Acta Astronautica 134 (2017) 333–344

Contents lists available at ScienceDirect

Acta Astronautica
journal homepage: www.elsevier.com/locate/actaastro

Numerical simulation of the gas-liquid interaction of a liquid jet in MARK

supersonic crossflow
⁎ ⁎
Peibo Li, Zhenguo Wang , Mingbo Sun , Hongbo Wang
Science and Technology on Scramjet Laboratory, National University of Defense Technology, Changsha 410073, China

A R T I C L E I N F O A BS T RAC T

Keywords: The gas-liquid interaction process of a liquid jet in supersonic crossflow with a Mach number of 1.94 was
Large eddy simulation investigated numerically using the Eulerian-Lagrangian method. The KH (Kelvin-Helmholtz) breakup model
Eulerian–Lagrangian was used to calculate the droplet stripping process, and the secondary breakup process was simulated by the
Supersonic crossflow competition of RT (Rayleigh-Taylor) breakup model and TAB (Taylor Analogy Breakup) model. A correction of
Liquid jet
drag coefficient was proposed by considering the compressible effects and the deformation of droplets. The
Droplet breakup
Spray
location and velocity models of child droplets after breakup were improved according to droplet deformation. It
Two-phase flow was found that the calculated spray features, including spray penetration, droplet size distribution and droplet
velocity profile agree reasonably well with the experiment. Numerical results revealed that the streamlines of air
flow could intersect with the trajectory of droplets and are deflected towards the near-wall region after they
enter into spray zone around the central plane. The analysis of gas-liquid relative velocity and droplet
deformation suggested that the breakup of droplets mainly occurs around the front region of the spray where
gathered a large number of droplets with different sizes. The liquid trailing phenomenon of jet spray which has
been discovered by the previous experiment was successfully captured, and a reasonable explanation was given
based on the analysis of gas-liquid interaction process.

1. Introduction
The scramjet engine is the key component of future hypersonic
aircraft. Compared with gaseous fuel, liquid fuel has a higher volu-
metric and mass energy density and is easier and safer to store and
transport [1]. So, the liquid fuel gets more and more attention in
scramjet engines, where it is often injected vertically into high-speed
crossflow and atomized into small droplets to promote the evaporation.
Due to the extremely high speed of the free stream, the combustion
efficiency and engine performance depend strongly on the consequent
fuel-air mixing process. Therefore, the study of the liquid jet in high-
speed crossflow has become an important research area.
A large amount of experimental work have been carried out since
the 1970s regarding liquid jet in high-speed crossflow [2–4]. However,
due to the lack of advanced experimental technology, earlier experi-
ments can only rely on high-speed photography and schlieren techni-
que to give a macroscopic description of spray field. With the
development of experimental techniques, detailed and reliable proper-
ties become possible. Wu et al. [5] and Lin et al. [6–8] employed the
phase Doppler particle analyzer (PDPA) to study the atomization of
liquid jets in respectively subsonic and supersonic air flows. The effects
of injector nozzle size, gas-liquid momentum ratio and injection angle
on jet spray were investigated. The penetration height of the spray was
measured, and corresponding correlation function was presented. Lee
et al. [9] and Olinger et al. [10] utilized the digital holographic
microscopy to study the rapidly developing region near the injector
of several sprays in subsonic crossflows, and it was found that the area
immediately downstream of the nozzle features a continuous distribu-
tion of droplets. Fei et al. [11] successfully combined the shadowgraph
with the planar laser-induced fluorescence (PLIF), and visualized the
wave structures and the droplets distribution of kerosene jet in a
supersonic cold flow. Considering the high-energy and short pulse
characteristics of the laser, Wu et al. [12] presented a kind of pulsed
laser background imaging method (PLBI), and successfully obtained
explicit transient images of spray in supersonic crossflows, accurately
visualized and extracted unstable fluctuation surface waves and the
inclined forward formations of the spray.
Although the existing experimental work have made explicit pro-
gress, an in-depth understanding of the two-phase flow physics in the
spray plume is still challenging due to the complexity of its flow field
and the limitation of experimental measurement. So, numerical
simulation becomes an effective tool to explore the flow characteristics

⁎
Corresponding authors.
E-mail addresses: zgwang@nudt.edu.cn (Z. Wang), wind_flowcfd@163.com (M. Sun).

