Combustion and Flame 237 (2022) 111742

Contents lists available at ScienceDirect

Combustion and Flame

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

A flamelet LES of turbulent dense spray flame using a detailed

high-resolution VOF simulation of liquid fuel atomization
Jian Wen a,∗, Yong Hu b, Takayuki Nishiie c, Jun Iino c, Assaad Masri d, Ryoichi Kurose a
a
Department of Mechanical Engineering and Science, Kyoto University, Kyoto daigaku-Katsura, Nishikyo-ku, Kyoto 615-8540, Japan
b
State Key Laboratory of Fire Science, University of Science and Technology of China, Hefei, Anhui 230026, China
c
Numerical Flow Designing Ltd., Higashigotanda, 1-10-10, Shinagawa City, Tokyo 141-0022, Japan
d
School of Aerospace Mechanical and Mechatronic Engineering, The University of Sydney, NSW 2006, Australia

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

Article history: A numerical framework used to model dense spray flames is proposed. In this framework, the liquid
Received 8 May 2021 fuel (acetone) atomization is solved by a detailed high-resolution VOF simulation, and the Eulerian com-
Revised 6 September 2021
ponents of liquid droplets are transformed into Lagrangian droplets, which are stored in a database at a
Accepted 6 September 2021
certain downstream cross-section. Then, the combustion process is solved by a LES/FPV (flamelet progress
Available online 28 September 2021
variable) adopting the pre-stored database of Lagrangian droplets (i.e., the position, size, and velocity
Keywords: of each droplet) as the inlet boundary conditions. This framework is a one-way coupling between a
Atomization VOF simulation and a combustion simulation. The validity of this approach is investigated by comparing
Dense spray combustion the computations with the experiments of the Sydney Piloted Needle Spray Burner. The VOF simulation
LES shows that the volume flux of the droplets at the nozzle exit fluctuates both temporally and spatially
Flamelet and the larger droplets tend to be located away from the center axis compared to the small droplets.
E-L transformation
The computed breakup length is in good agreement with the empirical correlation. In the database of
Needle spray burner
the Lagrangian droplets for the LES/FPV of spray flames, the location of the sampling cross-section, the
sampling time, and the threshold value for Eulerian–Lagrangian (E-L) transformation strongly affect the
properties of the Lagrangian droplets, and are critical for the successful use of the LES/FPV. Two spray
flames with different recess distances are computed using their optimal pre-stored droplets databases
and both show generally good agreement with the experiments in terms of the gas temperature and
droplet size distributions. The spray flame with a longer recess distance, which is more representative of
a dilute spray, is considered to have a longer and wider premixed core than that with a shorter recess
distance representing a dense spray. The discrepancy in the prediction of denser spray flames becomes
more evident leading to over-predictions of gas temperature further downstream. Reasons for this behav-
ior are discussed in the text.
© 2021 The Authors. Published by Elsevier Inc. on behalf of The Combustion Institute.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)

1. Introduction ing up with combustion. Therefore, the complexity involved makes

it difficult to clarify the detailed mechanism and relevant models
Owing to the increasing concerns regarding global warming underlying spray atomization and combustion.
and a shortage of energy, it is important for energy generators to As mentioned by Masri [1], studies on spray combustion are
achieve a high combustion efficiency and extremely low emissions. generally classified into dense and dilute sprays. In the latter case,
Therefore, liquid fuel spray combustion, which is widely used in the spray dynamics associated with atomization is neglected. In
gas turbine, gasoline, diesel, and rocket engines, merits detailed in- a dense spray region where atomization occurs, many generated
vestigations. Spray combustion includes extremely complex phys- droplets make it difficult for experimental diagnostics to acquire
ical phenomena, starting from the liquid fuel atomization along sufficient information [2], thereby leaving the atomization process
with the droplet evaporation and evaporated fuel-air mixing, end- unclarified, and hence the dense spray region remains vague nu-
merically [e.g., 3–30] and experimentally [e.g., 31,32–37,38]. For
∗
dilute spray flames, fuel sprays are regarded as a cluster of indi-
Corresponding author.
vidual droplets and are then solved by the Lagrangian approach to
E-mail address: [email protected] (J. Wen).

https://doi.org/10.1016/j.combustflame.2021.111742
0010-2180/© 2021 The Authors. Published by Elsevier Inc. on behalf of The Combustion Institute. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

consider an interaction with the surrounding gas phase. However, gas phases, respectively. The compressive velocity is considered
the Lagrangian approach still has difficulty in providing the initial only for the gas-liquid interface in the normal direction to avoid
droplet size distribution near the nozzle exit owing to the abun- a dispersion of the VOF value. This additional compressive term
dance of interplayed phenomena, namely, the processes of primary helps retain the mass conservation and convergence for the VOF
breakup, secondary breakup, and droplet coalescence/collision [e.g., advection. It also facilitates the simulation of a multiphase flow
39,40–46,47]. with a large liquid/gas density ratio. Because the present study is
To investigate such spray combustion located in the dense performed in cylindrical coordinates with unstructured grids, the
regime, Masri et al. [e.g., 48,49,50] at the University of Sydney HRIC scheme is implemented instead of the coupled level-set and
designed a canonical platform that can supply various spray inlet VOF method, which avoids the unnecessary complexity induced by
conditions, which is called the Sydney Piloted Needle Spray Burner the use of the level-set method.
(referred to as the Sydney Burner, hereafter) to stabilize repeat- A tagging method [56] is then utilized to transform the Eule-
able turbulent spray flames by placing two concentric tubes within rian liquid parts at a specific downstream cross-section into the
the pilot annulus. By varying the recess distance, which refers to Lagrangian spherical droplets with the droplet properties such as
the distance between the liquid fuel jet nozzle to the pilot out- the position, size, and velocity, which are saved in a database (re-
let, different types of sprays can be reproduced. However, there ferred to as E-L tagging and E-L transformation, respectively, here-
haven’t been any attempts to use a numerical simulation to cou- after). Then, the stored Lagrangian droplets are utilized as the inlet
ple the combustion process with the beginning atomization pro- boundary conditions for the combustion process.
cess together.
The purpose of this study is, therefore, to propose a numerical 2.2. Governing equations for dilute spray region
framework to model the coupling of atomization and combustion
of dense spray flames while maintaining reasonable computational The combustion process occurring in the dilute spray region is
cost. Results are compared with experimental data obtained from modeled by a LES, utilizing the governing equations for the mass,
the Sydney needle burner [49,50]. The concept is as follows. The momentum, and energy, and the detailed information can be found
liquid fuel atomization is solved by a detailed numerical simula- in studies [24,28,30]. To include the detailed chemical kinetics, a
tion, in which both continuum gas and liquid phases are strictly non-adiabatic version of flamelet/progress variable approach (FPV)
solved in a Eulerian framework, and the Eulerian components of [52,57] is used for the modeling of the turbulence-chemistry inter-
the liquid droplets are transformed into the Lagrangian droplets action, which results in the solution of the following equations,
at a certain downstream cross-section, i.e., sampling cross-section,
∂ ρ̄
whose information is stored in database. Then, the combustion +  · (ρ̄ u
˜ ) = Sρ , (4)
process is solved by a large eddy simulation (LES) with a flamelet ∂t
model adopting the pre-stored database of Lagrangian droplets,
namely, by a one-way coupling between a VOF simulation and a ∂ ρ̄ u˜
+  · (ρ̄ u
˜u˜ ) = −  P̄ +  · (τ̄ + τ̄ sgs ) + Sρ u , (5)
combustion simulation. ∂t
2. Mathematical models
∂ ρ̄ h˜ h
+  · (ρ̄ u
˜h˜ ) =  · [ρ̄ (D ˜ )] + qh + Qrad + Sρ h , (6)
In this work, all computations are performed using an unstruc- ∂t h

tured LES solver, i.e., the FrontFlow/Red extended by Kyoto Univer-

sity [23,51,52]. ∂ ρ̄ Z˜ Z  Z˜ )] + qZ + Sρ Z ,
+  · (ρ̄ u
˜ Z˜ ) =  · [ρ̄ (D (7)
2.1. Governing equations for dense spray region
∂t

