Hot surface ignition dynamics in premixed hydrogen-air near the
lean flammability limit
L.R. Boeck
∗
a
,
J. Melguizo-Gavilanes
a,b
,
J.E. Shepherd
a
a
Graduate Aerospace Laboratories, California Institute of Technology, Pasadena, CA 91125, USA
b
Institut Pprime, UPR 3346 CNRS, ISAE–ENSMA, BP 40109, 86961 Futuroscope–Chasseneuil Cedex,
France
Abstract
The dynamics of ignition of premixed hydrogen-air from a hot glow plug were investigated in a combined experi-
mental and numerical study. Surface temperatures during heating and at ignition were obtained from 2-color pyrometry,
gas temperatures were measured by high-speed Mach-Zehnder interferometry, and far-field effects were captured by
high-speed schlieren imaging. Numerical simulations considered detailed chemical kinetics and differential diffusion
effects. In addition to the known cyclic (puffing) combustion phenomenon, singular ignition events (single puff) were
observed near the lean flammability limit. Detailed analysis of the results of our numerical simulations reveal the ex-
istence of multiple combustion transients within the thermal boundary layer following the initial ignition event and,
at late times, sustained chemical reaction within a thermal plume above the glow plug. The results have significant
implications for ignition from hot surfaces within near-flammability limit mixtures at the edge of plumes resulting from
accidental release of hydrogen or within the containments of nuclear power plants during severe accidents.
Keywords: Experiments; Hot surface; Hydrogen; Ignition; Numerical simulation
1. Introduction
Combustion of premixed hydrogen-air atmospheres near the lean flammability limit exhibits a rich vari-
ety of behavior. The very low (less than 10 cm/s for hydrogen concentrations less than 10% [1, 2]) burning
speeds, high diffusivity of hydrogen and the negative Markstein lengths result in a strong coupling of chem-
istry and hydrodynamics in lean mixtures. This coupling [3] results in significant effect of flame stretch
as well as the potential for thermal-diffusive and Landau-Darrieus instabilities, resulting in local extinction
Preprint submitted to Combustion and Flame
July 1, 2019
Preprint,
published in
Combustion and Flame
210 (
2019
) 467-478.
https://doi.org/10.1016/j.combustflame.2019.09.002
and significant geometric distortion of the flame surface even in otherwise laminar flows [1].
The combustion chemistry of hydrogen-air mixtures has been examined by many researchers, Sanchez
and Williams [4] provide an extensive review with sections 6.3 and 6.6 specifically addressing lean com-
bustion issues, which involve the competition between OH and HO
2
and H
2
O
2
reactions. In mixtures near
the flammability limits, it has long been known [5, 6], that buoyancy strongly affects the flame propaga-
tion. The effect of buoyancy is pronounced in lean hydrogen-air mixtures and the interaction of propagating
flames with gravity in normal and micro-gravity conditions has been examined by a number of researchers
[7–10]. Near-limit mixtures exhibit a number of unusual combustion phenomena such as flame balls [11].
The topic of ignition by hot surfaces has been examined extensively from the beginnings of combustion
science in the 19th century and continuing today as the subject of both fundamental and applied research.
Analytical studies of ignition within the boundary layer of hot surfaces have focused on the balance between
energy release, heat transfer and reactant depletion within the gas [12–17] as well as the role of conjugate
heat transfer between gas and surface [18] and the role of thermal diffusion [19] by hydrogen. Recent
studies on thermal ignition by hot particles [20–22] in hydrogen-oxygen-diluent mixtures have focused on
examining the ability to both measure and predict through either numerical simulation or correlation the
dependence of ignition thresholds on particle material, size and heating rate.
Applied research studies have focused on measuring [23, 24] and predicting the minimum surface tem-
perature required to ignite a flammable mixture (i.e., the ignition threshold). This metric is a function of
the mixture composition, hot surface size, orientation, shape and material, as well as the relative velocity
between gas and hot surface [25]. Many of these studies have been motivated by nuclear reactor safety and
were carried out in the 1980s [26] in which hot surfaces (glow plugs) were used to ignite hydrogen at lean
conditions. Loss-of-coolant accidents in nuclear reactors can potentially result in the release of significant
quantities of hydrogen and the generation of flammable atmospheres within the containment, which oc-
curred at Three Mile Island Unit 2 in 1979 [27] and in Units 1, 3 and 4 at Fukushima Dai-ichi in 2011 [28].
