively assess the differences of mass fraction distribution for the
The effect of Factor A is computed as different column breakup coefficients and secondary breakup mod-
els. The l2 − norm expresses a cumulative difference between the
1
F actor A = [a + ab + ac + abc − (1 ) − b − c − bc], (39) numerical results and the experimental data for mass fraction dis-
4n tribution. Table 4 presents the deviations for simulations compared
i.e., the mean difference of the l2 norm for the simulations with to the experimental data. Minor differences on l2 − norm are seen
high and low factor values, C2 and C1 respectively. The same pro- for different primary and secondary breakup models. Thus, for low
cedure applies for Factors B and C. Weber numbers, such as in (C1), the droplet size distribution is
Since the evaluated l2 norm indicates how much the numeri- reasonably predicted by either column breakup coefficient or sec-
cal result disagrees with the experimental data, the factorial de- ondary breakup model analyzed in this study. However, the AB-TAB
sign shows that changing C1 to C2 decreases the numerical agree- model is yet more appropriate to solve the secondary breakup on
ment with the experimental data in 0.066. This value is the result LJIC, since the mass fraction distribution peak fitted well the exper-
of Eq. (39), which can be verified summing the four outcome val- imental data. This better prediction of the mass fraction distribu-
ues of Table 5, subtracting by the sum of the four outcome values tion may be related to the accurate conservation of linear momen-
of Table 4, and finally dividing by 4. tum, angular momentum and kinetic energy during the breakup,
Changing the primary breakup coefficient from 2.6 to 3.44 and provided by the AB-TAB model.
the secondary breakup model from TAB to AB-TAB increases the Fig. 15 shows the numerical results for the normalized mass
numerical agreement in 0.018 and 0.020, respectively. However, fraction distribution in C2. The distributions for all models simu-
these are the combined result considering the simulation setup lated are shifted to the left relative to the experiments. However,
ranging from C1 to C2: changing Factor B from 2.6 to 3.44 affects the simulation using the column breakup coefficient Cb = 3.44 and
the result at C1 configuration less and in a different manner than the AB-TAB model shows the best agreement among all simula-
at C2; the same applies to Factor C. To refine this statement, sep- tions for C2. In fact, the simulations using Cb = 3.44 improve the
arate analyses are presented onward to C1 and C2 through a 2k=2 prediction of the mass fraction distribution despite the secondary
factorial design with no replicate (n = 1), Fig. 13. breakup model chosen, as well as the simulations using AB-TAB
Factor A has the greatest effect over the l2 norm because the model improve the prediction of the mass fraction distribution de-
dimensionless parameters which define C1 and C2, i.e., gas-liquid spite the primary breakup coefficient.
momentum ratio, Reynolds and Weber numbers, differ signifi- In Table 5 the l2 − norm for C2 is shown, expressing the quanti-
cantly; while C1 is on the bag breakup regime, C2 is on a most tative differences between the models in the prediction of the size
severe, multi-mode regime composed by bag and shear breakups, distribution from a LJIC spray. Unlike C1, the secondary breakup
Fig. 10. model and the column breakup height coefficient affect the mass
Fig. 13 shows how different the effects of Factor B and C at C1 fraction distribution more significantly. Regarding case C2, the in-
and C2 are. Both cases present an improvement in the numerical stabilities that promote secondary breakup of the droplets are
agreement (i.e., a decrease of l2 norm) changing from TAB to AB- more susceptible to appear, since the air velocity is higher and
TAB. However, the primary breakup coefficient reports the oppo- consequently the relative velocity that a droplet experiences is also
site pattern, i.e., the coefficient Cb = 2.6 is the best at C1 but the higher. Therefore, extreme cases of spray formation in LJIC must
110 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Fig. 14. Droplet size distribution for C1 using different primary and secondary breakup models.