In the dense spray region where the atomization process oc- ∂ ρ̄C˜ C  C˜ )] + qC + ρ̄ ω
+  · (ρ̄ u
˜ C˜ ) =  · [ρ̄ (D ˜˙ C , (8)
curs, liquid and gas continuum phases are treated as incompress- ∂t
ible fluids and are both solved in a Eulerian framework. Their gov- where, the overbar, −, denotes the filtered mean value, and the
erning equations solved in this region include the conservation tilde, ∼, denotes the Favre averaged value. τsgs = ρ̄ (u ˜ −u
˜u u ) is the
equations of mass and momentum as follows, subgrid term of the stress tensor, h the enthalpy, Z the mixture
ρ  ·u = 0, (1) fraction. The mixture fraction is defined as the mass fraction of fuel
stream, such that Z = 1 means a fuel stream and Z = 0 means an
  oxidizer stream. Following the work [52,57], the progress variable
∂u C is defined as the summation of combustion products, i.e., C =
ρ + u · u = −  P +  · (2μS ) + F σ + g, (2)
∂t YH2 O + YH2 + YCO2 + YCO . Y is the mass fraction of chemical species.
Dh , DZ , and DC are diffusion coefficients of h, Z, and C, respectively.
Here, ρ is the local density, u the velocity vector, P the pressure,
Dh is the gaseous thermal diffusivity given by Dh = λ/(ρ c p ), DZ
μ the viscosity, S the rate-of-strain tensor Si j ≡ (∂i u j + ∂ j ui )/2, and
and DC are obtained by assuming the unity Lewis number. λ is
F σ is the source term of surface tension calculated by the contin-
the heat conductivity, c p is the specific heat capacity at constant
uum surface force (CSF) model [53].
pressure. qh , qZ , and qC are the subgrid-scale (SGS) scalar fluxes,
The high-resolution interface capturing (HRIC) scheme [54] is
qφ = ρ̄ (u ˜ φ˜ − u
φ ), (φ = h, Z, C ). Qrad is the radiation heat loss mod-
implemented into the volume of fluid (VOF) method in order to
eled by the weighted sum of gray gases (WSGG) model [58]. ω˙ C is
capture the gas-liquid interface and the atomization process, where
the source term of reaction progress variable. The eddy viscosity
the VOF advection function takes the following form Albadawi
approximation is used to determine the τ and q as follows,
et al. [55],
∂ψ τ̄sgs = μt [(u˜ ) + (u˜ )T )], (9)
+ u · ψ + um · ∇ [(1 − ψ )(ψ )] = 0, (3)
∂t
where, ψ is the VOF value within each grid, and um = ul − ug is
the compressive velocity. The subscripts l and g refer to liquid and qφ = ρ̄αt  φ˜ , (10)

2
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

where, μt and αt denote the turbulent viscosity and eddy diffu- d ud f1

= (u˜ − ud ), (20)
sivity, respectively, and are generally related in the formulation as dt τd
αt = μt /(ρ̄ Sc ) with a constant Schmidt number of Sc = 0.4 [59],
where μt is determined by the dynamic Smagorinsky–Lilly model
    
dTd Nu c˜p f2 1 dmd LV
[60]. = (T˜ − Td ) + , (21)
dt 3P r c p,d τd md dt c p,d
A non-adiabatic flamelet/progress variable approach (FPV),
which can consider the effect of the heat loss caused by the la-
dmd Sh md
tent heat of spray vaporization and radiation, is used. In order to =− ln(1 + BM ), (22)
generate the flamelet library, the following flamelet equations de-
dt 3Sc τd
pending on the unity Lewis number assumption with heat loss are Here, f1 and f2 are the correction coefficients for the Stokes
solved as follows, drag and heat transfer for the evaporating fuel droplet, τd is the
particle response time [13,63], T the gas temperature, LV the la-
∂ Yk ρχ ∂ 2Yk
ρ − − ω˙ k = 0, (11) tent heat of evaporation at Td , c p and c p,d the specific heat of
∂t 2 ∂ Z2 gas and fuel droplet, the Nusselt number and Prandtl number
     Nu = 2 + 0.522Re1sl/2 P r 1/3 and P r = μc p /λ, the Sherwood number
∂ T ρχ ∂ 2T 1 ∂ cp∂ T  ρχ ∂ Yk Yk ∂ W c p,k ∂T
ρ − + + + 1− and Schmidt number Sh = 2 + 0.552Re1sl/2 Sc1/3 and Sc = μ/(ρ Dk ),
∂t 2 ∂ Z2 c p ∂ Z∂ Z 2 ∂Z W ∂Z cp ∂Z
k the mass transfer number BM = (YF,s − YF )/(1 − YF,s ). The detailed
1  information of the droplet evaporation model can be found in our
+ hk ω˙ k + qloss = 0, (12)
cp previous studies [24,46,64,65]. The employed secondary breakup
k
model is the Taylor analogy breakup (TAB) model [66].
α
qloss = − hk ω˙ k , (13)
cp 3. Computational setup
k

where, the subscript k denotes the chemical species, χ is the scalar 3.1. Computational domains
dissipation rate, ω˙ k the reaction rate of species k, T the gas tem-
perature, W the mean molecular weight of mixture, c p,k the spe- A numerical framework is proposed in the present study, where
cific heat capacity of species k at constant pressure, qloss the heat the spray atomization and the subsequent combustion are simu-
loss, α the heat loss rate parameter which can be varied from 0 to lated using different methods following an assumption that these
1. Then a four dimensional flamelet library is obtained as, two sub-processes usually occur in separated domains. Therefore,
ϕ˜ = ϕ˜ (Z˜ , Z h ),
2 , C˜,  (14) two computational domains are adopted and are marked as atom-
ization and combustion domains, respectively, as shown in Fig. 1.
where, Z 2 is the variance of mixture fraction, h is the enthalpy In Fig. 1(a), the part depicted by the black line is the Sydney
defect due to heat loss, ϕ is the flame properties such as gas tem- Burner, which consists of two concentric tubes, 1 and 2, sur-
perature, species mass fraction, and reaction rate. Here, the adia- rounded by a pilot tube, 3. The inner tube, 1, is the liquid fuel noz-
batic enthalpy ha is calculated by Eq. (16) by assuming h = 0, i.e., zle with an inner diameter of Dl = 0.686 mm and wall thickness
ha = h˜ (Z˜ , Z h = 0 ), and thus the enthalpy defect h can be
2 , C˜, 
of 0.381 mm, and the outer tube, 2, is the air stream nozzle for
calculated by h = ha − h, where h is determined with Eq. (6). the liquid fuel atomization with an inner diameter of Dg = 10 mm
The influence of the evaporating droplets on the carrier gas and wall thickness of 0.5 mm. The pilot tube, 3, is used to sup-
flow is considered using the Particle-Source-In Cell (PSI-Cell) ply the hot combustion products through an inner diameter of
method [61]. Sρ , Sρ u , Sρ h , and Sρ Z , which are the source terms for D p = 25 mm and wall thickness of 0.2 mm. The concentric tubes
the mass, momentum, enthalpy, and mixture fraction originating are adjustable such that the distance from the liquid fuel nozzle to
from the dispersed droplets, respectively, are given as follows, the pilot outlet is variable, enabling the Sydney Burner to supply
1  dmd a dense or dilute spray for the combustion. The distance is called
Sρ = − , (15) the recess distance and is referred to as Lr in the present and re-
V dt
N lated studies [e.g., 48,49,50], ranging from 0 to 80 mm. The present
study selected two different flame cases, N-AF8-25 and N-AF8-80
1  dmd ud which hold different Lr values, Lr = 25 and 80 mm, respectively.
Sρ u = − , (16)
V dt However, for the atomization computation, the case of Lr =
N
80 mm requires a much higher computational cost than that of
1  dmd hd Lr = 25 mm. Therefore, only the N-AF8-25 case atomization pro-
Sρ h = − , (17) cess is calculated, which is depicted by the solid and dotted blue
V dt
N lines in Fig. 1(a), and is shown in Fig. 1(b) in detail. The atom-
ization domain consists of a region with a wall boundary having
1  dmd a length of 25 mm corresponding to the experimental recess dis-
Sρ Z = − for fuel, (18)
V dt tance and a free boundary of 25 mm. Two atomization-combustion
N
coupling (ACC) planes are set in the atomization domain. These are
where, V is the volume of the unit grid, N is the number of
utilized to transform the Eulerian components into the Lagrangian
droplets in the grid, and md , ud , and hd are the mass, velocity, and
droplets and serve as the inlet boundary conditions for the com-
specific enthalpy of the droplet, respectively.
bustion simulation. To investigate the influence of different ACC
Considering the non-equilibrium Langmuir–Knudsen evapora-
positions on the E-L transformation, one plane is set at 5 mm up-
tion model [24,62], the governing equations used to track the
stream of the pilot outlet (ACC1) and the other is 5 mm down-
droplet profiles such as the position, xd , velocity, ud , temperature,
stream of the pilot outlet (ACC2).
Td , and mass, md , are given,
The combustion domain is depicted by the red line in Fig. 1(a),
dxd the details of which are presented in Fig. 1(c). The entire combus-
= ud , (19)
dt tion domain holds a diameter of Dc = 104 mm. In the axial direc-
tion, the inlet boundary is placed at the ACC plane, whose distance

3
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 1. Schematics of computational domains.

