Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/374945301

GT2024-125957 Large Eddy Simulation of Ammonia-Hydrogen Non-Premixed


FlamesStabilized by a Bluff Body

Conference Paper · June 2024

CITATIONS READS

0 101

6 authors, including:

Yu Xia Ishan Verma


ANSYS ANSYS India
88 PUBLICATIONS 312 CITATIONS 58 PUBLICATIONS 192 CITATIONS

SEE PROFILE SEE PROFILE

Sourabh Shrivastava Pravin Nakod


ANSYS ANSYS
23 PUBLICATIONS 20 CITATIONS 58 PUBLICATIONS 254 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Yu Xia on 12 April 2024.

The user has requested enhancement of the downloaded file.


Proceedings of ASME Turbo Expo 2024
Turbomachinery Technical Conference and Exposition
GT2024
June 24-28, 2024, London, England, United Kingdom

GT2024-125957

LARGE EDDY SIMULATION OF AMMONIA-HYDROGEN NON-PREMIXED FLAMES


STABILIZED ON A BLUFF BODY BURNER

Yu Xia∗ Ishan Verma, Sourabh Shrivastava, Pravin Nakod


Ansys UK Ltd. Ansys Software Pvt. Ltd.
Springfield House, Springfield Road Hinjawadi, Pune
Horsham, West Sussex RH12 2RG Maharashtra 411057
United Kingdom India

David F Fletcher Bassam Dally


School of Chemical and Biomolecular Engineering King Abdullah University
The University of Sydney of Science and Technology
New South Wales 2006 Thuwal 23955-6900
Australia Saudi Arabia

ABSTRACT NH3 combustion. In this study, Large Eddy Simulation (LES) is


The demand for carbon-free fuels such as ammonia (NH3 ) utilized to predict a series of NH3 /H2 /N2 non-premixed flames
and hydrogen (H2 ) has been growing rapidly due to stricter stabilized on a bluff body burner. The accuracy of the Flamelet
environmental policies on carbon emissions. Since ammonia Generated Manifold (FGM) combustion model combined with
is easier to store and transport, it is considered the leading LES has been examined, using a commercial CFD solver.
alternative fuel for marine, aviation, and gas turbine industries. The different cracking levels of NH3 into H2 and N2 lead to
However, the narrow flammability, low reactivity, and potential simulation cases ranging from an H2 /N2 flame (i.e., “fully
emissions of nitrogen oxides (NOx ) are important challenges that cracked”), to a flame with 50% NH3 (i.e., “half cracked”).
need to be overcome if ammonia is to be used as a practical fuel The different flame shapes, temperature distributions, NOx and
for industrial use. Recent works have provided important experi- unburnt NH3 emissions have all been simulated, generally
mental data on the characteristics of NH3 flames under premixed matching well the available experimental data. The devel-
and non-premixed combustion modes, focusing on flame speed, oped model settings and numerical workflows may be applied
turbulence-chemistry interaction, and especially the dissocia- to simulating more complex combustors which use NH3 as a fuel.
tion/cracking of NH3 into H2 and N2 . These measurements have
encouraged the use of Computational Fluid Dynamics (CFD) Keywords: Ammonia Combustion; Blended Fuels; Ammo-
for the simulation and scaling of practical ammonia systems, nia Cracking; NOx Emissions; Large Eddy Simulation; Flamelet
evaluating the performance of different numerical models for Generated Manifold

∗ Address all correspondence to Yu Xia. Email: steven.xia@ansys.com


1 Copyright © 2024 by ASME
NOMENCLATURE H2 -carrier fuel [5] and has become a potential energy provider
for marine, aviation, and gas turbine industries.
Cs Smagorinsky model constant [-] However, the narrow flammability, low reactivity, and poten-
c Reaction progress variable [-] tial emissions of nitrogen oxides (NOx ) have all been important
cp Specific heat capacity at constant pressure [J/(kg·K)] challenges for ammonia to be used as a practical fuel in indus-
d Dimensions of the domain in radial direction [m] trial combustors. Given the fact that NH3 is usually blended with
ht Total enthalpy [J/kg] H2 and N2 , recent experiments (e.g., [6,7,8,9]) have investigated
kRES Kinetic energy of resolved eddies [m2 /s2 ] the fundamental characteristics of blended ammonia flames un-
kSGS Turbulent kinetic energy of sub-grid scale eddies [m2 /s2 ] der premixed and non-premixed combustion modes, focusing on
l Dimensions of the domain in axial direction [m] flame speeds, auto-ignition behavior, turbulence-chemistry inter-
lf Effective flame axial length [m] actions, and the impact of dissociation/cracking of NH3 into H2
M Pope criterion evaluating LES mesh quality [-] and N2 , etc. The measurements also range from canonical flames
Nsp Total number of species [-] (e.g., [10]) to flames within gas turbines (e.g., [11,12]) and spark-
r Radial coordinate [m] ignition engines (e.g., [13]).
Tm Mixture temperature [K] Based on the above experimental measurements, Computa-
t Time [s] tional Fluid Dynamics (CFD) simulations have also been per-
U Average velocity of the inflow [m/s] formed (e.g., [14, 15, 16]), where a series of numerical models
X Species volume fraction [-] have been compared and evaluated for different ammonia flames.
x Vector of coordinates, (x, y, z) [-] The validated combustion models and numerical workflows can
then be used to decrease the number of experiments needed,
Y Species mass fraction [-]
eventually reducing the total costs of designing NH3 -fueled
Z Mixture fraction [-]
burners. To predict the highly turbulent flame behavior, scale-
ZS Stoichiometric mixture fraction [-]
resolving models such as Large Eddy Simulation (LES) have
been proven to be more accurate than the Reynolds-Averaged
∆ Cube root of any mesh cell volume [m]
Navier-Stokes (RANS) approach [17, 18].
λ Thermal conductivity [W/(m·K)]
The above-mentioned studies have mainly focused on the
νt Sub-grid scale turbulent viscosity [m2 /s]
premixed ammonia flames blended with other fuels (e.g., H2 ).
ρ Density [kg/m3 ] However, for non-premixed ammonia combustion, very few in-
φ Local fuel/air equivalence ratio [-] vestigations have been made: for example, the experiments by
χc Scalar dissipation rate [s−1 ] Refs. [19, 20] have found that the cracking of NH3 into H2 and
ω̇ Mass reaction rate of a species [kg/s] N2 is similar to a blending process and can reduce the NOx and
unburnt NH3 emissions by improving the reactivity. More re-
⟨·⟩ Time-averaging operator cent measurements performed by Refs. [21, 22] have shown that
∇(·) Vector differential operator by adjusting the staged fuel injection for non-premixed NH3 /H2
flames, the formation of NO can be controlled, thus reducing the
CFD Computational Fluid Dynamics NOx emissions in the exhaust. It has been observed that the dis-
CFL Courant–Friedrichs–Lewy number sociation of NH3 into radicals would enable self-sustained reac-
FGM Flamelet Generated Manifold tions, which increase the flame stability and reduce the pollutant
LES Large Eddy Simulation levels. Since many industrial burners have used non-premixed
PDF Probability Density Function turbulent flames stabilized by bluff bodies and/or swirling flows
ppm Parts per million by volume (e.g., [23]), it is meaningful to perform CFD simulations of non-
RANS Reynolds Averaged Navier-Stokes model premixed ammonia flames to better understand their features and
SGS Sub-Grid Scale to assist the design of combustors using ammonia as a fuel.
URF Under-Relaxation Factor In this work, a series of bluff-body stabilized, NH3 /H2 /N2
blended non-premixed jet flames have been numerically simu-
1 INTRODUCTION lated, using a combination of the Flamelet Generated Manifold
The demand for carbon-free fuels such as ammonia (NH3 ) and (FGM) combustion model and the LES turbulence model. The
hydrogen (H2 ) has been growing rapidly in the last decade due commercial CFD solver, Ansys Fluent® 2024R1 [24], has been
to stricter environmental policies being applied on carbon emis- used. The solution accuracy has been evaluated by comparing
sions [1, 2, 3]. The storage and transport of NH3 in the liquid with the measurements from Alfazazi et al. [16] at King Abdul-
form, at a relatively low tank pressure, are comparatively easier lah University of Science and Technology (KAUST). The differ-
than those of H2 [4]. As a result, NH3 is considered an alternative ent cracking levels of NH3 into H2 and N2 result in three sim-

