Combustion and Flame 265 (2024) 113504

Contents lists available at ScienceDirect

Combustion and Flame

journal homepage: www.sciencedirect.com/journal/combustion-and-flame

Direct numerical simulation of a high-pressure hydrogen micromix

combustor: Flame structure and stabilisation mechanism
T.L. Howarth a ,∗, M.A. Picciani b,c , E.S. Richardson d , M.S. Day e , A.J. Aspden a
a
School of Engineering, Newcastle University, Newcastle-Upon-Tyne, NE1 7RU, UK
b
Reaction Engines Ltd., Culham Science Centre, Abingdon, OX14 3DB, UK
c
Sunborne Systems Ltd., The Lambourn, Wyndyke Furlong, Abingdon, OX14 1UJ, UK
d
Faculty of Engineering and Physical Sciences, University of Southampton, Southampton, SO17 1BJ, UK
e
Computational Science Center, National Renewable Energy Laboratory, Golden, CO 80401-3305, USA

ARTICLE INFO ABSTRACT

Keywords: A high-pressure hydrogen micromix combustor has been investigated using direct numerical simulation with
Direct numerical simulation detailed chemistry to examine the flame structure and stabilisation mechanism. The configuration of the
Hydrogen combustor was based on the design by Schefer et al. [1], using numerical periodicity to mimic a large square
Flame structure
array. A precursor simulation of an opposed jet-in-crossflow was first conducted to generate appropriate
Flame stabilisation
partially-premixed inflow boundary conditions for the subsequent reacting simulation. The resulting flame
Micromix combustor
can be described as an predominantly-lean inhomogeneously-premixed lifted jet flame. Five main zones
were identified: a jet mixing region, a core flame, a peripheral flame, a recirculation zone, and combustion
products. The core flame, situated over the jet mixing region, was found to burn as a thin reaction front,
responsible for over 85% of the total fuel consumption. The peripheral flame shrouded the core flame, had
low mean flow with high turbulence, and burned at very lean conditions (in the distributed burning regime).
It was shown that turbulent premixed flame propagation was an order-of-magnitude too slow to stabilise the
flame at these conditions. Stabilisation was identified to be due to ignition events resulting from turbulent
mixing of fuel from the jet into mean recirculation of very lean hot products. Ignition events were found to
correlate with shear-driven Kelvin–Helmholtz vortices, and increased in likelihood with streamwise distance.
At the flame base, isolated events were observed, which developed into rapidly burning flame kernels that
were blown downstream. Further downstream, near-simultaneous spatially-distributed ignition events were
observed, which appeared more like ignition sheets. The paper concludes with a broader discussion that
considers generalising from the conditions considered here.

1. Introduction in these combustors is influenced by several factors, including global

equivalence ratio, injection velocities, and combustor geometry.
Micromix combustor design seeks to meet the competing objectives While there have been extensive experimental and low-fidelity nu-
of low flashback susceptibility and low NO𝑥 emission by replacing merical studies (e.g. [4–7]) of the factors affecting flame stability
larger burners with a large number of smaller non-premixed fuel in-
and NO𝑥 emissions, a fundamental investigation into the structure of
jectors, such that NO𝑥 formation is limited by the short residence
the flame, and detailed stabilisation mechanism in particular, has not
time within the small flames. In particular, micromix has received
attention as a means to manage the elevated flashback risks asso- been reported experimentally or numerically for hydrogen micromix
ciated with burning hydrogen in aircraft, spacecraft, industrial and burners. Understanding both flame structure and stabilisation is crucial
domestic combustors [2]. Over the years, several generations of designs for selecting appropriate models in lower-fidelity simulations, as well
have been proposed; a comprehensive overview can be found in [3]. as for the conceptualisation and design of new micromix combustors.
Micromix combustors are designed based on the concept of a large The particular configuration examined in this paper follows the
number of small non-premixed injectors, and a sudden expansion to NASA design used by Schefer et al. [1], whereby the fuel injector plate
establish recirculating flow. Common designs use a jet-in-crossflow con-
is comprised of an array of circular air nozzles, with two diametrically
figuration, injecting fuel into an oxidiser crossflow. Flame behaviour

∗ Corresponding author.
E-mail address: [email protected] (T.L. Howarth).

https://doi.org/10.1016/j.combustflame.2024.113504
Received 9 September 2023; Received in revised form 2 May 2024; Accepted 7 May 2024
Available online 22 May 2024
0010-2180/© 2024 The Authors. Published by Elsevier Inc. on behalf of The Combustion Institute. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

an injector, and the configuration abbreviated to JICF. The JICF is

described non-dimensionally by the momentum flux ratio, given by
𝜌i 𝑢2i
𝐽= , (1)
𝜌c 𝑢2c
where subscripts i and c denote injector and crossflow quantities,
respectively, and 𝜌 and 𝑢 denote density and velocity, respectively. For
a given global equivalence ratio 𝜙, assuming ideal gas behaviour and
no change in pressure across the injector, a relation can be found

𝜙2 𝑊 c 𝑇i 𝑌o,c
2 𝐴2
c
𝐽= , (2)
𝑁 2 𝑊 i 𝑇c 𝑠2 𝑌f,i2 𝐴2i

where 𝑊 is the mean molecular weight, 𝑇 is the temperature, 𝑁 is the

number of fuel injectors per air port, 𝐴 is the area, 𝑌o,c is the oxygen
mass fraction in the air crossflow, 𝑌f,i is the fuel mass fraction in the fuel
injector, and 𝑠 is the mass stoichiometric ratio. This equation shows the
direct relationship between global equivalence ratio and momentum
flux ratio. At a lower momentum flux ratio, the fuel does not penetrate
as far into the crossflow, similarly at a higher momentum flux ratio the
fuel penetrates further, and may interact with an opposed fuel injector,
Fig. 1. Schematic of a micromix combustor in 2D, including nomenclature used depending on geometry. A review of the effect of momentum flux ratio
throughout. (and other factors) on JICF trajectory and mixing can be found in [9].
For a fixed geometry, the penetration 𝑦 typically scales with 𝐽 𝛼 ,
where 1∕3 < 𝛼 < 2∕3. Fluid regions in a canonical JICF configuration
opposed fuel jets within the nozzle. To enable direct numerical simula- can be said to be on the windward- or leeward-side which distinguish
tion (DNS) with detailed chemistry, periodicity is exploited to mimic between fluid flowing above or below the counter-rotating vortex pair
the behaviour of an individual injector within a large square array respectively. Furthermore, regions within a small distance downstream
of ports. A range of inlet temperatures and pressures are relevant for of the injector (typically 𝑥 < 0.3𝐽 𝑑i , for 𝑑i the injector diameter, [10])
hydrogen micromix combustion systems, from ambient temperatures are classed as the near-field, and further downstream as far-field. In
and pressures that might be found in a domestic boiler, to elevated tem- the case of far-field mixing, the scalar field fluctuations are directly
peratures and pressures resulting from compression within aerospace attributed to the Reynolds number of the injector (Rei = 𝑢i 𝑑i ∕𝜈i ), with
propulsion systems. The inlet conditions affect flame stabilisation, due improving mixedness up to Rei ∼ 20000 [11]. Naturally, increasing 𝐽
to their effect on flame speed and ignition behaviour, and strongly (or 𝜙) √will increase the Reynolds number through increased velocities
affect NO𝑥 production. The present study employs high-pressure high- Rei ∼ 𝐽 (∼ 𝜙). In the near field, mixing is driven by the formation
temperature operating conditions that might be encountered in a high of the counter-rotating vortex pair rather than turbulent mixing in the
power-density aerospace application, such as the preburner in Reaction far-field, with decay of scalar concentration along the JICF centreline
Engines’ SABRE cycle (see [8] for an outline of the thermodynamic scaling with arc length of the centreline to the power of −1.3 [10].
cycle). The aim of the study is to identify the flame structure and However, the study of JICF trajectories and mixing in variable-density
the stabilisation mechanism. A review of technical background is pro- JICF is still an active area of research (e.g. [12–14]), and more studies
vided in Section 2, followed by a description of the numerical solver are required.
and DNS configuration in Section 3. The simulations are analysed in Flame stabilisation mechanisms depend simultaneously on the fluid
parts. The first part considers both the steady (Section 4.1) and in- dynamics, flame regime and interaction with external factors, such as
stantaneous (Section 4.2) flame structure, the combustion regimes and walls or acoustics. Assuming no direct interaction with external sources,
turbulence-flame interactions. The second part considers the stabilisa- four loosely categorised lifted flame stabilisation theories are [15–21]:
tion mechanism, assessing the role of flame propagation (Section 5.1),
ignition events (Section 5.2) and entrainment and mixing (Section 5.3). 1. Premixed flame propagation: The base of the lifted flame is
The paper is concluded by a summary of the structure and stabilisation premixed, and hence stabilised by premixed flame propagation,
(Section 6.1), and a broader discussion of the factors affecting micromix where the propagation can either be driven by the laminar or
combustor design (Section 6.2). turbulent flame speed.
2. Critical scalar dissipation rate: The extinction of diffusion
2. Background flamelets controls the position of flame stabilisation, which can
be determined by the scalar dissipation rate dropping below
A schematic of the micromix combustor considered is shown in some critical value [17]. This theory has become less popular
Fig. 1, where opposing sub-millimeter injectors blow hydrogen at high in recent years [18].
speed into an air crossflow inside the region is referred to as the ‘air 3. Large eddy theory: Large scale structures in the fluid are able to
port’. Each fuel jet forms a counter-rotating vortex pair, which may or repeatedly transport either the flame, or its products upstream
may not interact depending on the level of penetration. Some distance to stabilise the reaction zone [19].
downstream from the fuel injection, there is a sudden expansion into 4. Edge flame propagation: This theory assumes that the flame is
the combustion chamber which forms a jet. The corner between the air partially premixed, and can propagate upstream in balance with
port and combustion chamber is referred to as the ‘lip’. the local flow-field, with a triple-flame structure observed [20].
A key contributing factor to the flame structure and stabilisation
is the bulk flow and turbulent mixing in the air port ahead of the [21] presents a visualisation of each theory (see Fig. 1 therein), and it is
combustion chamber. The hydrogen injection into the air stream forms noted that autoignition may also be a factor that can contribute to any
a jet-in-crossflow configuration; to avoid confusion with the main jet of these mechanisms. The stabilisation mechanism of lifted flames has
in the combustion chamber, the hydrogen inflow is referred to as been investigated using three-dimensional DNS in several studies, but

