Combustion and Flame 157 (2010) 421–435
Contents lists available at ScienceDirect
Combustion and Flame
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m b u s t fl a m e
Methane ignition catalyzed by in situ generated palladium nanoparticles
T. Shimizu a, A.D. Abid a, G. Poskrebyshev a, H. Wang a,*, J. Nabity b, J. Engel b, J. Yu b, D. Wickham c,
B. Van Devener d, S.L. Anderson d, S. Williams e
a
Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA 90089, United States
TDA Research, Inc., 12345 W. 52nd Ave, Wheat Ridge, CO 80033, United States
c
Reaction Systems, LLC, 19039 E. Plaza Drive, Suite 290, Parker, CO 80134, United States
d
Department of Chemistry, University of Utah, Salt Lake City, UT 84112, United States
e
Air Force Research Laboratory, Mail Stop RZA, 1950 Fifth Street, WPAFB, OH 45433, United States
b
a r t i c l e
i n f o
Article history:
Received 9 May 2009
Received in revised form 11 July 2009
Accepted 13 July 2009
Available online 10 August 2009
Keywords:
Catalytic combustion
Ignition
Gas-surface reaction
Kinetic modeling
Microscopy
Flow reactor
Particle size distribution
a b s t r a c t
Catalytic ignition of methane over the surfaces of freely-suspended and in situ generated palladium
nanoparticles was investigated experimentally and numerically. The experiments were conducted in a
laminar flow reactor. The palladium precursor was a compound (Pd(THD)2, THD: 2,2,6,6-tetramethyl3,5-heptanedione) dissolved in toluene and injected into the flow reactor as a fine aerosol, along with
a methane–oxygen–nitrogen mixture. For experimental conditions chosen in this study, non-catalytic,
homogeneous ignition was observed at a furnace temperature of 1123 K, whereas ignition of the same
mixture with the precursor was found to be 973 K. In situ production of Pd/PdO nanoparticles was confirmed by scanning mobility, transmission electron microscopy and X-ray photoelectron spectroscopy
analyses of particles collected at the reactor exit. The catalyst particle size distribution was log-normal.
Depending on the precursor loading, the median diameter ranged from 10 to 30 nm. The mechanism
behind catalytic ignition was examined using a combined gas-phase and gas-surface reaction model.
Simulation results match the experiments closely and suggest that palladium nanocatalyst significantly
shortens the ignition delay times of methane–air mixtures over a wide range of conditions.
Ó 2009 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
1. Introduction
Supersonic combustion has drawn attention in recent decades
with possible applications to hypersonic propulsion systems.
Hydrogen has long been used as a fuel for hypersonic systems
due to its short ignition delay and fast kinetics. However, its low
fuel density results in poor vehicle packaging. Conversely, liquid
kerosene based jet fuels package well but have additional complications with regard to fuel injection and droplet vaporization leading to reduced combustion performance. Furthermore, liquid jet
fuels tend to produce significant amounts of coke during endothermic cooling cycles. Thus, methane has gained interest for its potential to overcome most of these problems. It has greater heat sink
capacity and specific energy content than jet fuels and a much
higher energy density, if stored as a cryogenic liquid, than hydrogen [1]. However, the use of methane could lead to poor ignition
performance. The current work explores the question whether
ignition of methane can be enhanced catalytically by freely-suspended palladium (Pd) nanoparticles generated in situ from a suit* Corresponding author. Address: Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA 90089, United States. Fax: +1 213 740
8071.
E-mail address: [email protected] (H. Wang).
able fuel-soluble Pd precursor. Entrained in the mixture, palladium
particles are expected to provide catalytic surfaces to initiate gassurface reactions, leading to fuel ignition at lower temperatures
and/or with drastically reduced ignition delay times.
Catalytic oxidation of hydrogen and small hydrocarbons over
palladium has been studied quite intensely in recent years [2]. It
is known that the palladium exhibits complex hysteresis behavior
with respect to the rate of methane oxidation over its surface and
that the catalytic activity is influenced by the presence of water vapor [3] and can be sensitive to OH poisoning [4]. In general the catalytic reaction rate increases with an increase in temperature until
about 900 K, above which the reaction rate decreases. Above
1050 K, the catalytic reaction rate increases again [3]. This complex
behavior appears to be related to the variation of the bulk composition of palladium to be discussed below.
In an oxidizing environment, palladium oxide (PdO) is stable
thermodynamically for temperatures of relevance to catalytic oxidation. The equilibrium constant Kp of
2PdO ðsÞ ¼ 2Pd ðsÞ þ O2 ðgÞ
has been reported to be 2.2 103 atm at 909 K and 0.64 atm at
1124 K [5]. Hence the preferred thermodynamic state is palladium
oxide in air below 900 K, but it dissociates into Pd(s) and O2(g)
0010-2180/$ - see front matter Ó 2009 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2009.07.012
422
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
at temperatures above 1100 K. Kinetically, oxidation of the palladium surface can lead to penetration of oxygen atoms into the surface, creating local microstructures such as surface islands and
peninsulas [6]. Studies of oxygen adsorption and desorption on palladium nanoparticles supported on Fe3O4 show quite complex
behaviors with respect to reversible chemisorption and surface
oxide formation [7]. Hysteresis has also been observed for oxidation
and reduction of Pd supported on Al2O3 [8]. Computationally it was
shown that the hysteresis behavior is caused, at least in part, by the
competition between adsorption and desorption of oxygen from the
surface, coupled with the diffusion of oxygen into and from the
atomic layers below the Pd surface [9]. In general, the studies discussed above were conducted at temperatures near or below
1000 K. Only a few studies have been reported for Pd catalytic
kinetics above 1000 K. Notably, Rosen and coworkers [10,11] examined the reaction mechanism of hydrogen oxidation over a palladium surface in a stagnation flow reactor.