Fig. 15. Droplet size distribution for case C2 using different primary and secondary breakup models.

accurately account for the secondary breakup, as by the AB-TAB dicts better the primary breakup in the top of the liquid jet col-
model. umn. An explanation for this is that while in the original experi-
The column breakup coefficient for the column breakup height ments of Wu et al. (1997), the air inlet section was long enough
of LJIC is obtained by experimental visualizations (Wu et al., 1997; to ensure a fully turbulent boundary layer, in the experiments of
Sallam et al., 2003). In fact some differences in this value do exist. Sallam et al. (2003), special care was taken to ensure a uniform air
In the cases simulated in this work, the coefficient Cb = 3.44 pre- velocity profile upstream of the liquid jet. Consequently, the value
D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 111

The numerical results for droplet velocity using the different


primary and secondary breakup models are quite similar to each
other. Simulations using Cb = 3.44 showed a slight improvement
on the droplet velocity curve, that is closer to the experimen-
tal curve than by using Cb = 2.6. The simulations using differ-
ent secondary breakup models show nearly identical results for
the droplet velocity curves. The higher influence of the primary
breakup coefficient than the secondary breakup on the droplet ve-
locity was expected, since droplet velocity is basically related to
the drag force.
The release condition of the drop at the top of the liquid jet is
strongly dependent on the column breakup height. All numerical
results display a peak of droplet velocity for lower droplet diame-
ters, similarly to the experimental data. However, the peak of the
droplet velocity is better predicted when the column breakup co-
efficient Cb = 2.6 is used, regardless the secondary breakup model.
Since the liquid column jet using Cb = 2.6 is lower than that using
Cb = 3.44, a Lagrangian drop created from the primary breakup us-
ing Cb = 2.6 experiences an airflow more perpendicular than a La-
Fig. 16. Droplet velocity for C1 at the sampling plane 3.81 cm downstream of the grangian drop created from the primary breakup using Cb = 3.44,
liquid jet nozzle. generating smaller droplet diameters. Fig. 17 shows drops creation
from the Eulerian approach (VOF) for Cb = 2.6 and Cb = 3.44. The
obtained by Wu et al. (1997) for Cb was higher (3.44), and thus angle between liquid and gas flow paths is greater for Cb = 2.6
applicable to the test cases investigated in the current work. in the primary breakup region, than that using Cb = 3.44, as pre-
In order to provide a sound explanation for the droplet size dis- sented in Fig. 17. Therefore, the simulations for C1 show that the
tributions deviations, further investigation on the numerics must droplet velocity is not significantly influenced by different sec-
be carried. Because most of the elements in the method have been ondary breakup models, whereas the column breakup coefficient
individually validated, such as the Euler-Lagrange conversion, mass has a small effect on the droplet velocity for all diameters.
and momentum conservation, many questions still remain regard- Fig. 18 shows the droplet velocity for C2 using combinations of
ing the results of the turbulence model. As a next step in this re- the two secondary breakup models and the two primary breakup
search, other turbulence models, such as the recently optimized k- coefficients. For this higher Weber number case, C2, the numerical
epsilon model for jets in crossflows (Lefantzi et al., 2013) will be results of the droplet velocity depart more from the experimental
tested. data. The smaller droplets (Dp ≈ 10 μm) nearly reach the air veloc-
ity, while the larger droplets (Dp > 80 μm) are having their velocity
5.3.2. Droplet velocity decreased. This discrepancy between the numerical results and ex-
Fig. 16 shows the droplet velocity for C1 using combinations of periments suggest that the interaction of the air flow and the liq-
the two secondary breakup models and the two primary breakup uid droplets is not being correctly accounted. Therefore, it is worth
coefficients. investigating the effects of two-way coupling, which is carried out
In general, the numerical results of droplet velocity over- in the next section. Obviously, C2 is more complex than C1, and the
estimate the experimental data in C1. However, the numeri- interactions between the air flow and the liquid jet may change
cal results fit well to the experimental data for larger droplets the droplet velocity, consequently the droplet breakup, more in-
(150 μm < Dp < 160 μm). Furthermore, the experimental trend of tensely than in C1. Different drag correlations were experimented
the droplet velocity is well predicted numerically, with smaller with, unsuccessfully.
droplets faster than that of larger droplets. Still considering the Considering the differences on the droplet velocity for different
general results, due to the rare generation of smallest droplets primary and secondary breakup models in C2, the two different
(Dp < 20 μm) upstream of the sampling plane, the droplet velocity secondary breakup models present little changes on the numerical
curve in this diameter range is not so smooth. In fact, in C1 the results, in the same way as in C1. The simulations using Cb = 3.44
secondary breakup is not so intense, as expected in this regime, result in lower droplet velocity than that using Cb = 2.6, which is
W e = 11, so that smaller droplets are not created frequently.