2
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

almost exclusively focussing on non-premixed configurations stabilised Table 1

Parameters for the non-reacting JICF simulation.
through autoignition in hot coflows (e.g. [22–26]).
It is noted by Schefer et al. [1] that various flame behaviours are Quantity Value

observed in the prescribed micromix combustor including both stable Domain size 7 mm × 7 mm × 7 mm
Resolution (base) 256 × 256 × 256
and unstable lifted flames, flames attached to the lip, blow-off and
𝛥𝑥 (base) 27.3 μm
flashback. Flashback only occurred in low speed, higher equivalence Resolution (effective) 1024 × 1024 × 1024
ratio environments which are rarely used in micromix combustors. 𝛥𝑥 (finest) 6.84 μm
Blow-off occurs at very lean equivalence ratios, with this limit increas- Ambient pressure 24 atm
ing with air crossflow velocity. Mixedness will have a strong effect Crossflow velocity 𝑢̄ c 100 m/s
Injector velocity 𝑢̄ i 1360 m/s
on flame structure, stability and whether the flame becomes attached,
Crossflow temperature 𝑇c 750 K
all of which is poorly understood. Flame stabilisation in micromix Injector temperature 𝑇i 550 K
combustors has been attributed to ‘flame stabilising vortices’ (e.g. [3]), Crossflow density 𝜌c 11.27 kg/m3
although the precise mechanism by which this enables stabilisation has Injector density 𝜌i 1.04 kg/m3
not yet been investigated, nor has the combustion regime that lifted Crossflow rms velocity fluctuation 𝑢′c 3 m/s
Injector rms velocity fluctuation 𝑢′i 40 m/s
micromix flames burn in been explored. Momentum flux ratio 𝐽 = 𝜌i 𝑢2i ∕𝜌c 𝑢2c 17.2
Injector Reynolds number Rei = 𝑢i 𝑑i ∕𝜈i 48 021
3. Direct numerical simulation configuration

This section provides a description of the numerical solver, the

generation of micromix combustor inflow conditions using a precursor density to cover the injected fuel, and an additional level of AMR based
simulation of a non-reacting JICF, and the configuration for the reacting on hydrogen mass fractions to cover the potential core of each JICF.
combustor simulation. A further precursor simulation of maintained homogeneous isotropic
turbulence (following [41]) was first run to generate the crossflow
3.1. Numerical solver: PeleLMeX and injector inflow boundary conditions with a turbulence intensity
of approximately 3%. Hyperbolic tangent functions were employed
The simulations presented here were conducted with the PeleLMeX to smooth the inflow velocities to zero at the crossflow and injector
solver [27,28], developed as part of the US DoE Exascale Computing walls. The boundary layer thickness was estimated for this smoothing
project. PeleLMeX is the non-subcycling version of the well-established using correlations from [42]. Isothermal walls (𝑇wall = 𝑇𝑐 = 750 K)
code PeleLM, which has been extensively used to study hydrogen were used on the 𝑥 and 𝑦 boundaries outside of the injectors. The
flames in the canonical flame-in-a-box configuration in 2D (e.g. [29, parameters associated with the non-reacting simulation can be found
30]), 3D freely-propagating (e.g. [31,32]) and turbulent (e.g. [32– in Table 1. Note that the injector Reynolds number is high enough that
34]) flames, as well as in experimental configurations (e.g. [35,36]). improved mixedness would not be anticipated in the far-field with a
PeleLMeX is capable of running on massively parallel CPU- and GPU- higher equivalence ratio.
based computing systems, both of which were used for the present To construct the partially-premixed turbulent inflow for the com-
study. bustor simulation, planes of velocities, species and temperatures were
The equations of motion are based on a low-Mach-number for- sampled from the JICF simulation 3 mm downstream from hydrogen
mulation of the reacting Navier–Stokes equations, where the fluid is injectors; the sampling frequency was based on assuming an inflow
treated as a mixture of ideal gases [37]. A mixture-averaged model for velocity of approximately 100 m/s and computational cell size of the
differential species diffusion is used, where each species has its own reacting combustor simulation. The distance of the sampling plane is
temperature- and composition-dependent diffusivity; the Soret effect is just beyond the near-field criteria devised by [10] (in this case is 𝑥 =
neglected. The discretisation couples a multi-implicit spectral deferred 2.58 mm). A total of 784 planes were sampled and collated (covering
correction approach for the integration of mass, species and energy approximately 1.5 jet times 𝑡j = 𝑑j ∕𝑢j ), with the combustor inflow
equations with a density-weighted approximate projection method, condition cycling periodically through the planes, using quadratic in-
which has incorporated the equation of state through a velocity diver- terpolation between planes. The precursor simulation is not intended
gence constraint [38]. The resulting discretisation is integrated with to be a comprehensive DNS study of the upstream opposing JICF, and
timesteps determined by the advective transport through an advective as such does not fully resolve boundary layers, for example, and does
CFL number, with the faster diffusion and chemical processes treated not strictly satisfy the criteria for the low-Mach assumption; however,
implicitly. This scheme is embedded in a parallel adaptive mesh refine- the intention of this simulation was to generate partially-premixed
ment (AMR) algorithm framework based on a hierarchical system of reactants representative of the upstream flow to provide appropriate
rectangular grid patches. The complete integration algorithm is second- boundary conditions for the combusting simulation. The decoupling
order accurate in space and time; the reader is referred to [37–39] of the simulations exploits the difference in time scales between the
for further details. The transport coefficients, thermodynamic relation- two regions, with the non-reacting case being more cost intensive in
ships and chemical kinetics are obtained from a comprehensive H2 /O2 time (due to higher velocities), but the combustion chamber more
kinetic model [40], suitable for high-pressure hydrogen combustion. expensive due to a larger domain with finer resolution and chemistry.
The inflow velocity is large, both relative to the fluctuations and the
3.2. Jet-in-crossflow configuration interior velocities close to the inlet plane; while it may be possible
for there to be some upstream interaction close to the lip, it is not
To establish partially-premixed inflow boundary conditions suitable anticipated to be strong enough to affect the subsequent combusting
for the micromix combustor, a precursor simulation was conducted simulation.
using opposed hydrogen injectors in an air crossflow. A schematic for
the simulation can be seen at the top of Fig. 2. Two round injectors of 3.3. Combustor configuration
diameter 0.5 mm were set up 2 mm downstream from the crossflow
inflow face in a cubic domain with 7 mm sides. The mass flow rate of The combustor domain was a 14 mm × 14 mm × 84 mm cuboid,
hydrogen was determined from the global equivalence ratio specified with a 7 mm diameter circular inflow imposed on the bottom (𝑧 low)
for the combustor, using a crossflow velocity of 100 m/s. A base grid of face using the pre-generated inflow described above. Outside of the
2563 was used, with one level of AMR based on adjacent differences in circular inflow, a no-slip wall condition was imposed on the remainder

3
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Table 2
Parameters for the combusting simulation.
Quantity Value
Domain size 14 mm × 14 mm × 84 mm
Resolution (base) 256 × 256 × 1536
𝛥𝑥 (base) 54.7 μm
Resolution (effective) 2048 × 2048 × 12288
𝛥𝑥 (finest, at flame) 6.84 μm
𝛥𝑥 (inflow) 27.3 μm
Jet diameter 𝑑j 7 mm
Mean inflow velocity 𝑢̄ j 111 m/s
Inflow r.m.s. velocity fluctuation 𝑢′ 26.1 m/s
Inflow integral length scale 𝓁 1.0 mm
Inflow turbulent energy dissipation rate 𝜖 = 𝑢′3 ∕𝓁 1.79 × 107 m2 /s3
Mean inflow viscosity 𝜈̄ 3.33 × 10−5 m2 /s
Inflow Kolmogorov length scale 𝜂 = (𝜈̄ 3 ∕𝜖)1∕4 6.77 μm
Effective mean inflow viscosity 𝜈̄𝑒 5.90 × 10−5 m2 /s
Effective inflow Kolmogorov length scale 𝜂𝑒 = (𝜈̄𝑒3 ∕𝜖)1∕4 10.4 μm
Jet Reynolds number Rej = 𝑢̄ j 𝑑j ∕𝜈̄ 23 200
Inflow turbulent Reynolds number Ret = 𝑢′ 𝓁∕𝜈̄ 781
Mean inflow equivalence ratio 𝜙j 0.46
Mean inflow temperature 𝑇j 732 K
Mean thermal thickness 𝓁l (𝜙j , 𝑇j , 𝑝) 17.4 μm
Flame resolution 𝓁l ∕𝛥𝑥 2.54

assessment of resolution is provided in Appendix A for these particular

conditions.
The following strategy was used to establish the statistically-steady
flame. The simulation was first run at the base grid (i.e. without AMR)
for approximately 20 jet times. Ignition was induced synthetically by
depositing heat in four locations at (𝑥, 𝑦, 𝑧) = (±3, ±3, 30) mm for 3 μs
over a radius of 400 μm, which was sufficient to establish sustained
reactions. The simulation was then run for a further 95 jet times at
the base grid, over which time the flame propagated upstream to a
statistically-stationary location, and the temperature of the recircula-
tion zone settled to a statistically-stationary value. The AMR was then
enabled, and the simulation was run for another 8 jet times before
any statistics were taken; the subsequent statistical averaging period
was 4 jet times, which corresponds to approximately 25 flame times
(based on the mean reactant conditions). Animations of the statistically-
steady period used for analysis are included as supplementary material,
which provide valuable context and insight into the flame structure and
Fig. 2. Schematic of the JICF precursor simulation (top) and the reacting combustor stabilisation discussed in the following results sections.
simulation (bottom). A summary of the parameters for the combustor is given in Table 2,
where the integral length scale has been calculated as the mean integral
of the auto-correlation function for the velocities on each 2D inflow
of the bottom face with an isothermal condition matching the non- plane. Effective inflow viscosities and Kolmogorov length scales were
reacting simulation of 𝑇wall = 750 K. The domain is centred with calculated following [41].
(0,0,0) mm at the centre of the circular inflow. Periodic boundary
conditions were used laterally to mimic a Cartesian grid of inlet ports. 4. Flame structure
An outflow condition was employed at the top boundary. A schematic
is shown at the bottom of Fig. 2. A first level of AMR was defined to This section considers the general structure of the flow, first in
cover the entire jet, as defined by large values of 𝑢𝑧 , 𝑌h2 and magnitude a temporally-averaged sense in Section 4.1, and then through the
of vorticity |𝝎|, leading to a resolution of 27.3 μm inside the jet. Two evolution of instantaneous snapshots in Section 4.2.
further levels of AMR were added based on the intermediate species
𝑌ho2 to cover the flame, leading to a resolution in the most reactive parts 4.1. Temporally-averaged jet and flame structures
of the flame of 6.4 μm. The base grid is referred to as level 0, increasing
up to ‘‘maximum level 3’’ (ML3), giving an effective resolution in
Temporal averaging was performed according to
excess of 50 billion computational cells. The ML3 simulations were
𝑡
1 ∑
𝑓
run on ARCHER2, the UK national facility at EPCC. Given the varied
equivalence ratio of the inflow, there is no single thermal thickness that ⟨𝑞⟩ (𝑥, 𝑦, 𝑧) = 𝑞(𝑥, 𝑦, 𝑧, 𝑡), (3)
𝑁𝑡 𝑡=𝑡
𝑠
could be chosen before simulation, however 1D unstretched values of
thermal thickness obtained from Cantera [43] at the mean conditions where 𝑁𝑡 is the number of time points averaged over the period [𝑡𝑠 , 𝑡𝑓 ],
of the inflow suggest an approximate thickness of 17.4 μm. A further with 𝑡𝑠 = 7.53 ms, 𝑡𝑓 = 7.85 ms, 𝑁𝑡 = 65, and the corresponding Favre
simulation was conducted with an additional level of AMR (i.e. ML4) at average is defined as ⟨𝑞⟩f (𝑥, 𝑦, 𝑧) = ⟨𝜌𝑞⟩ ∕ ⟨𝜌⟩.
the flame to ensure statistical invariance. This resulted in a simulation To paint a picture ⟨of the
⟩ flame, profiles of average temperature
with an effective resolution of over 400 billion cells, which was run ⟨𝑇 ⟩, fuel mass fraction 𝑌h2 and streamwise velocity ⟨𝑢𝑧 ⟩f are shown
f
on Polaris at the Argonne Leadership Computing Facility (ALCF). An as two-dimensional slices through the centre of the jet in both the 𝑥