4
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 1. Continued

to the pilot outlet is marked as L1 . Since the recess distance of N- Table 1

Parameters for atomization simulation.
AF8-25 and N-AF8-80 are adjusted to 25 and 80 mm to produce
dense and dilute sprays, L1 holds two values of 10 and 65 mm Physical properties
for N-AF8-25 and N-AF8-80, respectively, and the other parameters Liquid fuel Acetone
L2 = 10 0 0 mm, L3 = 75 mm, and L4 = 135 mm are kept the same Gas&Air
in both cases. Liquid nozzle diameter, Dl (mm) 0.686
The cylindrical coordinate system with the unstructured grids Liquid jet velocity, ul (m/s) 2.57
Liquid jet viscosity, μl (Pa s) 3.33 × 10−4
is utilized for both atomization and combustion simulations. The
Liquid jet density, ρl (kg/m3 ) 786
atomization computational domain has a total of 24.6 million grid Liquid jet temperature, Tl (K) 300
points, with a non-uniform mesh size ranging from 8 to 100 μm Liquid jet Reynolds number, Rel (−) 4161
in the radial direction, and a uniform mesh size of 100 μm in Gas jet diameter, Dg (mm) 10
Gas jet velocity, ug (m/s) 48
the axial direction. The combustion computational domain has a
Gas jet viscosity, μg (Pa s) 1.81 × 10−5
total of 18 million grid points, with an increasing mesh size of Gas jet density, ρg (kg/m3 ) 1.20
0.13 to 3 mm in the radial direction, and a variable mesh size of Gas jet temperature, Tg (K) 300
0.15 to 3.5 mm in the axial direction. In addition, the Hinze scale Gas jet Reynolds number, Reg (−) 31, 823
ηH = σ /(ρgUg2 ) in the atomization computation is estimated to be Liquid-gas surface tension, σ (N/m) 2.37 × 10−2
Ambient pressure, P (MPa) 0.1
8.54 μm and the mesh size holds 1–10 ηH . As suggested in our
Aerodynamic Weber number, We (−) 80
previous study [46] and regarding this simulation reaches to the
experimental scale, it could be considered as a high-resolution VOF
simulation.

3.2. Computational conditions experimental conditions, as shown in Thomas and Lowe’s works
[49,50]. The parameters used in the atomization computation are
For the atomization computation, the temperature and pres- shown in Table 1.
sure are set to room temperature and atmospheric pressure, and After the atomization computation, the droplets are trans-
hence, both the liquid fuel and air have a temperature of 300 K, formed into the Lagrangian droplets, which are stored in a
which can be regarded as a cold state for only atomization, ne- database for the following combustion computation. Droplets
glecting the evaporation effect. Therefore, the air viscosity, μg , is recorded at the ACC plane of the atomization domain are directly
1.81 × 10−5 Pa s, and liquid viscosity, μl , is 3.33 × 10−4 Pa s, injected into the combustion domain at the ACC plane at a fixed
the liquid-gas surface tension, σ , is 2.37 × 10−2 N/m. At the inlet, time step determined by satisfying the step interval of the com-
both the liquid and carrier gas velocities, ul and ug , are assigned bustion computation, which are explained in the following section.
as 2.57 m/s and 48 m/s with a flat laminar velocity profile, respec- When these cold Lagrangian fuel droplets flow out of the nozzle
tively, according to the experiments [49,50]. Thus, in both config- and face the hot pilot gases, combustion occurs after the evapora-
urations, i.e., N-AF8-25 and N-AF8-80, the dimensionless param- tion and mixing with the combustion products from the pilot that
eters, including the aerodynamic Weber number, W e = ρg u2g Dl /σ , is in the stoichiometric condition and has a velocity of 1.5 m/s.
evaluated as 80, the liquid jet Reynolds number, Rel = 4161, as well Two different gas velocity profiles: (1) a uniform flat gas velocity
as the carrier gas Reynolds number, Reg = 31, 823, are close to the that equals to 48 m/s; (2) a gas velocity profile extracted from the

5
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 2. Algorithm of E-L tagging method.

atomization computation, are performed to study the influence of tion. In fact, Lowe et al. [50] reported that large numbers of lig-
different inlet gas velocity profiles. aments and irregular shaped objects are formed before secondary
The flamelet calculations for the flamelet library generation are breakup. Those ligaments and irregular shaped objects are consid-
conducted with FlameMaster code [67]. The numbers of grids set ered to more likely increase the turbulence in the flow field. Al-
for Z˜ , Z h are 100 × 20 × 100 × 10. The reaction mecha-
2 , C˜, and  though such liquid ligaments and irregular shaped objects are also
nism for acetone/air combustion proposed by Pichon et al. [68] is observed in the present simulation, those are forced to be trans-
employed, which consists of 81 species and 416 reactions. formed into Lagrangian sphere droplets using a E-L tagging method
The computational cost for single realization of atomization and owing to the limitation of the consideration of those shapes. The
combustion simulations, performed on the Kyoto University Super- influence of this will be discussed later.
computer (Cray XC40), are around 430k core·hours (840 h in real The E-L tagging method is used to recognize the dispersed Eule-
time using 512 cores) and 250–370k core·hours (230–340 h in real rian components generated by the primary breakup during the at-
time using 1088 cores), respectively. omization process and transfer their properties into the Lagrangian
droplets, which are later directly placed in the computational do-
4. Results and discussion main to replace the Eulerian components. To simply explain the
E-L tagging method employed in the present study, 2-dimensional
4.1. Atomization and E-L transformation schematics are given as shown in Fig. 2. First, a threshold for the
tagging method should be artificially given, with which the cells
In the experiments of the dense sprays, it is difficult to acquire satisfying the criterion would be tagged and then become one
confident droplet size distribution owing to the diagnostics limita- part of a transformed Lagrangian droplet. In the present study, the

6
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 3. Comparison of droplet position and size distributions between Eulerian components and Lagrangian droplets at 1.0 ≤ z/Dg ≤ 2.5 and −0.5 ≤ r/Dg ≤ 0.5 (Lagrangian
droplets are colored and scaled by diameter size).

Fig. 4. Comparison of axial velocity and position distributions between Eulerian components and Lagrangian droplets at 1.0 ≤ z/Dg ≤ 2.5 and −0.5 ≤ r/Dg ≤ 0.5 (Eulerian
components and Lagrangian droplets are colored by velocity).

volume fraction of gas phase V OID = 1 − ψ , is used for the tag- markers are switched to red markers to represent the inside of
ging method, and hence the threshold for the V OID value is set as the tagged droplet as shown in Fig. 2(c). By looping the steps from
V OIDcri . (b) to (c), a droplet would be filled by red markers and no further
For a cell satisfying V OID < V OIDcri , it would be tagged by a neighboring cells could be tagged by blue markers, then all mark-
red marker as shown in Fig. 2(a). The surrounding cells are then ers are switched to red markers as shown in Fig. 2(d), which would
checked and also the cells satisfying V OID < V OIDcri are tagged by later be tagged by a specific marker such as 1 shown in Fig. 2(e).
blue markers, which represent the edge of the tagged droplet as By employing the steps from (a) to (d), all the dispersed Eulerian
shown in Fig. 2(b). Later on, the cells neighboring the edge of the components would be tagged by the specific markers such as 1, 2,
tagged droplet are checked and tagged by blue markers to rep- and 3, as shown in Fig. 2(e). Then the details of each tagged Eu-
resent the new edge of the tagged droplet, and the original blue lerian component would be further checked. For example, for 3 in

7
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 2(e), some cells in it are neighboring the wall boundary such
that the detailed information could not be further checked, which
makes 3 unable to be recognized as a Lagrangian droplet as shown
in Fig. 2(f). Some more complex situations are provided in Fig. 2(g),
for example, how to deal with the cells marked by the question
marks since they should belong to two individual droplets, or how
to deal with the dispersed component with extremely slim struc-
ture marked by the red rectangle. Therefore, for more interests of
this tagging method, please refer to the work of Herrmann [56],
Zuzio et al. [69] as well as our previous work [46].
In the present study, the E-L transformation is triggered only
if the Eulerian droplet passes through the ACC plane, as shown
in Fig. 1(b), which is a cross-section at the downstream region in
the atomization computational domain. Therefore, a buffer region
is created upstream of the ACC plane, and the Eulerian compo-
nents in this buffer region that can pass through the ACC plane
within one sampling time interval are transformed into Lagrangian
droplets and then saved in the database for subsequent combus-
tion. Because the computational time interval used for the com-
bustion simulation is 1 × 10−5 s, and that for the atomization sim-
ulation is 5 × 10−8 s, the sampling time interval is thus set as 200
steps of the atomization computation, that is, 1 step of the com-
bustion simulation.
Figures 3 and 4 show comparisons of the position, size, and
axial velocity between the Eulerian components and the trans-
formed Lagrangian droplets. Generally, a good agreement is ob-
served. Specifically, it can be seen that the Lagrangian droplets
with a red color match the large Eulerian components in both the
droplet position and size. A good match is also shown for the small
droplets with a green color. Such an agreement of the axial veloc-
ity profiles can also be found in Fig. 4. Therefore, the present E-L Fig. 5. Atomization behavior and the Eulerian components distributions at ACC1
and ACC2 (V OIDcri = 0.9).
tagging method can work properly to transform Eulerian compo-
nents into the Lagrangian droplets.