Fig. 17. Liquid jet curvature for C1.


112 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Table 6
l2 − norm of the mass fraction distribution for C2.

Case Coupling l2

C1 One-way 3.57 × 10−2


C1 Two-way and collision 3.41 × 10−2
C2 One-way 6.56 × 10−2
C2 Two-way and collision 7.18 × 10−2

5.4. Two-way coupling and droplet collision effects

In the previous sections, results were obtained disregarding the


droplet effects on the continuous flow and droplet-to-droplet col-
lisions, i.e. one-way coupling was assumed. In this section, such
droplet effects are accounted for in order to assess their role in
the investigated cases. The two cases, C1 and C2, were solved con-
sidering droplet effects on the continuous flow and droplet-to-
Fig. 18. Droplet velocity for C2 at the sampling plane 3.81 cm downstream of the
droplet collisions, using the column breakup coefficient Cb = 3.44
liquid jet exit. and the AB-TAB as secondary breakup model. In general, the col-
umn breakup coefficient, Cb = 3.44, and the secondary breakup
model, AB-TAB, showed the best prediction of the mass fraction
also seen in the results for C1. Thus, independently of the Weber distribution and droplet velocity for both LJIC cases.
number, the droplet velocity is affected by the column breakup The size distribution considering two-way coupling and droplet
height, but negligibly influenced by the secondary breakup mod- collision also shows minor differences in the two cases, Fig. 19.
els. The l2 − norm of the mass fraction distribution of the two cases
The droplet velocity curves are not so smooth for large droplets for one-way coupling and two-way coupling with droplet collision
(Dp > 100 μm), because the physical condition of C2 (W e = 53) pro- are presented in Table 6, showing quantitative minor differences in
motes secondary breakup more frequently, such that the occur- using the two-way coupling and droplet collision modeling.
rence of larger droplets are rare. For the simulation using Cb = 2.6 Fig. 20 shows the results of the droplet velocity for the two
(lower height of the liquid column) and AB-TAB model, no droplets cases with one-way coupling and with two-way coupling and
with Dp > 102.3 μm cross the sampling plane. For other simula- droplet collisions. Two-way coupling and droplet collision results
tions, droplets with Dp > 100 μm cross the sampling plane with reveal minor changes on the droplet velocity. Actually, the droplet
low frequency, so that the droplet velocity curves in this diame- concentration in the LJIC domain is considered to be low enough
ter range display some spikes. so that the flow is dilute and the droplets are small (Dp < 160μm).

Fig. 19. Size distribution using Cb = 3.44 and AB-TAB model.


D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114 113

Acknowledgments

This study was financed in part by the Coordenação de Aper-


feiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Fi-
nance Code 001. The authors acknowledge the support given by
the Conselho Nacional de Desenvolvimento Científico e Tecnológico
(CNPq) and the Fundação de Amparo à Pesquisa de Minas Gerais
(FAPEMIG) to this research. Also, the authors would like to ac-
knowledge Dr. Marco Arienti for useful discussions on the hybrid
method.

Fig. 20. Droplet velocity using Cb = 3.44 and AB-TAB model.