One strategy initially considered for mitigating explosive overpressure was to perform deliberate ignition
by glow plugs located inside nuclear reactor containments. The emphasis in the last two decades has shifted
to passive autocatalytic recombiners for ultra-lean hydrogen combustion in nuclear power plants with con-
tributions from engineering studies [29, 30], analysis [31], numerical simulation [32], and surface science
[33].
Another motivation for studying near-limit combustion and ignition processes is the proposed use of
hydrogen as an energy carrier, and the continuing use of hydrogen in chemical processes and as a coolant in
large-scale electric power generation. Accidental releases and explosions pose an inherent potential hazard,
and ignition of flammable clouds created by accidental release of hydrogen from production, transportation
or storage facilities is of continued interest.
2
Time (s)
0
10
20
30
40
50
60
Temperature (K)
400
600
800
1000
1200
1400
pyrometer reading
curve fit
ignition
Figure 1: Glow plug surface temperature evolution during heating, measured by 2-color pyrometry.
In the presence of continuous thermal ignition sources, Boettcher et al. [34], observed cyclic (puffing)
flame propagation in near-limit premixed mixtures using hydrocarbon fuels. This regime is characterized by
the repeated formation of a flame near the hot surface followed by propagation upward within the thermal
plume above the thermal igniter. The present work is motivated by these observations and examines in-
depth the regime of lean, near-limit hydrogen-air mixtures (
X
H
2
= 5-6%,
φ
= 0.125-0.15), the limiting case
of the puffing behavior in which only a single puff is observed. This phenomenon was first noticed by
Boettcher [35] (see p. 127) and we refer to this as the “single-puff regime”. The objective of the present
work is the characterization of single-puff behavior using a combined experimental and numerical approach.
2. Experimental methodology
2.1. Combustion vessel and procedure
Ignition experiments were conducted in a prismatic vessel whose internal dimensions were 0.114 m
(W) x 0.114 m (D) x 0.165 m (H). The hot surface, a cylindrical Autolite 1110 glow plug with a stainless
steel 316 surface, was mounted vertically in the lower section of the vessel and equipped with a horizontal
stagnation plate at the bottom, used to provide a defined flow boundary condition. The glow plug diameter
was 5.1 mm, and the height above the stagnation plate exposed to the mixture was 9.3 mm. This setup has
been used in previous work and is described in detail in [34].
The experiment was performed as follows: the vessel was evacuated and filled with hydrogen, oxygen
and nitrogen (
X
O
2
:
X
N
2
= 1 : 3.76) using the method of partial pressures with a 10 Pa measurement accu-
racy. The components were mixed by a circulation pump and left to settle for 180 s. The initial conditions
before the start of glow plug heating were
P
o
= 101.3
±
0.1 kPa and
T
o
= 296
±
3 K. Supplying electrical
current to the glow plug initiated the heating. The glow plug current and voltage were chosen such that fast
initial heating was achieved and temperature was stabilized subsequently, see Fig. 1. Each experiment was
terminated after 60 s.
3
2.2. Diagnostics
Ignition was characterized in terms of glow plug surface temperature, measured by 2-color pyrometry,
gas temperature fields in the vicinity of the glow plug, measured by high-speed Mach-Zehnder interferom-
etry, and the far-field was captured by high-speed schlieren imaging.
2.2.1. Surface temperature measurement by 2-color pyrometry
A two-color pyrometer was used to measure the glow plug surface temperature as a function of time.
The construction and calibration of the pyrometer, using a black-body source, is described in [36]. Pyrome-
try allows for non-contact temperature measurement, not affecting the boundary layer near the hot surface;
measurement errors pertaining to thermocouple measurements of hot surface temperatures, related to radia-
tive, convective and conductive losses from the junction, thermal contact resistance, and surface reactions,
are eliminated. The pyrometry measurements were additionally validated using interferometry [36]. The
pyrometer field-of-view had a Gaussian sensitivity profile with a FWHM of 1.0 mm. Optical access to the
glow plug was restricted such that the pyrometer could only measure the temperature of the side wall near
the glow plug top,
z
≈
8.5 mm. Since ignition consistently took place on the symmetry axis above the
glow plug top face, by translating the pyrometer we determined that the top face temperature was about
2.5% lower than the temperature at the pyrometer measurement location. This deviation was included in
the ignition threshold measurement uncertainty along with emissivity corrections, measurement noise and
calibration uncertainty. The total estimated measurement uncertainty is represented by error bars on the
plots.