4
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

⟨ ⟩
Fig. 4. 𝑌h2 f slices through L: 𝑥∕𝑑j = 0 and 𝑦∕𝑑j = 0 and R: 𝑧∕𝑑j = 0, 1, 2, 3. The black
Fig. 3. ⟨𝑇 ⟩ slices through L: 𝑥∕𝑑j = 0 and 𝑦∕𝑑j = 0 and R: 𝑧∕𝑑j = 0, 1, 2, 3. The scale on and white circle on the 𝑧∕𝑑j = 0 image illustrates the airport. Notice the approximate
the left is streamwise distance in jet diameters. At heights below the flame stabilisation kidney shape taken by the hydrogen entering the domain due to the upstream jet in
height, outside the jet, the temperature of the fluid is consistently around 1200 K, with crossflow, which interact and rotate into circular profiles in the opposite axis, leading to
pockets of even hotter fluid near the wall closer to 1400 K. (For interpretation of the the asymmetry in the flame which can be seen in the temperature profile in Fig. 3. (For
references to colour in this figure legend, the reader is referred to the web version of interpretation of the references to colour in this figure legend, the reader is referred
this article.) to the web version of this article.)

and 𝑦 directions (the top third of the domain has been cropped) in Further planar averaging results in one-dimensional profiles as a
Figs. 3–5, respectively, along with horizontal slices at four streamwise function of streamwise distance, i.e.
locations. Colours are normalised by the adiabatic flame temperature 𝐿𝑦 𝐿𝑥
1
̂ =
𝑞(𝑧) ⟨𝑞⟩ (𝑥, 𝑦, 𝑧) d𝑥 d𝑦, (4)
of the mean inflow condition (𝜙 = 0.46, 𝑇𝑢 = 732 K, 𝑝 = 24 atm), the 𝐿𝑥 𝐿𝑦 ∫0 ∫0
stoichiometric mass fraction of hydrogen in air, and the velocity of the
with the natural Favre average 𝑞(𝑧)
̌ = 𝜌̂𝑞∕𝜌.̂
jet, allowing for negative velocities. In this context, the normalisation
Normalised profiles of 𝑇̂ and 𝑌̌oh are given in Fig. 6. Between the
relates to the value at the red colour; the extension up to pink and temperature and OH profile, it can be concluded that the majority of
white is 50% more than the normalisation value. The flame base can the flame sits 3–4.5 𝑑j (21–30.5 mm) above the jet inflow. Defining a
be seen to be lifted at a height approximately 3.5 𝑑j (≈ 25 mm) from the flame brush thickness as
bottom, with a higher temperature observed in the core of the flame. 𝑇𝑏 − 𝑇j
The flame is visibly asymmetric; there is a degree of asymmetry in each 𝐿fb = , (5)
max{d𝑇̂ ∕d𝑧}
panel, which is likely due to the short temporal averaging window, but
more importantly, the slices through 𝑥 = 0 and 𝑦 = 0 are different gives an approximate thickness of 𝐿fb = 1.2 cm (1.7 𝑑j ).
Despite the jet not being perfectly axisymmetric, for modelling and
from each other. This can also be seen in the slices taken through
averaging purposes, azimuthal averaging can also be performed in
the fuel mass fraction in Fig. 4. This asymmetry is inherently due to
terms of the radius 𝑟 and streamwise distance 𝑧
the inflow condition; the counter-rotating vortex pair originating from
the upstream JICF results in kidney-shaped structures in temperature, ⎧ 2𝜋
⎪ 1 ⟨𝑞⟩ (𝑟, 𝜃, 𝑧) d𝜃 for 𝑟 ≤ 𝐿𝑥 ∕2,
species and velocity (see 𝑧 = 0 slices in the bottom right of Figs. 3–5). ⎪ 2𝜋 0
∫
𝑞(𝑟, 𝑧) = ⎨∑ 3 𝛽𝑛 (6)
After entering the domain, the two pairs of vortices propagate towards ⎪ 1
⟨𝑞⟩ (𝑟, 𝜃, 𝑧) d𝜃 otherwise.
the 𝑥 = 0 plane, interact with each other, and then propagate away ⎪𝑛=0 𝛽𝑛 − 𝛼𝑛 ∫𝛼𝑛
⎩
from the 𝑦 = 0 plane. This results in larger mass fractions of fuel
and velocities visible in the slices through the 𝑥 planes, and therefore where
( ) ( )
the asymmetry observed. The interaction of the crossflow with the jet 𝑛𝜋 𝐿𝑥 (𝑛 + 1)𝜋 𝐿𝑥
𝛼𝑛 = + cos−1 , 𝛽𝑛 = − cos−1 (7)
can be seen in the 𝑧 = 0 slices in Fig. 5, where the fluid on the 2 2𝑟 2 2𝑟
windward side has been accelerated (the white crescent), and surrounds after converting from Cartesian to cylindrical coordinates, with corre-
slower-moving fluid on the leeward side of the JICF (the yellow/orange ̃ 𝑧) = 𝜌𝑞∕𝜌. Fig. 7 shows the profiles of 𝑇 , 𝑌̃h2
sponding Favre average 𝑞(𝑟,
mushroom-like structure). and 𝑢̃ 𝑧 at streamwise locations from 𝑧 = 0 to 5 𝑑j ; the black vertical line

5
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 5. ⟨𝑢𝑧 ⟩f slices through L: 𝑥∕𝑑j = 0 and 𝑦∕𝑑j = 0 and R: 𝑧∕𝑑j = 0, 1, 2, 3. Upon entry
into the domain, the wake vortices are still clear through the mushroom-shaped profile
leading from the wall of the air port. (For interpretation of the references to colour in
this figure legend, the reader is referred to the web version of this article.)

Fig. 7. Profiles of 𝑇 , 𝑌̃h2 and 𝑢̃ 𝑧 at 𝑧∕𝑑j = 0, 1, 2, 3, 4, 5. Note the jump in temperature

and drop in H2 between 𝑧∕𝑑j = 3 and 4 as expected from the flame height, while
the combination of jet spreading and thermal expansion cause almost no change in
streamwise velocity at these heights within the jet. The dashed lines are profiles from
planes between the explicitly labelled values of 𝑧.

Fig. 6. Integrated average profiles of normalised temperature and 𝑌oh . These profiles 4.2. Instantaneous flame structure
suggest the flame sits at a height of between 20 and 36 mm.

While the analysis above provides a temporally-averaged descrip-

tion of flame height and jet structure, an in-depth analysis of instan-
represents the point beyond which the radius exceeds half the domain taneous snapshots is required to understand the flame structure more
width, meaning profiles extend into the corners of the domain. completely. An example temperature field in 𝑥 and 𝑦 planes is provided
There is a clear recirculation region (𝑢̃ 𝑧 < 0), especially in the in Fig. 9. Fig. 10 contains 𝑥 and 𝑦 slices of mass fractions of HO2 .
corners, which extends down from flame stabilisation height. Crucially, In addition to highlighting the thin flame front in the middle of the
relatively-hot (partially) reacted fluid is transported upstream all the domain, HO2 is present in the recirculation zone in low (but non-
way to the inlet plane. The upstream transport of heat is further negligible) concentrations. It is also found in higher concentrations
illustrated by streamlines in Fig. 8, where recirculation can be seen in the region separating the regions of recirculated and downstream
outside of the jet up to approximately 3.6 𝑑j (≈ 25 mm). The background products; this suggests the existence of a region of distributed reactions.
is coloured by temperature, indicating fluid upwards of 1500 K is The presence of this burning region can be highlighted by considering
recirculated. measures of flame progress. Typically in the literature, a progress

6
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 9. Snapshots of 𝑇 in the L: 𝑥∕𝑑j = 0 and 𝑦∕𝑑j = 0, and R: 𝑧∕𝑑j = 0, 1, 2, 3 planes.