2 Copyright © 2024 by ASME


ulation cases with different NH3 /H2 /N2 blending ratios, ranging sion flamelets has been found to produce similar results as using
from an H2 /N2 flame (i.e., ”fully cracked”) to a flame with 50% the current 1-D premixed flamelets.
volume of NH3 (i.e., “half cracked”). The time-averaged flame
For the current hybrid NH3 /H2 /N2 combustion, the progress
shapes, temperature distributions, major species mass fractions,
variable (c) is defined as the normalized sum of H2 O and HO2
NOx and unburnt NH3 emissions, etc., have all been computed
mass fractions with different weights [30]:
and validated against available experimental data.
It should be noted that the KAUST experiments [16] also in-    
clude measurements of a blended flame with 72% NH3 , meaning YH2 O −YHu2 O + 10 · YHO2 −YHO
u
2
that the ammonia cracking level is as low as 28%. The numerical c=    , (1)
eq u eq u
simulation of this case is still underway, therefore the results are YH2 O −YH2 O + 10 · YHO2 −YHO2
not shown in this paper.
The rest of this paper is organized as follows: Section 2 where Y denotes the species mass fractions, the superscript u
briefly introduces the FGM combustion model implemented denotes the unburnt reactants at the flame inlet, and the super-
within the current CFD solver. Section 3 explains the computa- script “eq” means the chemical equilibrium state at the flame
tional domain and the key numerical settings. Section 4 discusses outlet. The progress variable (c) serves as a reduced represen-
the simulation results for all the investigated cases, including the tation of the multi-component mixture, which better captures the
flame shape, temperature, species fractions, and their validation time-scales of the mixture reactions. The use of YH2 O in Eq. 1
against the experimental data. The conclusions and the proposed captures well the major heat release reactions, while the inter-
future work are summarized in the final section. mediate species, HO2 , has been found as a good indicator of the
reaction initiation and ignition. Since the mass fraction of HO2 is
2 NUMERICAL METHODOLOGY much smaller than those of the stable major species such as H2 O
2.1 Reaction Progress Variable and H2 , the weight of HO2 has been increased by 10 times in
In the present study, the Flamelet Generated Manifold Eq. 1 to include its impact on c. This weight can be even higher,
(FGM) combustion model [25] has been used to model the chem- but since HO2 is a radical with both positive and negative rates,
ical reaction process and its interaction with the turbulence. The choosing a higher weight may destroy the monotonicity of the
FGM is a combustion chemistry reduction technique, capturing progress variable. The implemented weights in Eq. 1 have been
the most important aspects of a flame front’s internal structure optimized based on a series of numerical tests, and are consistent
with lower computational costs. It assumes that the scalar evolu- with those used in a recent study of lifted pure H2 flames [17,18].
tion, which defines the path of reaction progress in the thermo-
chemical manifold for a 3-D turbulent flame, may be approxi-
mated by the scalar evolution in a lower-dimensional, 1-D lam-
inar flame. The FGM model also represents the species and 2.2 Flamelet Equations and PDF Table Generation
mixture properties by using reduced variables, such as mixture
The above definition (i.e., Eq. 1) of the progress variable, c,
fraction (denoted by Z) and reaction progress variable (denoted
is then used to define the 1-D premixed flamelet equations in the
by c), and solves the transport equations of these variables in
reaction progress space (i.e., c-space) as follows [31, 32]:
a 3-D CFD simulation. In this work, the FGM model imple-
mented in the CFD solver has been proven accurate across a wide
∂Yk ∂Yk ∂ 2Yk
range of flames, by simulations validated against rich experimen- ρ + ω̇c = ρ χc 2 + ω̇k , (2)
tal data [26, 27, 28, 29, 17, 18]. It has been considered appropriate ∂t ∂c ∂c
for the present blended ammonia flame simulations.
Depending on the mixedness of the fresh fuel/air mix- ∂ Tm ∂ Tm ∂ 2 Tm 1
+ ω̇c = ρ χc − ht,k ω̇k
c p,m ∑
ρ
ture, the resulting flames can be treated as premixed, partially- ∂t ∂c ∂ c2 k
premixed or diffusion, and can be modeled by FGM based on 1- ! (3)
D premixed or diffusion flamelets. The FGM manifold is created ρ χc ∂ c p,m ∂Yk ∂ Tm
+ + ∑ c p,k ,
from steady state flamelet solutions plus one quenching flamelet, c p,m ∂c k ∂c ∂c
and is typically associated with one underlying flamelet configu-
ration, e.g., premixed or diffusion. In this study, the FGM model where ρ denotes fluid density, t the time, ω̇k , ht,k and c p,k are
is generated based on 1-D premixed flamelets, which are solved the mass reaction rate, total enthalpy and specific heat capacity
in the space of the reaction progress variable, i.e., c-space. A of the k-th species, respectively, with k = 1, · · · , Nsp and Nsp the
value of c = 0 refers to the fresh mixture before reaction, while total number of involved species. Tm denotes the mixture temper-
c = 1 corresponds to the fully burnt products after combustion. ature, c p,m the mixture heat capacity, and χc the mixture’s scalar
For the present blended ammonia flames, the use of 1-D diffu- dissipation rate, which is a function of c, and defined as:

3 Copyright © 2024 by ASME


TABLE 1: Numbers of grid points used by the five variables to TABLE 2: Properties of the fuel stream for each simulation case.
generate a PDF table for the FGM model.
Case Name FNH3 -0 FNH3 -25 FNH3 -50
Variable Number of grid points XNH3 [%] 0 25 50
Mixture fraction 64 XH2 [%] 75 56.25 37.5
Progress variable 64 XN2 [%] 25 18.75 12.5
Enthalpy 20 NH3 cracking [%] 100 75 50
Mixture fraction variance 19 Reynolds number [-] 5500 5500 5500
Progress variable variance 26 Momentum flux ratio [-] 5 3.03 2.06

λ under-relaxation factors, and the number of sub-iterations used


χc (c) = |∇c|2 , (4)
ρc p to achieve convergence within each time-step. More details of
the applied numerical settings can be found in Sec. 3.2.
with λ the thermal conductivity. χc (c) serves as an input to the Finally, the implemented FGM model utilizes the unity
above equation set, Eqs. 2 – 3, but the explicit expression of χc (c) Lewis number (Le) assumption in the combustion modeling.
is usually unknown, which can be estimated from the solutions This means that the Schmidt number (Sc) is always equal to
of 1-D laminar flames. the Prandtl number (Pr) in the current simulations, because Le
To account for the turbulent fluctuations, the present for- = Sc/Pr. The Schmidt number varies within the domain (mainly
mulation of FGM applies a non-adiabatic, beta-type Probabil- along the axial direction) depending on the local flow solutions
ity Density Function (β -PDF) lookup table to store the pre- and the inlet fuel composition, and its time-averaged value ranges
calculated variables, which are then used to solve the temper- from ∼0.20 to ∼0.67 for all the investigated simulation cases.
ature, density, and species mass and volume fractions with re-
duced computational costs. The numbers of grid points used by 3 KAUST AMMONIA-HYDROGEN FLAMES
the five variables to generate a PDF table are listed in Table 1. The series of turbulent non-premixed NH3 /H2 /N2 jet flames in-
The automated grid refinement is enabled for all the variables. vestigated in this work are known as the “KAUST Flames” [16],
As the PDF table is based on the non-adiabatic energy condition, which are operated at the atmospheric pressure. Detailed ex-
it is also a function of enthalpy. In this study, the impact of en- perimental data are available for time-averaged temperature and
thalpy (i.e., heat loss) is considered during density, specific heat multiple species volume and mass fractions (e.g., OH, NO, NO2 ,
capacity, and temperature computations, but is ignored on all the N2 O, NH2 and NH3 ). This section mainly introduces the com-
species concentration calculations, which makes the created PDF putational domain, mesh resolution, and numerical setup of the
table a pseudo 5-D table. current simulations for the KAUST flames.