An added complexity is that the catalytic activity and mechanism on the surface of nanosized Pd particles can be different from
a perfect Pd(1 1 1) surface [12]. These complexities indeed pose
some significant challenges for initiating methane ignition with
in situ generated Pd/PdO nanoparticles, especially when one considers the limitation of catalyst loading. The fundamental difficulty
to achieve fuel ignition below 900 K is that the catalytic oxidation
of the fuel must be designed in such a way that the reaction process transitions from a catalytic process with mild oxidation rates
at low temperatures to a flame process driven by gas-phase freeradical chain branching above about 1400 K. Heat release from
the catalytic process must occur, at least locally, to allow the gas
to heat up and reach the cross-over temperature above which radical chain branching becomes possible. At the same time, the catalytic reaction and heat release cannot be too fast, since the flame, if
ignited, must have sufficient fuel left to sustain and propagate
itself.
Meanwhile, the Pd/PdO nanoparticles require a finite time to
nucleate out of its precursor. This time must be shorter than the
transit time through the combustor to allow the catalytic process
to occur. On the other hand, if the nucleation time is too short compared to the transit time, particle–particle coagulation and coalescence would lead to a reduced surface area per unit mass of
catalyst injected and a reduced catalytic activity. Additionally as
the Pd/PdO particles grow in size and mass, they must pass
through a cluster size regime where the catalytic mechanism is
poorly understood.
Previously methane oxidation over a Pd surface has been studied in flow reactors [13,14]. In general, these studies used wallcoated Pd to accomplish low temperature oxidation (up to
1100 K) with the purpose of minimizing NOx formation. No studies
have been reported on fuel ignition on Pd nanoparticles suspended
in unburned fuel-oxidizer mixtures.
The objective of this study is to understand the catalytic ignition
behavior of methane with Pd nanoparticles generated in situ. The
study combines flow reactor experiments, aerosol and surface sciences with kinetic modeling using detailed gas-phase and gas-surface reaction kinetics to explore the underlying mechanisms of
methane ignition over Pd nanoparticles. A combined gas-phase
and gas-surface reaction model is proposed on the basis of previous studies [15,16]. Combined with microscopy, particle mobility
sizing and other chemical analyses of the Pd particles, the model
was used to explain the observations made in the flow reactor. This
is not meant to be a quantitative kinetic modeling study. The use of
a laminar flow reactor indeed limits our ability to account for local
reaction conditions in a precise manner. Nonetheless the results
described here are useful for designing a catalytic reaction process
suited for high-speed combustion.
2. Experimental
Generation of palladium particles in situ and the subsequent
ignition procedure is drawn in Fig. 1 schematically. The sequence
of the nanoparticle formation process is described as follows. At
the reactor inlet section, a Pd precursor solution diluted in toluene
is introduced in the form of micron sized aerosol carried by nitrogen stream. This flow was mixed with a stream of premixed CH4/
O2/N2 mixture. Due to the high temperature and small droplet size,
the aerosol quickly evaporates in the inlet section of the reactor.
Dissociation of the precursor follows due to heating, releasing Pd
atoms or clusters. At high enough atom or cluster concentrations,
Pd nucleates into small clusters which grow in size and mass by
coagulation and surface condensation. The clusters or the surfaces
of the nanoparticles become catalytically active at some point. The
gas mixture is then expected to undergo catalytic oxidation and
eventually ignite in the flow reactor. Here, we follow the concentrations of CH4 and CO2 to monitor fuel ignition.
2.1. Setup
A schematic flow diagram of the experimental setup is presented in Fig. 2. A flow tube reactor was built in conjunction with
a detector for measurement of CH4 and CO2 concentrations at the
exit of the reactor. The particles are also sampled at the reactor
exit, enabling us to simultaneously measure the particle size distribution functions and catalytic activity during experiments.
The flow reactor is made of a quartz tube (1.9 cm OD, 1.7 cm ID,
and 94 cm in length). Although by visual inspection the tubes were
usually clean after each experiment, they were replaced after each
Furnace
Fuel/Oxidizer/
Catalyst Solution
Ignition
Exhaust
Vaporization
Nucleation
Coagulation/Agglomeration
CO2
CH4
Particle Size
Distributions/
Species Profiles
Pd(v)
Pd2
Pdi
Flow direction and reaction time
Pd nanoparticles
x, t
~ 20 nm
Fig. 1. Conceptual drawing of in situ generation of nanoparticles followed by catalytic ignition of a fuel-oxidizer mixture in a flow reactor.