http://dx.doi.org/10.1016/j.actaastro.2016.12.025
Received 12 November 2016; Accepted 20 December 2016
Available online 23 December 2016
0094-5765/ © 2017 IAA. Published by Elsevier Ltd. All rights reserved.
P. Li et al.
and present valuable insights into the complex flow structures in the
spray plume. Currently, two general types of computational models
have been developed for numerical simulation of two-phase flows. One
is the droplet tracking method based on the Lagrangian framework,
and the other is interface tracking method based on the Eulerian
framework. By capturing the movement of the gas-liquid interface, the
latter method can capture the atomization and fragmentation of the jet
and droplets finely [13–15]. However, this approach is computationally
expensive and has not yet been able to simulate the complete jet
atomization and evolutionary mixing process. While in the droplet
tracking method, the spray is represented by discrete droplets, which
are injected and tracked in the computational domain individually
according to the Newton's second law of motion. The advantage of this
method is that it can flexibly use models to simulate the breakup,
collision, evaporation and combustion of the droplets according to
specific conditions. Thus it is widely applied in liquid spray combustion
system [16–22].
In the present work, a coupled Eulerian (for the gas-phase flow) and
Lagrangian (for the liquid-phase flow) formulation is used to represent
the spray dynamics and the interaction between the two phases. The
Eulerian-Lagrangian model has been widely applied in the simulation
of coaxial liquid jets in stationary gasses [17,18,23] and swirl liquid jets
in high-speed gasses [16,20,24], and has provided sufficient reference
data for the design of internal combustion (IC) engines and gas
turbines. For the present study, reasonable droplet motion model
and droplet atomization model are critical for the simulation of liquid
jet in supersonic crossflow. Considering the compressibility effects on
drag coefficient, Henderson [25] presented a correlation that reduces in
the limit to certain analytical expressions and shown good agreement
with experimental data. Moin et al. [26] pointed out that the droplet
deformation affects the drag force significantly, and developed a
correlation of drag force according to the amount of droplet deforma-
tion in the form of ellipticity. Based on the experimental data,
researchers have proposed a variety of droplet breakup models. The
most successful and widely used models for spray atomization includes
TAB model [27], wave breakup model [28] and various modified or
hybrid models based on the above two models [22,29,30]. Patterson
et al. [30] suggested that the droplet breakup is the result of
competition between the KH instability and RT instability, and
successfully used the KH-RT hybrid model to simulate the liquid jet
in diesel. Im et al. [22] modified the KH-RT hybrid breakup model by
considering the compressible effects and simulated the liquid jet in
supersonic crossflow. Reasonable spray penetration height and droplet
size distribution were given, but the droplet velocity distribution was
not given. In spite of a certain gap compared with the experimental
results, these numerical work suggest that the atomization models
which stemmed from low speed can be applied to supersonic condition
after reasonable modification. Unfortunately, these work did not give
an extensive analysis on the physical phenomena in spray plume. To
our knowledge, few numerical work has been carried out to explain the
gas-liquid interaction process of the liquid jet in supersonic crossflow.
The fundamental driving force of the atomization process and the
mixing process between droplets and free stream air comes from the
gas-liquid interaction. Therefore, it is important to study the gas-liquid
interaction process to explain the experimental observations and to
understand the essence of atomization and mixing process in the two-
phase flow field.
Wu et al. [12,31] have mentioned that there exist inclined forward
formations of the spray in his experimental observation. Moreover, in
our supersonic low noise wind tunnel [32], we used Planer Laser
Scattering (PLS) system again observed the liquid trailing phenomenon
of jet spray as shown in Fig. 1. After liquid column breaks up into
fragment or droplet groups, there appears some ribbon-shaped spray
structure on the leeward side of the jet plume. Detailed experimental
results will be discussed in a later article. Here, we just use the
numerical simulation to explain this interesting phenomenon.
334
Acta Astronautica 134 (2017) 333–344
Fig. 1. The liquid trailing phenomenon of jet spray observed by experiment.
In this paper, the two-phase compressible LES (Large Eddy
Simulation) code based on the Eulerian-Lagrangian scheme was
developed. The drag coefficient and the spray breakup model were
modified by considering the compressible effects and the deformation
of droplets. The primary objective of this paper is to study the
interaction process between a liquid jet and a supersonic crossflow
and explain the above interesting phenomenon. This paper is organized
as follows: The physical models for gas-phase, liquid-phase, interphase
exchange terms and corresponding numerical approaches are pre-
sented in Section 2. In Section 3, the problem and computation setup
are described and followed by the validation of numerical models to
make sure the physical models and numerical methods are reliable.
After that, the characteristics of gas-phase flow field and spray field are
presented as the foundation to analyze the liquid trailing phenomenon
of jet spray. Finally, key findings of this study are summarized in
Section 5.
2. Mathematical models
Three main mathematical components are contained in the two-
phase compressible LES model: (1) the Eulerian filtered gas-phase
governing equations with turbulence model, (2) the Lagrangian droplet
trajectory tracking and spray breakup model, and (3) the interphase
exchange terms with two-way coupling effects.
2.1. Gas-phase governing equations
The Favre-filtered compressible Navier-Stokes conservation equa-
tions of continuity, momentum and total energy, with the contribution
of the dispersed phase included, for the gas-phase can be written as
below:
∂ρ ∂(ρ u∼i )
+ =0
∂t ∂xi (1)
∂(ρ u∼i ) ∂[ρ u∼i u∼j + p δij − τij + τijsgs] ∼
+ = Fṡ , i
∂t ∂xj (2)
∼ ∼
∂ρ E ∂[(ρ E + p ) u∼i + qi − u∼j τji + Hisgs + σijsgs] ∼
+ = Q̇ s
∂t ∂xi (3)
∼ ∼
Where ρ is the filtered density, ui is the filtered velocity, E is the filtered
total energy, p is the filtered pressure, τij is the filtered laminar viscous
stress tensor, qi is the energy flux caused by heat conduction and mass
diffusion. Here, ‘-’ denotes the spatial filtering and ‘~’ denotes the Favre
filtering. The Favre filtering of flow parameter f can be defined by
∼
f = ρf / ρ . τijsgs is the unclosed (sub-grid scale) stress tensor, Hisgs is the
unclosed enthalpy flux vector and σijsgs is the unclosed heat flux vector.
∼ ∼
Fṡ , i and Q̇s are respectively the momentum and energy source terms
caused by the effect of droplet motion.
P. Li et al.
The sub-grid scale terms in the Favre-filtered equations require
closure by establishing turbulence model. Due to high mesh resolution
required, it is difficult for LES to accurately simulate the flows in the
near-wall region at high Reynolds numbers. So, a hybrid RANS/LES
method blending the Spalart-Allmaras (S-A) RANS model and
Yoshizawa Sub-Grid Scale (SGS) model is adopted, the former is used
in near-wall regions, and the latter is used in the regions away from the
solid wall. The details of turbulence modeling process can be seen in
Ref. [33]. In fact, the droplet dispersion will affect the sub-grid scale
turbulence, but for well resolved LES of gas-phase, the interaction of
droplet dispersion and sub-grid scale velocity was shown to be small
[34,35], then it can be neglected.
The fifth-order WENO (Weighed Essentially Non-Oscillation)
scheme [36] is adopted for the inviscid fluxes, and viscous fluxes are
calculated by the fourth-order central difference scheme. The time
integration is performed by means of a third-order accuracy total-
variation-diminishing (TVD) Runge-Kutta method. The CFL (Courant-
Friedrichs-Lewy) number is fixed to be 0.5.
Notably, the numerical methods and models of gas-phase inte-
grated into the present two-phase code have been validated through
early simulation studies of gas-phase jet mixing in supersonic crossflow
[33], supersonic cavity flow [37], supersonic jet combustion [38] and
other supersonic flow or combustion problems [39–41].
2.2. Liquid-phase equations
2.2.1. Governing equations of droplet motion
The liquid-phase is represented by Lagrangian droplets, carrying
the information of diameter and velocity, which are tracked by the
trajectory method. Since real spray contains a very large number of
droplets, tracking and solving of every droplet evolution in space and
time within the gas field can be computationally expensive. Therefore, a
sampling technique is employed, whereby a characteristic group of
droplets is gathered into a single unit termed as ‘computational
droplet’. Each computational droplet represents a group of real
droplets having the same diameter and velocity. This method is feasible
for it can give reasonable statistic results of the spray characteristic,
and it has been widely used in the simulation of two-phase flow
[24,35,42].
For simplicity, the void fraction occupied by spray droplets is not
taken into account. The interactions among droplets (collision and
coalescence processes) are neglected due to the assumption of the
dilute flow in the present study. The droplet motion is simulated using
the Basset-Boussinesq-Oseen (BBO) equations [43]. The gravity,
Coriolis force, centrifugal force, Basset force and virtual mass effect
are all neglected, only the drag force is considered in the present
simulation. Under these assumptions, the Lagrangian droplet equa-
tions for the position and the velocity become:
d Xl
= Ul
dt (4)
dU
ml l = Dl (Ug − Ul)
dt (5)
Where Xl is the position vector of the droplet centroid, Ul is the droplet
velocity vector, Ug is the gas-phase velocity interpolated to the droplet
location, subscript l denotes the parameters of liquid-phase, subscript g
denotes the parameters of gas-phase. Dl (Ug − Ul) is the drag force
putting on a droplet, and it is modeled by drag coefficient Cd :
Dl = (πdl2 Cd ρg Ug − Ul )/8
(6)
Where dl is the diameter of droplets, ρg is the gas-phase density. Drag
coefficient, Cd , plays a significant role in the prediction of the droplet
trajectory. However, in most spray research, the drag coefficient was
given under the condition of subsonic [17,24] or the hypothesis of the
rigid sphere [21,22,44]. When droplets move in supersonic flow, their
335
Acta Astronautica 134 (2017) 333–344
deformation and the effect of compressibility may affect the drag
coefficient significantly, so an empirical model developed by Henderson
[25] was utilized, where the drag coefficient of the rigid sphere (Cds )
was calculated based on the local relative Mach number Md and droplet
Reynolds number Red :
Cds = f (Md , Red ) (7)
Where Md is evaluated based on the relative velocity between droplet
and surrounding gas Ur = Ug − Ul and local sonic speed of gas-phase,
Red is estimated based on Ur , droplet diameter, free stream density and
viscosity:
Ur ρg dl Ur
Md = , Red = μg
ag (8)
On the other hand, the droplets undergo significant flattening and
are no longer spherical as soon as they enter the supersonic air flow. To
quantify the effect of droplet deformation on drag coefficient, the
distortion parameter Q in the TAB model [27] was adopted. The
normalized radial droplet deformation Q = x /(Cb r ) indicates the level of
droplets deviation from spherical forms, and its time evolution can be
described by the forced, damped linear harmonic oscillator:
2ρg (Ug − Ul)2
d 2Q 8σl 5μl dQ
2
= − Q−
dt 3ρl r 2 ρl r 3 ρl r 2 dt (9)
Where x denotes the radius increase of the equator from its equilibrium
position, r is the droplet radius and Cb = 0.5 is determined from
theoretical considerations and experimental results. In Eq. (9), the
external forcing term is given by the aerodynamic droplet-gas interac-
tion, the damping is due to the liquid viscosity ( μl ) and the restoring
force is supplied by the surface tension (σl ). When there is a distortion
Q , the effect of droplet deformation can be modeled by using an
effective droplet cross-sectional area Sf = πr 2 /(1 − Cb Q), and a simple
correction of drag coefficient was proposed as:
Cd = Cds /(1 − Cb Q) (10)
In the present study, the droplet temperature is assumed to be
constant without evaporation effects as the temperature difference
between phases is very small, so that the droplet energy equation is not
considered, which is a realistic assumption based on the experimental
conditions of Lin et al. [8].
The new position and velocity of droplets are integrated from Eqs.
(4) and (5) using a fourth-order Runge-Kutta time stepping algorithm.
After obtaining the new position, an efficient droplet-locating algorithm
[45] was adopted to relocate droplets on corresponding control volume
cells. The node-based linked list of droplets [46] was created to reduce
the computational time consumption for searching and judgment
between the two phases. The inter-processor strategy of droplets’
sharing and transfer [47] was implemented to perform the parallel
computing of droplets.
2.2.2. Spray breakup model
The breakup of fuel jet is indispensable in practical spray combus-
tion system in order to create large liquid surface-to-volume ratio
which facilitates fast evaporation. The TAB model [27,48,49] based on
the oscillation and deformation of a droplet and the KH/RT hybrid
breakup model [22,30,50] based on linearized stability analysis of jet
column are widely used in spray researches. The linearized stability
process, the deformation and oscillation all exist in the liquid jet when
it is injected vertically into supersonic crossflow. Therefore, in the
present study, the KH model was used to calculate the droplet stripping
process, the TAB model was adopted to simulate the oscillation and
deformation, and after the critical breakup time, RT breakup model
and TAB model competed to simulate the droplet secondary breakup
process.
For KH breakup model, from the numerical solution of a general
dispersion equation, the maximum growth rate, ΩKH and its corre-
P. Li et al. Acta Astronautica 134 (2017) 333–344