2.3 Definition and Optimization of FGM Model 3.1 Computational Domain and Boundary Conditions
The present FGM model is then defined using two streams Firstly, a 3-D cylindrical computational domain represent-
as inputs: one is a “fuel stream”, which contains a jet mixture ing the KAUST experimental rig has been created, as shown in
of NH3 , H2 , and N2 . The second stream is an “oxidizer stream”, Fig. 1. The blended fuel jet enters the domain through a long and
which refers to a co-flow of pure air. For each simulation case, narrow fuel pipe, before mixing with the surrounding air co-flow
i.e., FNH3 -0, FNH3 -25 and FNH3 -50, the species volume frac- at the pipe exit, and then being ignited and stabilized by the bluff
tions within the fuel stream (i.e., XNH3 , XH2 , and XN2 ) have been body. The fuel pipe’s exit plane coincides with the bluff body’s
listed in Table 2, which are used to define the FGM model for top surface. Two domain inlets have been defined, which are:
each case. Since the ideal gas assumption has been used for the (i) a “Fuel Inlet”, which injects a fresh mixture of NH3 , H2 , and
mixture in this study, the volume fraction of any species is equal N2 at very high velocities; (ii) a “Co-flow Inlet”, which injects
to its mole fraction. pure air at a low speed. Both inlets have an ambient temperature
For all the present simulations, the compressible version of of 300 K, and are defined as “Velocity Inlets” in the present CFD
the FGM model has been applied for better solution convergence, solver [24].
and the Finite-Rate Closure Model [30] has been used to com- The inflow axial velocity profiles at the fuel inlet are shown
pute the source term in progress variable equations. The alge- in Fig. 2 for all the three simulation cases: FNH3 -0, FNH3 -25
braic method [30] is used to solve the variable variances. The and FNH3 -50, which are obtained based on separate simulations
global solution controls of the CFD solver have also been opti- of fully-developed turbulent flows within the fuel pipe. The max-
mized for the FGM model, such as the computational time-step, imum velocity magnitudes can be found at the centerline (i.e., r

4 Copyright © 2024 by ASME


FIGURE 2: Inflow axial velocity profiles at the fuel inlet for all
three simulation cases.

divided by the air co-flow momentum, ρU 2 coflow , with U the




corresponding average inflow velocity. All the fuel jet properties


listed in Table 2 are consistent with the experimental settings in
Ref. [16].
The air inflow at the co-flow inlet has a constant and uniform
velocity of Ucoflow = 12.84 m/s. The co-flow would be gradually
entrained into the central jet zone and mix with the fuel, espe-
cially beyond the jet’s mixing layer. The mixture fraction, Z,
is set as unity at the fuel inlet, as no oxidizer exists within the
NH3 /H2 /N2 jet. For the co-flow inlet, Z is set as zero, as the co-
flow only contains pure air. The value of progress variable (c)
is defined as zero at both fuel and co-flow inlets, as no chemical
reactions should occur at any inlet.
The Synthetic Turbulence Generator [33, 34] is used at the
fuel inlet to reconstruct the turbulent fluctuations from the ap-
plied velocity profiles. The turbulence intensity and turbulent
viscosity ratio at the fuel inlet have been selected as 10% and 10,
respectively. These applied turbulence settings have been found
to provide sufficient perturbations to the fuel inflow after some
sensitivity analyses. No turbulent perturbations have been de-
FIGURE 1: Computational domain with boundary types: (a) view
fined at the co-flow inlet, as a perturbation to the co-flow has
from the bottom inlets; (b) view of an instantaneous flame on the
been found to have negligible impacts on the jet flame solutions.
mid-plane.
All solid walls in the domain, including the bluff body sur-
faces and the fuel pipe wall (see Fig. 1), have been defined as no-
= 0 m), which are {64.84, 45.18, 34.01} m/s for the three in- slip and adiabatic in this study. As the experimental rig is open
flow profiles. The average jet velocities for the three profiles to the ambient air, i.e., measuring the “open flames”, the domain
are computed based on their mass flow rates, and are equal to exit, side boundaries, and the outer bottom surface, have all been
Ujet = {53.24, 37.01, 27.82} m/s for the three LES cases. defined as “Pressure Outlets” in the solver at atmospheric pres-
The detailed species compositions at the fuel inlet have been sure, as shown in Fig. 1. An example simulation of a desired jet
provided by Table 2 for each LES case, being consistent with the flame stabilized on the bluff body can be seen in Fig. 1(b).
FGM fuel stream definition used for the same case. Table 2 also In this work, the entire computational domain is meshed
lists the Reynolds number of the fuel jets, which is fixed as 5500 with poly-hexcore cells, as shown in Fig. 3, using the CFD
for all the simulation cases. The “momentum flux ratio” pro- solver’s inbuilt meshing tools [24]. The total cell count of the
vided by Table 2 is defined as the fuel jet momentum, ρU 2 jet ,

mesh is around 15.86 million. Mesh refinements have been

5 Copyright © 2024 by ASME


TABLE 3: Key dimensions of the computational domain.

Dimension d1 d2 d3 d4 d5
Value [mm] 4.6 50.8 104 150 225
Dimension d6 l1 l2 l3
Value [mm] 450 36 90 380

current mesh is appropriate for LES simulations. The criterion is


defined as follows:

kSGS (xx,t)
M (xx,t) = , (5)
kSGS (xx,t) + kRES (xx,t)

where x denotes the vector of coordinates (x, y, z). kSGS is the


turbulent kinetic energy of the sub-grid scale (SGS) eddies. Since
this work uses the Smagorinsky model [38, 39, 40] for LES, kSGS
FIGURE 3: Poly-hexcore mesh with a total of 15.86 million cells. can be approximated as:
The mesh cell sizes are: Zone 1: 0.16 mm; Zone 2: 0.64 mm;  2
Zone 3: 2.56 mm; Zone 4: ∼10 mm. 1 νt (xx,t)
kSGS (xx,t) = , (6)
0.3 Cs · ∆ (xx)
placed along the domain’s centerline, covering the entire fuel
pipe and the jet’s width. As indicated in Fig. 3, the smallest with νt the SGS turbulent viscosity, and ∆ the cubic root of any
cell size (0.16 mm) can be found in “Zone 1”, which has a di- mesh cell volume. Cs is a constant and equal to 0.1 in this work.
ameter slightly larger than the fuel pipe diameter (denoted d1 ). kRES is the kinetic energy of the resolved eddies, defined as:
A slightly coarser cell size of 0.64 mm is used for the rest of
the core reaction zone, denoted “Zone 2”. Larger mesh cells 1h i
have been applied within the outer co-flow domain, ranging from kRES (xx,t) = (u − ⟨u⟩)2 + (v − ⟨v⟩)2 + (w − ⟨w⟩)2 , (7)
2
2.56 mm in “Zone 3” to ∼10 mm in “Zone 4”, as moving out-
wards along the radial direction. The key dimensions of the com- where (u, v, w) are velocity components in (x, y, z) directions,
putational domain are defined in Fig. 3, with their values listed and ⟨·⟩ is the time-averaging operator.
in Table 3.
According to Refs. [36, 37], an accurate LES simulation
The present mesh has been found to provide a sufficient so- aims at resolving at least 80% of the turbulent energy spectrum,
lution accuracy, and a detailed mesh sensitivity analysis has been which means the magnitude of M (xx,t) should not exceed 20% in
performed in Sec. 4.3 of this paper. The use of poly-hexcore the computational domain. As shown in Fig. 4, the core fuel/air
mesh type is based on its efficiency in mesh generation and mixing and reaction zone (i.e., “Zone 1” and “Zone 2” in Fig. 3)
its accuracy solving a wide range of reacting flow applications generally has M < 0.2, while the outer zones (e.g., “Zone 3” and
(e.g., [29, 17, 18]). “Zone 4”) have some areas of M ≥ 0.2. As the flow in the outer
In this work, the full 360◦ domain (see Fig. 1) has always zones usually does not have large impacts on the inner zone mix-
been used for all the simulations, in order to ensure that any po- ing and reactions, the present mesh resolution may be considered
tential non-symmetric mixing and flame behavior can be accu- as sufficient and appropriate for the current LES simulations.
rately captured during the LES calculations, especially for those
flow features in the tangential directions. As shown in Ref. [35],
using a symmetric portion (i.e., “sector”) of the full 360◦ geome- 3.2 Physical Models and Numerical Settings
try would usually give similar results for the RANS calculations, The physical models and numerical settings have all been
but it could bring considerable solution differences when running carefully selected. The pressure-based CFD solver [24, 30] is
LES simulations. This is due to the non-isotropic flow field under used for all the simulations. The LES model is used for turbu-
LES which generates non-symmetric flow features. lence, which applies the dynamic Smagorinsky-Lilly model [38,
The Pope criterion [36, 37] has been used to evaluate if the 39, 40] for the sub-grid scale modeling. The Flamelet Generated

6 Copyright © 2024 by ASME


TABLE 4: Under-relaxation factors (URFs) for all simulations.
Variable Pressure Density
URF [-] 0.75 0.75
Variable Body Forces Momentum
URF [-] 1 0.75
Variable Energy Temperature
URF [-] 1 1
Variable Progress Variable Mean Mixture Fraction
URF [-] 0.9 1