4.2. Atomized droplets database databases can be acquired, as presented in Table 2. Note that the
atomization simulation is only calculated once under the condition
4.2.1. Database concept and droplet size distribution shown in Table 1, and the droplet databases are obtained during
To confidently build a database serving as the inlet bound- the simulation by setting different ACCs and thresholds.
ary conditions for the combustion computation, the critical factors Figure 5 shows the atomization behavior in the front view and
affecting the atomization properties should be carefully checked. at the cross-sections of ACC1 and ACC2, the colored surface of
The present E-L tagging method includes three important parame- which is the iso-surface of the E-L tagging method threshold of
ters, i.e., the downstream distance from the fuel nozzle to the ACC 0.9, i.e., the liquid-phase volume fraction of the grids is larger than
plane, Z0 , the threshold value of the E-L tagging method, and the 10%. The liquid column starts to show instability around the down-
total sampling time for recording the droplets profiles. Regarding stream at z/Dg = −2.25, and twists at approximately z/Dg = −1.75.
the downstream distance Z0 , with a smaller Z0 , where the liquid When it passes z/Dg = −1.5, a breakup behavior can be observed.
jet is not fully developed, the E-L tagging method cannot transform The liquid column becomes discontinuous, and smaller droplets
the ligaments and irregular shaped objects into larger Lagrangian and some ligaments and irregular shaped objects can be found
blobs, which is unrealistic, resulting in mass loss; however, with around the z/Dg = −1.0. Beyond z/Dg = −0.5, most of the visible
a larger Z0 , where less ligaments exist and Eulerian droplets are liquid phase is due to dispersed droplets, which tend to flow away
easily transformed into the Lagrangian droplets, the E-L tagging from the center axis. By comparing the cross-sections of ACC1 and
method also loses its accuracy owing to the numerical diffusion ACC2, the existing droplets are found to be denser at ACC1, and the
since the cells might hold diffused VOF values and the unexpected downstream droplets at ACC2 reach further in the radial direction.
transformed Lagrangian droplets. Therefore, two different ACCs, i.e., Figure 6 shows the droplet profiles in different databases, plot-
one is 5 mm upstream of the pilot outlet (ACC1) and another is ted with droplet diameters which are binned over 10 μm interval
5 mm downstream of the pilot outlet (ACC2), which are shown for droplets up to 50 μm, 20 μm interval for droplets from 50 to
in Fig. 1(b), are exhibited to check the confidence of the position. 150 μm, and 50 μm interval for droplets with size above 150 μm.
On the other hand, considering the threshold values used in the For Fig. 6(a), with the ACC plane going downstream and increas-
E-L tagging method, a larger threshold, which can recognize the ing the threshold of the E-L tagging method, the peak of the PDF
tiny Eulerian droplets, increases the risk of numerical error caused value changes from 70 to 30 μm. By comparing the droplet size
by numerical diffusion, some of which are simply numerical noise distributions of A1 and A2 in three different clusters, i.e., 0–50 μm,
and referred to as fake droplets. However, a smaller threshold ne- 50–150 μm, and 150–200 μm, there are few differences between
glects those droplets with a small scale in comparison to the local both cases, indicating occurrences of further breakup. Moreover,
grid size, losing realistic small droplets, which have a significant the threshold of the E-L tagging method seems to have a more sig-
influence on the subsequent evaporation and combustion proper- nificant influence on the droplet size distribution than the ACC po-
ties. Thus, three different thresholds, V OIDcri = 0.9, 0.95, and 0.99, sition by comparing the differences between A2 and A4 as well as
are selected and their validities are examined. Hence, four different A1 and A2. By applying a larger threshold, more droplets with the

8
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Table 2
Databases acquired in atomization simulation.

Database ACC plane Threshold for E-L transformation Volumetric particle loading

A1 ACC1 0.9 7.40

A2 ACC2 0.9 9.28
A3 ACC2 0.95 9.84
A4 ACC2 0.99 11.22

Fig. 6. Droplet size distribution in terms of (a) PDF and (b) accumulated volume of different atomization databases (A1–A4).

Fig. 7. Radial distributions of droplet size in (a) D32 and (b) D10 for different atomization databases (A1–A4).

diameter less than 50 μm seem to be captured with the value of S indicates the distance between the centers of two neighboring
0.99. By comparing the total droplet volume of different databases droplets, and d denotes the diameter of droplet. With S/d ≥ 10, the
shown in Fig. 6(b), all databases show a similar total droplet mass, two-way coupling method could be employed in this study instead
which means that almost of the Eulerian components are captured of four-way coupling. When the ACC plane is set further down-
by the present E-L tagging method. In addition, one thing worth stream, the droplets can reach further in the radial direction, which
noting is that for a large Lagrangian droplet, which might be com- can refer to the cross-sections of ACC1 and ACC2 in Fig. 5, and thus
parable to the mesh size, the mass and momentum interactions A2 shows a larger value than A1, indicating a more dilute spray
between gas phase and the Lagrangian droplet are decided by the compared with A1.
surrounding cells instead of the local cell such that the Lagrangian
droplet could be still considered as a point by the PSI-Cell method. 4.2.2. Droplet spatial and temporal size distributions
Because this study is focused on dense spray, the volumetric In traditional combustion simulations, the sizes of droplets is-
particle loading of droplets in the computational domain is further sued from the inlet boundary are often given using a simple at-
checked, as shown in Table 2. Here, the ratio of droplet distance omization model or a presumed droplet size distribution. In this
and droplet diameter is presented instead of simple volume frac- study, on the other hand, those are taken from a database resulting
tion, which can refer to as S/d in Elghobashi’s work [70], where, from the atomization computation. Therefore, the sampling time of

9
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 8. Time variations of droplet volume through ACC1 plane for the database A1: (a) in total, (b) each quadrant.

atomization computation should be carefully discussed. To obtain a the Sauter mean diameter (D32 ) and arithmetic mean diameter
symmetrical flame structure without any preferences in any direc- (D10 ) radial distributions obtained from the different databases.
tions, the spatial distribution of the droplets must be checked such The droplet size for either D32 or D10 increases as the droplets flow
that the position profile should be spatially homogeneous with- further away from the central axis, which means that the droplet
out any biases in any directions. For example, many more larger size has a strong correlation with the radial distribution, and thus
droplets can be found in π ≤ θ < 3π /2 and fewer droplets can be it is sufficient to base the analysis for azimuthal homogeneity on
found in 0 ≤ θ < π /2 at ACC1, and larger droplets are observed on investigating the droplet size with different angles θ in the radial
the right side at ACC2, as shown in Fig. 5, thus the droplets distri- plane. The droplet size can be used to represent the radial dis-
bution shows some preferences if the sampling time is insufficient. tance between the droplet and the center axis, and the angle θ
The complexity of the azimuthal homogeneity analysis in- can be used to represent different radial directions. In addition, re-
creases if the droplet size is further considered. Figure 7 shows ferring to the experimental study [50] based on the Sydney Burner

10
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 11. Conducted cases and schematic of locations of recess distance, Lr, and ACC
planes, ZACC , for combustion in detail.

To obtain a sufficient number of reliable data for a homoge-

neous analysis, the sampling time and interval are discussed. The
sampling interval is set 1 × 10−5 s, which is 200 steps for the at-
omization computation and 1 step for the combustion computa-
tion, and thus the combustion simulation can simply read the data
in a step-by-step manner. Figure 8 shows the time variations of
droplet volume through ACC1 plane and through each quadrant
of ACC1 plane, from the 120,0 0 0th step, in which the liquid jet
is thought to be fully developed and starts to steadily breakup, to
the 30 0,0 0 0th step, which is thought to be long enough for sam-
pling. Five periods can be observed in Fig. 8(a) showing that the
atomization process holds a periodic characteristic with breakup
and non-breakup periods. In Matas’s study [71,72], the frequency
of the liquid jet breakup is found to have a strong relation to the
gas and liquid velocities as well as the nozzle size, and the empir-
ical correlations are shown as follows.
 
ρg δl
f = u + ul /Dg , (23)
ρl δg g
Fig. 9. Time variations of D32 and D10 of droplets through each quadrant of ACC1  
dv
plane for the database A1: (a) D32 , (b) D10 .
δ = v0 /max . (24)
dr
where δ is the thickness of the vorticity layer, v denotes the radial
velocity, and v0 is the radial velocity measured near the Sydney
Burner nozzle, e.g., z/Dg = 0. The thickness of the vorticity layer of
liquid and gas phases, δl and δg are directly acquired in the simu-
lation. By this correlation, the calculated result is 547 Hz, and the
mean period shown in Fig. 8(a) is simply calculated as 556 Hz.
Therefore, the present detailed numerical simulation and the E-
L tagging method are considered to be reasonable to reproduce
the atomization phenomena. Figure 8(b) shows the variations of
droplet volume through each quadrant of ACC1 plane, and it could
be observed that the droplet volume of each quadrant shows to-
tally different values in different breakup periods. For example,
for the quadrant of 3π /2 − 2π , it holds larger values in the later
time in each breakup period except for the 3rd period between 3
and 5 ms. Figure 9 also shows the time variations of D32 and D10
through each quadrant of ACC1 plane in each a half breakup pe-
riod. It could be observed that the values of D32 and D10 in each a
half breakup period are totally different such that the droplet sam-
pling time for acquiring a homogeneous distribution need to be
further checked.
By comparing the D32 and D10 distributions in Fig. 7, the
Fig. 10. Total values of different databases A1–A4 with different sampling time.
droplet size has a strong correlation with the radial distance, that
is, the closer the droplets to the center axis, the smaller the
droplets are. Therefore, the angle θ in the radial plane and the
of N-AF8-25 and N-AF8-80, which only supplies the D32 distribu- droplet size can be used to analyze the droplet spatial distribution
tions along the center axis (see Fig. 12), D32 is always less than by the Chi-square homogeneity check, the detail of which can be
60 μm. Given that the database of A4 holds the D32 value less than found in Appendix A.
60 μm, whereas those of other three cases are larger than 60 μm, Table 3 shows the calculated chi-square values of the database
the database A4 provides a better droplet size distribution accord- A1 with a sampling time starting at the 120,0 0 0th step and end-
ing to the experiment. ing up with the 30 0,0 0 0th step, covering five full-breakup peri-

11
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 12. Instantaneous gas temperature iso-surface and droplet distribution for Cases 3 and 6.