Ignition events were identified using high-speed interferograms, which clearly show the incipient igni-
tion kernel, compared to no-ignition cases (see, for example, ignition kernel above the glow plug in Fig.
3,
t
=
t
ref
). The ignition threshold was determined by reading the surface temperature of the glow plug,
measured by 2-color pyrometry, at the time the ignition kernel first became visible in the interferograms.
2.2.2. Gas temperature measurement by high-speed Mach-Zehnder interferometry
The gas temperature in the vicinity of the glow plug was inferred from interferograms obtained from a
Mach-Zehnder interferometer and high-speed video camera, operating at a framing rate of 2000 fps and a
resolution of 800
×
800 px
2
, covering a field-of-view of about 25
×
25 mm
2
. The interferometer measured
variations in refractive index inside the combustion vessel. Temperature variations were computed from
interferograms using the methodology described in [36, 37].
2.2.3. Far-field characterization by high-speed schlieren imaging
A classical Z-type schlieren setup was used to qualitatively image the gas-density gradients associated
with thermal plume and flame, in the entire optically accessible region of the vessel. This extends the
4
smaller region observed by the Mach-Zehnder interferometer and enables direct comparison with numerical
simulations, which cover the entire vessel domain.
3. Computational methodology
The motion, transport and chemical reaction in the gas surrounding the glow plug were modeled using
the low Mach number, variable-density reactive Navier-Stokes equations with temperature-dependent trans-
port properties [38]. Differential diffusion effects were taken into account using a constant but non-unity
Lewis number for each species [38]. The Lewis numbers were computed for the specific investigated con-
ditions. A detailed description and validation of the model can be found in [39]. The governing equations
were solved in an axisymmetric 2-D geometry (a 5 degree wedge with its axis of rotational symmetry lo-
cated along the
z
-axis at
x
= 0
) using the OpenFOAM toolbox [40]. The chemistry was modeled using
M
́
evel’s mechanism for hydrogen oxidation, which includes 9 species and 21 reactions [41, 42]. A detailed
comparison of the ignition delay time performance of M
́
evel’s mechanism with others commonly used in
the literature is provided in [43, 44]. The main challenge in simulating very lean mixtures lies in the fact that
the concentrations of interest (
X
H
2
= 5%) fall outside the range in which these mechanisms are typically
validated ( 9.5%
≤
X
H
2
≤
63%). However, as will be shown below, this kinetic mechanism does a very
good job at reproducing the experimentally observed dynamics before, during and after ignition without
requiring any modification.
The computational domain was discretized with 200,000 cells, compressed near the wall of the glow
plug with a minimum cell size of 80
μ
m, to resolve the thermal and hydrodynamic boundary layers. A
resolution study resulted in the grid size chosen. Additionally, a thorough validation of the ability of our
numerical model to predict the heat transfer was performed in [37, 39] via a side-to-side comparison of
experimental and numerical temperature fields and profiles around the hot surface and thermal plume. The
initial conditions were
P
o
= 101 kPa,
T
o
= 300 K,
U
o
= (0,0) m/s, and mass fractions
Y
H
2
= 0.00364,
Y
O
2
=
0.23216,
Y
N
2
= 0.7642, corresponding to a hydrogen mole fraction of
X
H
2
= 5%. A no-slip boundary con-
dition and constant temperature,
T
wall
=
T
o
, were imposed on the vessel walls. On the glow plug surface, the
experimental temperature-time history measured by pyrometry as shown in Fig. 1 was imposed uniformly
on the entire glow plug surface. The simulation was carried out as an initial value problem, integrating in
time starting from quiescent initial conditions with ignition and flame propagation being computed as part
of the solution, rather than artificially igniting the mixture or imposing a prescribed flame speed as was done
in [34]. The numerical ignition criterion was defined as
T
ign
=
T
surf
+ 150
K.
5
4. Results
Section 4.1 provides ignition thresholds and temperature fields measured experimentally, and Sec. 4.2
shows numerical results in comparison with experimental results, analyzing the ignition process in detail.