423
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
To reactor
From liquid
pump
Liquid
Atomizer
Liquid Pool
N2
Liquid
Pump
Catalyst
Solution
PI-1
N2
N2/ O2/ CH4
Temperature
Controlled
Furnace
1.9cm OD×1.7cm ID×94cm L
Quartz Tube
Zone I
TC-1
SMPS System
Tf-1
TSI 3080 EC
L = 76cm
Kr
85
Zone II
n-DMA
TC-2
Tf-2
TSI 3025A
UCPC
ΔP
Vent
TC-3
Vent
N2
TEM Grid
NDIR CO2/CH4
Analyzer
Vent
Fig. 2. Schematic of the experimental apparatus.
experiment to minimize the influence of wall deposits. An
insulated furnace (Mellen SV12) with the test section length (L)
76 cm was used for heating. The furnace has two separate zones.
The temperature in each zone was measured using a thermocouple
(OMEGA K-type). The measured temperatures are denoted as Tf-1
(TC-1, Zone I) and Tf-2 (TC-2, Zone II), respectively. These
temperature values were acquired by the LabVIEWTM program
and used to control the furnace temperature. The temperature
profiles inside the reactor were measured using 122 cm long
K-type thermocouples (OMEGA) inserted from the reactor exit
(TC-3).
The gas mixtures were prepared by mixing CH4, N2, and O2 from
cylinders. The flow rates were controlled by mass flow controllers
(Porter Model 201) calibrated for each of the gases. The mixture
composition tested and the total gas velocity U0 selected are summarized in Table 1. The residence time in the reactor was about 1 s
at the temperature of reactor operation. The reactor exhaust was
diluted eight times by a cold stream of N2. The flow rate was controlled by a mass flow controller (Porter Model 202). The concentrations of CH4 and CO2 at the exit of the reactor were
continuously monitored using an NDIR gas analyzer (California
Analytical Instruments Model 200), which withdraws 2 L/min of
the diluted exhaust gas.
The precursor atomizer was designed to produce micron sized
droplets suspended in a N2 flow. The atomizer utilized a dual, concentric nozzle design, as shown schematically in Fig. 2, whereby a
compressed N2 flow passes through the inner nozzle (2 mm ID
glass tubing with a tip diameter of 0.5 mm). Under this condition,
the Venturi effect draws the precursor solution from a liquid pool
through the outer nozzle (4 mm ID glass tubing with a tip diameter
of 1 mm), forming a stream of aerosol.
The atomizer outlet used 6.53 mm ID tubing and an elbow to
form a series of impactors, thus eliminating the passage of large
droplets into the reactor. The large droplets collected onto the impactors were returned to the liquid pool. The aerosol production rate
depends on the liquid pool level, the flow rate of N2, and the position of the impactors. The N2 flow rate through the atomizer was
controlled manually with a needle valve and an upstream pressure
indicator (PI-1). The liquid pool was replenished with a single piston syringe pump (Chrom Tech ISO-1000), resulting in steady operation of the atomizer. The atomizer was cooled and insulated to
avoid evaporation of the droplets collected on the inner wall,
which would precipitate and crystallize the precursor onto the
wall.
Pd(THD)2 (THD: 2,2,6,6-tetramethyl-3,5-heptanedione) (molecular weight 475 g/mol, sublimation temperature 150 °C at
Table 1
Summary of flow and mixture conditions.
No.
F-0
F-1
F-2
F-3
a
Total gas velocity (STP), U0 (cm/s)
52.9
52.9
52.5
52.5
Mole fraction
C7H8
CH4
O2
N2
–
–
0.007
0.007
–
0.083
0.084
–
–
0.417
0.420
0.420
1.000
0.500
0.490
0.573
Calculation of the equivalence ratio includes toluene.
Equivalence ratio, /a
Purpose
0.00
0.40
0.55
0.15
Background temperature
Non-catalytic ignition
Catalytic ignition
TEM analysis
424
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
0.01 Torr and melting point 239 °C [17]) was synthesized at TDA
and used as the Pd precursor. Pd(THD)2 is a coordination complex,
and as such the binding energy between Pd and THD is small. Upon
heating, it dissociates readily, releasing a Pd atom. The precursor
was dissolved in toluene (J.T. Baker 99.7% purity). The concentration of Pd(THD)2 in toluene solution was less than 1.9 wt% in the
experiments.
The size distributions of Pd nanoparticles were measured at the
reactor exit using a scanning mobility particle sizer (TSI SMPS
3090). A nominal flow of 0.3 L/min of the diluted exhaust was sampled into the SMPS. The residence time from the reactor exit to the
SMPS inlet was 0.6 s, which is short enough to prevent coagulation
of particles. The SMPS is composed of an electrostatic classifier
(Model 3080) and a particle counter (Model 3025A). The classifier
was equipped with a nano differential mobility analyzer (Model
3085). The diameter of the impactor nozzle was 0.0457 cm. The
particle size distribution was scanned over the range of 4.5–
160 nm in diameter. Each scan time took 55 s.