Fig. 8. Streamlines of velocity in 𝑟 − 𝑧 space, with blue lines originating in the jet Again, the scale is streamwise distance in jet diameters. (For interpretation of the
inflow, and red lines originating in the recirculation region. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
references to colour in this figure legend, the reader is referred to the web version of this article.)
this article.)

variable for partially premixed combustion (e.g. [44]) is defined as a

function of fuel mass fraction and mixture fraction:
⎧1 − 𝑌f where 𝑍 ≤ 𝑍st ,
⎪ 𝑍
𝑐f = ⎨ 1 − 𝑍 𝑍 − 𝑌 (8)
⎪ st f
otherwise.
⎩ 𝑍st 1 − 𝑍
A progress variable based on product mass fraction can also be derived,

⎧ 𝑌p 𝑍st
⎪ where 𝑍 ≤ 𝑍st ,
⎪ 𝑍(𝑌o,cf [1 − 𝑍st ] + 𝑍st )
𝑐p = ⎨ (9)
⎪ 𝑌p (1 − 𝑍st )
otherwise,
⎪ (1 − 𝑍)(𝑌o,cf [1 − 𝑍st ] + 𝑍st )
⎩
where the mixture fraction 𝑍 is given by
𝑠𝑌f − 𝑌o + 𝑌o,cf
𝑍= , (10)
𝑠 + 𝑌o,cf
with 𝑠 = 8 the stoichiometric ratio, and 𝑌o,cf = 0.232 the oxygen mass
fraction in the crossflow.
An 𝑥 and 𝑦 slice of the fuel- and product-based progress variable
are given in Fig. 11. The two progress variables are naturally similar,
and both identify a thin flame in the middle of the domain, as well as
what appears to be a distributed flame in the periphery (there is a broad
spatial region with intermediate values of progress). Both variables also
identify the recirculation region as products (hot/wet and absent of
fuel). It is worth noting that near the jet inlet, there is a region of Fig. 10. Snapshots of 𝑌ho2 in the L: 𝑥∕𝑑j = 0 and 𝑦∕𝑑j = 0 planes, and R: 𝑧 slices at
near-pure air surrounding the fuel, which is identified differently by the 𝑧∕𝑑j = 0, 1, 2, 3. The black and white circle on the 𝑧∕𝑑j = 0 image illustrates the airport.
two progress variables; the fuel-based progress variable identifies this
region as products (no fuel) and the products-based progress variable
identifies it as reactants (cold/dry).

7
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 11. A comparison between the fuel-based and products-based progress variables
(L: 𝑐f , R: 𝑐p ). There are differences between the two where pure air is present, and
identifies both a very thin core flame and significant turbulent mixing on the periphery.
(For interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this article.)

4.2.1. Flame zones

A zonal classification can be defined to segregate the flow into five
regions as described by the following criteria, ordered by priority (note
that the zone is assigned by the first criterion met):
Fig. 12. Zone classification L: 𝑥 slice, R: 𝑦 slice. Orange corresponds to the core flame,
1. Core flame: 𝑌ho2 > 4 × 10−4 ; products are dictated by red, peripheral flame by green, recirculating fluid is light blue
2. Products: 𝑇 > 1600K (or 𝑧 > 3.8 cm); and the jet is dark blue. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
3. Jet: 𝑇 < 800 K;
4. Peripheral flame: 𝑌ho2 > 1 × 10−4 ;
5. Recirculation otherwise.

Furthermore, a buffer region is added to each of the core and peripheral

flame zones to ensure comprehensive coverage of the reacting regions,
while avoiding spurious classification away from the flames. Specifi-
cally, any point within 24 (fine-level) computational cells of a point
satisfying criterion 1 is also classified in the core flame zone; similarly
8 cells within a point satisfying criterion 4 is classified in the peripheral
flame zone.
To determine the combustion regime in each zone, a flame index is
formulated following [45,46] as
∇𝑌f ⋅ ∇𝑌o
𝐹𝐼 = , (11)
|∇𝑌f ∥ ∇𝑌o |
conditioned on flame location and is shown in Fig. 13. In this case,
the flame index is evaluated conditional on 𝑐f < 0.25 and conditional
on location within either the core or peripheral zone. This value was
chosen to isolate gradients in the preheat region and avoid division
Fig. 13. Flame index in preheat region of the core and peripheral flame.
by small numbers close to the products. The flame index is essentially
unity everywhere, meaning the flame burns in a premixed regime,
but may burn at varying equivalence ratio across the flame surface,
as discussed below; the regime can be described as ‘inhomogeneously are shown separately in Fig. 14. The jet region has an equivalence
premixed’. ratio below 𝜙 = 0.3 almost everywhere, but low probabilities extend
To explore the burning regimes further, distributions of local equiv- up to about 1.3; this suggests that it may be possible to observe
alence ratio is considered in the different zones. The local equivalence diffusion flames (in the context of an edge flame) but they should
ratio is defined as 𝜙 = 2𝜉h ∕𝜉o , where 𝜉𝑖 denotes the molar concentration not be anticipated to be prevalent. In the core flame, preferential
of the elements O and H. Probability density functions (PDFs) of 𝜙 diffusion of H2 alters the local equivalence ratio distribution and so
conditioned on each zone in the non-reacting and reacting regions may not be representative of the unburned/fully-burned equivalence

8
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 14. PDFs of local equivalence ratio 𝜙 conditioned on each zone in the non-reacting Fig. 15. JPDFs of temperature and local equivalence ratio in the core (top) and
(top) and reacting (bottom) regions. peripheral (bottom) flame. Magenta lines are profiles from 1D Cantera simulations
with equivalence ratios from 0.3–1.2 in the case of the core flame, and 0.2–0.36 in the
peripheral flame. (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
ratio; consequently, the 𝜙 distribution doubly-conditioned on zone and
𝑐f < 0.01 is also shown in Fig. 14. This suggests that the actual range of
unburned equivalence ratios in the core flame is around 0.4–1.2, in an
as lower injector Reynolds number of momentum flux ratios, where
approximately Gaussian distribution centred at 𝜙 ≈ 0.8. On the other
insufficient mixing may occur.
hand, the equivalence ratio distribution in the peripheral flame is far
narrower ranging from approximately 0.18–0.3, with the distribution
4.2.2. Turbulent burning regimes
centred around 𝜙 ≈ 0.22.
Turbulence-flame interactions in the core and peripheral flames are
The variation in equivalence ratio through the flame can be exam-
very different. The core flame appears as a distinct, if corrugated, flame
ined in more detail in a joint probability distribution function (JPDF)
surface, with burning cells characteristic of low Lewis number effects.
of temperature and local equivalence ratio, shown in Fig. 15. One-
On the other hand, the peripheral flame is spatially distributed over a
dimensional premixed flame profiles (from Cantera) are superimposed
large region with no clear flame surface.
on the JPDF at equivalence ratios 0.3–1.2 (in 𝜙 increments of 0.2) and Based on an equivalence ratio distribution, and unburned tempera-
unburned temperature of 750 K, which show the decrease in equiva- ture of 750 K (which appears to be the case for almost all the flamelets
lence ratio due to preferential diffusion effects. The decrease in equiv- observed), and a pressure of 24 atm, maximum and minimum values of
alence ratio in the JPDF is more pronounced than the one-dimensional characteristic flame speed and thickness can be obtained from Cantera.
profiles, which may be due to differential/preferential diffusion later- Assuming a range of turbulent characteristics of 𝑢′ = 5–20 m/s (𝑢′ =
ally due to the inhomogeneously-premixed reactants. The correspond- 26 m∕s at the inflow and decays to approximately 7 m∕s along the
ing JPDF of 𝑇 and 𝜙 in the peripheral flame is given in Fig. 15, with centreline) and 𝓁 = 1–2 mm, regions of the premixed turbulent flame
magenta lines from 1D Cantera simulations at equivalence ratios 0.2 regime diagram can be highlighted using these values. From the PDFs,
and 0.36 at an unburned temperature of 750 K. a range of 0.4 < 𝜙 < 1.2 is observed in the core flame, and 0.18 <
In summary, there is a broad range of equivalence ratios (i.e. not 𝜙 < 0.3 in the peripheral flame; corresponding indicative regions are
fully-premixed) but those equivalence ratios are almost-everywhere be- shown in Fig. 16. These two regions imply Karlovitz number ranges
low unity (i.e. not non-premixed); therefore, this flame can be classified of 0.05 < Ka < 22 for the core flame and 350 < Ka < 66, 000 for
as burning in an ‘essentially-lean inhomogeneously-premixed’ mode. the peripheral flame. This would place the core flame either in the
At the operating conditions considered, there is a low probability of thin reaction zone or corrugated/wrinkled flame regime, whereas the
finding diffusion flamelets; this may differ at other conditions, such peripheral flame would be in the distributed burning regime. The large

9
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

where 𝑠 is some reference local flame speed (i.e. laminar or freely-

propagating [30]). By taking turbulent intensity 𝐼 = 𝑢′ ∕𝑢, ̄ turbulent
flame propagation can only stabilise the flame if
𝑠
𝑢̄ < . (13)
1 − 2𝐼
Given an area of the jet 𝐴j and area of the total bottom plate repre-
sentative of jet separation 𝐴p , the velocity could decay from the inlet
velocity of 𝑢j to 𝐴j 𝑢j ∕𝐴p . This would be the minimum possible flow
velocity. Therefore, taking a maximum laminar flame speed of 𝑠max
(for the range of temperatures and equivalence ratios found in the
configuration), the flame can be stabilised by flame propagation where