sponding wavelength, ΛKH are related to pertinent properties of liquid the unit vector in the direction of Ur .
and gas as [28]: From the results of high-speed photography [51,52], it is found that
the child droplets are always at the downstream locations of the parent
⎛ ρ r 3 ⎞0.5 (0.34 + 0.38Weg1.5)
ΩKH ⎜⎜
l p
⎟⎟ = droplet, indicating that the child droplets move faster than the parent
⎝ σl ⎠ (1 + Oh )(1 + 1.4TP0.6 ) (11) droplet. The resultant velocity (Ul ) of droplet group after breakup can
be calculated by Eq. (5). It can be considered as a combination of two
ΛKH (1 + 0.45Oh0.5)(1 + 0.4TP0.7) components: Ul = Ujet + Udg , where Ujet is the jet exit velocity vector, Udg
= 9.02
rp (1 + 1.67 0.6
0.87Weg ) (12) is the sub-component velocity vector obtained from air flow accelera-
tion. The droplet stripping process of a liquid jet mainly occurs in the
The parameters above are defined as: near-injector region, where the relative velocity between the two phases
ρg Ur2 rp ρl Ur2 rp ρl dl Ur Wel is very higher. Compared to the aerodynamic drag force, surface
Weg = , Wel = , Rel = , Oh = , TP = Oh Weg tension and viscous force can almost be negligible, so that the convex
σl σl μl Rel
bulge generated on the surface of parent droplet acquires a larger
Where rp is the parent droplet radius, Weg and Wel are the gas and liquid acceleration than that of parent droplet. According to Eqs. (5) and (6),
Weber number, respectively, Rel is the liquid Reynolds number, Oh is the droplet acceleration is inversely proportional to the droplet
the Ohnesorge number, TP is the Taylor number. This model postulates diameter, so, at the breakup instant, the velocity relationship between
that if the wavelength of the fastest-growing wave meets the condition child and parent droplets is estimated as Uc, dg / Up, dg = dp /dc . Using
of B0 ΛKH ≤ rp , there will be mass stripped from the parent droplet. And velocity redistribution strategy and according to the momentum
the rate of change of the droplet radius is given by: conservation law at breakup moment, sub-component velocity Udg
was redistributed to child droplets and parent droplet as:
drp (rp − rc )
=−
dt τKH (13) Ndc3 + dp3
Uc, dg = Udg, Uc = Ujet + Uc, dg
Where rc is the child droplet radius, τKH = 3.726B1 rp /(ΛKH ΩKH ) is the KNdc3 + dp3 (16)
droplet breakup time, B1 = 1.73 is a constant. The liquid mass shedded
3 3
from the parent droplet is tracked during the computation. New 1 Ndc + dp
Up, dg = Udg, Up = Ujet + Up, dg
droplets are produced when the shed mass reaches or exceeds 3% of K KNdc3 + dp3 (17)
the parent droplet mass, and the radius of the new droplets is assumed
proportional to the wavelength of the fastest growing wave: Where K = dp / dc , N is the number of child droplets stripped from the
rc = 0.61ΛKH . parent droplet, subscript c denotes parameter of child droplet, and
Existing researches rarely mention how to determine the velocity subscript p denotes parameter of parent droplet.
and position of child droplets after breakup. Our numerical tests found There are a lot of researches have been carried out to discuss the
that the location and velocity of child droplets have obvious effects on droplet breakup characteristic time [54,55]. But, to our knowledge,
final results. Here, the deformation of the droplet is concerned to most of which are only applied to a single droplet breakup simulation.
evaluate the location of child droplets, and a velocity redistribution A critical breakup time which has been empirically validated by
strategy is proposed to estimate the droplet velocity after KH breakup. experimental data in the liquid column breakup process [22], was
In KH breakup mode, child droplets are stripped from the edge of adopted in the present model as the active boundary for the RT model:
the parent droplet, and always viewed on the leeward side [51–53], so tbRT = 1.9(Weg − 12)−0.25 (1 + 2.2Oh1.6 ) (18)
it is not reasonable to locate the child droplets in the center of parent
droplet. A simple expression proposed here is: If the dimensionless time after start-of-injection is greater than the
critical breakup time, tbRT , the RT mode is activated for each droplet.
Xc = Xp + a nf ⊥ + b nf (14)
The RT breakup model also uses the fastest growing disturbances to
rp determine when and how droplets will break up. The growth rate ΩRT
a= , b = rp (1 − Cb Q)
1 − Cb Q (15) and the wavelength ΛRT of the most unstable surface waves are:

As the schematic diagram showed in Fig. 2, the parent droplet ⎛ 2 [−ap (ρ − ρ )]1.5 ⎞0..5
ΩRT = ⎜⎜ ⎟⎟
l g
deforms into disk shape under the impact of gas-phase. Xc and Xp
⎝ 3 3σs ρl + ρg ⎠ (19)
denote the position vector of child droplet and parent droplet,
respectively. a and b denotes the semi-major axis and the semi-minor
−ap (ρl − ρg )
axis of the elliptical parent droplet. nf ⊥ is the unit vector randomly ΛRT = 2πCRT
distributed in a plane normal to the relative velocity vector Ur , and nf is 3σl (20)

Where ap is the droplet acceleration, CRT is an adjustable constant equal

Fig. 2. Schematic diagram of child droplets location.
to 0.3. The wavelength ΛRT is compared to the distorted droplet
diameter, and if the wavelength is smaller than the droplet diameter,
RT waves are assumed to be growing on the surface of the droplet. The
amount of time that the waves grow is tracked and compared to the
droplet breakup time, τRT = Cτ / ΩRT , where Cτ is a constant equal to 1. If
the RT waves have been growing for a time greater than the droplet
breakup time, the parent droplet breaks up into a collection of smaller
droplets that have radius: rc = 0.5ΛRT .
TAB model was adopted to model the deformation and oscillation
of the droplets and simulate the secondary breakup process competing
with RT model. The governing equation for an oscillating, distorting
droplet has been given in Eq. (9). This model assumes that a droplet
will breakup as long as the normalized deformation of the droplet
grows to, Q = 1. The Sauter mean diameter (SMD) of the child droplets
can be determined from the conservation of droplet energy during

336
P. Li et al. Acta Astronautica 134 (2017) 333–344

breakup:

rp 7 ρ r3
= + l Q̇ 2
rc 3 8σl (21)

Rosin-Rammler distribution is adopted for the droplet sizes after

breakup. The initial deformation condition for new droplets is
Q n +1 = Q̇ n +1 = 0 , and its velocity equal to the combination of parent
velocity and a velocity component Cb rQ̇ which is randomly distributed
in a plane normal to the path of the parent droplet.
Unlike the stripping process in KH mode, both RT mode and TAB
mode are explosion-type breakup modes in which the droplets formed
are all significantly smaller than the original droplet. So the position
vector of child droplets can be given by considering the droplet
deformation as Xc = Xp + ϕa nf ⊥, where ϕ ∈ [0, 1] is a random para- Fig. 4. Schematic diagram of the computational domains and experimental conditions.
meter.
⎧∼ Nd
⎪ Fṡ, i = − 1 ∑ Dl (ug, i − ul, i, k ) Df
⎪ ΔV
2.3. Interphase exchange terms ⎨ k
⎪∼ 1
Nd

⎪Q̇s = − ΔV ∑ Dl (Ug − Ul, k ) Ul, k Df

The presence of the interphase exchange terms accounted for two-
way coupling between the continuous and the dispersed phases. The
gas-phase governing equation was calculated in the Eulerian coordi-
nate system, while the droplet motion equation was calculated in the
Lagrange coordinate system. Droplets are not restricted to lie on the
Eulerian grid points where the gas-phase properties are known. So, a
tri-linear interpolation method was adopted to estimate the gas-phase
properties at individual droplet location. Firstly, the control volume
where droplet located in should be known, and the flow field
parameters F (xg ) of its eight grid point should be acquired, then the
interpolation function S (xl , xg ) determined by relative position of
droplets and control volume was used as the following general form:
8 experiment of Lin et al. [8]. Fig. 4 shows a schematic diagram of the
F (xl ) = ∑ S (xl , xg) F (xg)
g =1
Where xg is the Eulerian grid point and F is the gas-phase property
known at the grid point xg or should be calculated at the droplet
position xl . For example, the schematic diagram of two-dimensional
data exchange between coordinates is given in Fig. 3.
For the reaction and evaporation are not concerned in the present
study, the interphase source terms that appear on the right-hand side
of the gas-phase governing equations only remain the momentum
∼ ∼
source term Fṡ , i and the energy source term Q̇s . They are obtained from
the equations governing droplets dynamics (see in Eq. (5)). Let Df
denote the number of real droplets contained in the computational
droplet, then the volume-averaged source terms caused by the effect of
droplet motion are computed by summing the contribution from all
droplets as:
⎩ k (23)
Where i = x, y, z , ΔV is the volume of the control cell, the summation
index k is over all droplets located in the control volume. Nd is the total
number of computational droplets in the control volume.
3. Results and discussion
3.1. Computational conditions
The present simulation was performed to investigate the gas-liquid
interaction and mixing mechanism of a liquid jet in supersonic cross-
flow. To validate the code, all test conditions were based on the
computational domain and experimental conditions. Water was used as
(22) the test liquid in the research, which has a density of 988 kg/m3,
viscosity of 2.67×10−3 kg/(m s) and surface tension of 0.072 N/m. The
water was injected into a crossflow of Mach number M∞ = 1.94 , the
static temperature and pressure of which are 304.1 K and 29 kPa
respectively, and the free stream velocity U∞ is 678.13 m/s. The gas-
liquid momentum ratio is q0 = ρl vl2 / ρ∞ v∞2 = 7.0 , and the injection
velocity vj can be derived as 32.73 m/s. The corresponding total mass
of spray is 6.415×10−3 kg/s. The initial temperature of liquid water jet
is 300 K. Detailed parameters of jet-exit and crossflow are shown in
Table 1.
Considering the computational cost, the computational domain was
confined to a limited local region near the injector nozzle, which
comprises of a solid surface that represents the flat plate with a circular
hole as the injector. The computational domain was designed to be
L x × L y × L z =200 mm×40 mm×40 mm in three dimensions. The in-
jector nozzle with a diameter of 0.5 mm is located in the center of the
bottom floor and 50 mm downstream from the leading edge. The mesh
has been refined near the injector nozzle and the near-wall region. The
number of grid points is 481×201×201 in the x, y and z directions,
which leads to a grid resolution of Δx+ ≈ 10–50, Δy+ ≈ 1–50 and
Δz+ ≈ 10–50 in the focused region basing on the wall stress τw at the

Table 1
Physical parameters in the experiment of Lin et al. [8].

Supersonic crossflow Liquid jet-exit flow

Mach number, M∞ 1.94 Gas-liquid momentum ratio, 7.0

q0
Velocity, U∞ 678.13 m/s Injection velocity, vj 32.73 m/s
Static pressure, P∞ 29 kPa Water density, ρl 998 kg/m3
Static temperature, T∞ 304.1 K Injector nozzle diameter, d 0.5 mm
Fig. 3. Schematic diagram of two-dimensional data exchange between coordinates.