The particles were also collected on grids (Electron Microscopy
Science HC200-CU) and analyzed by Transmission Electron Microscopy (TEM, Philips EM420). Each grid, affixed on top of an aluminum rod (0.6 cm in diameter) with aluminum tape, was exposed
to a diluted exhaust stream flow parallel to the grid surface. The
linear flow rate of the aerosol was 13 m/s over the grid. Under this
condition, the flow is turbulent and the principal mechanism of
particle collection is diffusion of the particles across the laminar
boundary layer near the surface, which may bias against large particles. High-resolution TEM, STEM/EDX, and X-ray photoelectron
spectroscopy (XPS) analyses of particle structure and chemical
composition were carried out for particles from selected experiments. The details of the particle characterization are described
elsewhere [18].
2.2. Procedure
Two different types of tests were performed as shown in Table
2. In the first type (Runs 3 and 4), the concentration of Pd(THD)2
was kept constant at 1.92 wt% in the atomizer, corresponding to
4.5 104 moles Pd per mole CH4 in the main reactor flow. The
furnace temperature was transient and increased at a constant rate
of 20 K/min. This allowed us to determine the furnace temperature
at the point of ignition with a specified catalyst condition. In the
second type the furnace temperature was held constant, while
the Pd(THD)2 concentration was transient and raised gradually,
from 0 to 1.92 wt% until ignition was observed (Run 7). This allowed for determination of the minimum catalyst concentration
for ignition at a specified furnace temperature. Additional experiments were carried out to measure baseline methane ignition under non-catalytic conditions (Runs 1, 2 and 6).
A key experimental challenge was to minimize the effect of wall
deposits. The effect of the Pd deposited on the reactor wall was
carefully investigated in Run 5. The test was carried out without
the use of a Pd precursor but with a quartz tube repetitively used
in prior catalytic experiments.
In addition, Runs 8 and 9 were designed to produce Pd/PdO
nanoparticles for TEM sampling. To avoid fuel ignition, which results in large changes in gas temperature and concentration, methane was omitted from the O2/N2 flows, and their combined flow
rate was equal to those of catalytic experiments. The mass loading
of the Pd(THD)2 in toluene was 1.2 wt%, which is about the same as
that in Run 7.
3. Detailed kinetic modeling
3.1. Mathematical descriptions
The underlying reacting flow problem was simulated as an initial value problem. SURFACE CHEMKIN [19] was utilized to solve
the system of equations and to calculate the chemical reaction
rates for both gas-phase and gas-surface processes. The model is
capable of simulating the evolution of temperature and species
along the flow axis. Simulations were performed at 1 atm pressure
with and without a catalyst. Convective time t was obtained from
the local linear gas velocity u and the axial position x of the flow
Rx
reactor by the relationship, t ¼ 0 dx=u. Since the flow reactor
operates under the laminar flow condition, the flow is inherently
two-dimensional; and the velocity profile is parabolic. Nonetheless, the current analysis assumes this to be a plug flow problem
and neglects the radial transport – a simplification in comparison
to the actual experiments. Sensitivity analyses, experimentally
and computationally, were performed to examine this model
assumption.
The gas-surface reaction rate constant is described as
sffiffiffiffiffiffiffiffiffiffiffi
8kB T 2
k¼c
pr
pmi d
ð1Þ
where c is the reaction probability or sticking coefficient, kB the
Boltzmann constant, mi the molecular mass of ith gas-phase species,
and rd the particle radius. The species conservation is written as
dY i
M
_ s;i Þ w;i
_ g;i þ fx
¼ ðx
dt
q
ð2Þ
Table 2
Summary of experimental condition and key results.
No.
Experimental
type
Flow condition
of Table 1
Furnace temperature
Tf-1 (K)
Tf-2 (K)
Transient furnace temperature at +20 K/min heating rate, constant precursor injection rate
Run 1
Non-catalytic
F-1
1149
1077
Run 2
Non-catalytic
F-1
1142
1073
Run 3
Catalytic
F-2
998
970
Run 4
Catalytic
F-2
954
916
b
F-1
1179
1117
Run 5
Non-catalytic
Constant furnace temperature, increased precursor loading over time
Run 6
Non-catalytic
F-1
1113
Run 7
Catalytic
F-2
973
Controlled particle formation for TEM analysis
Run 8
No ignition
F-3
Run 9
No ignition
F-3
a
b
c
773
973
Measured at the reactor exit.
The reactor tube used was contaminated by wall coating.
Molar loading, particle diameter and surface area at the point of ignition.
Moles Pd/mole CH4
1113
973
773
973
4.3 104
4.7 104
2.1 104,c
Median diameter,
hDpi (nm)
a
Surface-to-volume
ratio,a f (cm1)
–
–
24
28
–
–
0.023
0.027
–
–
19c
–
0.013c
26
23
0.016
0.016
425
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
where C is the catalytic surface site density.
to C = 1.95 109 mol/cm2, i.e., the average of the literature values
for the Pd/PdO phases [9,15,30–32].
The sticking coefficient of O2 (R1f) and the surface coverage (hO)
dependent energy of desorption (R1b) are based on the temperature-programmed desorption (TPD) experiments [33]. These rate
parameters values were used also by Wolf et al. [9] and Jackson
and coworkers [30,31]. The rate parameters of H2O adsorption
and desorption (R2f and R2b) were adopted from a reaction model
by Deutschmann et al. [32], who studied catalytic ignition of
hydrogen over a Pd catalyst in a stagnation flow burner. Cao and
Chen [34] reported a Density Functional Theory (DFT) calculation
for water adsorption on Pd(1 1 1). The critical energies adopted
in our model were taken from their DFT study.