(i.e., ∼0.04 - 0.06 s), in order to eliminate any initial transients,


and then for another ∼5 - 6 flow-through times (i.e., ∼0.1 -
0.12 s) to collect the statistical data (e.g., time-averaged values).
FIGURE 4: Pope criterion: the instantaneous magnitude of
M (xx,t) in the computational domain for LES case FNH3 -25. 4 SIMULATION RESULTS
In this section, the simulation results of all the investigated LES
Manifold (FGM) model [24] is applied for combustion modeling, cases have been presented and discussed. The solution accuracy
as defined in detail in Sec. 2. has been evaluated by comparing with the available experimental
The chemical mechanism for the blended NH3 /H2 reactions data. Three simulation cases, FNH3 -0, FNH3 -25, and FNH3 -50,
used in this work is the skeletal, 33 species (214 reaction steps) as defined in Table 2, have been validated, which refer to 0%,
reaction scheme available within Ansys Model Fuel Library® 25%, and 50% ammonia in the fuel, respectively. The volume
(MFL) [41]. Sensitivity analyses on different ammonia reac- fractions of the other species (e.g., H2 , N2 ) in the fuel jet are de-
tion mechanisms (e.g., Nakamura [42], Okafor [8], Otomo [43], termined based on the assumption that NH3 dissociates into H2
Zhang [44] and Glarborg [45]) have also been performed across and N2 in a 3:1 volume ratio, i.e., NH3 ⇒ 3H2 + N2 . The inves-
multiple operating conditions and equivalence ratios, on vari- tigated cases represent different ammonia dissociation scenarios.
ables such as laminar flame speed and adiabatic flame temper- For example, case FNH3 -0 refers to a situation that all the am-
ature. The present Ansys MFL® mechanism has been consis- monia has been cracked, resulting in 75% volume of H2 and 25%
tently found as one of the most accurate mechanisms, and has volume of N2 in the fuel jet.
therefore been used in this work. A detailed comparison of the
above mechanisms is, however, beyond the scope of this paper.
For the spatial discretizations, the Least Squares Cell-Based 4.1 Flame Structures and Effective Flame Lengths
(LSCB) method is used for all the gradients, the second-order The computed flame structures for all the simulation cases
Bounded Central Differencing (BCD) scheme for momentum, are shown in Fig. 5. The time-averaged flow temperature, ⟨T ⟩, of
the Second-Order method for pressure, and the Second-Order the three flames have been presented, indicating a similar flame
Upwind scheme for all the other variables (e.g., density, energy structure consisting of three zones from upstream fuel exit to
and mean mixture fraction). The Bounded Second-Order Im- downstream domain outlet: (i) a mixing and recirculation zone
plicit scheme is used for all the temporal discretizations. The near the bluff body (Zone “A”); (ii) a narrowed “neck zone”
SIMPLEC scheme is applied for the pressure-velocity coupling. (Zone “B”); (iii) an expanded jet flame (Zone “C”). This com-
The under-relaxation factors (URFs) for different flow variables bined flame structure is consistent with those measured in the
have also been optimized, with their values listed in Table 4 for experiments (e.g., [46, 47, 16]). The axial length of each flame
all the present LES simulations. zone has been listed in Table 5 for all the LES cases, which shows
A transient solver with a fixed time-step of 1×10−5 s is ap- the increase of axial length for Zone “A” as more NH3 is being
plied, ensuring a good balance between a reasonably small CFL added to the domain. Such recirculation zone “A” is vital for
number and limited computational costs. The number of sub- the flame stabilization, as it recirculates hot burnt products back
iterations per time step is set as 12, which ensures a sufficient to the bluff-body surfaces, and serves as a continuous ignition
pressure-velocity coupling and flow solution convergence, with- source for the incoming fuel stream, ensuring the flame does not
out significantly increasing the computational costs. All the LES lift or blow off.
cases have been firstly computed for ∼2 - 3 flow-through times In order to better compare the axial flame lengths, an “effec-

7 Copyright © 2024 by ASME


cases (with less than 8% difference, see Table 5), which suggests
that the effective flame length may be linearly proportional to the
jet velocity in general.
(ii) The recirculation region (i.e., Zone “A”) is becoming
hotter with an increase of NH3 concentration, which may be at-
tributed to the increases of hot burnt products (e.g., H2 O) and
unburnt NH3 fragments being recirculated in this region.
(iii) In contrast, the temperatures of the neck zone (Zone
“B”) and downstream jet flame (Zone “C”) have largely reduced
when ammonia increases, suggesting weaker reaction activities
would occur within these regions. This is associated with a larger
portion of the fuel being burnt in the recirculation zone “A”.
(iv) Since the jet velocity decreases as XNH3 increases
(see Fig.
 2), the  momentum flux ratio of jet-to-coflow, i.e.,
ρU 2 jet / ρU 2 coflow (see Table 2), is also decreasing as the co-
flow velocity remains unchanged, which radially shifts the high-
temperature region within Zone “A” from the inner shear layer
next to the fuel jet, to the outer shear layer close to the co-flow
(see Fig. 5(c)). This alters the overall temperature distribution
FIGURE 5: Time-averaged flow temperature, ⟨T ⟩, on the mid-
within the recirculation zone “A”, with lower temperatures being
plane for LES simulation cases: (a) FNH3 -0; (b) FNH3 -25; found close to the core jet, but hotter in the outer mixing zone.
(c) FNH3 -50. Zone “A” refers to the recirculation zone; Zone The above simulated trends from case FNH3 -0 to case
“B” is the “neck zone”; Zone “C” is the expanded jet flame. l f FNH3 -50 are consistent with the experiments [16], suggesting
denotes the effective flame axial length. the reductions of both jet velocity and ammonia cracking level
would lead to a more compact flame with a lower mixture reac-
TABLE 5: Lengths of flame zones and effective flame lengths. tivity (also known as “burning rate”). It has also been noted that
the change of the flame behavior (e.g., l f ) from 0% to 25% NH3
Case FNH3 -0 FNH3 -25 FNH3 -50
(see Figs. 5(a, b)) is more significant than that from 25% to 50%
Zone “A” [mm] 30.2 38.3 45.9 NH3 (see Figs. 5(b, c)).
Zone “B” [mm] 51.9 51.9 51.9 In order to better evaluate the simulated flames and the reac-
tion zones, the mass fractions of species OH have been compared
Zone “C” [mm] 207.9 199.8 192.2 in Fig 6. It has been found that for the case with no ammonia in
l f [mm] 250.9 167.7 142.5 the fuel (i.e., FNH3 -0), the axial distribution of OH is largely ex-
tended (see Fig. 6(a)), indicating a long burning zone which is
l f /Ujet [ms] 4.71 4.53 5.12
in line with the long flame seen in Fig. 5(a). By adding NH3
into the fuel jet (e.g., case FNH3 -25), most OH would now be
concentrated within the recirculation zone “A” (see Fig. 6(b)),
tive flame length”, denoted l f , has been defined in this simulation suggesting the shortening of the burning zone and more concen-
work as the axial distance between the bluff-body top surface and trated reaction activities. After further increasing the ammonia
the top edge of the main hot region, as indicated in Fig. 5. The at the inlet (e.g., case FNH3 -50, see Fig. 6(c)), all the reactions
values of l f are {250.90, 167.68, 142.49} mm for the three LES have been weakened in the domain, confirming the reduction of
cases, as listed in Table 5. the fuel/air mixture’s reactivity with higher ammonia in the fuel.
By comparing the flames shown in Fig. 5, some key phe-
nomena have been observed, which are similar to those measured
in the experiments [16]: 4.2 Mixture Fractions and Local Equivalence Ratios
(i) The effective flame length, l f , is decreasing with an in- The time-averaged mixture fraction, ⟨Z⟩, is plotted in Fig. 7
crease of NH3 volume fraction at the fuel inlet (e.g., from 0% to for the three flames. The central fuel jet (with ⟨Z⟩ = 1) penetrates
50%). This may be related to the decrease of the fuel jet velocity through the recirculation zone “A” and mixes with the air co-flow
as defined in Fig. 2, which reduces the momentum flux ratio (see (with ⟨Z⟩ = 0), before being ignited and generating a flame. The
Table 2) and increases the dominance of the air co-flow over the jet’s mixture fraction gradually decreases when moving down-
fuel jet. Furthermore, the ratio of l f to the average jet velocity, stream, as more fuel mixes and reacts with the air co-flow. This
Ujet , i.e., l f /Ujet , has a very close value between the three LES trend is consistent with that observed in a recent numerical study

8 Copyright © 2024 by ASME


FIGURE 6: Time-averaged OH mass fraction, ⟨YOH ⟩, on the mid- FIGURE 7: Time-averaged mixture fraction, ⟨Z⟩, on the mid-
plane for LES cases: (a) FNH3 -0; (b) FNH3 -25; (c) FNH3 -50. plane for LES cases: (a) FNH3 -0; (b) FNH3 -25; (c) FNH3 -50.
Please refer to the caption of Fig. 5 for more information. Please refer to the caption of Fig. 5 for more information.

of pure hydrogen flames [17, 18].