337
P. Li et al. Acta Astronautica 134 (2017) 333–344

inlet bottom floor. It may be coarse for a strict wall-resolved LES but is
suitable for the hybrid RANS/LES approach. Since the droplets are
discrete and have no fixed positions, there is no exact number of
droplets in the calculation region. After the two-phase flow-field is
established, the total number of computational droplets tracked in
computational domain is approximately 1 million. And every computa-
tional droplet represent four real droplets which is very close to the
maximal precision for the spray representation according to the
research of Smirnov et al. [56].
For the gas-phase, a no-slip, no-penetration adiabatic condition was
imposed at the wall, supersonic inflow condition was used at the inlet
section, and the other boundaries were treated by extrapolation
method. For the liquid-phase, the Tsuji empirical formula [57] was
used to calculate the droplet-wall interaction.
Estimating the stochastic errors accumulation is necessary for
numerical simulation of an unsteady flow-field. Smirnov et al.
[58,59] has shown that the accumulated error depends on the spatial
resolution and time integration steps. The relative errors Serr of
integration in three directions is assumed to be
3 Fig. 6. Averaged SMD at different streamwise locations in the central plane, dashed lines
Serr = ∑ (ΔL i /L i )k+1 represent experimental data, whereas solid lines indicate simulation.
i =1 (24)
where ΔL i / L i denotes the mean ratio of cell size to the domain size in
the direction of integration. k indicates the order of accuracy of
numerical scheme. Let us set an allowable value of total error S max =
1% to ensure the simulation precision. Then the maximal number of
time steps for solving present problem can be estimated by the
following formula
n max = (S max / Serr )2 (25)
A evaluation parameter Rs , denotes the ratio of maximal allowable
number of time steps for the problem and the actual number of time
steps through the simulation, is defined to characterize the reliability of
results
Rs = n max / n (26)
The actual number of integration steps n performed before obtain-
ing the results is 9×105. Based on the mesh intensity and time
integration scheme, the evaluation parameter Rs in present simulation
is approximately 1705 which is high enough, therefore the accumulated
error is very low and the calculated results are of a satisfactory level in
precision. Fig. 7. Averaged ul /a at different streamwise locations in the central plane, dashed lines
represent experimental data, whereas solid lines indicate simulation.
3.2. Code validation
h / d = 4.73q00.3 (x /d )0.3 (27)
The spray penetration height is an important parameter to evaluate
the mixing characteristic of the liquid jet with the free stream air. Fig. 5 Where h denotes the spray penetration height at the position of x from
shows the penetration height of the numerical result compared with injector nozzle, d is the nozzle diameter. It should be mentioned that
experimental data. The black dots in the illustration represent the the definition of the penetration height in experiment is bounded as
calculated instantaneous water droplets, the dashed line represents the locations where the liquid volume flux is 0.01 cc /(s⋅cm2 ). So, the present
averaged jet boundary of numerical result, and the solid line represents numerical result use the same threshold in consistent with the
the experimental correlation function of spray penetration height experiments. It can be seen that the spray penetration height from
developed by Lin et al. [8]: the calculation is in good agreement with the experiment on a wide

Fig. 5. Comparison of simulated spray penetration height with the experiment of Lin et al. [8] in the central plane.

338
P. Li et al. Acta Astronautica 134 (2017) 333–344

Fig. 8. Instantaneous snapshot of velocity contour in the central plane.

Fig. 9. Averaged distributions of gas velocity: (a) along the x-axis at different y positions, (b) along the y-axis at different x positions.

range, except in the near-injector region where the calculation is

Fig. 10. Streamlines of gas-phase and vectors of droplets in the central plane.
slightly higher than the experimental value.
Centerline distribution profiles of droplet properties in the free
stream direction can be normalized by its local penetration height to
obtain universal curves. Fig. 6 compares the normalized distribution
profiles of Averaged SMD (Sauter Mean Diameter) at different loca-
tions. In general, the calculation results of SMD are in good agreement
with the experimental results and reproduce the S-shaped distribution
observed in the experiment. Consistent with the experimental observa-
tions, the droplet SMD at the position of x=100 mm is smaller than
that of x=50 mm in the central region of spray, while greater than that
of x=50 mm in the near-wall region. The overall SMD at the position of
x=50 mm downstream from the injector nozzle is calculated as 14.8 μm
and agrees well with the experimental result which is 15.3 μm .
The normalized distribution profiles of averaged droplet velocity are
shown in Fig. 7. The droplet velocity ul is normalized by the sonic speed

Fig. 11. The instantaneous distribution of gas-liquid relative velocity in the central plane.

339
P. Li et al. Acta Astronautica 134 (2017) 333–344

Fig. 12. Averaged distributions of gas-liquid relative velocity: (a) along the x-axis at different y positions, (b) along the y-axis at different x positions.

Fig. 13. The instantaneous distribution of normalized droplet deformation in the central plane.

Fig. 14. Averaged distributions of normalized droplet deformation: (a) along the x-axis at different y positions, (b) along the y-axis at different x positions.

of the free stream to compare with the experimental data. For a fixed
position of x, the droplet velocity is larger at the upper edge and smaller
at the bottom. The droplet velocity at the position of x=100 mm is
obviously greater than that of x=50 mm, which is consistent with the
experiment. In general, the calculated results successfully predict the
entire distribution profile of droplet velocity even though the results are
slightly smaller than the experimental results.
The present results show that the jet spray structure, averaged
droplet size, and velocity distribution are in good agreement with the
experimental data, indicating that the physical models and numerical
methods are reliable.