The reactions and their rate parameters of R6–12 were taken
mainly from Sidwell et al. [15]. The activation energy of dissociative adsorption of methane on oxide surface sites (R11) was reduced from 38 to 20 kJ/mol. This downward revision is
supported by the theoretical result of Broclawik et al. [35,36].
Although a thermodynamically consistent approach [37] may be
employed for assigning the surface reaction rate parameters, this
was not attempted here because the exact mechanism for the surface reaction of CH4 remains quite uncertain.
3.2. Chemical kinetic models
4. Results and discussion
The gas-phase reaction model employs USC Mech II [16]. The
reaction model was developed for H2/CO/C1–C4 hydrocarbon combustion through a series of kinetic modeling studies over the last
decade [16,20–29]. The model is composed of 784 reactions and
111 species, and has been validated over a variety of experiments
including ignition delays, laminar flame speeds, and species profiles under the conditions relevant to the present flow reactor
experiments.
The surface reaction mechanism was largely derived from the
work of Jackson and coworkers [9,30,31]. The resulting model is
shown in Table 3. It consists of five reversible and seven irreversible reactions with nine surface species. The total surface site density C available for the catalytic reactions on a Pd particle was set
4.1. Reactor temperature profiles
where Y is the mass fraction, Mw the molecular weight, q the gas
mass density, f the surface area-to-gas volume ratio of the catalyst,
x_ g and x_ s are the gas-phase and gas-surface molar production
rates, respectively.
The rate of temperature change in the reactor is obtained by
considering the heat release due to chemical reactions in addition
to the convective heating/cooling
dT dT b
1 X
_ g;i þ fx
_ s;i Þhi Mw;i
ðx
¼
dt
dt
qcp i
ð3Þ
where h is the specific enthalpy, cp the specific heat capacity of the
gas mixture. The first term on the right-hand side of the equation
expresses the experimental temperature gradient obtained from a
direct measurement for a non-reacting flow. As will be shown later,
the particles formed are substantially smaller than the mean free
path of the gas. Hence the particles are treated as being in free molecule regime. The time evolution of the site fraction hi for the ith
surface is given as,
_ s;i
dhi x
¼
dt
C
ð4Þ
Under the current experimental condition the flow inside of the
reactor is laminar (568 P Re P 236). The resulting parabolic velocity profile creates a radial gradient in the gas temperature. In the
axial direction, a temperature gradient also exists due to the finite
time needed to equilibrate the gas flow with the wall temperature.
For this reason, the gas temperature was measured both at the centerline Tb,center and adjacent to the wall Tb,wall, in a pure N2 flow at a
rate equal to the ignition tests (see, mixture F-0 of Table 1). The results are presented in Figs. 3 and 4, respectively. The uncertainty
for the temperature measurement is estimated to be ±15 K. As
seen, the gas in the reactor undergoes heating in the first two-
Table 3
Surface reaction model.
No.
1f
1b
2f
2b
3f
3b
4f
4b
5f
5b
6
7
8
9
10
11
12
a
b
c
d
e
f
g
h
Reactionb
O2 + 2Pd(S) ? 2O(S)
2O(S) ? O2 + 2Pd(S)
H2O + Pd(S) ? H2O(S)
H2O(S) ? H2O + Pd(S)
H(S) + O(S) ? OH(S) + Pd(S)
OH(S) + Pd(S) ? H(S) + O(S)
H(S) + OH(S) ? H2O(S) + Pd(S)
H2O(S) + Pd(S) ? H(S) + OH(S)
2OH(S) ? H2O(S) + O(S)
H2O(S) + O(S) ? 2OH(S)
CO2(S) ? CO2 + Pd(S)
CO(S) + O(S) ? CO2 + Pd(S)
C(S) + O(S) ? CO(S) + Pd(S)
CH4 + 2Pd(S) ? CH3(S) + H(S)
CH3(S) + 3Pd(S) ? C(S) + 3H(S)f
CH4 + Pd(S) + O(S) ? CH3(S) + OH(S)
CH3(S) + 3O(S) ? C(S) + 3OH(S)h
Rate parametersa
Reference/comments
A
b
0.8c
5.13 1021d
0.75c
1.00 1013
5.13 1021d
5.13 1021d
5.13 1021d
5.13 1021d
5.13 1021d
5.13 1021d
5.00 1010
5.13 1021d
5.13 1021d
4.00 105c
5.13 1021d
4.20 102c
5.13 1021d
0.5
E
230–115hO
44
94.5
113.8
31.1
88.8
14.5
32.8
29
76
62.8
196c
85.1
20c
25.1
Arrhenius parameters for the rate constants written in k = ATb exp (E/RT). The units of A are given in terms of mol, cm3/s. E is in kJ/mol.
The surface coverage is specified as the site fraction h. Total site density of Pd is C = 1.95 109 mol/cm2.
Sticking coefficient.
The frequency factor was estimated from vibration frequency (1013 s1) and scaled by the total site density C.
The activation energies are based on the DFT results of Cao and Chen [34].
The reaction order is 1 for Pd(S).