TABLE 6: Approximate average axial lengths and diameters of
As the volume fraction of NH3 increases, the flame becomes
the ammonia cracking (or hydrogen reaction) regions for all LES
more compact with a reduced axial length, meaning more fuel is
cases. All regions start from the fuel pipe exit plane.
now being burnt within the recirculation zone “A”. These obser-
vations are consistent with those seen in Fig. 5 and in a previous
Case FNH3 -0 FNH3 -25 FNH3 -50
CFD study of the same flames [16].
As the jet-to-coflow momentum flux ratio reduces from 5 Axial length [mm] 120 110 90
for case FNH3 -0 to 2.06 for case FNH3 -50 (see Table 2), a more Diameter [mm] 10 15 30
complete fuel/air mixing and reaction would happen in Zone “A”,
leading to a slower transition of ⟨Z⟩ from unity to zero, as mov-
ing radially outwards from the jet to the co-flow within the re-
circulation zone (see Fig. 7(c)). It has also been found that the
ammonia cracking mainly occurs between the fuel jet and the co- Z (1 − ZS )
φ= , (8)
flow within Zones “A” and “B”. As more NH3 is being added to ZS (1 − Z)
the fuel, the cracking region would become wider but shorter, in
line with the simulated flame variations. The approximate am- where ZS is the stoichiometric mixture fraction, which corre-
monia cracking region average lengths and diameters have been sponds to the locations where both fuel and oxygen are consumed
provided in Table 6 for all the LES cases. For case FNH3 -0, no completely. The value of ZS is stored within the PDF table gen-
ammonia exists in the fuel jet, so the “ammonia cracking region” erated by the FGM model for each LES case.
now refers to the main hydrogen reaction zone. As shown in Fig. 8, the time-averaged local equivalence ra-
Since the fuel and air are not premixed, the fuel/air mix- tio, ⟨φ ⟩, has been compared across the three LES cases. Similar
ture would always burn at its stoichiometric state in the reaction to the comparison of the mean mixture fraction, ⟨Z⟩ (see Fig. 7),
zone, so using a global average equivalence ratio for the mixture rich fuel/air mixture with ⟨φ ⟩ > 1 can be found near the cen-
may not be meaningful, as the burning now occurs for each local terline, while lean mixtures (⟨φ ⟩ < 1) can be seen in the outer
mixture and not globally. Each location in the domain now has mixing zones. The rich and lean regions are separated by the iso-
its own “local equivalence ratio” based on the local mixing level surface of ⟨φ ⟩ = 1 (see black lines in Fig. 8). As more and more
and the flow structure. In this work, such local fuel/air equiva- ammonia is being added to the fuel, the mixture in the upstream
lence ratio (denoted φ ) is defined based on the mixture fraction, Zone “A” is becoming richer and richer, suggesting more and
Z, as: more reactions are now taking place in this recirculation zone.

9 Copyright © 2024 by ASME


FIGURE 9: Time-averaged temperature, ⟨T ⟩, along the centerline
FIGURE 8: Time-averaged local equivalence ratio, ⟨φ ⟩, on the for LES cases: FNH3 -0 (solid line); FNH3 -25 (dash-dotted line);
mid-plane for LES cases: (a) FNH3 -0; (b) FNH3 -25; (c) FNH3 - FNH3 -50 (dotted line), compared with experimental data [16]:
50. Please refer to the caption of Fig. 5 for more information. FNH3 -0 (•); FNH3 -25 (■); FNH3 -50 (▲). The origin of axial
The black lines indicate ⟨φ ⟩ = 1. coordinate refers to the fuel pipe exit plane.

4.3 Temperature and Species Concentration Profiles burnt products (e.g., H2 O, NH3 fragments) from an ammonia-
The time-averaged flow temperatures shown in Fig. 5 have rich combustion would contribute to a higher thermal radiation
now been validated against the experimental data [16]. As shown and hence more heat losses, which can considerably bring down
in Fig. 9, the predicted ⟨T ⟩ axial profiles along the centerline gen- the measured flow temperature in the reaction zones.
erally match well the measurements for all the simulation cases, To better evaluate the flow temperature drop with an increase
including the magnitudes and the trends, despite small underpre- of ammonia (see Fig. 9), the adiabatic flame temperatures of the
dictions near the fuel pipe exit plane (i.e., x ≃ 0 – 80 mm). investigated NH3 /H2 /N2 mixtures have been computed using An-
It has also been observed in Fig. 9 that the axial location sys Chemkin-Pro® 2024R1 [48]. As the present flame is non-
of the highest temperature would monotonically shift upstream premixed, the fuel/air mixture would always burn at the stoichio-
as the volume fraction of NH3 increases, accompanied with a metric state, leading to adiabatic flame temperatures of {2257.23,
considerable drop in the temperature magnitude, especially for 2187.39, 2138.44} K for the three LES cases: {FNH3-0, FNH3-
the maximum temperature peaks and in the downstream regions. 25, FNH3-50}. The possible reason for the above-computed adi-
These ⟨T ⟩ trends are consistent with the flame variations shown abatic flame temperatures to be very similar between the different
in Fig. 5, where the flame is becoming shorter and the flow be- cases is that ammonia would crack into H2 and N2 in a 3:1 vol-
comes cooler in the central jet zone as ammonia increases. ume ratio, and an adiabatic flame temperature may be reached
The above temperature reduction may be due to the decrease when all the ammonia in the fuel has been fully cracked, which
of the fuel/air mixture’s reactivity and hence the heat release rate eventually gives identical volume fractions of 75% H2 and 25%
from the flame, when more NH3 but less H2 are being added to N2 for all the three simulation cases (see Table 2). The small re-
the fuel. This is because the ammonia itself has a much lower duction of the computed adiabatic flame temperature when am-
burning rate and a smaller energy density (∼18.8 MJ/kg) than monia increases may be attributed to an increased number of en-
those of hydrogen (which is highly reactive with an energy den- dothermic ammonia cracking reactions [49], which absorb more
sity of ∼120 MJ/kg), and the partial dissociation of NH3 into H2 energy from the surroundings.
cannot compensate the direct reduction of H2 within the fuel jet. Therefore, the main possible reason for the CFD-simulated
Another possible reason for the experimentally measured flow temperature drop in Fig. 9 is that the ammonia has a much
flow temperature drop in Fig. 9 is the increase of flame thermal lower reactivity and energy density than those of hydrogen.
radiation when ammonia increases in the fuel. As reported in Please note, no radiation model has been used in the present CFD
the KAUST experiments [16], the intermediate species and the simulations.

10 Copyright © 2024 by ASME


Another finding from Fig. 9 is the inflection of the ⟨T ⟩
profiles, where the slope of the curves suddenly decreases at
∼40 mm downstream of the fuel exit in the experiments (i.e.,
x ≃ 40 mm). This may correspond to the beginning of the nar-
rowed neck zone “B” as shown in Fig. 5, where higher strain
rates between the fuel and air have been measured. For the cur-
rent LES simulations, the slope change in ⟨T ⟩ profiles has been
captured, which also occurs at the x ≃ 40 mm location for case
FNH3 -0, but slightly further upstream at x ≃ 30 mm for cases
FNH3 -25 and FNH3 -50. For both measurements and simula-
tions, the ⟨T ⟩ profile slope change is more pronounced when
more ammonia is being added to the fuel jet.
It is also noticed in Fig. 9 that for all of the LES cases, the
flow temperature remains at 300 K for a short distance of x ≃ 0 –
20 mm before being heated-up, a phenomenon which has not
been measured by the experiments. This may suggest that the
mixing between the fuel and air does not happen immediately
after the fuel injection in the simulations, which consequently
causes the delay of ignition and may explain the underpredictions
of flow temperature between x ≃ 0 – 50 mm for case FNH3 -0 and
x ≃ 0 – 80 mm for cases FNH3 -25 and FNH3 -50. Further inves-
tigations on improving the inlet turbulence conditions in order to
enhance the initial fuel/air mixing are now underway.
For the species mass fractions, ⟨Y ⟩, the experiments and the
simulations have been compared in Fig. 10 for two LES cases:
FNH3 -25 and FNH3 -50. No experimental data are available for
the H2 /N2 flame (i.e., case FNH3 -0). As shown in Fig. 10, for
the intermediate species OH and NH2 , their predicted mass frac-
tions, ⟨YOH ⟩ and ⟨YNH2 ⟩, are both lower than the corresponding
measurements, while the computed unburnt NH3 mass fraction
(i.e., ⟨YNH3 ⟩) matches better the experimental profile. The possi-
ble reason is that the minor species, e.g., OH and NH2 , are low
in concentration and take a longer time to evolve than the major
species such as NH3 . As the progress variable of the FGM model
mainly focuses on major species, additional transport equations
may need to be solved for these slow-forming, minor species. FIGURE 10: Time-averaged mass fractions of (blue) OH, ⟨YOH ⟩,
This investigation is, however, beyond the scope of this paper. (green) NH2 , ⟨YNH2 × 10⟩, and (red) NH3 , ⟨YNH3 ⟩, along the
By comparing the two LES cases, FNH3 -25 and FNH3 -50, centerline for LES cases (dash-dotted lines): (a) FNH3 -25 and
in Fig. 10, one could see that the level of OH is lower in the (b) FNH3 -50, compared with experiments [16] (solid lines).
richer case with 50% NH3 , while oppositely the volume of NH2
is higher in the case of 50% ammonia, for both measurements computed 1-D profiles along the centerline have also been com-
and simulations. The mechanism here is that NH3 cracks into pared with the experiments, as shown in Fig. 11 for case FNH3 -
NH2 , and then NH2 reacts with the oxidizer to generate OH. 25, and in Fig. 12 for case FNH3 -50. For both LES cases, the
Although a higher ammonia concentration would produce more predicted ⟨XO2 ⟩ profiles generally match well the measurements,
NH2 as shown in Fig. 10(b), it would also impose termination despite small underpredictions mainly in the upstream areas. The
reactions which suppress the formation of active radicals such as calculated ⟨XNO ⟩, ⟨XNO2 ⟩ and ⟨XNH3 ⟩ for both cases also match
H, OH, and O in the flame. This may explain the lower OH mass the experimental trends well, but their predicted magnitudes have
fraction in an ammonia-richer flame (see Fig. 10(b)). The peaks considerable deviations from the experiments, especially in the
of OH and NH2 mass fractions both move upstream with an in- upstream zones. Given the fact that the measurements of very
creased supply of NH3 , in line with the shortening of the flame low-in-volume species have been very challenging themselves,
length seen in Fig. 5. the current predictions of ⟨XNOx ⟩ and ⟨XNH3 ⟩ in ppm concentra-
For the volume fractions of O2 , NOx , and unburnt NH3 , their tions may still be acceptable.