Supplementary material

Supplementary material associated with this article can be


For this reason, droplet collision rarely happens, and the small found, in the online version, at doi:10.1016/j.ijmultiphaseflow.2019.
droplets change very little the air flow. 02.009.
Considering the results obtained in this work, at the Weber and
Reynolds numbers investigated, the two-way coupling and droplet
References
collision modeling are unnecessary to solve the spray formation in
LJIC. The differences between the experimental and numerical re- Apte, S.V., Gorokhovski, M., Moin, P., 2003. LES Of atomizing spray with stochas-
sults on droplet velocity may be overcome using a more accurate tic modeling of secondary breakup. Int. J. Multiphase Flow 29 (9), 1503–1522.
turbulence closure modeling, such as Large Eddy Simulations (LES). doi:10.1016/S0301-9322(03)00111-3.
Arienti, M., Madabhushi, R.R., Slooten, P.R.V., Soteriou, M.C., 2006. Aerodynamic
Considering all comparisons and analyses carried out in this blockage effect on the spray characteristics of a liquid jet atomized by cross-
work, the hybrid model using Cb = 3.44 as the primary breakup flowing air. Proceedings of. ASME Turbo Expo 2006 Power for Land, Sea and Air
coefficient and the AB-TAB secondary breakup model for spray for- 1 (May), 1–10.
Arienti, M., Soteriou, M.C., 2007. Dynamics of pulsed jet in crossflow. Proceedings
mation in LJIC is the most advanced model hybrid model for low
of. GT2007 ASME Turbo Expo 2007 1–12.
computational cost and low Weber numbers (W e = 11 and W e = Arienti, M., Sussman, M., 2015. A high-Fidelity study of high-Pressure diesel injec-
53) in the literature. tion. 2015 SAE Int.
Arienti, M., Wang, L., Corn, M., Li, X., Soteriou, M.C., Shedd, T.a., Herrmann, M.,
2011. Modeling wall film formation and breakup using an integrated interface-
Tracking/discrete-Phase approach. J. Eng. Gas Turbine. Power 133 (3), 031501.
doi:10.1115/1.4002019.
6. Final remarks Arunajatesan, S., Laboratories, S.N., 2012. Evaluation of two - Equation RANS models
for simulation of jet - in - Crossflow problems. 50th AIAA Aerospace Sciences.
Meeting. Including the New Horizons Forum and Aerospace Exposition 1–14.
This work aims at presenting an improved hybrid model to nu- Balasubramanyam, M.S., Chen, C.P., 2008. Modeling liquid jet breakup in high speed
merically solve LJIC spray at low computational cost. The Euler ap- cross-flow with finite-conductivity evaporation. Int. J. Heat Mass Transf. 51 (15–
16), 3896–3905. doi:10.1016/j.ijheatmasstransfer.2007.11.054.
proach through VOF method was used to describe the jet-air inter- Becker, J., Hassa, C., 2002. Breakup and atomization of a kerosene jet in crossflow
actions, whereas droplets created from the liquid jet were solved at elevated pressure. doi:10.1615/AtomizSpr.v12.i123.30.
as discrete Lagrangian droplets. Two column breakup coefficients Brackbill, J.U., Kothe, D.B., Zemach, C., 1992. A continuum method for modeling
surface tension. J. Comput. Phys. 100 (2), 335–354. doi:10.1016/0021-9991(92)
(Cb = 3.44 and C= 2.6), two secondary breakup models (TAB model 90240-Y.
and AB-TAB model) and two-way coupling with droplet collision Bravo, B.L., Ivey, C.B., Kim, D., 2014. High-fidelity simulation of atomization in diesel
were evaluated in order to establish an improved hybrid model engine sprays. Center Turbul. Res. Proc. Summer Progr. 2014.
Chen, X., Ma, D., Yang, V., Popinet, S., 2013. High-fidelity simulations of imping-
suitable to solve liquid jet in crossflow at low computational cost. ing jet atomization. Atomizat. Spray 23 (12), 1079–1101. doi:10.1615/AtomizSpr.