4.1. Experimental ignition thresholds and temperature fields
Figure 2 presents ignition thresholds as a function of hydrogen mole fraction between the flammability
limits. Ignition thresholds increase with increasing hydrogen mole fraction, from 1010 K at
X
H
2
= 5%
to 1100 K at
X
H
2
= 70%, and an additional steep increase occurs near the upper flammability limit, up
to 1170 K at
X
H
2
= 73%. As previous research summarized in [43] has shown, ignition thresholds of
hydrogen-air mixtures for small hot surfaces with dimensions on the order of 1-10 mm range above 1000 K
and increase with increasing hydrogen concentration. As shown by Kutcha et al. [23], ignition thresholds
increase with decreasing hot surface size, due to increased heat losses and reduced residence time of gas
near the hot surface. As the temperatures increase above 900-1000 K, chain-branching reactions become
increasingly more relevant as compared to reactions forming peroxides [39, 43]. This leads to a sharp
decrease in ignition delay time with increasing surface temperature and enables ignition to occur when
ignition delay time and gas residence time near the hot surface become comparable. Numerical simulations
indicate that the increase in ignition threshold with increasing hydrogen concentration may be related to the
removal of H and HO
2
through surface reactions [43].
Between the flammability limits, we observed the following ignition regimes: (i) at
X
H
2
≥
10%, ex-
panding flames occurred with a steady consumption of the entire vessel volume; (ii) at
X
H
2
= 9%, a buoyant
flame occurred that propagated towards the vessel top and recirculated along the vessel side walls towards
the vessel bottom; (iii) at 7%
≤
X
H
2
≤
8% cyclic (puffing) combustion took place. In this regime ignition
takes place periodically at the hot surface. Boettcher et al. [34] described the fluid mechanics of this regime
in detail for
n
-hexane-air mixtures; (iv) at 5%
≤
X
H
2
≤
6%, we observed single-puff ignition events, which
had previously been observed, but not examined in detail by Boettcher [35]. We define a single-puff as a
single ignition event, in contrast to the periodic ignition of fresh mixture near the hot surface in the case of
cyclic (puffing) flames. The single-puff first manifests itself as an ignition kernel near the hot surface and
subsequently produces a region of high temperature attached to the hot surface. In contrast to expanding
flames, this high-temperature region does not expand laterally into the fresh mixture, but is confined to a
narrow vertical region bounded by the thermal plume above the hot surface. No further ignition kernels
appear near the hot surface within the observed time frame of 60 s. The detailed mechanism of single-puff
ignition is described next.
Gas temperature fields inferred from Mach-Zehnder interferometry are presented in Fig. 3 for
X
H
2
=
5%. The first frame,
t
=
t
ref
−
12.5 ms, corresponds to a glow plug surface temperature of 1010 K, where
6
Hydrogen mole fraction (%)
02
04
06
08
0
Ignition threshold (K)
800
900
1000
1100
1200
1300
1400
no ignition
expanding flame
puffing flame
single puff
10
12
800
1000
1200
1400
8
6
Φ
= 1
Figure 2: Ignition thresholds and ignition regimes (markers) as a function of hydrogen mole fraction.
t
=
t
ref
refers to the first appearance of an ignition kernel in the experiment. The leading edge of the flame
subsequently propagates upwards within the thermal plume, while the bottom part of the flame propagates
downwards along the glow plug sides, reaching a minimum height of about
z
≈
4 mm, at
t
=
t
ref
+ 12.5 ms.
At
t
=
t
ref
+ 25 ms, a characteristic concave flame section can be seen at
z
≈
10 mm and
x
≈
4 mm. This
section travels upwards subsequently. At later times,
t > t
ref
+ 1 s, a plume of hot gas is visible in the
temperature fields above the glow plug. The gas in this region is at distinctly higher temperature than the
gas within the initial thermal plume above the glow plug, at
t
=
t
ref
−
12.5 ms. Numerical simulations will
help evaluate this situation in more detail.
4.2. Numerical results and comparisons against experiments
Figure 4 shows a comparison of experimental and numerical schlieren sequences for a single-puff event
in a 5% hydrogen-air mixture. Numerical schlieren images display the gradient of the density field; the
optical details of the experimental schlieren imaging process were not computed. As a result, differences in
intensity between the experimental and numerical images are expected, particularly on the centerline of the
fields. The widths of the thermal plume and boundary layer, above and near the surface of the glow plug,
respectively, are predicted with reasonable accuracy for all the time instances shown. At
t
=
t
ref
, there are
only slight discrepancies regarding the location of the tip of the flame during the main ignition transient; for
t > t
ref
, the flame shape during its propagation along the plume is captured by the simulation. Additional
experimental schlieren sequences are included in the supplemental material, providing an overview of the
different ignition regimes referred to in Fig. 2, and demonstrating that single-puff ignition can also be
obtained from smaller ignition sources heated at higher rates, such as a heated wire, and is therefore not
specific to the geometry at hand.