11 Copyright © 2024 by ASME


FIGURE 13: Time-averaged volume fraction of NOx , ⟨XNOx ⟩ =
⟨XNO ⟩ + ⟨XNO2 ⟩, along the centerline for LES cases (dash-dotted
lines): (a) FNH3 -25 and (b) FNH3 -50, compared with experi-
ments [16] (solid lines).

NO2 have been measured in the experiments for case FNH3 -50
(see Figs. 12(b, c)), mainly due to the doubling of NH3 volume
fraction in the fuel jet. On the other hand, the current simula-
tions fail to distinguish between the two LES cases on ⟨XNO ⟩ and
⟨XNO2 ⟩ calculations. Therefore, the prediction error on the NOx
emissions of the FNH3 -50 flame is greater than the error on the
FIGURE 11: Time-averaged volume fractions of (a) O2 , ⟨XO2 ⟩;
FNH3 -25 flame.
(b) NO, ⟨XNO ⟩; (c) NO2 , ⟨XNO2 ⟩; (d) NH3 , ⟨XNH3 ⟩, along the It has also been confirmed that a large amount of unburnt
centerline for LES case FNH3 -25 (symbol ◦ + dash-dotted line), NH3 would exist in the domain and be transported downstream
compared with experiments [16] (symbol • + solid line). by the flow for case FNH3 -50 (see Fig. 12(d)), while in the lower
NH3 case FNH3 -25, all the ammonia has been consumed before
reaching the domain outlet (see Fig. 11(d)). Such difference on
the measured ⟨XNH3 ⟩ has been successfully predicted by the sim-
ulations, as shown in Figs. 11(d) and 12(d).
Finally, the sum of ⟨XNO ⟩ and ⟨XNO2 ⟩, which may be con-
sidered as the time-averaged NOx volume fraction (i.e., ⟨XNOx ⟩),
has been computed and compared between simulations and ex-
periments, as shown in Fig. 13 for both FNH3 -25 and FNH3 -50.
As the measured and the simulated ⟨XNO ⟩ values are both orders
of magnitude higher than those of NO2 for the two cases, their
comparisons on ⟨XNOx ⟩ (see Fig. 13) are very similar to their
comparisons on ⟨XNO ⟩ (see Figs 11(b) and 12(b)).
Tables 7 and 8 have listed the measured and the simulated
values (and their differences in %) of ⟨XO2 ⟩, ⟨XNO ⟩, ⟨XNO2 ⟩,
⟨XNOx ⟩ and ⟨XNH3 ⟩ at all the measurement locations for cases
FNH3 -25 and FNH3 -50, respectively. It is confirmed that the
prediction errors of both LES cases are relatively small on ⟨XO2 ⟩,
but are significantly larger for ⟨XNOx ⟩ and ⟨XNH3 ⟩.
Please note, the Ansys MFL® mechanism [41] used in this
work includes oxidation reactions from NO to NO2 , such as:
FIGURE 12: Time-averaged volume fractions of (a) O2 , ⟨XO2 ⟩; (i) NO + O ⇒ NO2 , and (ii) NO + HO2 ⇒ NO2 + OH. Therefore,
(b) NO, ⟨XNO ⟩; (c) NO2 , ⟨XNO2 ⟩; (d) NH3 , ⟨XNH3 ⟩, along the the present LES/FGM simulations are able to predict individual
centerline for LES case FNH3 -50 (symbol □ + dash-dotted line), species volume fractions including both XNO and XNO2 .
compared with experiments [16] (symbol ■ + solid line). In the experiments [16], the measurements of the above
species fractions (e.g., NO, NO2 , NH3 and O2 ) were taken at sev-
By comparing cases FNH3 -25 and FNH3 -50 in Figs. 11 eral locations on the centerline where the O2 volume fraction was
and 12, it has been found that higher concentrations of NO and measured to be 14%, 16%, 18%, 19%, and ∼21%, respectively,

12 Copyright © 2024 by ASME


TABLE 7: Comparisons on species volume fractions for case TABLE 8: Comparisons on species volume fractions for case
FNH3 -25 at different measurement locations between experi- FNH3 -50 at different measurement locations between experi-
ments [16] and present LES simulations. The prediction error ments [16] and present LES simulations. The prediction error
= (simulated value - measured value)/measured value ×100%. = (simulated value - measured value)/measured value ×100%.
Location [mm] 144 158 200 240 279 Location [mm] 126 139.2 163 196 279
EXP: ⟨XO2 ⟩ [%] 14 16 18 19 19.43 EXP: ⟨XO2 ⟩ [%] 14 16 18 19 20
LES: ⟨XO2 ⟩ [%] 12.14 13.82 16.70 18.56 19.19 LES: ⟨XO2 ⟩ [%] 11.15 13.82 15.99 17.99 19.28
Error on ⟨XO2 ⟩ [%] -13.29 -13.63 -7.22 -2.32 -1.24 Error on ⟨XO2 ⟩ [%] -20.36 -13.63 -11.17 -5.32 -3.60
EXP: ⟨XNO ⟩ [ppm] 1000 643 330 220 170 EXP: ⟨XNO ⟩ [ppm] 1775 1100 526 353 176
LES: ⟨XNO ⟩ [ppm] 488.58 359.93 158.26 21.81 2.03 LES: ⟨XNO ⟩ [ppm] 433.95 313.37 186.40 53.94 6.35
Error on ⟨XNO ⟩ [%] -51.14 -44.02 -52.04 -90.09 -98.80 Error on ⟨XNO ⟩ [%] -75.55 -71.51 -64.56 -84.72 -96.39
EXP: ⟨XNO2 ⟩ [ppm] 20 9 5 2 1 EXP: ⟨XNO2 ⟩ [ppm] 56 23 6 3 2
LES: ⟨XNO2 ⟩ [ppm] 0.87 0.76 0.58 0.30 0.16 LES: ⟨XNO2 ⟩ [ppm] 1.05 0.90 0.75 0.34 0.16
Error on ⟨XNO2 ⟩ [%] -95.63 -91.54 -88.42 -85.18 -83.72 Error on ⟨XNO2 ⟩ [%] -98.13 -96.08 -87.56 -88.67 -92.17
EXP: ⟨XNOx ⟩ [ppm] 1020 652 335 222 171 EXP: ⟨XNOx ⟩ [ppm] 1831 1123 532 356 178
LES: ⟨XNOx ⟩ [ppm] 489.45 360.69 158.84 22.10 2.19 LES: ⟨XNOx ⟩ [ppm] 435.00 314.27 187.14 57.28 6.51
Error on ⟨XNOx ⟩ [%] -52.01 -44.68 -52.59 -90.04 -98.72 Error on ⟨XNOx ⟩ [%] -76.24 -72.01 -64.82 -84.75 -96.34
EXP: ⟨XNH3 ⟩ [ppm] 50 0 0 0 0 EXP: ⟨XNH3 ⟩ [ppm] 100 100 200 170 140
LES: ⟨XNH3 ⟩ [ppm] 265.00 178.97 60.48 0.74 0.073 LES: ⟨XNH3 ⟩ [ppm] 5990.15 2342.8 537.32 220.79 66.39
Error on ⟨XNH3 ⟩ [%] 430 Inf Inf Inf Inf Error on ⟨XNH3 ⟩ [%] >1000 >1000 168.66 29.88 -52.58