The counter-rotating vortices, related to the air-jet interactions, 2013007619.
were solved through the VOF method. Such interactions are not re- Dahms, R.N., Oefelein, J.C., 2016. The significance of drop non-sphericity in sprays.
Int. J. Multiphase Flow 86, 67–85. doi:10.1016/j.ijmultiphaseflow.2016.07.010.
solved by the use of Euler-Lagrange approach solely. On the other
De Souza, F.J., De Vasconcelos Salvo, R., De Moro Martins, D.A., 2012. Large eddy
hand, Lagrangian droplets originating from the Euler-Lagrange con- simulation of the gas-particle flow in cyclone separators. Separat. Purif. Technol.
version reduced dramatically the computational cost required in a 94, 61–70. doi:10.1016/j.seppur.2012.04.006.
pure Euler-Euler simulation, without loss of physics of the prob- De Souza, F.J., Silva, A.L., Utzig, J., 2014. Four-way coupled simulations of the gas-
particle flow in a diffuser. Powder Technol. 253, 496–508. doi:10.1016/j.powtec.
lem. Thus, the hybrid approach was crucial to obtain a good repre- 2013.12.021.
sentation of the LJIC physics without the high mesh count usually Deepe, J.M., 2006. Effect of weber number on the fuel transfer function for mod-
required for LJIC simulations. ulated liquid jets. Thesis 119. http://search.proquest.com/docview/28834514?
accountid=2909.
In general, the numerical results of C1 for the mass fraction and Denner, F., 2013. Balanced-Force two-Phase flow modelling on unstructured and
droplet velocity agree well with experimental data. For case C2 adaptive meshes. thesis 237.
(higher Weber number) the AB-TAB model combined with a col- Dhakal, T.P., Walters, D.K., Strasser, W., 2014. Numerical study of gas-cyclone airflow:
an investigation of turbulence modelling approaches. Int. J. Comut. Fluid Dyn.
umn breakup coefficient Cb = 3.44 was the most suitable model 28 (1–2), 1–15. doi:10.1080/10618562.2013.878800.
in the prediction of the droplet size distribution. Two-way cou- Duarte, C.A.R., de Souza, F.J., dos Santos, V.F., 2015. Numerical investigation of mass
pling with droplet collision presented little influence on the mass loading effects on elbow erosion. Powder Technol. 283, 593–606. doi:10.1016/j.
powtec.2015.06.021.
fraction and droplet velocity predictions. In fact, droplet concen-
Duarte, C.A.R., de Souza, F.J., de Vasconcelos Salvo, R., dos Santos, V.F., 2017. The
tration in these two LJIC cases can be categorized as dilute, since role of inter-particle collisions on elbow erosion. Int. J. Multiphase Flow 89, 1–
the average droplet volumetric concentration is low, of the order 22. doi:10.1016/j.ijmultiphaseflow.2016.10.001.
Feng, Z.-G., Michaelides, E.E., 2001. Drag coefficients of viscous spheres at inter-
of 10−5 , and the maximum value is about 0.03, at 20 jet diam-
mediate and high reynolds numbers. J. Fluids Eng. 123 (4), 841. doi:10.1115/1.
eters downstream of the jet inlet. Therefore, the hybrid approach 1412458.
using Cb = 3.44 and the AB-TAB model for LJIC sprays in this We- Ferziger, J.H., Perić, M., 2002. Computational Methods for Fluid Dynamics, third edi-
ber number range (W e = 6 and W e = 11) is an improved model to tion Springer Berlin Heidelberg, Germany.
Fontes, D.H., Duarte, C.A.D., de Souza, F.J., 2018. Numerical simulation of a water
predict satisfactorily droplet size distribution and velocity at low droplet splash: effects of density interpolation schemes. Mech. Res. Commun.
computational cost. 90, 18–25. doi:10.1016/j.mechrescom.2018.04.003.
114 D.H. Fontes, V. Vilela and L.d. Souza Meira et al. / International Journal of Multiphase Flow 114 (2019) 98–114