𝐴p 𝑠max
𝑢j < . (14)
𝐴j (1 − 2𝐼)
In the present setup, the area ratio 𝐴p ∕𝐴j = 5.09, the flame speed
of a stoichiometric hydrogen flame with unburned temperature 750 K
and pressure 24 atm is 8.5 m/s. While Lewis number effects are present
(as seen in the variation of equivalence ratio through the flame),
Fig. 16. Turbulent premixed flame regime diagram, with the core and peripheral flame
thermodiffusive effects at these conditions are not expected to accel-
shown by the red and blue blocks respectively. The black arrow indicates the effect of
increasing pressure. (For interpretation of the references to colour in this figure legend,
erate the flame, primarily due to the stabilising effect of the high
the reader is referred to the web version of this article.) unburned temperature of the reactants (see [30,32,48]). Evaluation of
the turbulent intensity suggests a maximum value of 22% reached near
the inflow before decaying to approximately 10% further downstream.
Combining all of these conservative estimates, the upper bound on
range of burning regimes is only present due to the high pressure of
the jet velocity is approximately 77 m/s. This would require the bulk
the system; if an identical assessment were performed at atmospheric
flow have reached the theoretical minimum, with the highest feasible
pressure, there would be less of a distinction between the regimes,
turbulent intensity, and the flame burning everywhere at the highest
as indicated by the pale regions in Fig. 16. In the case of the core
feasible speed, but still results in a mean inflow velocity that is 30%
flame, the Karlovitz number is mostly unchanged by an increase in
smaller than the actual value.
pressure because the flame speed increases and thickness decreases For completeness, the actual mean local flame speed in the core
along an approximately-constant Karlovitz line; Damköhler number, on flame region can be calculated as follows. Given a flame surface 𝛤 ,
the other hand, increases significantly with increasing pressure. The with area 𝐴𝛤 , a mean local flame speed can be determined by
lower value of 𝓁∕𝓁f in the core flame also indicates significantly less
−1
resolution would be required to resolve the flame. This is not the case 𝑠= 𝜌𝜔̇ d𝛺, (15)
(𝜌𝑌h2 )j 𝐴𝛤 ∫𝛺 h2
in the peripheral flame, where the flame speed is substantially reduced
at increased pressure at lean conditions, which moves the flame from where (𝜌𝑌h2 )j is the mean density-weighted fuel mass fraction at the
being moderately-turbulent to the distributed regime at high pressure. inflow. Furthermore, the mean local flame speed in the core flame of
interest can be found as
5. Flame stabilisation mechanism −1
𝑠𝑐 = 𝜌𝜔̇ h2 d𝛺𝑐 , (16)
(𝜌𝑌h2 )j 𝐴𝛤 ,𝑐 ∫𝛺𝑐
There are several mechanisms by which a flame can stabilise, de- where 𝛺𝑐 denotes the core zone defined in Section 4.2 and 𝐴𝛤 ,𝑐 denotes
pending on the thermochemical composition and mixedness of the flow. the area of the instantaneous isosurface inside the core zone.
The present flame is inhomogeneously premixed with elevated reactant An isosurface can be defined using the progress variable constructed
temperature, situated in a recirculatory flow. Therefore, it may stabilise in Section 4.2, with 𝑐f = 0.9; the fuel-based progress variable and
through purely deflagrative flame propagation, autoignition, or in some particular value of 0.9 was chosen based on obtaining accurate mean
mixed mode whereby deflagrative or ignitive burning is accelerated by local flame speeds in lean hydrogen flames (see [30,49]). Calculating
both pre-ignition reactions and pre-mixing with combustion products. a mean local flame speed based using this surface and then ensemble
averaging over time gives a value of 𝑠𝑐 = 1.75 m/s (compare this
Stabilisation through deflagrative flame propagation is reliant on diffu-
with the value of 8.5 m/s used above for 𝑠max ). Using the relation
sive and turbulent transport from the burned to unburned side of the
above, and assuming a turbulent intensity of 10%, the maximum jet
flame. On the other hand, autoignition allows isolated flame kernels to
velocity allowing for stabilisation through flame propagation would
develop ahead of the flame base.
only be 11.3 m/s (an order-of-magnitude lower than the actual value).
Comparing the integral of fuel consumption in the core region to the
5.1. Flame propagation overall fuel consumption also suggests that 86.7% of fuel consumption
occurs in the core.
The feasibility of flame stabilisation by purely deflagrative turbulent In summary, the flame cannot be stabilised by turbulent premixed
flame propagation is assessed by comparing relevant flow and flame flame propagation at these conditions, despite the vast majority of the
speeds. First, a comparison is made between the turbulent flame speed fuel being consumed by deflagration in the core flame zone.
and jet velocity to establish the feasibility of flame stabilisation by
flame propagation; specifically, by estimating the upper bound on the 5.2. Ignition events
jet inlet velocity that can be stabilised by turbulent flame propagation
To examine the evolution of ignition events, a line of sight diagnos-
without blowing off. Following [47], in the large-scale turbulence
tic has been constructed, taking a one-dimensional integral of a reaction
limit (i.e. giving the highest feasible enhancement of surface area), an
term, i.e.
estimate for turbulent flame speed can be written as
𝐿𝑖
1
𝑠𝑇 = 𝑠 + 2𝑢′ , (12) 𝜔̇ los = 𝜔̇ 𝑘 d𝑥𝑖 , (17)
𝐿𝑖 ∫ 0

10
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

where 𝑖 could be either 𝑥 or 𝑦, and 𝜔̇ 𝑘 could be a species consumption

or heat release rate. This removes the possibility of misleading out-of-
plane effects, and so allows for easy identification of ignition spots and
tracking them as they move through multiple planes.
Fig. 17 shows the formation and evolution of an isolated kernel. The
left-hand column shows the line of sight diagnostic (using heat release
rate as the reaction rate variable), where a small flame kernel appears
spontaneously and grows with time (from top to bottom). The forma-
tion and growth of the flame kernel can also be seen in the temperature
slices in the middle column. The accumulation of HO2 seen in the right-
hand column demonstrates the presence of low-temperature chemistry
and eventual thermal runaway associated with ignition. Additional
ignition events can be observed at later times in the lower-right corner
of the images in Fig. 17. The kernel formation appears to coincide with
Kelvin–Helmholtz (KH) rollers; large-scale shear-driven mixing brings
fuel into contact with recirculated heat, leading to ignition. Not all large
scale KH mixing events lead to ignition, but when an ignition event
occurs, it appears to coincide with a KH roller. Ignition events happen
more frequently with increasing height; sporadic kernels are seen to
appear below the flame stabilisation point, whereas more frequently
formed kernels form adjacent to the peripheral flame. Ignition events
can also form in clusters or sheets; Fig. 18 shows the simultaneous
formation of several flame kernels, which propagate towards the core of
the jet, evolving into a flame sheet. The left column is again the line of
sight where several small kernels appear before connecting and forming
a relatively-stable flame sheet on the side of the jet. This can be seen
further in the temperature and 𝑌ho2 slices, and this sheet persists for far
longer than the sporadic kernels that appear. All of these phenomena
can be seen more clearly in the accompanying animations provided as
supplementary materials.
Animations of the flame (provided as supplementary material) show Fig. 17. Evolution of an igniting kernel from formation to flame sheet. The circles
that the core flame is stabilised by these repeated, sometimes isolated, point to the formation of the initial kernel and the secondary kernels that form in
the KH roller. The left column is the line of sight (17), the middle column is the
ignition events. The peripheral flame itself is approximately situated in
temperature field and the right column is the heat release rate. Time is increasing
a (highly-turbulent) stagnation region. However, the convection time from top to bottom, and the time difference between slices 𝛥𝑡 = 10.3 μs. The axis on
from the fuel injector to the flame base (around 0.2 ms based on cen- the left and top indicates the height and depth respectively in terms of jet diameters.
treline velocity) is orders-of-magnitude smaller than the autoignition
delay time of hydrogen at the mean inflow conditions (𝜏𝑖𝑔𝑛 > 9 s), pre-
cluding flame stabilisation by autoignition of the jet fluid alone. Rather, • OT - Mixing region between the pure air region (O) and hot
the appearance of isolated ignition kernels can be explained by mixing recirculation region (T).
between the jet fluid and recirculated lean low-temperature combustion Defined by 𝑅3 (𝑧) < 𝑟 < 𝑅4 (𝑧) (where 𝑧 < 𝑧𝑐 ), and by 𝑅2 (𝑧) < 𝑟 <
products. This mixing and the appearance of ignition kernels occurs in 𝑅4 (𝑧) (where 𝑧 > 𝑧𝑐 ).
the region where the spread of unburned fuel within the jet intersects • FOT - Region where there is a flammable mixture of species
with the recirculation zone. spread from the core of the jet and high temperatures entrained
from the recirculation region. Ignition is possible in this region.
5.3. Entrainment and mixing Defined by 𝑅3 (𝑧) < 𝑟 < 𝑅2 (𝑧) and 𝑧 > 𝑧𝑐 .

Since flame ignition and stabilisation appears to be driven by the The height 𝑧𝑐 is defined to be the point where 𝑅2 (𝑧) = 𝑅3 (𝑧), and the
mixing of fuel from the jet with recirculated heat, this section considers radius definitions are given by:
the mechanisms by which mixing occurs. Since the combustor is lean 𝑅1 (𝑧) = 𝑟(𝑌h2 = 0.0133), (18)
everywhere, there is an abundance of oxygen, and so any ignition
events and hence flame stabilisation point is determined by the mixing 𝑅2 (𝑧) = 𝑟(𝑌h2 = 0.003), (19)
of fuel and heat. Based on azimuthally and temporally averaged data, 𝑅3 (𝑧) = 𝑟(𝑇 = 850𝐾), (20)
different mixing zones can be classified as follows:
𝑅4 (𝑧) = 𝑟(𝑇 = 1150𝐾). (21)
• F - Unburned fuel–air mixture with 𝑌h2 > 0.0133.
The mass fractions are associated with the mean equivalence ratio and
Defined by 0 < 𝑟 < 𝑅1 (𝑧).
flammability limit of hydrogen, which also dictates the lower temper-
• O - Essentially pure air which acts as a non-reactive buffer be-
ature limit. The upper limit of the temperature is set to reflect the
tween the regions with fuel and hot recirculating fluid.
temperature of the recirculation zone. Fig. 19 shows the partitioning
Defined by 𝑅2 (𝑧) < 𝑟 < 𝑅3 (𝑧) and 𝑧 < 𝑧𝑐 . of the domain after applying the conditioning.
• 𝑇 - Hot recirculating products from the flame above, mostly For each zone 𝛺𝑖 , the integrated mass , species  and heat  fluxes
comprised of steam and air. are given by
Defined by 𝑟 > 𝑅4 (𝑧). 𝑅r,𝑖
• FO - Mixing region between the inner fuel/air region (F) and pure 𝛺𝑖 (𝑧) = 𝜌𝑢𝑧 𝑓 (𝑟) d𝑟, (22)
∫𝑅l,𝑖
air region (O).
Defined by 𝑅1 (𝑧) < 𝑟 < 𝑅2 (𝑧) (where 𝑧 < 𝑧𝑐 ), and by 𝑅1 (𝑧) < 𝑟 < 𝑅r,𝑖
𝛺𝑖 (𝑧) = 𝜌𝑌h2 𝑢𝑧 𝑓 (𝑟) d𝑟, (23)
𝑅3 (𝑧) (where 𝑧 > 𝑧𝑐 ). ∫𝑅l,𝑖

11
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

constant as hydrogen moves to the FOT region. The recirculation region

(T) loses mass and heat to the OT region, which can be seen through the
negative (positive) values of mass and heat flux in the 𝑇 (OT) region;
there is no transport of fuel in either of these regions. Beyond 𝑧𝑐 the
OT region transfers mass and heat to the FOT region.
In each panel in Fig. 20, the totals are essentially constant; there is
practically no fuel consumption or heat release since the analysis region
is upstream of the flame base, and supports the use of temperature as an
approximation for heat flux. Therefore, the type and direction of species
and heat transfer can be examined by integrating the cylindrical form
of the continuity, species and temperature equation (for simplicity,
assuming constant specific heat capacities, no reactions, and neglecting
molecular diffusion), giving
[ ]𝑅r,𝑖
d
 = 𝜌𝐮 ⋅ 𝐧 , (26)
d𝑧 𝛺𝑖 𝑅l,𝑖
[ ]𝑅r,𝑖
d
 = 𝜌𝑌h2 𝐮 ⋅ 𝐧 , (27)
d𝑧 𝛺𝑖 𝑅l,𝑖
[ ]𝑅r,𝑖
d
 = 𝑇𝐮 ⋅ 𝐧 , (28)
d𝑧 𝛺𝑖 𝑅l,𝑖