340
P. Li et al.
Fig. 15. The liquid trailing phenomenon of jet spray revealed by simulation.
Fig. 16. The mechanism of the liquid trailing phenomenon.
3.3. Characteristics of gas-phase flow field
The characteristics of gas-phase flow field and liquid-phase flow
field may be the two key factors that lead to the liquid trailing
phenomenon of jet spray. They will be analyzed in this section and
the following section respectively.
In supersonic two-phase flow, the aerodynamic force provides the
driving force for the breakup and the movement of droplets, so the gas-
phase flow field has a critical influence on the evolution of liquid jet.
The gas-phase flow pattern determines the droplet movement pattern,
at the same time it is also significantly affected by the liquid jet. To
study the interaction between the two phases, it is necessary to identify
the essential characteristics of the gas-phase flow field firstly.
The instantaneous snapshot of velocity contour in the central plane
is given in Fig. 8. Apparently, a bow shock appears in the front of the
liquid jet. It is observed that the air flow velocity decreases significantly
after passing the liquid jet. There may be two reasons. One is the shock
which leads to a sharp decline in gas-phase velocity. The other is the
kinetic energy transfer from gas-phase field to liquid-phase field.
Fig. 9 shows the averaged results of gas-phase velocity which is
normalized by the free stream velocity. Fig. 9(a) gives the averaged
distributions of gas velocity in the central plane at different y positions.
It can be seen that the gas velocity decreases rapidly after the shock.
And the velocity has a stronger slowdown when the position gets closer
to the wall-region. After that, the gas velocity rebounds a little, and
then decreases rapidly due to the air stream flows into the liquid spray
region where gas-liquid momentum exchange is very strong. After the
droplets obtain a large enough momentum, the gas-liquid momentum
exchange becomes weaker, and then the gas velocity gradually recovers
to higher value. The second drop of gas velocity starts at the front
region of the spray, and the gas velocity at the leading edge of the spray
increases quickly with increasing y. Fig. 9(b) gives the averaged
distributions of gas-phase velocity in the central plane at different x
341
Acta Astronautica 134 (2017) 333–344
positions. The continuous decrease of the gas velocity from the free
stream to liquid spray zone can be seen clearly. But for the position of
x=25 mm and x=50 mm, the gas velocity acquires a local increase in
the near wall region, which is mainly caused by the effect of three-
dimensional flow around the liquid jet. In the middle region of the
spray, it can be clearly seen that the gas velocity increases with
increasing x.
To further explore the gas-liquid interaction process and under-
stand the mixing and dispersion of the liquid spray, Fig. 10 gives the
streamlines of gas-phase and velocity vectors of droplets in the central
plane. Due to the effect of bow shock, the streamlines are deflected
upward when they just encounter the spray zone. But before long, the
air stream flows obliquely downward and is deflected towards the near-
wall region by the effects of droplets. The velocity vectors of droplets
show that the droplets move obliquely upward. This result is also
confirmed in the experiment [60,61]. In the central plane, due to the
discrete characteristics of the liquid jet after breakup, the flow
streamlines may intersect with the trajectory of droplets, which is
different from the gas jet in supersonic crossflow [62,63]. Thus when
there is higher pressure at the front region of the spray and a relatively
low pressure inside the liquid spray, there will be a deflection tendency
for air flow towards the spray interior. The deflection of air flow will in
turn causes an aerodynamic force on droplets and lets the droplets
acquire an inclined downward acceleration. This may be the first basic
condition of the liquid trailing phenomenon.
3.4. Characteristics of the spray field
The aerodynamic force acting on the droplets determines the
evolution process and breakup process of the droplets, so it can be
used to deepen the understanding of mixing and distributing process of
droplets. The aerodynamic force acting on the droplets can be
expressed as Dl = (πdl2 Cd ρg Ug − Ul )/8, where the greatest influence
factor is the relative velocity Ug − Ul . Therefore, this relative velocity
can be used to analyze the strength of gas-liquid interaction and reflects
the potential position of breakup.
Fig. 11 shows the instantaneous distribution of gas-liquid relative
velocity in the central plane, where all droplets are colored and sized by
the relative velocity magnitude and the relative velocity is normalized
by the free stream velocity. Results show that the droplets with a
normalized relative velocity larger than 0.5 are mainly distributed in
the region of x=5–20 mm, and mainly in the leading edge of the spray
field. While in the inner region or far downstream region, the relative
velocity is relatively lower. Fig. 12 shows the averaged distributions of
gas-liquid relative velocity at different locations. For a fixed position of
y, the maximum of relative velocity always appears in the leading edge,
and these maximum values increase firstly with increasing y, reaching
their peak values around y=15 mm and then decrease. From the
longitudinal distribution, it can be seen that the relative velocity
profiles tend to be more uniform with increasing x, which indicates
that the gas-liquid interaction becomes weaker.
Larger deformation of droplets indicates larger possibility of break-
up. Fig. 13 shows the instantaneous distribution of normalized droplet
deformation in the central plane, where all droplets are colored and
sized by the droplet deformation. And Fig. 14 shows the averaged
distributions of normalized droplet deformation at different locations.
It indicates that the droplets with normalized deformation Q ≥ 0.8 are
mainly distributed in the region of x < 10 mm. For a fixed position of y,
the droplet deformation declines rapidly with increasing x, and
becomes stable while x > 10 mm. In the region of x > 10 mm, due to
the reduction of gas-liquid relative velocity, the droplet deformation
decreases gradually and becomes steady under the action of viscous
force and tension. This is consistent with the conclusion that the
atomization process mainly occurs within the region of x < 50 mm
noted by Lin et al. in his experimental study [8]. It can also be found
that the droplets with larger deformation are mainly distributed in the
P. Li et al.
Fig. 17. The evolution process of liquid trailing phenomenon ( Δt = 24μs ).
leading edge region of the spray field, while droplets in the inner region
of spray field have smaller deformation due to the lower relative
velocity and weaker interaction between the two phases.
The results of gas-liquid relative velocity and droplet deformation
show that the strongest gas-liquid interaction takes place at the front
region of the spray field, so does the breakup of droplets, which forms
the second basic condition of the liquid trailing phenomenon.
3.5. Liquid trailing phenomenon of jet spray
On the basis of the analysis above, We try to explain the formation
process of the liquid trailing phenomenon using numerical results.
Fig. 15 shows the streamlines of air flow and the distribution of droplet
in the central plane, where all droplets are colored and sized by their
diameter. The experimentally observed liquid trailing phenomenon of
jet spray is well captured. As shown in this figure, under the strong
influence of supersonic air flow, the liquid jet rapidly breaks up into
small droplets with different sizes, and the small droplet groups mainly
concentrate at the front region of the jet. On the other hand, the air
stream flows obliquely downward when they encounter the liquid spray
due to the effect of pressure difference. The inclined downward
movement of the air stream will let the droplets acquire an inclined
downward acceleration. Since the droplets after breakup are different
in size, and thus different in acceleration, the droplet groups shape into