Haselbacher, a., Najjar, F.M., Ferry, J.P., 2007. An efficient and robust particle- Mazallon, J., Dai, Z., Faeth, G. M., 1999. Primary breakup of nonturbulent round liq-
localization algorithm for unstructured grids. J. Comput. Phys. 225 (2), 2198– uid jets in gas crossflows. 10.1615/AtomizSpr.v9.i3.40.
2213. doi:10.1016/j.jcp.2007.03.018. Mazallon, J., Dai, Z., Faeth, G.M., Arbor, A., 1998. Aerodynamic primary breakup at
Herrmann, M., 2010. A parallel eulerian interface tracking/lagrangian point particle the surface of nonturbulent round liquid jets in crossflow. AIAA J. 15.
multi-scale coupling procedure. J. Comput. Phys. 229 (3), 745–759. doi:10.1016/ Montgomery, D.C., 2013. Design and Analysis of Experiments, 8th ed John Wiley &
j.jcp.20 09.10.0 09. Sons, Inc, 111 River Street, Hoboken, NJ, 07030-5774.
Hiroyasu, H., Kadota, T., 1974. Fuel droplet size distribution in diesel combustion O’Rourke, P., 1981. Collective drop effects on vaporizing liquid sprays. Thesis.
chamber. SAE Paper 2615–2624. O’Rourke, P.J., Amsden, A.A., 1987. The TAB method for numerical calculation of
Hirt, C., Nichols, B., 1981. Volume of fluid (VOF) method for the dynamics of spray droplet breakup. International Fules and Lubricants Meeting and Expo-
free boundaries. J. Comput. Phys. 39 (1), 201–225. doi:10.1016/0021-9991(81) sition, Toronto, Ontario.
90145-5. Reitz, R., 2006. 4. Spray processes. Spray Course 1–44. http://www.papers2://
Hojnacki, J.T., 1972. Ramjet engine fuel injection studies. Technical report to Aero publication/uuid/BEE087C4- 3B90- 4D5F- 963E- 8C28C55B1ECB.
Propulsion Lab., AFAPL-TR-72-76, 1 (September), 754–852. Reitz, R.D., 1987. Modeling atomization processes in high-Pressure vaporizing
Inamura, T., 1997. Spray characteristics of liquid jet traversing. J. Propul. Power 13 sprays. Atomizat. Sprays 3, 309–337.
(2). Reynolds, O., 1895. On the dynamical theory of incompressible viscous fluids and
Karvinen, A., Ahlstedt, H., 2005. Comparison of turbulence models in case of jet in the determination of the criterion. Philos. Trans. R Soc.London A 186, 123–164.
crossflow using commercial cfd code. In: Rodi, W., Mulas, M. (Eds.), Engineer- doi:10.1098/rsta.1895.0 0 04.
ing Turbulence Modelling and Experiments 6. Elsevier Science B.V., Amsterdam, Rhie, C.M., Chow, W.L., 1983. Numerical study of the turbulent flow past an airfoil
pp. 399–408. doi:10.1016/B978-008044544-1/50038-8. with trailing edge separation. 3rd Joint Thermophysics, Fluids, Plasma and Heat
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent flows. Transfer Conference 21 (11), 1525–1532. doi:10.2514/3.8284.
Comput. Methods Appl. Mech. Eng. 3, 269–289. Sallam, K.A., Aalburg, C., Faeth, G.M., 2003. Breakup of round nonturbulent liquid
Lefantzi, S., Ray, J., Arunajatesan, S., Dechant, L., 2013. Tuning a RANS k  model for jets in gaseous cross flows. 41st Aerospace Sci. Meet. Exhibit 1–16.
jet-in-crossflow simulations. report. Sallam, K.a., Ng, C.-L., Sankarakrishnan, R., Aalburg, C., Lee, K., 2006. Breakup
Li, P., Wang, Z., Sun, M., Wang, H., 2017. Numerical simulation of the gas-liquid in- of turbulent and non-Turbulent liquid jets in gaseous crossflows. 44th AIAA
teraction of a liquid jet in supersonic crossflow. Acta Astronaut. 134 (November Aerospace Sciences Meeting and Exhibit; Reno, NV; USA; 9–12 Jan. 2006 1–13.
2016), 333–344. doi:10.1016/j.actaastro.2016.12.025. doi:10.2514/6.2006-1517.
Li, X., Arienti, M., Soteriou, M.C., 2010. Towards an efficient, high-Fidelity methodol- Shedd, T.A., 2009. Liquid film formation by an impinging jet in a high- velocity air
ogy for liquid jet atomization computations. Aiaa 092407. stream. 47th AIAA Aerospace Sciences Meeting Including The New Horizons Fo-