Table 3 be five breakup periods from the 120,0 0 0th step to the 30 0,0 0 0th
Chi-square homogeneity check for the database A1.
step.
A1 0 − π /2 π /2 − π π − 3π /2 3π /2 − 2π Total In total, for the atomization computation, the physical time is
0–50 μm 0.12 0.27 4.88 1.77 7.04 15 ms, in which the first 6 ms is used for the liquid jet develop-
50–150 μm 0.11 0.02 0.78 0.16 1.07 ment to reach a steady state, and the last 9 ms is used for the
150–300 μm 0.56 0.22 0.80 1.32 2.9 droplet database sampling discussed in this section. For the com-
Total 0.79 0.51 6.46 3.25 11.01 bustion computation, the physical time is 100 ms such that the
particle injection is cycled about 11 times.

4.3. Combustion characteristics

ods. Compared to other cells, there is an odd value particularly
larger than the others, which is marked as red in π ≤ θ < 3π /2 4.3.1. Simulation cases and flame features
with a diameter of less than 50 μm. It means that in the region As mentioned in Section 3.1, two different configurations, N-
π ≤ θ < 3π /2, more droplets with a diameter of less than 50 μm AF8-25 and N-AF8-80, are utilized with representative recess dis-
exist compared to the other three regions. Because the number of tance of 25 mm and 80 mm. In addition, two different inlet gas ve-
droplets with diameter of less than 0–50 μm is very small in the locity profiles, one flat and the other extracted from the atomiza-
database A1 (see Fig. 6(a), black line), it is reasonable to observe tion computation, are incorporated to study the sensitivity of the
such a bias that some region holds a different number of droplets dense spray flame to the inlet boundary conditions. Further, con-
because the droplets generated by the breakup fluctuate spatially sidering the computational cost, out of the four databases of A1
and temporally. However, the total value of this case is 11.01, which to A4 two databases, A1 and A4, are selected to investigate the
is less than the critical value of 12.59, ensuring that the spatial dis- influence of the droplet inlet boundary conditions on the flame
tribution of the droplets has no obvious bias by sampling the five characteristics because A1, A2, and A3 display similar droplet size
breakup periods from the 120,0 0 0th step to the 30 0,0 0 0th step. profiles of D32 and D10 , whereas A1 holds different gas velocity
Figure 10 shows the total values of each database by employing compared to A2–A4. Therefore, a total of six cases are discussed as
different sampling times from one to five periods. The total val- shown in Table 4, Fig. 11 provides a clear schematic of the con-
ues of A1 and A2 continue decreasing when the sampling time is ducted six cases. The orange lines depicted in the figure indicate
increased from one to five periods, out of which the total values the ACC planes, where the droplets are recorded in the atomiza-
of A4 continue fluctuating because the number of small droplets tion computation and are applied in the combustion computations
with a diameter of 0–50 μm of A4 is larger than those of the other as the inlet boundary conditions. In addition, the axial distance of
three cases, thus those odd values (an example is shown in red in the ACC plane ZACC relative to the pilot outlet, is displayed for a
Table 3) are decreased such that the total value is much smaller better understanding.
than those of the other three cases. By sampling five breakup pe- In dense sprays, laser diagnostic measurements are difficult ow-
riods, all four databases hold reasonable total values to ensure the ing to the co-existence of liquid fragments and abundant atomized
azimuthal homogeneity, and the sampling time is thus selected to droplets, which reflect and absorb the light. Therefore, a chirped-

12
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 13. Comparison of radial distributions of time-averaged of gas temperature at different axial downstream locations of z/Dg = 3, 5, 10, and 20 between combustion
simulation and experiment for Cases 1–6. Left is for dense spray and right is for dilute spray.

probe-pulse femtosecond coherent anti-Stokes Raman spectroscopy sprays with ethanol and acetone fuels. The gas temperature dis-
(CPP-fs-CARS) with a repetition rate of 5 kHz was employed by tributions in both the dense and dilute sprays with the acetone
Thomas and Lowe [49,50] to measure the gas temperature distri- fuel are considered in this work to validate the proposed numeri-
bution based on the platform of the Sydney Burner, which sup- cal simulations.
plies various gas temperature profiles ranging from dense to dilute

13
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 14. Comparison of radial distributions of RMS of gas temperature at different axial downstream locations of z/Dg = 3, 5, 10, and 20 between combustion simulation and
experiment for Cases 1–6. Left is for dense spray and right is for dilute spray.

14
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Table 4
Conducted cases for combustion simulation.

Case Velocity profile Atomization database Configuration

1 Flat A1 Dense
2 Atomization A1 (N-AF8-25)
3 Atomization A4
4 Flat A1 Dilute
5 Atomization A1 (N-AF8-80)
6 Atomization A4

Figure 12 shows the instantaneous gas temperature profiles and

fuel droplets distribution for Cases 3 and 6. The double reaction
zone, where a premixed core is inside a surrounding non-premixed
zone, can be observed. The premixed cores are extended in both
the span and axial directions despite the recess distance changes
from 25 to 80 mm. Meanwhile, each temperature iso-surface in
Case 3 is shorter than that of Case 6, which means that the di-
lute spray (Case 6) with a longer recess distance can enhance the
mixing and combustion processes.

4.3.2. Comparisons with experiments

Figures 13 and 14 show the radial distributions of the time-
averaged and RMS of gas temperature at different axial down-
Fig. 15. Comparison of distribution of time-averaged of gas temperature for Cases
stream locations, z/Dg = 3, 5, 10, and 20, where the experimen-
1–6.
tal data are compared with simulations of Cases 1–6 detailed in
Table 4. The influences of the inflow gas velocity profiles and in-
flow droplet profiles are discussed.
The computed results of the mean gas temperature depict a
symmetric unimodal distribution at different axial locations, and at
the downstream of z/Dg = 20, the peak temperature is suppressed.
However, based on the experiments at z/Dg = 20, the symmetric
unimodal distribution is absent for the dense case, and is still
distinguished in the dilute case. The reason for this difference is
thought to be the result of a higher proportion of fragments and
filaments presented in the dense case [49,50], which leads to a
less stable and lower mean temperature. By contrast, at the up-
stream locations of z/Dg = 3, the flame is more dominated by the
pilot flame such that the reaction zone is less affected by the spray
properties and a similar temperature profile is observed for both
dense and dilute flame cases. Regarding the simulation results, the
atomization velocity presents its priory in comparison to the flat
one, showing a good agreement with experiments at the upstream
region. Even for the z/Dg = 10 of the dilute cases, the tempera-
ture distribution still matches very well. For both cases with at-
omization velocity profiles, the temperature distributions do not
show much difference though the D32 , D10 , and PDF distribution of
droplet diameter show a significant difference (see Figs. 6(a) and
7(b)). However, some discrepancies can still be observed, i.e., cases
with atomization gas velocity and smaller droplets (Cases 3 and
6) have intermediate values between the cases with flat gas veloc- Fig. 16. Comparison of distribution of RMS of gas temperature for Cases 1–6.
ity and larger droplets (Cases 1 and 4), and cases with atomization
gas velocity and larger droplets (Cases 2 and 5). This is because the
turbulence generated by the primary breakup during the atomiza- An obvious discrepancy is seen between the simulation re-
tion process of Cases 3 and 6 (5 mm downstream of the pilot out- sults and the experimental data at the downstream location of
let) is weaker than that of Cases 2 and 5 (5 mm upstream of the z/Dg = 20 of a dense spray, where the temperatures of simula-
pilot outlet), but stronger than the flat ones. In addition, the differ- tion are overpredicted. In contrast to which, in a dilute spray, the
ence disappears when the recess distance is increased from 25 to temperature profiles of simulation match those of experiment very
80 mm based on a comparison of the differences between a dense well and only drift away at the downstream location of z/Dg = 20,
spray (Cases 2 and 3) and a dilute spray (Cases 5 and 6) because which is still not that obvious compared to that in a dense spray.
the ACCs, where the turbulence in the atomization computation is The good agreement in the upstream region is due to the well-
recorded, are 5 mm upstream and 5 mm downstream of the pilot controlled boundary conditions which are also the interest of the
outlet for a dense spray, but are 60 and 50 mm upstream of the present study. In addition, a suitable flamelet model or a better
pilot outlet for a dilute spray, as shown in Fig. 11. Therefore, the combustion mechanism may help to improve the difference ob-
difference in turbulence owing to different ACCs finally disappears served in the downstream region for the dilute spray. However, for
at the pilot outlet for a dilute spray. the great discrepancy in the dense spray, the situation is totally