𝑅
where [𝑞(𝑟)]𝑅2 = 𝑞(𝑅2 ) − 𝑞(𝑅1 ) is used to denote the difference in
1
horizontal fluxes in and out of the region.
The equations can be decomposed further by specifying the normal
as
[ ] [ ]
𝑛 ±∇𝑐 ±1 𝜕𝑟 𝑐
𝐧= 𝑟 = = √ , (29)
𝑛𝑧 |∇𝑐| 2
(𝜕𝑟 𝑐) + (𝜕𝑧 𝑐)2 𝜕 𝑧𝑐

where 𝑐 is the field chosen to define the boundary, either 𝑌h2 or 𝑇 in this
particular case; the sign of the normal is chosen such that the direction
is always to the right (i.e. 𝑛𝑟 ≥ 0). The resulting decomposition can be
written as
[ ]𝑅r,𝑖
d
 = (𝜌̄𝑢̃ 𝑧 )𝑛𝑧 + (𝜌̄𝑢̃ 𝑟 )𝑛𝑟 , (30)
d𝑧 𝛺𝑖 𝑅l,𝑖
[( ) ( ) ]𝑅r,𝑖
d
 = 𝜌̄𝑌̃h2 𝑢̃ 𝑧 + 𝜌̄𝑌̃ ′′ ′′ ̃ ̃ ′′ ′′
h2 𝑢𝑧 𝑛𝑧 + 𝜌̄𝑌h2 𝑢̃ 𝑟 + 𝜌̄𝑌h2 𝑢𝑟 𝑛𝑟 𝑅 , (31)
d𝑧 𝛺𝑖 l,𝑖
[( ) ( ) ]𝑅r,𝑖
d
 = 𝑇̄ 𝑢̄ 𝑧 + 𝑇 ′ 𝑢′𝑧 𝑛𝑧 + 𝑇̄ 𝑢̄ 𝑟 + 𝑇 ′ 𝑢′𝑟 𝑛𝑟 . (32)
d𝑧 𝛺𝑖 𝑅l,𝑖

for Favre-averaged mean and fluctuations 𝑞̃ = 𝜌𝑞∕𝑞̄ and 𝑞 ′′ = 𝑞 − 𝑞, ̃

respectively. The exchanges of particular interest are the species fluxes
flowing from the F to FO to FOT zones, as well as the heat flux flowing
from 𝑇 to OT to FOT zones. Fig. 21 shows the evaluation of the bound-
ary hydrogen fluxes between the F and FO, and FO and FOT zones. In
Fig. 18. Formation of multiple flame kernels simultaneously leading to a sheet that
both cases, the main driver of species transport is the turbulent radial
propagates inwards and blows downstream much more slowly compared to isolated
kernels. As with Fig. 17, the left column is the line of sight (17), the middle column is term (𝜌̄𝑌̃′′ ′′ ̃
h2 𝑢𝑟 𝑛𝑟 ), followed by mean radial transport driven (𝜌̄𝑌h2 𝑢̃ 𝑟 𝑛𝑟 ),
the temperature and the right column is the heat release rate. Time is increasing from with very little streamwise transport. From Fig. 22, the transport of
top to bottom and the time difference between slices 𝛥𝑡 = 15.6 μs. The axis on the left heat from the 𝑇 to OT region (below 𝑧∕𝑑j = 2) is mostly driven by
and top indicates the height and depth respectively in jet diameters.
mean transport (both 𝑇̄ 𝑢̄ 𝑟 𝑛𝑟 and 𝑇̄ 𝑢̄ 𝑧 𝑛𝑧 ), with little contribution from
the turbulent terms. Finally, the heat transport between the OT and
𝑅r,𝑖 FOT region is driven by mean streamwise transport (𝑇̄ 𝑢̄ 𝑧 𝑛𝑧 ).
𝛺𝑖 (𝑧) = 𝑇 𝑢𝑧 𝑓 (𝑟) d𝑟, (24)
∫𝑅l,𝑖

where 𝑅r,𝑖 and 𝑅l,𝑖 are the radii associated with the left and right 6. Discussion and conclusions
boundaries of zone 𝛺𝑖 , respectively, and the function 𝑓 (𝑟) is given by
A high-pressure hydrogen micromix combustor has been investi-
⎧ 𝐿𝑥
⎪2𝜋𝑟 if 𝑟 < , gated using three-dimensional DNS with detailed chemistry to examine
𝑓 (𝑟) = ⎨ ( ) 2 (25)
𝐿𝑥 the flame structure and stabilisation mechanism. The combustor ge-
⎪2𝜋𝑟 − 8𝑟 cos−1 otherwise.
⎩ 2𝑟 ometry follows the archetypal NASA micromix design from Schefer
By partitioning the domain in this way, mass, species and heat et al. [1], and the high-pressure and temperature operating condi-
exchange between each zone can be evaluated; integrated fluxes are tions are relevant to aerospace applications. The flame stabilisation
presented in Fig. 20. In the fuel region (F), the mass and heat flux process at these conditions is summarised in Section 6.1, and shown
stay constant, while the hydrogen flux decreases as it moves to the FO schematically in Fig. 23. A more general discussion of the main features
region. In the FO region, the mass and heat flux again are relatively of micromix flame stabilisation is given in Section 6.2, along with
constant, while the hydrogen flux increases up to 𝑧𝑐 and remains implications for design across different operating conditions.

12
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 19. Zonal arrangement of the F/FO/FOT/O/OT/T regions. (For interpretation of

the references to colour in this figure legend, the reader is referred to the web version
of this article.)

6.1. Flame structure and stabilisation mechanism

The present simulation employs opposed jet-in-crossflow injection

of hydrogen fuel with momentum ratio 𝐽 = 17.2. At these conditions,
the hydrogen jets meet at the mid-plane (𝑦 = 0) and roll up into stream-
wise vortex tubes. At the inlet plane, the resulting profiles of fuel
mass fraction are shaped like two kidneys near the centreline (Fig. 4),
with a clear air gap between the fuel and the walls of the air port.
Planar averaging gives one-dimensional streamwise profiles (Fig. 6)
that indicate a liftoff height for the flame base between approximately 3
and 4.5 jet diameters (21 to 31.5 mm). Azimuthally-averaged profiles
in 𝑟 − 𝑧 space (Fig. 7) and streamlines (Fig. 8) indicate the existence Fig. 20. Integrated mass, species and thermal fluxes as a function of streamwise
of regions of recirculation in the corners of the domain (also see distance. Colour matches that used in Fig. 19. (For interpretation of the references to
Fig. 5b), with the recirculation bubble reaching up to 4 jet diameters colour in this figure legend, the reader is referred to the web version of this article.)

downstream. Importantly, the recirculation region extends sufficiently

far to result in upstream transport of combustion products (i.e. heat).
Instantaneous temperature and intermediate species fields (Figs. 9 peripheral flame burns in the narrower and leaner range of 0.18–
and 10) suggest there are multiple burning regimes in the domain 0.3 (Fig. 14). PDFs of normalised flame index demonstrate that no
(Fig. 16). Construction of fuel- and products-based progress variables diffusion flamelets are present (Fig. 13). The core flame was classified
(Fig. 11) concur with these findings, suggesting vastly different in the thin reaction zone, and the peripheral flame in the distributed
turbulence-flame interactions occur within each flame. The progress burning regime. The stark difference in turbulence-flame interaction
variables also agree that a non-reacting air buffer is present at the inlet arises due to the high pressure of the system, which has resulted in
and a short distance into the domain and that combustion products are a large Damköhler number in the core flame, and a large Karlovitz
present in the recirculation zone. There is a thin core flame situated number in the peripheral flame.
directly above the main jet region, accounting for over 85% of the total Given the essentially-premixed nature of the flame, the role of
fuel consumption, and a much thicker peripheral flame shrouding the turbulent burning velocity on the flame stabilisation is assessed. By
core. By partitioning the domain into 5 categories (jet fluid, core flame, estimating an upper bound for jet velocity that would allow flame sta-
peripheral flame, recirculating fluid and products, (Fig. 12), conditional bilisation through flame propagation, it was concluded that this flame
analyses were performed to examine the combustion regimes. It was cannot be stabilised through turbulent premixed flame propagation.
found that the core flame burns in an inhomogeneously-premixed Evaluation of mean local flame speeds and turbulent intensities sup-
mode, with equivalence ratios varying from 0.4–1.2, whereas the ports this argument, and suggests that the jet is an order-of-magnitude

13
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. 21. Hydrogen flux across F/FO and FO/FOT boundary. Fig. 22. Heat flux across OT/T and FOT/OT boundary.

too fast for flame propagation to be able to stabilise the flame at turbulent mixing in the radial direction (𝜌̄𝑌̃ ′′ ′′
h2 𝑢𝑟 𝑛𝑟 ; Fig. 21). On the other
these conditions. The flame speed computed is comparable to the one- hand, the transport of heat was shown to be driven by the mean flow
dimensional flame speed, implying that, although the core flame is not (𝑇̄ 𝑢;
̄ Fig. 22) from the recirculation to the ignition zone. Hypothetically,
stabilising, it is deflagrative (rather than ignitive) in nature. a one-dimensional model could be formulated to predict the structure of
The autoignition delay time of the inlet condition is orders-of- this zonal arrangement using appropriate models for turbulent mixing
magnitude too large to allow for the incoming mixture to autoignite of fuel inside the jet and entrainment of hot fluid, and will be the
with the prescribed domain size and inlet speed. However, sporadic subject of future work.
ignition kernels are observed ahead of the flame base and appear An idealised schematic (assuming symmetry across the left bound-
to be correlated with shear-driven Kelvin-Helmoltz vortices. By us- ary) of the combustor is shown in Fig. 23.
ing a line-of-sight diagnostic (Fig. 17), infrequent and isolated events
are observed, which grow into rapidly-burning flame kernels. Fur- 6.2. Principles for micromix combustor design
ther downstream, as the fuel spreads into the recirculation region,
the likelihood of ignition increased, and it was possible to observe 6.2.1. Present configuration
near-simultaneous ignition events more spatially distributed, which In the specific configuration presented here, the momentum flux
appeared more like an ignition sheet (Fig. 18). These sheets are situated ratio 𝐽 is a critically-important factor affecting lift-off height and
between the peripheral flame and core flame and result from the general flame dynamics. Ignoring direct interaction with other fuel
entrainment of the distributed reactions into the jet. injectors, increasing the value of 𝐽 will naturally increase the size of
Since there is an abundance of oxygen everywhere (due to the the air annulus and affect the point at which fuel encounters necessary
lean global equivalence ratio), the formation of ignition spots relies heat to ignite. It will also, as discussed in Section 2, increase the degree
on mixing fuel from the main jet with heat from the recirculation of mixing occurring before entering the combustor and within the jet.
zone. Partitioning the domain below the flame stabilisation point (using As such, times associated with the ignition process would be expected
values of hydrogen mass fraction and temperature corresponding to to decrease as hot fluid will encounter richer mixtures more readily
the flammability limits) identified five regions: fuel (F), oxidiser (O), upon entrainment into the jet. The precise relationship between 𝐽 and
temperature (T), fuel+oxidiser (FO), oxidiser+temperature (OT), and the lift-off height is beyond the scope of this paper and would be
fuel+oxidiser+temperature (FOT). This partitioning was used to anal- more suited to either large eddy simulation, experimental measurement
yse boundary fluxes between each of the zones, which showed that the or direct numerical simulation at lower pressures where resolution
spreading of hydrogen to the point of ignition primarily results from requirements are less stringent. However, flame stability will also be