The sticking coefficient was taken from Sidwell et al. [15]. The activation energy was deduced from Broclawik et al. [35,36].
The reaction order is 1 for O(s).
[9,30,31,33]
[9,30,31,33]
[32]
[32]
e
e
e
e
e
e
[15]
[15]
[15]
[15]
[15]
g
[15]
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
1200
150
1273 K
1173 K
1000
873 K
Tg,max (K)
1073 K
773 K
Tf
Centerline temperature, Tb,center (K)
426
973 K
800
600
Centerline
100
50
Reactor Wall
Tf =673 K
400
0
0
10
20
30
40
50
60
600
70
700
800
Distance from reactor inlet, x (cm)
Wall temperature, Tb,wall (K)
1000
1100
1200
1300
Furnace temperature, Tf (K)
Fig. 3. Background, centerline temperature Tb,center measured in an N2 flow (F-0 in
Table 1) at several furnace temperatures Tf. Symbols are experimental data and
solid lines are fitted to data. The dashed lines were obtained from extrapolating the
data from the measured data points for Tf = 1073 K and 1273 K.
1200
900
Fig. 5. Difference between the furnace temperature (Tf) and the peak gas temperature Tg,max as a function of the furnace temperature Tf.
flow problem. Because of the computational difficulties involved,
this was not attempted in this study. Rather we used both the centerline and near-wall temperatures as the two limiting cases for
one-dimensional numerical simulation.
1273 K
1173 K
1000
4.2. Catalytic ignition
1073 K
973 K
800
873 K
773 K
600
Tf =673 K
400
0
10
20
30
40
50
60
70
Distance from reactor inlet, x (cm)
We first examine the test cases in which the furnace temperature increased at 20 K/min (Runs 1–4). The furnace temperatures,
Tf-1 and Tf-2, at the point of ignition are recorded and shown in Table 2. The experiments were found to be reproducible (cf. Runs 1
and 2). The variations of CH4 and CO2 concentrations are presented
as a function of the furnace temperature in Fig. 6. It is seen that at
the point of ignition the CH4 mole fraction drops quite abruptly
while that of CO2 rises accordingly. For the catalytic case, the rise
of the CO2 mole fraction prior to the disappearance of CH4 was
caused by oxidation of toluene. This oxidation process is seen to
be gradual and non-explosive. Experimentally, we found that under noncatalytic conditions toluene itself did not affect methane
ignition.
Fig. 4. Background, wall temperature profiles Tb,wall measured in a N2 flow (F-0 in
Table 1) at several furnace temperatures Tf. Symbols are data and solid lines are
fitted to data. The dashed lines were obtained from extrapolating the data from
measured data points for Tf = 1173 K and 1273 K.
0.10
Non-Catalytic
Run 1
Mole fraction
0.08
thirds of the reactor and cooling in the remaining section. As expected, the centerline and wall temperature profiles can differ
from each other quite significantly, by as much as 200 K. Along
the centerline the temperature rises more gradually than that close
to the wall; and it reaches its peak value further downstream than
the temperature adjacent to the wall. The delay in temperature rise
is certainly caused by the finite heating rate under the laminar flow
condition. In comparison, Fig. 4 shows that heating along the inner
wall is more rapid and reaches the highest temperatures closer to
the inlet than along the centerline. The difference between the furnace temperature Tf and the peak gas temperature Tg,max is plotted
as a function of Tf in Fig. 5. The maximum centerline temperature
Tcenter,max never reached the actual furnace temperature because of
cooling towards the end of the reactor.
The inhomogeneous temperature makes it difficult to interpret
the data quantitatively. A realistic modeling approach would have
to simulate the problem as a two-dimensional, laminar reacting
CH4
0.06
0.04
0.02
CO2
Catalytic
Run 4
Catalytic
Run 3
0.00
900
1000
1100
Furnace temperature, Tf-1 (K)
Fig. 6. CH4 and CO2 mole fractions observed for Runs 1 and 3 as a function of the
furnace temperature of Zone I, Tf-1.
427
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
The results demonstrate that methane ignites noncatalytically
at around 1100 K furnace temperature, corresponding to a maximum gas temperature of roughly 950 K along the centerline of
the reactor. The use of catalyst generated in situ leads to a reduction of the furnace temperature by about 150 K at the point of ignition, using a Pd loading of 0.025 cm1 (catalyst surface area to
total gas volume), as shown in Table 2. Under this condition, the
atomic Pd to CH4 ratio is around 4.5 104, corresponding to
about 3 mg Pd/(g CH4).
Runs 3 and 4 were designed to test the operation of the atomizer. In Run 3 the atomizer was covered by ice packs, reducing its
temperature to 278 K. For Run 4 the atomizer was operated at
a somewhat higher temperature (283 K). This small difference in
temperature was sufficient to change the loading of aerosol by
10%. The impact was captured by the experiment. With a higher
catalyst loading (by about 10%), the furnace temperature at ignition was reduced by about 50 K, as shown in Table 2 and Fig. 6.
4.3. Particle size distribution
The particle size distribution functions (PSDFs) were determined at the exit of the reactor. Fig. 7 presents several PSDFs obtained during Run 3. These samples were collected with a
dilution ratio of 10 and an aerosol transmission time of 0.6 s.