For
X
H
2
= 5%, the simulation predicts an ignition threshold of 938 K, 7.1% lower than the value we
observe experimentally (1010 K). Due to this difference in ignition temperature, it is not informative to
present experimental and numerical temperature fields side-by-side. In [39] a thorough investigation of the
ignition dynamics was performed in which we studied the physical and chemical mechanisms that play a
7
t
=
t
ref
−
12.5 ms
t
=
t
ref
t
=
t
ref
+ 12.5 ms
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
t
=
t
ref
+ 25 ms
t
=
t
ref
+ 37.5 ms
t
=
t
ref
+ 50 ms
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
t
=
t
ref
+ 1 s
t
=
t
ref
+ 2 s
t
=
t
ref
+ 3 s
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
0
2
4
6
8
10
x (mm)
0
5
10
15
20
z (mm)
Temperature (K)
300
450
600
750
900
1050
1200
Figure 3: Single-puff ignition in 5% hydrogen-air mixture. Temperature fields inferred from experimental finite-fringe interferograms.
role in thermal ignition, including chemical pathways. Additionally, in [43], we discussed various effects
that could affect the experimentally reported and numerically predicted thresholds, such as non-uniform
heating of the hot surface and surface reactions in the experiments, and the choice of reaction mechanism
in the simulation. Different reaction mechanisms compared in [43] showed large discrepancies in terms of
ignition delay time predictions near the present ignition thresholds (
∼
900
−
1000
K). Rows 1 and 2 in Fig. 5
show numerical temperature and velocity (magnitude) fields, along with velocity vectors revealing the flow
dynamics inside the combustion vessel. Row 3 presents corresponding OH and HO
2
mass fractions. Ignition
occurs at the glow plug top evidenced by an increase in temperature and OH and HO
2
mass fractions. At
t
=
t
ref
, a flame can be seen, propagating upward within the thermal plume. The maximum in velocity
remains inside the plume indicating weak expansion of the gas during the ignition/flame propagation event.
However, it does induce an appreciable horizontal velocity that disturbs the streamlines as shown at
t
=
t
ref
compared to
t
=
t
ref
−
12.5 ms. The flame remains anchored at the outer top edge of the glow plug
throughout the majority of the process. The temperature and velocity fields, and streamlines show the
8
t
=
t
ref
- 12.5 ms
t
=
t
ref
t
=
t
ref
+ 25 ms
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
t
=
t
ref
+ 50 ms
t
=
t
ref
+ 75 ms
t
=
t
ref
+ 100 ms
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
−
10
0
10
x
(mm)
0
10
20
30
40
50
z
(mm)
Figure 4: Side-by-side comparison of schlieren fields for a single-puff event in a 5% hydrogen-air mixture. Left: experimental; right:
numerical.
formation of a vortical structure at the edge of the flame, see for instance at
t
=
t
ref
+ 25 ms at
z
≈
20 mm and
x
≈
−10 mm, which leads to a localized concave deformation of the flame as observed in the experiments.
This is the result of the horizontal velocity induced by the ignition event and the vertical buoyancy flow
induced by the heating of the glow plug. This vortex gets advected upwards, and results in complete
detachment of the top portion of the flame at later times (not shown). The early stages of the vortex-flame
interaction can be observed at
t
ref
+ 25 ms
≤
t
≤
t
ref
+ 50 ms around
z
= 25 mm in Fig. 5. The evolution just
described is consistent with the experimental observations in Fig. 3; the experiment shows additional side
ignitions at
t
=
t
ref
+ 12.5 ms and
t
ref
+ 25 ms. While side ignitions do take place numerically (see Fig. 7)
they do not seem to be as strong or to occur at the same time as in the experimental fields shown in Fig. 3.
This discrepancy is very likely due to the non-uniformity in surface temperature in the experimental glow
plug alluded to earlier. In the surface temperature range at hand (
∼
1000
K), the experimental temperature
non-uniformity of 2.5 % quoted above (
∼
25
K) induces an appreciable change in ignition delay time that
make side ignitions more likely to occur experimentally than numerically [43].
Figure 6 shows OH and HO
2
mass fractions (row 1), and H mass fraction and hydrogen concentration
(row 2), at late times after the initial ignition transient, 1 s, 2 s and 3 s after ignition. A vertical plume of
9