using a series of 0.5 mm diameter probes. This measurement TABLE 9: Cell sizes in different zones for the three meshes.
approach provides a consistent basis for comparing species such
Mesh Coarse Medium Fine
as NOx and unburnt NH3 from all the investigated cases. More
details on the measurement techniques can be found in Ref. [16]. Zone 1 [mm] 0.64 0.32 0.16
For the above observed discrepancies between the numerical Zone 2 [mm] 2.56 1.28 0.64
solutions and the experiments, one possible reason may be due
to the thermal boundary conditions used for the bluff body walls. Zone 3 [mm] 2.56 2.56 2.56
Despite being consistent with a previous numerical study [16], Zone 4 [mm] ∼10 ∼10 ∼10
the current utilization of adiabatic walls and neglecting the radi- Cell count [million] 0.89 2.76 15.86
ation from the domain may need to be re-evaluated. Given the
fact that the bluff body temperature was not measured in the ex-
periments [16], a reasonable estimation of the wall temperature cells respectively) have therefore been created. As listed in Ta-
and a sensitivity analysis of its impact on the solutions with and ble 9, the “Coarse” and “Medium” meshes both have larger cell
without the radiative heat flux may be required for future sim- sizes in “Zone 1” and “Zone 2” of the domain, compared with
ulations. Nevertheless, the present LES/FGM work successfully the currently used “Fine” mesh (with 15.86 million cells). Please
captures the global flame features and the axial variations of mul- note, the topologies of the three meshes are the same. Since the
tiple major and minor species, for flames with different ammo- flame and the main reaction activities only occur within “Zone
nia/hydrogen blending ratios which represent different ammonia 1” and “Zone 2”, the mesh cell sizes in the outer zones “3” and
cracking levels. “4” are the same between the three meshes.
For the LES case FNH3 -25, the axial profiles of the time-
4.4 Mesh Sensitivity Analyses averaged flow temperature, ⟨T ⟩, along the centerline have been
Finally, the present mesh (see Fig. 3) has been further ana- compared in Fig. 14 across the three meshes. It has been ob-
lyzed to evaluate the impact of mesh density on the solution ac- served that, the “Coarse” mesh slightly overpredicts the tempera-
curacy. Two coarser meshes (with 0.89 million and 2.76 million ture magnitude downstream of x ≃ 130 mm, while the “Medium”

13 Copyright © 2024 by ASME


port equations for the slower-evolving species in addition to the
progress variable equation; (iv) applying more detailed ammo-
nia reaction mechanisms. Nevertheless, the numerical workflow
and model settings developed by this study have paved the way
for future work, which may be useful for the design and cer-
tification of complicated industrial combustors which may use
blended ammonia as a fuel.

ACKNOWLEDGMENT
The authors thank Abhijit Patil, Karthik Puduppakkam, Rakesh
Yadav and Reza Farokhi from Ansys, and Jiajun Li from
KAUST, for all of their kind help and discussion.

REFERENCES
[1] Brady, K. B., Hui, X., and Sung, C.-J., 2015. “Effect of hy-
drogen addition on the counterflow ignition of n-butanol at
atmospheric and elevated pressures”. International Journal
of Hydrogen Energy, 40(46), pp. 16618–16633.
FIGURE 14: Time-averaged flow temperature, ⟨T ⟩, along the cen- [2] Xin, Z., Zhang, J., Sordakis, K., Beller, M., Du, C.-X., Lau-
terline for LES case FNH3 -25: Coarse mesh with 0.89 million renczy, G., and Li, Y., 2018. “Towards hydrogen storage
cells (solid line); Medium mesh with 2.76 million cells (dash- through an efficient ruthenium-catalyzed dehydrogenation
dotted line); Fine mesh with 15.86 million cells (dotted line), of formic acid”. ChemSusChem, 11(13), pp. 2077–2082.
compared with experiments [16] (symbol ■). [3] Sarathy, S. M., Brequigny, P., Katoch, A., Elbaz, A. M.,
Roberts, W. L., Dibble, R. W., and Foucher, F., 2020. “Lam-
mesh consistently underpredicts the temperature value along the inar burning velocities and kinetic modeling of a renewable
entire flame length. It is clear that the currently used “Fine” mesh e-fuel: Formic acid and its mixtures with H2 and CO2”.
is the most accurate, and therefore should be used for the present Energy & Fuels, 34(6), pp. 7564–7572.
LES/FGM simulations. [4] Takizawa, K., Takahashi, A., Tokuhashi, K., Kondo, S.,
and Sekiya, A., 2008. “Burning velocity measurements
5 CONCLUSIONS AND FUTURE WORK of nitrogen-containing compounds”. Journal of Hazardous
The present numerical work investigates a series of turbulent Materials, 155(1-2), pp. 144–152.
non-premixed NH3 /H2 /N2 blended flames stabilized on a bluff [5] Elbaz, A. M., Wang, S., Guiberti, T. F., and Roberts, W. L.,
body burner. The Large Eddy Simulation (LES) and the Flamelet 2022. “Review on the recent advances on ammonia com-
Generated Manifold (FGM) combustion model have been used bustion from the fundamentals to the applications”. Fuel
for the simulations, based on optimized mesh resolutions and nu- Communications, 10, p. 100053.
merical settings. The investigated volume fractions of NH3 range [6] Pfahl, U., Ross, M., Shepherd, J., Pasamehmetoglu, K., and
from 0% to 50%, resulting in three flames representing different Unal, C., 2000. “Flammability limits, ignition energy, and
ammonia cracking levels. The increase of NH3 and decrease of flame speeds in H2–CH4–NH3–N2O–O2–N2 mixtures”.
jet velocity shorten the flame and reduce the flow temperature, Combustion and Flame, 123(1-2), pp. 140–158.
bringing down the reactivity of the mixture due to the reduction [7] Lee, J., Lee, S., and Kwon, O., 2010. “Effects of ammonia
of hydrogen. The simulation results generally match well the ex- substitution on hydrogen/air flame propagation and emis-
perimental data and trends across all the studied cases, for the sions”. International Journal of Hydrogen Energy, 35(20),
time-averaged flow temperature and multiple species mass and pp. 11332–11341.
volume concentrations (e.g., O2 , OH, NH2 , NOx and unburnt [8] Okafor, E. C., Naito, Y., Colson, S., Ichikawa, A., Kudo,
NH3 ), despite having some considerable magnitude discrepan- T., Hayakawa, A., and Kobayashi, H., 2018. “Experimen-
cies in the upstream flame zones. tal and numerical study of the laminar burning velocity of
Several further improvements to the present study are cur- CH4–NH3–air premixed flames”. Combustion and Flame,
rently being investigated, including: (i) using other thermal 187, pp. 185–198.
boundary conditions for the bluff body surfaces, such as a fixed [9] Han, X., Wang, Z., Costa, M., Sun, Z., He, Y., and Cen, K.,
or a spatially varying wall temperature; (ii) adding a radia- 2019. “Experimental and kinetic modeling study of laminar
tive heat transfer model to the simulations; (iii) solving trans- burning velocities of NH3/air, NH3/H2/air, NH3/CO/air