Since the number density of the particles is of the order of
108 cm3 at the exit of the reactor, this density drops to 107 cm3
after dilution. Assuming the coagulation rate constant to be the
collision limit (109 cm3/s), we find the characteristic coagulation
time to be 1/(107 109) 100 s. Hence, little to no coagulation
losses are expected for the aerosol sampled.
As seen the median diameter of the particles falls between 20
and 30 nm. Without injecting the Pd(THD)2 precursor solution as
an aerosol, the same measurement yields only a trace of particles
with a number density similar to the ambient background.
Although the particle diameters are mobility diameters, they
should be very close to the true diameters for particles in the observed size range [38–40]. Over the period of 20 min during which
the furnace temperature was increased, the PSDFs remained almost identical. This indicates that the precursor injection rate
was steady, and that the processes of aerosol evaporation and precursor decomposition followed by particle nucleation and growth
were independent of the temperature for Tf > 800 K. It also suggests that these processes were complete or almost complete near
the inlet region of the reactor and at temperatures substantially
below the peak temperature of the gas.
All of the PSDFs observed are roughly log-normal with median
diameter as expected for particle size growth dominated by particle–particle coagulation. Additionally, the tail towards the small
size end of the distribution was also expected, and is indicative
of persistent particle nucleation, perhaps due to the residue Pd vapor in the flow tube. The small shoulder towards the large particle
size is probably indicative of the presence of aggregates. This
behavior in PSDFs was commonly seen in homogeneous soot
nucleation and growth [41–48], and is not indicative of particles
produced from aerosol drying or breakup. Given the mass fraction
of Pd(THD)2 in toluene and the initial aerosol droplet size, the particles would be about 100 nm in diameter if they were produced
purely from toluene evaporation. If the particles were produced
from a combination of aerosol drying and breakup, the size distribution would be significantly broader than observed.
The homogeneous nucleation mechanism for particle formation
is also supported by second order of coagulation kinetics. If we
start with Pd(v) as a monomer at a concentration of
N0 = 2 1014 cm3 , uniformly dispersed in the gaseous reactant
mixture, (calculated from the initial Pd(THD)2 loading for Runs 3
and 4) and assuming a unity sticking probability and a bimolecular
coagulation rate constant of b = 1 109 cm3/s, we estimate that
after 1 s, the number density of Pd particle is
N
N0
109 cm3
1 þ bN 0 t
ð5Þ
This is consistent with the observation shown in Fig. 8. In addition, the particle diameter may be estimated from the number of
atoms in each particle (2 105) and an assumed mass density
1x109
1050
(K
)
1000
850
r, D
p
(nm
)
100
800
m
pe
ra
le
Dia
me
te
te
900
10
-
e
Pa
rtic
Tf
tu
re
,
950
1
0
Fu
rn
ac
(#/c
g(D p)
dN/dlo
3
m)
2x109
Fig. 7. Selected particle size distributions observed for Run 3 at several furnace temperatures.
dN/dlogDp (#/cm3)
1010
Tf-1 = 911 K
Nucleation mode
N1 = 5.8x109 (cm-3)
σ1 = 1.34
<Dp>1 = 2.3 nm
109
Growth Mode
N2 = 7.5x108 (cm-3)
σ2 = 1.53
<Dp>2 = 23 nm
108
Aggregates
N3 = 4.2x106 (cm-3)
107
<Dp>3 = 98 nm
0.07
35
0.06
Run 4
30
0.05
25
0.04
Run 3
20
0.03
15
0.02
10
5
0
σ3 = 1.36
10
40
Ignition
Run 4
700
800
900
Ignition
Run 3
1000
1100
0.01
Area-to-volume ratio, ζ (cm-1)
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
Median diameter, <Dp> (nm)
428
0.00
Furnace temperature, Tf-1 (K)
100
Particle diameter, Dp (nm)
Fig. 8. Representative particle size distributions (symbols) measured during Run 3.
Lines are fitted into a tri-log-normal function whereas thinner lines indicate the
three separate particle size modes.
of 10 g/cm3. This gives a mean diameter of 20 nm, which is close to
the experimental observation (25 nm). Of course, the initial nonuniform distribution of the aerosol could cause the local Pd(v) concentration to exceed the initial concentration of 2 1014 cm3.
This would result in an effectively shortened coagulation time for
particle production.
The measured PSDFs can be fitted into a tri-log-normal
distribution,
(
)
3
X
dN
Nk
½logðDp =hDp ik Þ2
pffiffiffiffiffiffiffi
¼
exp d log Dp k¼1 2p log rk
2ðlog rk Þ2
ð6Þ
where Dp is the particle diameter, Nk, hDp ik and rk are the number
density, the median diameter, and the geometric standard deviation
of the kth size mode, respectively. Fig. 8 presents a sample of the fit
to data taken from Run 3. Aside from the fact that the PSDF parameters are roughly invariant with respect to the furnace temperature
(especially the dominant, second mode), the most important feature
here is that r2 = 1.53, which is slightly larger but close to that of a
self-preserved distribution function of 1.46 [49]. Hence, the particles measured are conclusively those from nucleation and growth
out of the gas-phase precursors.