14
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

2. The lean flammability isosurface of the fuel plume should in-

tersect with the recirculation zone; otherwise, the flame could
easily blow off.
3. The equivalence ratio of the recirculated fluid should be suffi-
ciently high that there is enough heat to ignite the fuel as it is
entrained into the jet.
4. For low-NO𝑥 operation, the fuel–air premixture should be
flammable but overall fuel-lean, preferably without any rich or
stoichiometric mixture, by the point that it reaches the core
flame.

CRediT authorship contribution statement

T.L. Howarth: Conceptualization, Data curation, Formal analy-

sis, Investigation, Methodology, Writing – original draft, Writing –
review & editing. M.A. Picciani: Conceptualization, Funding acquisi-
tion, Project administration, Supervision, Writing – review & editing.
E.S. Richardson: Supervision, Writing – review & editing. M.S. Day:
Software, Supervision, Writing – review & editing, Resources. A.J.
Aspden: Conceptualization, Funding acquisition, Methodology, Project
administration, Supervision, Writing – review & editing, Resources.

Fig. 23. Idealised half-plane schematic of the present micromix combustor.

Declaration of competing interest

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared to
dictated by the heat loss at the sides of the combustor, which are
influence the work reported in this paper.
not considered here through the periodic boundary conditions, where
extinguishing of outer flames may lead to full failure of the flame.
Acknowledgements
Furthermore, the usage of a low Mach number solver has neglected
the effect that thermoacoustic excitation will have on the flame. Future
This project is funded by an EPSRC Industrial CASE Studentship and
work could focus on any of these numerous factors that can affect flame
Reaction Engines Ltd. (grant number: EP/R51309X/1-2282954); the
stability in micromix combustors.
authors are grateful for discussions with Dr. Vladeta Zmijanovic. The
authors are grateful to EPSRC and ARCHER2 for computational support
6.2.2. General case as a part of their funding to the UK Consortium on Turbulent Reacting
While the present simulation represents a single combustor design Flows (grant number: EP/R029369/1). This research made use of the
and operating conditions, its analysis points to several principles of Rocket High Performance Computing Service at Newcastle University.
flame stabilisation in micromix combustors that are likely applicable Additionally, this research used resources of the Argonne Leadership
more generally. Given the design goal of low-NO𝑥 emission, a general Computing Facility, which is a DOE Office of Science User Facility sup-
objective is to achieve satisfactory flame stabilisation at low overall ported under Contract DE-AC02-06CH11357. This work was authored
equivalence ratios, and to provide short fluid residence time in any re- in part by the National Renewable Energy Laboratory, operated by
gions with locally elevated temperatures. Flame stabilisation behaviour Alliance for Sustainable Energy, LLC, for the U.S. Department of Energy
is satisfactory if a design provides a low risk of flashback/blowoff or (DOE) under Contract No. DE-AC36-08GO28308. This research was
flame attachment, acceptable ignition behaviour, and low susceptibility supported in part by the Exascale Computing Project (17-SC-20-SC), a
to thermoacoustic excitation. collaborative effort of the U.S. Department of Energy Office of Science
Hydrogen micromix combustion can be implemented in many dif- and the National Nuclear Security Administration. The views expressed
ferent ways: the fuel can be introduced into the airflow by any number in the article do not necessarily represent the views of the DOE or
of jets, which may be aligned with or angled into the airflow; the air the U.S. Government. The U.S. Government retains and the publisher,
jets need not be circular in cross-section, and might be arranged as by accepting the article for publication, acknowledges that the U.S.
a regular array (e.g. square or close-packed), or some other pattern Government retains a nonexclusive, paid-up, irrevocable, worldwide
intended to aid manufacture or improve performance; and the fuel/air license to publish or reproduce the published form of this work, or
ratio of individual air jets may in principle be controlled in order to allow others to do so, for U.S. Government purposes.
manage the overall emissions, heat release, and stability.
Using established experimental and computational methods, it is Appendix A. Resolution independence
feasible to tailor various different types of micromix combustor de-
signs to achieve desired mixing features. Even simple models could To establish resolution effects on the results presented in the paper,
be constructed configuration for both the recirculation bubble size one-dimensional flames were simulated at three representative equiva-
(e.g. [50,51]), or potentially the lift-off height by adapting models lence ratios each at three resolutions. The three resolutions presented
used for premixed bluff-body stabilised flames (e.g. [52]). For all of correspond to ML3 (𝛥𝑥 = 6.84 μm) that was used for data analysis,
the different design possibilities, the common objective is to achieve one extra level at ML4 and finally with two further levels ML6, which
recirculation zones and an inlet fuel–air premixture that provides the was taken as the reference for comparison. Figs. A.24(a), A.24(b) and
following features: A.24(c) show the profiles of the mass fractions of H, HO2 , OH and heat
release rate 𝑄 against temperature at three different equivalence ratios
1. At the inlet to the combustor, the fuel–air premixture should be (𝜙 = 0.4, 0.7 and 1.0). In the leanest case shown (𝜙 = 0.4, Fig. A.24(a))
adequately surrounded by a non-flammable mixture (e.g. air) to the thermal thickness is 23.2 μm which corresponds to 𝛥𝑥∕𝓁𝐿 = 3.39.
prevent flame attachment or flashback to the combustor lip. The one-dimensional profiles appear to be well-matched at both the

15
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

Fig. A.24. Profiles of 𝑌h , 𝑌ho2 , 𝑌oh and heat release rate 𝑄 as functions of temperature for 𝜙 = 0.4, 0.7, 1.0. Reasonable agreement with high-resolution profile is seen throughout,
bar small lags in the heat release profiles in the higher equivalence ratio cases.

ML3 and ML4 resolutions. In the middle case (𝜙 = 0.7, Fig. A.24(b)), nature of the numerical algorithm guarantees that the temperature of
the thermal thickness is 10.8 μm which corresponds to 𝛥𝑥∕𝓁𝐿 = 1.58. the products (and therefore recirculation region) are accurate.
Again, the profiles in both ML3 and ML4 data appear to match well
(despite the sparsity of data points), with a slight lag in the heat release Appendix B. Supplementary animations
noted. Finally, at the highest equivalence ratio (𝜙 = 1, Fig. A.24(c)),
the thermal thickness of 10.7 μm which corresponds to 𝛥𝑥∕𝓁𝐿 = 1.56.
Supplementary material related to this article can be found online
Similar observations are clear, with a nearly identical profile to the
at https://doi.org/10.1016/j.combustflame.2024.113504. Three ani-
𝜙 = 0.7 case, with good agreement, save for the slight lag in the heat
mations are included as supplementary materials. The first (named
release profiles.
‘‘Slices’’) presents diagonal slices through the domain (exploiting pe-
In each case, the flame speed was calculated as
riodicity to stitch the two together); the fields presented in turn are
∞
−1 vorticity magnitude, streamwise velocity, fuel mass fraction, oxygen
𝑠l,lev = 𝜌𝜔̇ d𝑥. (A.1)
𝜌𝑢 𝑌h2 ,𝑢 ∫−∞ h2 mass fraction, temperature, OH mass fraction, HO2 mass fraction, and
The flame speeds at 𝜙 = 0.4 at ML3, 4 and 6 were 0.709, 0.729 and heat release rate. The second presents (named ‘‘Combo’’) combines
0.724 m/s, respectively (i.e. a difference of approximately 2%). The temperature, heat release rate and fuel mass fraction in the red, green
peak differences at 𝜙 = 0.7 and 𝜙 = 1.0 were 0.2% and 1.4%, re- and blue channels, respectively; this highlights the mixing of fuel and
spectively. These small differences will lead to no different conclusions heat, resulting in ignition events. The third (named ‘‘Line of sight’’)
when examining flame speed statistics (i.e. Section 5.1). Furthermore, is the line-of-sight diagnostic, zoomed in on the region where ignition
the specific details of the internal flame structure are largely irrelevant events are observed; the diagnostic eliminates out-of-plane effects, and
for the global structure and stability, and the inherent conservative clearly demonstrates spontaneous ignition.