Li, X., Soteriou, M.C., 2016. High fidelity simulation and analysis of liquid jet atom- rum and Aerospace Exposition.
ization in a gaseous crossflow at intermediate weber numbers. Phys. Fluid 28 Shinjo, J., Umemura, A., 2011. Detailed simulation of primary atomization mecha-
(8), 082101. doi:10.1063/1.4959290. nisms in diesel jet sprays (isolated identification of liquid jet tip effects). Proc.
Lin, S.P., Reitz, R.D., 1998. Drop and spray formation from a liquid jet. Annu. Rev. Combust. Inst. 33 (2), 2089–2097. doi:10.1016/j.proci.2010.07.006.
Fluid Mech. 30, 85–105. Sommerfeld, M., 2001. Validation of a stochastic lagrangian modelling approach for
Ling, Y., Fuster, D., Tryggvason, G., Zaleski, S., 2018. A two-phase mixing layer be- inter-particle collisions in homogeneous isotropic turbulence. Int. J. Multiphase
tween parallel gas and liquid streams: multiphase turbulence statistics and in- Flow 27 (10), 1829–1858.
fluence of interfacial instability. J. Fluid Mech. 268–307. doi:10.1017/jfm.2018. Strasser, W., 2007. Discrete particle study of turbulence coupling in a confined jet
825. gas-liquid separator. J. Fluids Eng. 130, 011101. doi:10.1115/1.2816008.
Ling, Y., Fuster, D., Zaleski, S., Tryggvason, G., 2017. Spray formation in a quasiplanar Strasser, W., 2019. Oxidation-assisted pulsating three-stream non-newtonian slurry
gas-liquid mixing layer at moderate density ratios: a numerical closeup. Phys. atomization for energy production. Chem. Eng. Sci. 196, 214–224. doi:10.1016/j.
Rev. Fluid 2, 014005. doi:10.1103/PhysRevFluids.2.014005. ces.2018.10.055.
Liovic, P., Lakehal, D., 2007. Interfaceturbulence interactions in large-scale bubbling Sussman, M., Smereka, P., Osher, S., 1994. A level set approach for computing solu-
processes. Int. J. Heat Fluid Flow 28 (1), 127–144. The International Confer- tions to incompressible two-phase flow. 10.1006/jcph.1994.1155.
ence on Heat Transfer and Fluid Flow in Microscale (HTFFM-05). doi: 10.1016/j. Tanner, F.X., 1997. Liquid jet atomization and droplet breakup modeling of non-
ijheatfluidflow.20 06.03.0 03. evaporating diesel fuel sprays. In: SAE Technical Paper. SAE International, p. 16.
Ma, M., Lu, J., Tryggvason, G., 2015. Using statistical learning to close two-fluid doi:10.4271/970050.
multiphase flow equations for a simple bubbly system. Phys. Fluid. 27 (9). Tryggvason, G., Bunner, B., Esmaeeli, a., Juric, D., Al-Rawahi, N., Tauber, W., Han, J.,
doi:10.1063/L4930 0 04. Nas, S., Jan, Y.J., 2001. A front-Tracking method for the computations of multi-
Ma, M., Lu, J., Tryggvason, G., 2016. Using statistical learning to close two-fluid mul- phase flow. J. Comput. Phys. 169 (2), 708–759. doi:10.10 06/jcph.20 01.6726.
tiphase flow equations for bubbly flows in vertical channels. Int. J. Multiphase Tryggvason, G., Ma, M., Lu, J., 2016. DNS Assisted modeling of bubbly flows in ver-
Flow 85, 336–347. doi:10.1016/j.ijmultiphaseflow.2016.06.021. tical channels. Nucl. Sci. Eng. 184 (3), 312–320. doi:10.1364/OE.19.015955.
Madabhushi, R.K., 2003. A model for numerical simulation of breakup of a liquid jet Ubbink, O., 1997. Numerical prediction of two fluid systems with sharp interfaces.
in crossflow. Atomizat. Spray. 13, 413–424. Thesis 69. doi:10.1145/1774088.1774119.
Maric, T., Marschall, H., Bothe, D., 2013. Vofoam - A geometrical volume of fluid Wu, P.-k., Kirkendall, K.A., Fuller, R.P., 1997. Breakup processes of liquid jets in sub-
algorithm on arbitrary unstructured meshes with local dynamic adaptive mesh sonic crossflows. J. Propul. Power 13 (1), 64–73.
refinement using openfoam. report 1–30.

You might also like