15
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

Fig. 17. Comparison of streamwise distributions of time-averaged droplet sizes (D32 and D10 ) and gas temperature at different radial locations of r/Dg = 0, 0.4, and 0.6
between combustion simulation and experiment for Cases 3 and 6. Only droplets diameter less than 0.1 mm are counted.

different. According to the experiment [50], a speculation is pro- consumed, and those larger liquid components generated by the
posed that the combustion is finished in the downstream location primary breakup easily cause more unpredicted turbulence when
of z/Dg = 20 even there exists fuel droplets such that the temper- they flow downstream and breakup into smaller droplets. There-
ature distribution does not present the symmetric unimodal distri- fore, the local flow field is extremely unstable in the downstream
bution. On the other hand, the combustion is still happening in the region, which is different from the upstream region where the dis-
simulation such that an overestimation as well as a totally differ- turbances of those ligaments and irregular shaped objects have not
ent structure could be observed. The dense spray region has much developed. Therefore, the burning process might not occur much or
more larger ligaments and irregular shaped objects that cannot be just be finished, while the fuel is still present there, resulting in a

16
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

lower local temperature distribution. However, in the present sim- measure sphere droplets and most sphere droplets holds relatively
ulation, the larger ligaments and irregular shaped objects are just smaller diameter to maintain the shape, which leads to a decrease
simply transformed into Lagrangian droplets, failing to consider the in droplet size. However, there are still some shared properties for
realistic interaction between the gas phase and liquid fuel, as well the droplet size in simulation and experiment. Both D32 and D10
as the further breakup behavior. Therefore, the Lagrangian droplets increase from center to side despite increasing the recess distance
continue to be consumed when they reach further downstream, from 25 to 80 mm. For the dense spray, the D32 values increase
ensuring a continuous burning without any negative influences on slightly further downstream from z/Dg = 0.4–20, whereas for the
the combustion, and thus a higher temperature distribution could dilute spray, the D32 values decrease slightly from z/Dg = 10–20,
be observed. However, the unstable flow field caused by those liga- which also match the experiment results at the radial locations
ments and irregular shaped objects may have significant influences of r/Dg = 0.4 and 0.6. The D10 values of the simulation keep in-
on the fuel droplet evaporation and fuel vapor heat release. An- creasing in a dense spray and keep fluctuating in a dilute spray,
other possibility is that the larger ligaments and irregular shaped whereas the D10 values of the experiment keep fluctuating in a
objects tend to disperse further in the radial direction and thus dense spray and keep increasing in a dilute spray. This opposite
the heavier ligaments and irregular shaped objects finally fall when development is considered to be a result of omitting the larger
they flow out of the flame zones. However, the total energy of the Lagrangian droplets artificially. Since D10 is more affected by the
simulation is conserved such that the temperature distribution of portion of small droplets, and the measured droplets in the ex-
the experiment is lower than that of the simulation. periment tend to hold smaller diameter whereas the transformed
The profiles of the RMS of gas temperature show a twin peak droplets in the simulation may hold larger diameter, hence it is
on the both sides of the centerline until z/Dg = 10, which is less difficult for the simulation to predict the D10 values. Regarding the
distinct for the dense case. This twin peak structure is also well- temperature development along with the axial direction, the di-
captured by the present study for all six cases, and the Cases 3 lute spray is considered to have a good agreement with the exper-
and 6 still have the intermediate values. The computational results iment at different radial locations. Whereas, for the dense spray,
show a reasonable match with the experiments near the center even good agreement can be observed at the center axis, it fails
axis, and some discrepancies away from the center axis. In the ex- to capture the downstream temperatures especially for the outer
periment work, the temperature is strongly affected by the pilot locations of r/Dg = 0.6. This is probably because of insufficiency of
flame in the outer space, such that the temperature distribution the consideration of effect of liquid ligaments and irregular shaped
out of −1.25 ≤ r/Dg ≤ 1.25, outside the edge of the pilot flame, objects existing away from the center axis, as mentioned before.
presents some discrepancies. In addition, owing to the larger lig-
aments and irregular shaped objects mentioned in the last para- 5. Conclusions
graph, it is difficult to capture the accurate temperature distribu-
tions in the dense spray cases. Therefore, the discrepancies tend In this study, a numerical framework which is a one-way cou-
to become eliminated when the recess distance is increased from pling between a VOF simulation and a combustion simulation was
25 to 80 mm around the nozzle centerline from r/Dg = −1 to proposed, and the validity was investigated for the dense spray
r/Dg = 1, except for the location at z/Dg = 20, where the combus- flames. Atomization was simulated by a detailed high-resolution
tion of dense spray in the experiment is considered to be com- VOF simulation, in which both continuum gas and liquid phases
pleted. were strictly solved in a Eulerian framework, and the Eulerian
Figures 15 and 16 show the distributions of the time-averaged components of the liquid droplets were transformed into the La-
and RMS of gas temperature until the downstream location of grangian droplets at a certain downstream cross-section, i.e., sam-
z/Dg = 50 for all six cases. It can be seen that, compared to the pling cross-section, whose information was stored in the database.
results on the assumption of a uniform inlet boundary condition, Then, the combustion process was solved by a LES/FPV adopting
the use of inlet velocities determined from the atomization com- the pre-stored database of Lagrangian droplets (i.e., the position,
putation can enhance the mixing and establish the combustion at size, and velocity of each droplet) as the inlet boundary condi-
the most upstream locations. On the other hand, the profiles of the tions. Computations were validated against measurement made in
fluctuating temperature indicate that the differences of droplets in the Sydney Piloted Needle Spray Burner, which can generate both
databases A1 and A4 may not have a significant influence on the dilute and dense spray flames by varying the recess distance from
combustion characteristics when comparing Cases 2 and 3 as well the liquid fuel jet nozzle to the pilot outlet.
as Cases 5 and 6. In addition, the symmetrical flame structures in Regarding the detailed high-resolution VOF simulation of liq-
all cases prove that the droplet databases sampled from the atom- uid fuel atomization, the volume flux of the droplets at the exit
ization computation have no biases in any directions. This confirms of the nozzle was observed to fluctuate both temporally and spa-
that the proposed sampling method coupling the atomization and tially, which meant that there exist periods of breakup and non-
combustion computations is successful. breakup during the atomization process. It was also found that the
Figure 17 shows the streamwise distributions of the time- breakup period was in good agreement with an existing empiri-
averaged droplet sizes (D32 and D10 ) and gas temperature of Cases cal correlation, and that, compared to the small droplets, larger
3 and 6 at different radial locations of r/Dg = 0, 0.4, and 0.6 in droplets tended to be located away from the center axis. Mean-
comparison with the experiment results. Due to the limitation while, in the database of the Lagrangian droplets for the LES/FPV of
of the experiment measurement of Phase Doppler Anemometry spray flames, the location of the sampling cross-section, sampling
(PDA), the measured droplet size is limited to 100 μm. Therefore, time, and threshold value for a Eulerian–Lagrangian (E-L) trans-
the calculations of D32 and D10 omit the Lagrangian droplets with a formation were found to strongly affect the properties of the La-
diameter of larger than 100 μm. However, the droplet size distribu- grangian droplets, and to be critical for the success of the use of
tions between simulation and experiment still show large discrep- the LES/FPV of present spray flames.
ancies, and the discrepancies might result from two reasons from By use of the optimal pre-stored droplets database, the re-
the views of simulation and experiment. From the view of simula- sults of LES/FPV of two flame cases with different recess distances
tion, the E-L transformation utilized in the present study might im- showed generally similar trends with the experiment in terms of
properly transform the larger ligaments and irregular shaped ob- the gas temperature and droplet size distributions. The spray flame
jects into Lagrangian droplets leading to an increment in droplet with a longer recess distance, which represents a dilute spray, was
size. From the view of experiment, the PDA has severe criterion to considered to have a longer and wider premixed core than that