16
T.L. Howarth et al. Combustion and Flame 265 (2024) 113504

References [28] M.T. Henry de Frahan, L. Esclapez, J. Rood, N.T. Wimer, P. Mullowney, B.A.
Perry, L. Owen, H. Sitaraman, S. Yellapantula, M. Hassanaly, M.J. Rahimi, M.J.
[1] R.W. Schefer, T.D. Smith, C.J. Marek, Evaluation of NASA Lean Premixed Martin, O.A. Doronina, S.N. A., M. Rieth, W. Ge, R. Sankaran, A.S. Almgren, W.
Hydrogen Burner, Tech. Rep. SAND2002-8609, Sandia National Laboratories, Zhang, J.B. Bell, R. Grout, M.S. Day, J.H. Chen, The Pele Simulation Suite for
United States, 2003. Reacting Flows at Exascale, in: Proceedings of the 2024 SIAM Conference on
[2] H.H.-W. Funke, N. Beckmann, S. Abanteriba, An overview on dry low NOx Parallel Processing for Scientific Computing, 2024, pp. 13–25.
micromix combustor development for hydrogen-rich gas turbine applications, Int. [29] J.B. Bell, R.K. Cheng, M.S. Day, I.G. Shepherd, Numerical simulation of Lewis
J. Hydrog. Energy 44 (13) (2019) 6978–6990. number effects on lean premixed turbulent flames, Proc. Combust. Inst. 31 (1)
[3] H.H.-W. Funke, N. Beckmann, J. Keinz, A. Horikawa, 30 Years of dry-low-NOx (2007) 1309–1317.
micromix combustor research for hydrogen-rich fuels - an overview of past and [30] T.L. Howarth, A.J. Aspden, An empirical characteristic scaling model for
present activities, J. Eng. Gas Turbines Power 143 (7) (2021). freely-propagating lean premixed hydrogen flames, Combust. Flame 237 (2022).
[4] A. Haj Ayed, K. Kusterer, H.H.-W. Funke, J. Keinz, C. Striegan, D. Bohn, Ex- [31] A.J. Aspden, M.S. Day, J.B. Bell, Characterization of low lewis number flames,
perimental and numerical investigations of the dry-low-NOx hydrogen micromix Proc. Combust. Inst. 33 (1) (2011) 1463–1471.
combustion chamber of an industrial gas turbine, Propulsion Power Res. 4 (3) [32] T.L. Howarth, E.F. Hunt, A.J. Aspden, Thermodiffusively-unstable lean premixed
(2015) 123–131. hydrogen flames: Phenomenology, empirical modelling, and thermal leading
[5] H.H.-W. Funke, N. Beckmann, J. Keinz, S. Abanteriba, Numerical and experimen- points, Combust. Flame 253 (2023) 112811.
tal evaluation of a dual-fuel dry- low-NOx micromix combustor for industrial gas [33] A.J. Aspden, M.S. Day, J.B. Bell, Turbulence-chemistry interaction in lean
turbine applications, J. Thermal Sci. Eng. Appl. 11 (1) (2019). premixed hydrogen combustion, Proc. Combust. Inst. 35 (2) (2015) 1321–1329.
[6] G. Lopez-Ruiz, I. Alava, J.M. Blanco, Study on the feasibility of the micromix [34] A.J. Aspden, M.S. Day, J.B. Bell, Towards the distributed burning regime in
combustion principle in low NOx H2 burners for domestic and industrial boilers: turbulent premixed flames, J. Fluid Mech. 871 (2019) 1–21.
A numerical approach, Energy 236 (2021) 121456. [35] J.B. Bell, M.S. Day, M.J. Lijewski, Simulation of nitrogen emissions in a premixed
[7] S. Boerner, H.H.-W. Funke, P. Hendrick, E. Recker, R. Elsing, Development and hydrogen flame stabilized on a low swirl burner, Proc. Combust. Inst. 34 (1)
integration of a scalable low NOx combustion chamber for a hydrogen-fueled (2013) 1173–1182.
aerogas turbine, Progr. Propulsion Phys. 4 (2013) 357–372. [36] M.S. Day, S. Tachibana, J.B. Bell, M. Lijewski, V. Beckner, R.K. Cheng, A
[8] J. Zhang, Z. Wang, Q. Li, Thermodynamic efficiency analysis and cycle optimiza- combined computational and experimental characterization of lean premixed
tion of deeply precooled combined cycle engine in the air-breathing mode, Acta turbulent low swirl laboratory flames II. Hydrogen flames, Combust. Flame 162
Astronaut. 138 (2017) 394–406. (5) (2015) 2148–2165.
[9] K. Mahesh, The interaction of jets with crossflow, Annu. Rev. Fluid Mech. 45 [37] M.S. Day, J.B. Bell, Numerical simulation of laminar reacting flows with complex
(2013) 379–407. chemistry, Combust. Theory Model. 4 (4) (2000) 535.
[10] S.H. Smith, M.G. Mungal, Mixing, structure and scaling of the jet in crossflow, [38] A. Nonaka, J.B. Bell, M.S. Day, C. Gilet, A.S. Almgren, M.L. Minion, A deferred
J. Fluid Mech. 357 (1998) 83–122. correction coupling strategy for low mach number flow with complex chemistry,
[11] J.W. Shan, P.E. Dimotakis, Reynolds-number effects and anisotropy in Combust. Theory Model. 16 (6) (2012) 1053–1088.
transverse-jet mixing, J. Fluid Mech. 566 (2006) 47–96. [39] A. Nonaka, M.S. Day, J.B. Bell, A conservative, thermodynamically consistent
[12] L. Gevorkyan, T. Shoji, D.R. Getsinger, O.I. Smith, A.R. Karagozian, Transverse numerical approach for low mach number combustion. Part I: Single-level
jet mixing characteristics, J. Fluid Mech. 790 (2016) 237–274. integration, Combust. Theory Model. 22 (1) (2017) 156–184.
[13] L. Gevorkyan, T. Shoji, W.Y. Peng, A.R. Karagozian, Influence of the velocity [40] M.P. Burke, M. Chaos, Y. Ju, F.L. Dryer, S.J. Klippenstein, Comprehensive H2/O2
field on scalar transport in gaseous transverse jets, J. Fluid Mech. 834 (2018) kinetic model for high-pressure combustion, Int. J. Chem. Kinet. 44 (7) (2012)
173–219. 444–474.
[14] D.R. Getsinger, L. Gevorkyan, O.I. Smith, A.R. Karagozian, Structural and stability [41] A.J. Aspden, N. Nikiforakis, S.B. Dalziel, J.B. Bell, Analysis of implicit LES
characteristics of jets in crossflow, J. Fluid Mech. 760 (4) (2014) 342–367. methods, Commun. Appl. Math. Comput. Sci. 3 (1) (2008) 103–126.
[15] K.M. Lyons, Toward an understanding of the stabilization mechanisms of lifted [42] H. Schlichting, J. Kestin, Boundary layer theory, vol. 121, Springer, 1961.
turbulent jet flames: Experiments, Prog. Energy Combust. Sci. 33 (2) (2007) [43] D.G. Goodwin, R.L. Speth, H.K.M. Weber, B. W., Cantera: An object-oriented
211–231. software toolkit for chemical kinetics, thermodynamics, and transport processes,
[16] C.J. Lawn, Lifted flames on fuel jets in co-flowing air, Prog. Energy Combust. 2018.
Sci. 35 (1) (2009) 1–30. [44] K. Bray, P. Domingo, L. Vervisch, Role of the progress variable in models
[17] N. Peters, F.A. Williams, Liftoff characteristics of turbulent jet diffusion flames, for partially premixed turbulent combustion, Combust. Flame 141 (4) (2005)
AIAA J. 21 (3) (1983) 423–429. 431–437.
[18] N. Peters, Turbulent combustion, Cambridge University Press, 2000. [45] H. Yamashita, M. Shimada, T. Takeno, A numerical study on flame stability at
[19] J.E. Broadwell, W.J. Dahm, M.G. Mungal, Blowout of turbulent diffusion flames, the transition point of jet diffusion flames, Symp. (International) Combust. 26
Symp. (International) Combust. 20 (1) (1985) 303–310. (1) (1996) 27–34.
[20] J. Buckmaster, R. Weber, Edge-flame-holding, Symp. (International) Combust. 26 [46] P. Domingo, L. Vervisch, K. Bray, Partially premixed flamelets in LES of
(1) (1996) 1143–1149. nonpremixed turbulent combustion, Combust. Theory Model. 6 (4) (2002) 529.
[21] S. Karami, E.R. Hawkes, M. Talei, J.H. Chen, Mechanisms of flame stabilisation [47] N. Peters, The turbulent burning velocity for large-scale and small-scale
at low lifted height in a turbulent lifted slot-jet flame, J. Fluid Mech. 777 (2015) turbulence, J. Fluid Mech. 384 (1999) 107–132.
633–689. [48] L. Berger, A. Attili, H. Pitsch, Intrinsic instabilities in premixed hydrogen flames:
[22] C. Pantano, Direct simulation of non-premixed flame extinction in a methane–air parametric variation of pressure, equivalence ratio, and temperature. Part 2 –
jet with reduced chemistry, J. Fluid Mech. 514 (2004) 231–270. non-linear regime and flame speed enhancement, Combust. Flame 240 (2022)
[23] C.S. Yoo, R. Sankaran, J.H. Chen, Three-dimensional direct numerical simulation 111936.
of a turbulent lifted hydrogen jet flame in heated coflow: flame stabilization and [49] L. Berger, K. Kleinheinz, A. Attili, H. Pitsch, Characteristic patterns of thermod-
structure, J. Fluid Mech. 640 (2009) 453–481. iffusively unstable premixed lean hydrogen flames, Proc. Combust. Inst. 37 (2)
[24] S.G. Kerkemeier, C.N. Markides, C.E. Frouzakis, K. Boulouchos, Direct numerical (2019) 1879–1886.
simulation of the autoignition of a hydrogen plume in a turbulent coflow of hot [50] J.C. Massey, I. Langella, N. Swaminathan, A scaling law for the recirculation
air, J. Fluid Mech. 720 (2013) 424–456. zone length behind a bluff body in reacting flows, J. Fluid Mech. 875 (2019)
[25] S. Karami, E.R. Hawkes, M. Talei, J.H. Chen, Edge flame structure in a turbulent 699–724.
lifted flame: A direct numerical simulation study, Combust. Flame 169 (2016) [51] D.P. Kallifronas, P. Ahmed, J.C. Massey, M. Talibi, A. Ducci, R. Balachandran,
110–128. N. Swaminathan, K.N. Bray, Influences of heat release, blockage ratio and swirl
[26] C.S. Yoo, E.S. Richardson, R. Sankaran, J.H. Chen, A DNS study on the on the recirculation zone behind a bluff body, Combust. Theory Model. (2022).
stabilization mechanism of a turbulent lifted ethylene jet flame in highly-heated [52] D. Most, F. Dinkelacker, A. Leipertz, Lifted reaction zones in premixed turbulent
coflow, Proc. Combust. Inst. 33 (1) (2011) 1619–1627. bluff-body stabilized flames, Proc. Combust. Inst. 29 (2) (2002) 1801–1808.
[27] L. Esclapez, M. Day, J. Bell, A. Felden, C. Gilet, R. Grout, M. Henry de Frahan,
E. Motheau, A. Nonaka, L. Owen, B. Perry, J. Rood, N. Wimer, W. Zhang,
PeleLMeX: an AMR Low Mach Number Reactive Flow Simulation Code without
level sub-cycling, J. Open Source Softw. 8 (90) (2023) 5450.

17