a series of liquid trailing structures along the local air flow direction. To
show the trailing phenomenon of droplets clearly, the mechanism of
liquid trailing process is concluded in Fig. 16. The solid arrows in the
schematic diagram represent the direction of gas-phase while dotted
arrows represent the direction of droplet movement.
342
Acta Astronautica 134 (2017) 333–344
To explain the liquid trailing phenomenon in detail, the movements
of all child droplets from the same parent droplet are tracked and
labeled with the same notation namely ‘family number’. The distribu-
tion of child droplets with the same family number at different
moments is shown in Fig. 17. There are three characteristic droplet
families which are colored and sized by their diameter, and the
background is the corresponding instantaneous spray plume in the
central plane. This figure shows that many child droplets with different
sizes are formed after breakup of the parent droplet. Due to the inertia
effect, the child droplets still move obliquely upward. Based on the
above analysis, these child droplets will acquire an extra acceleration
under the influence of the local air flow, and its direction is inclined
downward. Since the sizes of child droplets are different, the smaller
droplets with better response characteristic move faster along the local
air flow direction than the larger droplets. As a result, all the child
droplets from the same parent droplet are arranged according to their
sizes along the direction of the local air flow as shown in Fig. 17.
Comparing the droplet distribution of different moments, it can be
found that the trailing angle of child droplets is gradually increasing,
and this is due to the fact that the larger droplets have a greater
inclined upward momentum, while the inertia of the smaller droplets is
relatively small, so that the smaller droplets are easier to move
obliquely downward with the local air flow. Consequently, the larger
droplets are mainly distributed in the outer region of the spray field,
and the smaller droplets mainly locate in the inner and bottom of the
spray field.
However, as the group of droplets moves downstream, the larger
droplets mainly locate in the outer region of the liquid spray where the
gas-liquid interaction is strong. Under the influence of air stream, the
P. Li et al.
larger droplets will further break up into small droplets. Whereas, the
smaller droplets mainly locate in the center of the spray field where the
gas-liquid interaction is weaker, so they hardly experience further
breakup. Consequently, in the downstream region, the relatively larger
droplets within the same droplet family mainly locate in the central
region, and finally result in the SMD profile shown in Fig. 6.
4. Conclusions
This research mainly focuses on the interaction process between a
liquid jet and a supersonic crossflow. The Eulerian-Lagrangian ap-
proach and modified breakup model were adopted to numerically study
the droplet deformation, breakup and mixing in supersonic crossflow.
Significant conclusions can be summarized as follows:
(1) The drag coefficient and the spray breakup model were modified by
considering the compressible effects and the deformation of
droplets. It was found that the calculated spray features, including
spray penetration, droplet size distribution and droplet velocity
profile agree reasonably well with the experiment, which indicates
that the physical models and numerical methods are reliable.
(2) The gas-phase velocity declines twice when the air flow passes
through the shock and the spray zone. The streamlines of air flow
could intersect with the trajectory of droplets and are deflected
towards the near-wall region after they enter into spray zone
around the central plane. This is the first basic condition of the
liquid trailing phenomenon.
(3) The strongest gas-liquid interaction process occurs around the
front region of the spray where the gas-liquid relative velocity and
the normalized droplet deformation all reach their maximum
value. Then, the gas-liquid relative velocity decreases gradually
with increasing x, and the interaction between air flow and
droplets weakens, resulting in a gradual decrease of the normalized
droplet deformation. The results of gas-liquid relative velocity and
droplet deformation show that the breakup of droplets mainly
occurs around the front region of the spray where gathered a large
number of droplets with different sizes, which forms the second
basic condition of the liquid trailing phenomenon.
(4) The calculated results successfully revealed the liquid trailing
phenomenon of jet spray which has been discovered by the
previous experiment. Under the influence of local air flow, all the
child droplets from the same parent droplet are arranged accord-
ing to their sizes, and the direction of arrangement is similar to the
local air flow. In the process of downstream movement, the trailing
angle of the child droplets group is increasing, and the larger
droplets located external will further break up due to the strong
gas-liquid interaction. These results are useful for understanding
the mixing process of liquid fuels in supersonic air flow.
Acknowledgments
This work was supported by the National Natural Science
Foundation of China under Grant No. 51406232, No. 11472305. The
computation was performed using the computer facilities provided by
the National Supercomputing Tianjin Center.
References
[1] J.P. Waltrup, Upper bounds on the flight speed of hydrocarbon-fueled scramjet-
powered vehicles, J. Propuls. Power 17 (2001) 1199–1204.
[2] T.H. Chen, C.R. Smith, D.G. Schommer, Multi-zone behavior of transverse liquid
jet in high-speed flow, in: Proceedings of the 31st Aerospace Sciences Meeting and
Exhibit, 1993.
[3] A. Sherman, J. Schetz, Breakup of liquid sheets and jets in a supersonic gas stream,
AIAA J. 9 (1971) 666–673.
[4] R.H. Thomas, J.A. Schetz, Distributions across the plume of transverse liquid and
slurry jets in supersonic airflow, AIAA J. 23 (1985) 1892–1901.
[5] P.-K. Wu, K.A. Kirkendall, R.P. Fuller, A.S. Nejad, Breakup processes of liquid jets
343
Acta Astronautica 134 (2017) 333–344
in subsonic crossflows, J. Propuls. Power 13 (1997) 64–73.
[6] K.-C. Lin, P.J. Kennedy, Spray penetration heights of angle-injected aerated-liquid
jets in supersonic crossflows, in: Proceedings of the 38th Aerospace Sciences
Meeting and Exhibit, 2000.
[7] K.-C. Lin, P.J. Kennedy, T.A. Jackson, Penetration heights of liquid jets in high-
speed crossflows, in: Proceedings of the 40th AIAA Aerospace Sciences Meeting and
Exhibit, 2002.
[8] K.-C. Lin, P.J. Kennedy, T.A. Jackson, Structures of water jets in a Mach 1.94
supersonic crossflow, in: Proceedings of the 42nd AIAA Aerospace Sciences
Meeting and Exhibit, 2004.
[9] J. Lee, K.A. Sallam, K.-C. Lin, C.D. Carter, Spray structure in near-injector region of
aerated jet in subsonic crossflow, J. Propuls. Power 25 (2009) 258–266.
[10] D.S. Olinger, K.A. Sallam, K.-C. Lin, C.D. Carter, Digital holographic analysis of the
near field of aerated-liquid jets in crossflow, J. Propuls. Power (2014) 1–10.
[11] L.S. Fei, S.L. Xu, C.J. Wang, Q. Li, S.H. Huang, Experimental study on atomization
phenomena of kerosene in supersonic cold flow, Sci. China Ser. E: Technol. Sci. 51
(2008) 145–152.
[12] L.Y. Wu, Z.G. Wang, Q.L. Li, C. Li, Study on transient structure characteristics of
round liquid jet in supersonic crossflows, J. Vis. 19 (2016) 1–5.
[13] E. Meillot, S. Vincent, C. Caruyer, D. Damiani, J.P. Caltagirone, Modelling the
interactions between a thermal plasma flow and a continuous liquid jet in a
suspension spraying process, J. Phys. D: Appl. Phys. 46 (2013) 1–11.
[14] F. Xiao, Large eddy simulation of liquid-jet primary breakup in air crossflow, AIAA
J. 51 (2013) 2878–2893.
[15] M. Herrmann, M. Arienti, M. Soteriou, The impact of density ratio on the liquid
core dynamics of a turbulent liquid jet injected into a crossflow, J. Eng. Gas.
Turbines Power 133 (2011) 1–9.
[16] N. Patel, S. Menon, Simulation of spray–turbulence–flame interactions in a lean
direct injection combustor, Combust. Flame 153 (2008) 228–257.
[17] M. Jangi, R. Solsjo, B. Johansson, X.-S. Bai, On large eddy simulation of diesel