17
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

with a shorter recess distance representing a dense spray. This azimuthal homogeneity of the droplet distribution. The calculation
behavior is believed to be due to the enhanced mixing between algorithm is very simple, which is briefly introduced. For the cell i
the evaporated fuel and oxidizer. For the dense spray flame, the in a certain row and column, there is a statistic Fi by counting the
discrepancy in the gas temperature between the prediction and number of droplets which satisfy the conditions (i.e., droplet size
experiment tended to become more evident, and when moving and angle), and thus a total value of the row or column can be
downstream, the over-predictions were observed. This was consid- calculated. Based on the statistic of cell i and total values of rows
ered to be attributed to the fact that for the LES/FPV of the dense and columns, the expected value of cell i could be calculated as Ei .
spray flame, the relatively large and non-spherical liquid compo- Then, the total value of all cells, X 2 , of one database can be cal-

nents were regarded as the Lagrangian spherical droplets in the culated as X 2 = ni=1 ((Ei − Fi )2 /E 2i ). By using this Chi-square ho-
present E-L transformation, and therefore, their further breakup mogeneity check, if the total value is less than the critical value
and influence on the turbulence development and flame evolution of 12.59, it can be said that the droplets are homogeneously dis-
were neglected. tributed in the four mathematical quadrants (0 ≤ θ < π /2, π /2 ≤
In summary, the numerical framework proposed in this study is θ < π , π ≤ θ < 3π /2, and 3π /2 ≤ θ < 2π ) with smaller droplets
capable of reproducing the spray atomization and the gas temper- close to the center axis and larger droplets away from the center
ature distributions in dilute spray flames of Sydney Piloted Nee- axial such that the database can be considered to have no special
dle Burner with a relatively low computational cost and without preferences existing in a specific direction.
any atomization model or presumed initial droplet size distribu-
tion. Some failures of gas temperature distributions in dense spray References
flames and droplet size distributions are discussed in detail which
[1] A.R. Masri, Turbulent combustion of sprays: from dilute to dense, Combust. Sci.
requires further investigations. The improvement of the E-L tagging Tech. 188 (2016) 1619–1639.
method in generating droplet databases for relatively dense sprays [2] M. Linne, Imaging in the optically dense regions of a spray: a review of devel-
merits future work. oping techniques, Prog. Energy Combust. Sci. 39 (2013) 403–440.
[3] H. Chiu, T. Liu, Group combustion of liquid droplets, Combust. Sci. Tech. 17
(1977) 127–142.
Declaration of Competing Interest [4] A. Umemura, Interactive droplet vaporization and combustion: approach from
asymptotics, Prog. Energy Combust. Sci. 20 (1994) 325–372.
[5] S. Li, Spray stagnation flames, Prog. Energy Combust. Sci. 23 (1997) 303–347.
The authors declare that they have no known competing finan-
[6] E. Gutheil, W. Sirignano, Counterflow spray combustion modeling with de-
cial interests or personal relationships that could have appeared to tailed transport and detailed chemistry, Combust. Flame 113 (1998) 92–105.
influence the work reported in this paper. [7] A.Y. Klimenko, R.W. Bilger, Conditional moment closure for turbulent combus-
tion, Prog. Energy Combust. Sci. 25 (1999) 25–62.
[8] V. Gopalakrishnan, J. Abraham, Effects of multicomponent diffusion on pre-
Acknowledgments dicted ignition characteristics of a n-heptane diffusion flame, Combust. Flame
136 (2004) 557–566.
This work was partially supported by MEXT as “Program [9] H. Pitsch, M. Ihme, An Unsteady/Flamelet Progress Variable Method for LES of
non-Premixed Turbulent Combustion, AIAA Paper No. 2005–557.
for Promoting Researches on the Supercomputer Fugaku” (Digital [10] J. Reveillon, L. Vervisch, Analysis of weakly turbulent dilute-spray flames and
Twins of Real World’s Clean Energy Systems with Integrated Uti- spray combustion regimes, J. Fluid. Mech 537 (2005) 317–347.
lization of Super-simulation and AI) and by JSPS KAKENHI Grant [11] M. Nakamura, F. Akamatsu, R. Kurose, M. Katsuki, Combustion mechanism of
liquid fuel spray in a gaseous flame, Phys. Fluids 17 (2005) 1–14.
number 19H02076. YH acknowledges the support of the National [12] J. Reveillon, F.X. Demoulin, Evaporating droplets in turbulent reacting flows,
Natural Science Foundation of China (Grant no. 52006151) and Proc. Combust. Inst. 31 (2007) 2319–2326.
“the Fundamental Research Funds for the Central Universities” [13] H. Watanabe, R. Kurose, S.M. Huang, F. Akamatsu, Characteristics of flamelets
in spray flames formed in a laminar counterflow, Combust. Flame 148 (2007)
(YJ201943). AM is supported by the Australian Research Council. 234–248.
The authors also thank Dr. A. Lowe for providing us the detailed [14] N. Abani, A. Munnannur, R.D. Reitz, Reduction of numerical parameter depen-
configuration of the Sydney Burner. dencies in diesel spray models, J. Eng. Gas Turbines Power 130 (2008) 032809.
[15] Y. Baba, R. Kurose, Analysis and flametlet modelling for spray combustion, J.
Fluid. Mech. 612 (2008) 45–79.
Appendix A. Details of Chi square homogeneity check for [16] M. Ihme, Y.C. See, Prediction of autoignition in a lifted methane / air flame us-
droplet spatial distribution analysis ing an unsteady flamelet/progress variable model, Combust. Flame 157 (2010)
1850–1862.
[17] R. Bilger, A mixture fraction framework for the theory and modeling of
The homogeneity analysis for droplet spatial distribution is dis- droplets and sprays, Combust. Flame 158 (2011) 191–202.
cussed in details. As mentioned in Section 4.2.2, the angle θ in the [18] K. Luo, H. Pitsch, M.G. Pai, O. Desjardins, Direct numerical simulations and
radial plane and the droplet size can be used to analyze the droplet analysis of three-dimensional n-heptane spray flames in a model swirl com-
bustor, Proc. Combust. Inst. 33 (2011) 2143–2152.
spatial distribution. The space can be divided into four mathemat- [19] M.R. Turner, S.S. Sazhin, J.J. Healey, C. Crua, S.B. Martynov, A breakup model
ical quadrants, i.e., 0 ≤ θ < π /2, π /2 ≤ θ < π , π ≤ θ < 3π /2, and for transient diesel fuel sprays, Fuel 97 (2012) 288–305.
3π /2 ≤ θ < 2π , to check the preference in different radial direc- [20] C. Bajaj, M. Ameen, J. Abraham, Evaluation of an unsteady flamelet progress
variable model for autoignition and flame lift-off in diesel jets, Combust. Sci.
tions, and the droplet size can be divided into three clusters, i.e., Tech. 185 (2013) 454–472.
0–50 μm, 50–150 μm, and 150–300 μm to check the distance bias [21] S. Ukai, A. Kronenburg, O.T. Stein, LES-CMC of a dilute acetone spray flame,
in a certain radial direction. The Chi-square homogeneity validation Proc. Combust. Inst. 34 (2013) 1643–1650.
[22] S. De, S.H. Kim, Large eddy simulation of dilute reacting sprays: droplet evap-
is a method generally used to analyze the data homogeneity in dif- oration and scalar mixing, Combust. Flame 160 (2013) 2048–2066.
ferent clusters, and the data can be regarded as homogeneous if [23] S. Tachibana, K. Saito, T. Yamamoto, M. Makida, T. Kitano, R. Kurose, Experi-
the accumulated value is satisfied. First, the degree of freedom for mental and numerical investigation of thermos-acoustic instability in a liquid–
fuel aero-engine combustor at elevated pressure: validity of large-eddy simu-
the Chi-square is calculated as df = (row − 1 ) × (col − 1 ), where df
lation of spray combustion, Combust. Flame 162 (2015) 2621–2637.
indicates the degree of freedom, row is the number of rows, with [24] T. Kitano, K. Kaneko, R. Kurose, S. Komori, Large-eddy simulations of gas- and
three droplet size clusters herein; col is the number of columns, liquid-fueled combustion instabilities in back-step flows, Combust. Flame 170
(2016) 63–78.
i.e., the 4 mathematical quadrants, as shown in Table 3. There-
[25] L. Ma, D. Roekaerts, Numerical study of the multi-flame structure in spray
fore, the degree of freedom is df = (3 − 1 ) × (4 − 1 ) = 6. The crit- combustion, Proc. Combust. Inst. 36 (2017) 2603–2613.
ical value (with 95% confidence level) under a degree of freedom [26] B. Wang, A. Kronenburg, G.L. Tufano, O.T. Stein, Fully resolved DNS of droplet
of 6 can be found to be 12.59 according to the statistics, which array combustion in turbulent convective flows and modelling for mixing filed
in inter-droplet space, Combust. Flame 189 (2018) 347–366.
means that if the total value of all cells is less than 12.59, then [27] A. Pillai, R. Kurose, Combustion noise analysis of a turbulent spray flame using
the data can be confirmed to have a 95% possibility to ensure the a hybrid DNS/APE-RF approach, Combust. Flame 200 (2019) 168–191.