14 Copyright © 2024 by ASME


and NH3/CH4/air premixed flames”. Combustion and 2021. “Characteristics of NH3/H2/air flames in a com-
Flame, 206, pp. 214–226. bustor fired by a swirl and bluff-body stabilized burner”.
[10] Hayakawa, A., Goto, T., Mimoto, R., Arakawa, Y., Kudo, Proceedings of the Combustion Institute, 38(4), pp. 5129–
T., and Kobayashi, H., 2015. “Laminar burning velocity 5138.
and Markstein length of ammonia/air premixed flames at [22] Pacheco, G. P., Rocha, R. C., Franco, M. C., Mendes,
various pressures”. Fuel, 159, pp. 98–106. M. A., Fernandes, E. C., Coelho, P. J., and Bai, X.-S., 2021.
[11] Valera-Medina, A., Pugh, D., Marsh, P., Bulat, G., and “Experimental and kinetic investigation of stoichiometric to
Bowen, P., 2017. “Preliminary study on lean premixed rich NH3/H2/Air flames in a swirl and bluff-body stabilized
combustion of ammonia-hydrogen for swirling gas turbine burner”. Energy & Fuels, 35(9), pp. 7201–7216.
combustors”. International Journal of Hydrogen Energy, [23] Khateeb, A. A., Guiberti, T. F., Zhu, X., Younes, M.,
42(38), pp. 24495–24503. Jamal, A., and Roberts, W. L., 2020. “Stability lim-
[12] Valera-Medina, A., Gutesa, M., Xiao, H., Pugh, D., Giles, its and NO emissions of technically-premixed ammonia-
A., Goktepe, B., Marsh, R., and Bowen, P., 2019. “Pre- hydrogen-nitrogen-air swirl flames”. International Journal
mixed ammonia/hydrogen swirl combustion under rich fuel of Hydrogen Energy, 45(41), pp. 22008–22018.
conditions for gas turbines operation”. International Jour- [24] Ansys Inc., 2024. “Ansys Fluent® User’s Guide. Release
nal of Hydrogen Energy, 44(16), pp. 8615–8626. 2024R1”. Canonsburg, PA, USA.
[13] Westlye, F. R., Ivarsson, A., and Schramm, J., 2013. “Ex- [25] Van Oijen, J., and De Goey, L., 2000. “Modelling of pre-
perimental investigation of nitrogen based emissions from mixed laminar flames using flamelet-generated manifolds”.
an ammonia fueled SI-engine”. Fuel, 111, pp. 239–247. Combustion Science and Technology, 161(1), pp. 113–137.
[14] Bioche, K., Bricteux, L., Bertolino, A., Parente, A., and [26] Xia, Y., Verma, I., Zore, K., and Sharkey, P., 2020.
Blondeau, J., 2021. “Large eddy simulation of rich ammo- “SBES/FGM simulation of forced response of a premixed
nia/hydrogen/air combustion in a gas turbine burner”. Inter- bluff-body stabilized flame”. In AIAA SciTech 2020 Fo-
national Journal of Hydrogen Energy, 46(79), pp. 39548– rum. Paper No. AIAA 2020-0175, Orlando, FL.
39562. [27] Xia, Y., Sharkey, P., Orsino, S., Kuron, M., Menter, F.,
[15] Wiseman, S., Rieth, M., Gruber, A., Dawson, J. R., and Verma, I., Malecki, R., and Sen, B., 2020. “SBES/FGM
Chen, J. H., 2021. “A comparison of the blow-out behavior simulation of film-cooled surface heat transfer and near-
of turbulent premixed ammonia/hydrogen/nitrogen-air and wall reaction”. In ASME Turbo Expo 2020. Paper No.
methane–air flames”. Proceedings of the Combustion Insti- GT2020-14717, Online, 21-25 September.
tute, 38(2), pp. 2869–2876. [28] Xia, Y., Sharkey, P., Orsino, S., Kuron, M., Menter, F.,
[16] Alfazazi, A., Elbaz, A. M., Li, J., Abdelwahid, S., Verma, I., Malecki, R., and Sen, B., 2021. “Stress-blended
Im, H. G., and Dally, B., 2023. “Characteristics eddy simulation/flamelet generated manifold simulation of
of ammonia-hydrogen nonpremixed bluff-body-stabilized film-cooled surface heat transfer and near-wall reaction”.
flames”. Combustion and Flame, 258, p. 113066. Journal of Turbomachinery, 143(1), p. 011008.
[17] Xia, Y., Verma, I., Nakod, P., Yadav, R., Orsino, S., and [29] Xia, Y., Stopford, P., Sharkey, P., and Verma, I., 2021. “Dy-
Li, S., 2022. “Numerical simulations of a lifted hydrogen namic mesh adaption for scale-resolving reacting flow sim-
jet flame using flamelet generated manifold approach”. In ulations”. In ASME Turbo Expo 2021. Paper No. GT2021-
ASME Turbo Expo 2022. Paper No. GT2022-80733, Rot- 59100, Online, 7-11 June.
terdam, The Netherlands, 13-17 June. [30] Ansys Inc., 2024. “Ansys Fluent® Theory Guide. Release
[18] Xia, Y., Verma, I., Nakod, P., Yadav, R., and et al., 2022. 2024R1”. Canonsburg, PA, USA.
“Numerical simulations of a lifted hydrogen jet flame us- [31] Nguyen, P.-D., Vervisch, L., Subramanian, V., and
ing flamelet generated manifold approach”. Journal of En- Domingo, P., 2010. “Multidimensional flamelet-generated
gineering for Gas Turbines and Power, 144(9), p. 091009. manifolds for partially premixed combustion”. Combustion
[19] Ciccarelli, G., Jackson, D., and Verreault, J., 2006. and Flame, 157(1), pp. 43–61.
“Flammability limits of NH3–H2–N2–air mixtures at ele- [32] Lodier, G., Vervisch, L., Moureau, V., and Domingo, P.,
vated initial temperatures”. Combustion and Flame, 144(1- 2011. “Composition-space premixed flamelet solution with
2), pp. 53–63. differential diffusion for in situ flamelet-generated mani-
[20] Ryu, K., Zacharakis-Jutz, G. E., and Kong, S.-C., 2014. folds”. Combustion and Flame, 158(10), pp. 2009–2016.
“Performance enhancement of ammonia-fueled engine by [33] Robert, H. K., 1970. “Diffusion by a random velocity
using dissociation catalyst for hydrogen generation”. In- field”. Physics of Fluids, 13(1), p. 22.
ternational Journal of Hydrogen Energy, 39(5), pp. 2390– [34] Smirnov, A., Shi, S., and Celik, I., 2001. “Random
2398. flow generation technique for large eddy simulations and
[21] Franco, M. C., Rocha, R. C., Costa, M., and Yehia, M., particle-dynamics modeling”. Journal of Fluids Engineer-

15 Copyright © 2024 by ASME


ing, 123(2), pp. 359–371. solar thermal (CST) systems”. In Advances in concentrat-
[35] Shrivastava, S., Verma, I., Yadav, R., Nakod, P., and Orsino, ing solar thermal research and technology. Woodhead Pub-
S., 2021. “Comparison of performance of flamelet gen- lishing, Thorston, UK, pp. 247–267.
erated manifold model with that of finite rate combustion
model for hydrogen blended flames”. In Turbo Expo:
Power for Land, Sea, and Air, Vol. 84959, American So-
ciety of Mechanical Engineers, p. V03BT04A040.
[36] Pope, S. B., 2004. “Ten questions concerning the large-
eddy simulation of turbulent flows”. New Journal of
Physics, 6(1), p. 35.
[37] Pope, S. B., 2014. Turbulent flows. Cambridge Univ Press.
[38] Lilly, D. K., 1992. “A proposed modification of the Ger-
mano subgrid-scale closure method”. Physics of Fluids A:
Fluid Dynamics, 4(3), pp. 633–635.
[39] Germano, M., Piomelli, U., Moin, P., and Cabot, W. H.,
1996. “Dynamic subgrid-scale eddy viscosity model”. In
the Summer Program. Center for Turbulence Research
(CTR), University of Stanford, CA, USA.
[40] Kim, S.-E., 2004. “Large eddy simulation using unstruc-
tured meshes and dynamic subgrid-scale turbulence mod-
els”. AIAA Paper, 2548, p. 2004.
[41] Ansys Inc., 2024. “Ansys Model Fuel Library® . Release
2024R1”. Canonsburg, PA, USA.
[42] Nakamura, H., Hasegawa, S., and Tezuka, T., 2017. “Ki-
netic modeling of ammonia/air weak flames in a micro flow
reactor with a controlled temperature profile”. Combustion
and Flame, 185, pp. 16–27.
[43] Otomo, J., Koshi, M., Mitsumori, T., Iwasaki, H., and Ya-
mada, K., 2018. “Chemical kinetic modeling of ammonia
oxidation with improved reaction mechanism for ammo-
nia/air and ammonia/hydrogen/air combustion”. Interna-
tional Journal of Hydrogen Energy, 43(5), pp. 3004–3014.
[44] Zhang, X., Moosakutty, S. P., Rajan, R. P., Younes, M., and
Sarathy, S. M., 2021. “Combustion chemistry of ammo-
nia/hydrogen mixtures: Jet-stirred reactor measurements
and comprehensive kinetic modeling”. Combustion and
Flame, 234, p. 111653.
[45] Glarborg, P., 2023. “The NH3/NO2/O2 system: Constrain-
ing key steps in ammonia ignition and N2O formation”.
Combustion and Flame, 257, p. 112311.
[46] Dally, B. B., Fletcher, D. F., and Masri, A. R., 1998. “Flow
and mixing fields of turbulent bluff-body jets and flames”.
Combustion Theory and Modelling, 2(2), pp. 193–219.
[47] Rowhani, A., Sun, Z. W., Medwell, P. R., Alwahabi, Z. T.,
Nathan, G. J., and Dally, B. B., 2022. “Effects of the
bluff-body diameter on the flow-field characteristics of non-
premixed turbulent highly-sooting flames”. Combustion
Science and Technology, 194(2), pp. 378–396.
[48] Ansys Inc., 2024. “Ansys Chemkin-Pro® . Release
2024R1”. Cannonsburg, PA, USA.
[49] Irwin, L., Stekli, J., Pfefferkorn, C., and Pitchumani, R.,
2017. “Thermochemical energy storage for concentrating

16 Copyright © 2024 by ASME

View publication stats

You might also like