spray for internal combustion engines, Int. J. Heat. Fluid Flow. 53 (2015) 68–80.
[18] A. Irannejad, F. Jaberi, Numerical study of high speed evaporating sprays, Int. J.
Multiph. Flow. 70 (2015) 58–76.
[19] F. Genin, S. Menon, LES of supersonic combustion of hydrocarbon spray in a
scramjet, in: Proceedings of the 40th AIAA/ASME/SAE/ASEE Joint Propulsion
Conference and Exhibit, 2004.
[20] S.V. Apte, K. Mahesh, P. Moin, J.C. Oefelein, Large-eddy simulation of swirling
particle-laden flows in a coaxial-jet combustor, Int. J. Multiph. Flow. 29 (2003)
1311–1331.
[21] X. Gu, S. Basu, R. Kumar, Dispersion and vaporization of biofuels and conventional
fuels in a crossflow pre-mixer, Int. J. Heat. Mass Transf. 55 (2012) 336–346.
[22] K.-S. Im, K.-C. Lin, M.-C. Lai, M.S. Chon, Breakup modeling of a liquid jet in cross
flow, Int. J. Automot. Technol. 12 (2011) 489–496.
[23] A. Irannejad, F. Jaberi, Large eddy simulation of turbulent spray breakup and
evaporation, Int. J. Multiph. Flow. 61 (2014) 108–128.
[24] W.P. Jones, A.J. Marquis, K. Vogiatzaki, Large-eddy simulation of spray combus-
tion in a gas turbine combustor, Combust. Flame 161 (2014) 222–239.
[25] C.B. Henderson, Drag coefficients of spheres in continuum and rarefied flows, AIAA
J. 14 (1976) 707–708.
[26] P. Moin, S.V. Apte, Large-eddy simulation of realistic gas turbine combustors, AIAA
J. 44 (2006) 698–708.
[27] P.J. O'Rourke, A.A. Amsden, The Tab method for numerical calculation of spray
droplet breakup, in: International Fuels and Lubricants Meeting and Exposition,
1987.
[28] R. Reitz, R. Diwakar, Structure of high-pressure fuel sprays, in: SAE Technical
Paper Series 870598, 1987.
[29] A.B. Liu, D. Mather, R.D. Reitz, Modeling the effects of drop drag and breakup on
fuel sprays, in: International Congress and Exposition, 1993.
[30] M.A. Patterson, R.D. Reitz, Modeling the effects of fuel spray characteristics on
diesel engine combustion and emissions, in: SAE Technical Paper Series 980131,
1998.
[31] Z.G. Wang, L.Y. Wu, Q.L. Li, C. Li, Experimental investigation on structures and
velocity of liquid jets in a supersonic crossflow, Appl. Phys. Lett. 134102 (2014)
1–4.
[32] Y.H. Zhao, J.H. Liang, Y.X. Zhao, Non-reacting flow visualization of supersonic
combustor based on cavity and cavity-strut flameholder, Acta Astronaut. 121
(2016) 282–291.
[33] H.B. Wang, Z.G. Wang, M.B. Sun, N. Qin, Large eddy simulation based studies of
jet–cavity interactions in a supersonic flow, Acta Astronaut. 93 (2014) 182–192.
[34] J. Pozorski, S.V. Apte, Filtered particle tracking in isotropic turbulence and
stochastic modeling of subgrid-scale dispersion, Int. J. Multiph. Flow. 35 (2009)
118–128.
[35] S.V. Apte, K. Mahesh, M. Gorokhovski, P. Moin, Stochastic modeling of atomizing
spray in a complex swirl injector using large eddy simulation, Proc. Combust. Inst.
32 (2009) 2257–2266.
[36] G.S. Jiang, C.W. Shu, Efficient implementation of Weighted ENO schemes, J.
Comput. Phys. 126 (1996) 917–923.
[37] H.B. Wang, M.B. Sun, N. Qin, H.Y. Wu, Z.G. Wang, Characteristics of oscillations in
supersonic open cavity flows, Flow., Turbul. Combust. 90 (2013) 121–142.
[38] H.B. Wang, N. Qin, M.B. Sun, H.Y. Wu, Z.G. Wang, A hybrid LES (large eddy
simulation)/assumed sub-grid PDF (probability density function) model for
supersonic turbulent combustion, Sci. China Technol. Sci. 54 (2011) 2694–2707.
[39] H.B. Wang, Z.G. Wang, M.B. Sun, N. Qin, Numerical study on supersonic mixing
and combustion with hydrogen injection upstream of a cavity flameholder, Heat.
Mass Transf. 50 (2014) 211–223.
[40] H.B. Wang, Z.G. Wang, M.B. Sun, N. Qin, Combustion characteristics in a
P. Li et al.
supersonic combustor with hydrogen injection upstream of cavity flameholder,
Proc. Combust. Inst. 34 (2013) 2073–2082.
[41] M.B. Sun, H. Geng, J.H. Liang, Z.G. Wang, Flame characteristics in supersonic
combustor with hydrogen injection upstream of cavity flameholder, Proc. Inst.
Mech. Eng., Part G: J. Aerosp. Eng. 24 (2008) 688–696.
[42] N.N. Smirnov, V.B. Betelin, A.G. Kushnirenko, V.F. Nikitin, V.R. Dushin,
V.A. Nerchenko, Ignition of fuel sprays by shock wave mathematical modeling and
numerical simulation, Acta Astronaut. 87 (2013) 14–29.
[43] C. Crowe, M. Sommerfeld, Y. Tsuji, Multiphase Flows With Droplets and Particles
(FL), CRC Press, Boca Raton, 1998.
[44] S.H. Yang, J.L. Le, W. He, L. Chen, Fuel atomization and droplet breakup models
for numerical simulation of spray combustion in a scramjet combustor, in:
Proceedings of the 18th AIAA/3AF International Space Planes and Hypersonic
Systems and Technologies Conference, 2012.
[45] R. Chorda, J.A. Blasco, N. Fueyo, An efficient particle-locating algorithm for
application in arbitrary 2D and 3D grids, Int. J. Multiph. Flows 28 (2002)
1565–1580.
[46] E. Shams, J. Finn, S.V. Apte, A numerical scheme for Euler-Lagrange simulation of
bubbly flows in complex systems, Int. J. Numer. Methods Fluids 1 (2010) 1–41.
[47] J. Capecelatro, O. Desjardins, An Euler–Lagrange strategy for simulating particle-
laden flows, J. Comput. Phys. 238 (2013) 1–31.
[48] H.P. Trinh, C.P. Chen, M.S. Balasubramanyam, Numerical simulation of liquid jet
atomization including turbulence effects, J. Eng. Gas. Turbines Power 129 (2007)
920–928.
[49] F.X. Tanner, G. Weisser, Simulation of liquid jet atomization for fuel sprays by
means of a cascade drop breakup model, in: International Congress and Exposition,
1998.
[50] S. Hossainpour, A.R. Binesh, Investigation of fuel spray atomization in a DI heavy-
duty diesel engine and comparison of various spray breakup models, Fuel 88 (2009)
799–805.
[51] T.G. Theofanous, G.J. Li, T.N. Dinh, C.-H. Chang, Aerobreakup in distrubed
subsonic and supersonic flow fields, J. Fluid Mech. 593 (2007) 131–170.
[52] G.J. Li, T.N. Dinh, T.G. Theofanous, An experimental study of droplet breakup in
supersonic flow: the effect of long-range interactions, in: Proceedings of the 42nd
344
Acta Astronautica 134 (2017) 333–344
AIAA Aerospace Sciences Meeting and Exhibit, 2004.
[53] Y. Kim, J.C. Hermanson, Breakup and vaporization of droplets under locally
supersonic conditions, Phys. Fluids 24 (2012) 1–24.
[54] N.N. Smirnov, V.F. Nikitin, V.R. Dushin, Y.G. Filippov, V.A. Nerchenko, J. Khadem,
Combustion onset in non-uniform dispersed mixtures, Acta Astronaut. 115 (2015)
94–101.
[55] M. Pilch, C.A. Erdman, Use of breakup time data and velocity history data to
predict the maximum size of stable fragments for acceleration-induced breakup of a
liquid drop, Int. J. Multiph. Flow. 13 (1987) 741–757.
[56] V.B. Betelin, N.N. Smirnov, V.F. Nikitin, V.R. Dushin, A.G. Kushnirenko,
V.A. Nerchenko, Evaporation and ignition of droplets in combustion chambers
modeling and simulation, Acta Astronaut. 70 (2012) 23–35.
[57] Y. Tsuji, Y. Morikawa, T. Tanaka, Numerical simulation of gas-liquid two-phase
flow in a two-dimensional horizontal channel, Int. J. Multiph. Flow. 13 (1987)
671–684.
[58] N.N. Smirnov, V.B. Betelin, V.F. Nikitin, L.I. Stamov, D.I. Altoukhov, Accumulation
of errors in numerical simulations of chemically reacting gas dynamics, Acta
Astronaut. 117 (2015) 338–355.
[59] N.N. Smirnov, V.B. Betelin, R.M. Shagaliev, V.F. Nikitin, I.M. Belyakov,
Y.N. Deryuguin, S.V. Aksenov, D.A. Korchazhkin, Hydrogen fuel rocket engines
simulation using LOGOS code, Int. J. Hydrog. Energy 39 (2014) 10748–10756.
[60] R. Ragucci, A. Picarelli, G. Sorrentino, P.D. Martino, Validation of droplets
behavior model by means of PIV measurements in a cross-flow atomizing system,
in: Proceedings of the 32nd Meeting on Combustion in Italian Section of the
Combustion Institute, 2009.
[61] Y. Gopala, B. Oleksandr, E. Lubarsky, B.T. Zinn, Liquid jet in crossflow-measure-
ment of the velocity field of the near field continuous medium, in: Proceedings of
the 47th AIAA Aerospace Sciences Meeting, 2009.
[62] X.C. Chai, K. Mahesh, Simulations of high speed turbulent jets in crossflow, in:
Proceedings of the 49th AIAA Aerospace Sciences Meeting including the New
Horizons Forum and Aerospace Exposition, 2011.
[63] S. Kawai, S.K. Lele, Dynamics and mixing of a sonic jet in a supersonic turbulent
crossflow, in: Annual Research Briefs, Center for Turbulence Research, 2009.