18
J. Wen, Y. Hu, T. Nishiie et al. Combustion and Flame 237 (2022) 111742

[28] Y. Hu, R. Kurose, Nonpremixed and premixed flamelets LES of partially pre- [51] H. Moriai, R. Kurose, H. Watanabe, Y. Yano, F. Akamatsu, S. Komori, Large-eddy
mixed spray flames using a two-phase transport equation of progress variable, simulation of turbulent spray combustion in a subscale aircraft jet engine com-
Combust. Flame 188 (2018) 227–242. bustor-predictions of NO and soot concentrations, J. Eng. Gas Turbines Power
[29] Y. Hu, R. Kurose, Partially premixed flamelet in LES of acetone spray flames, 135 (2013) 091503.
Proc. Combust. Inst. 37 (2019) 3327–3334. [52] A. Kishimoto, H. Moriai, K. Takenaka, T. Nishiie, M. Adachi, A. Ogawara,
[30] Y. Hu, R. Kai, R. Kurose, E. Gutheil, H. Olguin, Large eddy simulation of a par- R. Kurose, Application of a non-adiabatic flamelet/progress-variable approach
tially pre-vaporized ethanol reacting spray using the multiphase DTF/flamelet to large eddy simulation of H2 /O2 combustion under a pressurized condition,
model, Int. J. Multiph. Flow 125 (2020) 103216. J. Heat Transf. 139 (2017) 124501.
[31] Y. Hardalupas, A. Taylor, J.H. Whitelaw, Mass flux, mass fraction and concen- [53] J.U. Brackbill, D.B. Kothe, C. Zemach, A continuum method for modeling surface
tration of liquid fuel in a swirl-stabilized flame, Int. J. Multiph. Flow 20 (1994) tension, J. Comput. Phys. 100 (1992) 335–354.
233–259. [54] S. Muzaferija, M. Peric, Computation of free-surface flows using interface–
[32] G. Chen, A. Gomes, Dilute laminar spray diffusion flames near the transition tracking and interface-capturing methods, in: O. Mahrenholtz, M. Markiewicz
from group combustion to individual droplet burning, Combust. Flame 110 (Eds.), Nonlinear Water Wave Interaction, Computational Mechanics Publica-
(1997) 392–404. tions, Southampton (1998), pp. 59–100.
[33] M. Alden, J. Bood, Z. Li, M. Richter, Visualization and understanding of combus- [55] A. Albadawi, D.B. Donoghue, A.J. Robinson, D.B. Murray, Y.M.C. Delauré, Influ-
tion processes using spatially and temporally resolved laser diagnostic tech- ence of surface tension implementation in volume of fluid and coupled vol-
niques, Proc. Combust. Inst. 33 (2011) 69–97. ume of fluid with level set methods for bubble growth and detachment, Int. J.
[34] D. Cavaliere, J. Kariuki, E. Mastorakos, A comparison of the blow-off behaviour Multiph. Flow 53 (2013) 11–28.
of swirl-stabilized premixed, non-premixed and spray flames, Flow Turbul. [56] M. Herrmann, A parallel Eulerian interface tracking/Lagrangian point particle
Combust. 91 (2013) 347–372. multi-scale coupling procedure, J. Comput. Phys. 229 (2010) 745–759.
[35] M. Chrigui, J. Gounder, A. Sadiki, J. Janicka, A.R. Masri, Acetone droplet be- [57] C.D. Pierce, P. Moin, Progress-variable approach for large-eddy simulation of
havior in reacting and non reacting turbulent flow, Flow Turbul. Combust. 90 non-premixed turbulent combustion, J. Fluid Mech. 504 (2004) 73–97.
(2013) 419–447. [58] N. Lallemant, A. Sayre, R. Weber, Evaluation of emissivity correlations for
[36] H. Luo, K. Nishida, S. Uchitomi, Y. Ogata, W. Zhang, T. Fujikawa, Effect of tem- H2 O-CO2 -N2 /air mixtures and coupling with solution methods of the radiative
perature on fuel adhesion under spray-wall impingement condition, Fuel 234 transfer equation, Prog. Energy Combust. 22 (1996) 543–574.
(2018) 56–65. [59] H. Pitsch, H. Steiner, Large-eddy simulation of a turbulent piloted methane/air
[37] A. Verdier, J.M. Santiago, A. Vandel, G. Godard, G. Cabot, B. Renou, Local ex- diffusion flame (Sandia flame D), Phys. Fluids 12 (20 0 0) 2541.
tinction mechanisms analysis of spray jet flame using high speed diagnostics, [60] D.K. Lilly, A proposed modification of the Germano subgridscale closure
Combust. Flame 193 (2018) 440–452. method, Phys. Fluids A 4 (1992) 633–635.
[38] I.A. Mulla, G. Godard, G. Cabot, F. Grisch, B. Renou, Quantitative imaging of [61] C.T. Crowe, M.P. Sharma, D.E. Stock, The particle-source-in cell (PSI-CELL)
nitric oxide concentration in a turbulent -heptane spray flame, Combust. Flame model for gas-droplet flows, J. Fluid Eng. Trans. ASME 99 (1977) 325–332.
203 (2019) 217–229. [62] R.S. Miller, K. Harstad, J. Bellan, Evaluation of equilibrium and non-equilibrium
[39] D.R. Guildenbecher, C. Lopez-Rivera, P.E. Sojka, Secondary atomization, Exp. evaporation models for many-droplet gas-liquid flow simulations, Int. J. Mul-
Fluids 46 (2009) 371–402. tiph. Flow 24 (1998) 1025–1055.
[40] X. Jiang, G. Siamas, K. Jagus, T. Karyiannis, Physical modelling and advanced [63] R. Kurose, H. Makino, S. Komori, M. Nakamura, F. Akamatsu, M. Katsuki, Effects
simulations of gas-liquid two-phase jet flows in atomization and sprays, Prog. of outflow from the surface of a sphere on drag, shear lift, and scalar diffusion,
Energy Combust. Sci. 36 (2010) 131–167. Phys. Fluids 15 (2003) 2338–2351.
[41] M. Herrmann, Detailed numerical simulations of the primary atomization of a [64] T. Kitano, J. Nishio, R. Kurose, S. Komori, Evaporation and combustion of mul-
turbulent liquid jet in crossflow, J. Eng. Gas Turbines Power 132 (2010) 061506. ticomponent fuel droplets, Fuel 136 (2014) 219–225.
[42] N. Ashgriz, Atomization of a liquid jet in a crossflow, AIP Conf. Proc. 1440 [65] T. Kitano, J. Nishio, R. Kurose, S. Komori, Effects of ambient pressure, gas tem-
(2012) 33–46. perature and combustion reaction on droplet evaporation, Combust. Flame 161
[43] X. Li, M.C. Soteriou, Detailed numerical simulation of liquid jet atomization in (2014) 551–564.
crossflow of increasing density, Int. J. Multiph. Flow 104 (2018) 214–232. [66] P.J. O’Rourke, A.A. Amsden, The TAB Method for Numerical Calculation of Spray
[44] A. Umemura, J. Shinjo, Detailed SGS atomization model and its implementation Droplet Breakup, SAE Paper, 1987.
to two-phase flow LES, Combust. Flame 195 (2018) 232–252. [67] H. Pitsch, Flamemaster: a C++ computer program for 0D combustion and 1D
[45] T.-W. Lee, J.E. Park, R. Kurose, Determination of the drop size during atomiza- laminar flame calculations, Cited in, 1998, p. 81.
tion and liquid jets in cross flows, At. Sprays 28 (2018) 241–254. [68] S. Pichon, G. Black, N. Chaumeix, M. Yahyaoui, J.M. Simmie, H.J. Curran,
[46] J. Wen, Y. Hu, A. Nakanishi, R. Kurose, Atomization and evaporation process R. Donohue, The combustion chemistry of a fuel tracer: measured flame
of liquid fuel jets in crossflows: a numerical study using Eulerian/Lagrangian speeds and ignition delays and a detailed chemical kinetic model for the oxi-
method, Int. J. Multiph. Flow 129 (2020) 103331. dation of acetone, Combust. Flame 156 (2009) 494–504.
[47] A. Pillai, J. Nagao, R. Awane, R. Kurose, Influences of liquid fuel atomization and [69] D. Zuzio, J.E. ezes, B. DiPierro, An improved multiscale Eulerian–Lagrangian
flow rate fluctuations on spray combustion instabilities in a backward-facing method for simulation of atomization process, Comput. Fluids 176 (2018)
step combustor, Combust. Flame 220 (2020) 337–356. 285–301.
[48] J.D. Gounder, A. Kourmatzis, A.R. Masri, Turbulent piloted dilute spray flames: [70] S. Elghobashi, Particle-laden turbulent flows: direct simulation and closure
flow fields and droplet dynamics, Combust. Flame 159 (2012) 3372–3397. models, Appl. Sci. Res. 48 (1991) 301–314.
[49] L.M. Thomas, A. Lowe, A. Satija, A.R. Masri, R.P. Lucht, Five kHz thermometry [71] J.-P. Matas, S. Marty, A. Cartellier, Experimental and analytical study of the
in turbulent spray flames using chirped-probe pulse femtosecond CARS, part I: shear instability of a gas-liquid mixing layer, Phys. Fluids 23 (2011) 094112.
processing and interference analysis, Combust. Flame 200 (2019) 405–416. [72] J.-P. Matas, A. Delon, A. Cartellier, Shear instability of an axisymmetric air-wa-
[50] A. Lowe, L.M. Thomas, A. Satija, R.P. Lucht, A.R. Masri, Five kHz thermometry ter coaxial jet, J. Fluid Mech. 843 (2018) 575–600.
in turbulent spray flames using chirped-probe pulse femtosecond CARS, part
II: structure of reaction zones, Combust. Flame 200 (2019) 417–432.

19