Fig. 9. Particle median diameter hDpi and area-to-volume ratio f (STP) measured at
the exit of the flow reactor during Run 3 and Run 4, as a function of the furnace
temperature of Zone I, Tf-1. Squares and circles represent Run 3 and Run 4,
respectively.
As we discussed earlier, the PSDFs are not sensitive to the temperature for Tf > 800 K. Below that temperature, however, this is
not the case. Fig. 9 shows that below 800 K the median particle
diameter hDpi rises to 40 nm in some cases, whereas the surface
area-to-volume ratio f (integrated from the distribution function)
remains relatively constant. The rise in the median diameter at
low temperatures was probably caused by coke formation or carbon deposition, since the aerosol was injected into the reactor in
a nitrogen stream and locally the precursor flow was extremely
fuel rich for a period of time. Meanwhile, toluene undergoes pyrolysis either homogeneously or catalyzed by dissociation of the
Pd(THD)2 and Pd/PdO clusters formed thereafter. Regardless, the
observed variations in the median size below 800 K do not affect
our interpretation of the experiment, since ignition was observed
above that temperature.
4.4. Particle composition
To confirm that the particles observed are palladium and to
examine the fine structure of these particles, we carried out TEM
analysis on particles collected in Runs 8 and 9, in which the furnace
temperature was held constant and equal to 773 and 973 K,
respectively. Fig. 10 shows the TEM images from the two runs.
Fig. 10. TEM images of nanoparticles sampled in Run 8 (left) and Run 9 (right). The furnace temperature Tf was kept at Tf = 773 K and 973 K, respectively. The images for the
two runs look alike with particles of nearly identical size and dispersion.
429
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
Fig. 11. HRTEM images of nanoparticles sampled in Run 8 (left) and Run 9 (right). The furnace temperature Tf was kept at Tf = 773 K and 973 K, respectively. Most particles
were found to be rich in Pd metal.
ratio of the reactor. Hence, our effort was directed at minimizing
the effect of wall catalysis.
The precursor atomizer was designed specifically for that purpose. Because of the use of multiple impactors and hence the relatively small Stokes’ number, the aerosol follows the flow closely;
and the droplets have little possibility to impinge onto the reactor
wall before they disintegrate. In addition, the unburned reactant
gas (CH4/O2/N2) acts as a sheath surrounding the aerosol injection
tube at the inlet of the reactor. This design further prevents the
aerosol from reaching the reactor wall, at least before coming in
contact with the unburned gas. Finally, as nanoparticles begin to
form in the gas flow, the fact that the wall temperatures are higher
than the centerline temperature means that thermophoretic effects tend to drive particles away from the walls. Under ignition
conditions, heat release in the gas-phase may reverse this tendency
to some extent, but because the experiments always were scanned
from non-reactive conditions toward ignition, thermophoresis
should help prevent catalyst deposition up to the ignition point,
and thus the ignition temperatures or concentrations should not
be strongly affected by wall catalysis. Nonetheless, there still remains a minor possibility that the actual ignition is induced by catalytic reaction on the reactor wall.
0.10
CH4
0.08
Mole fraction
Both TEM grids were exposed to the exhaust flow for 6 min. The
two images were almost identical; an observation consistent with
the mobility size measurement. Most of the particles are smaller
than the median diameter hDpi observed by SMPS. This could be
due to the fact that the current particle collection mechanism relies
on the particle diffusion towards the grid. The collection efficiency
would be, therefore, biased against larger particles. More detailed
particle structure and composition were probed by high resolution
TEM and XPS, as described in a separate paper [18]. Only key results are summarized here. Fig. 11 shows HRTEM images collected
for particles representative of the samples. The fact that the images
show lattice spacings that clearly correspond to the Pd(1 1 1) interplane spacing indicates that bulk of the particles is crystalline,
metallic Pd. XPS analyses, however, show that an oxide coating
several atomic layers thick is present on the surface of the particles
collected in the 973 K experiment. In the 773 K experiment, the
particles have nearly identical bulk crystalline Pd structure, but almost no surface oxide [18]. This observation suggests that oxide
growth is kinetically controlled, and that formation of the oxide
may be a critical step in ignition, which occurs at 973 K, but not
773 K. This observation is consistent with the literature on oxide
formation for bulk Pd surfaces over much longer residence time
[8,9], and also with the observation that PdO is more catalytically
active than Pd [2]. An appreciable amount of SiO2 or partially oxidized silicon was also detected in the particle sample [18]. Control
experiments show that the source of the silicon contamination was
the toluene solvent used to dissolve the Pd(THD)2 precursor. Because silica is a stable, chemically inert oxide, its participation directly in the catalysis process is quite unlikely. This is also because
the Pd particles are relatively large. The presence of this silica contamination should not have much effect on the catalytic chemistry
of the Pd particles. Indeed, noncatalytic runs with the toluene solvent only show the ignition temperature to be nearly identical to
that without toluene injection.
0.06
0.04
0.02
CO2
4.5. Possible wall effect
SMPS, TEM and XPS evidence all points to a mechanism of ignition induced by the catalytic activity of the Pd/PdO nanoparticles
produced in situ from the decomposition of the Pd(THD)2/toluene
aerosol. An issue that remains to be addressed is the role of condensed matter deposition on the wall and the subsequent wall
catalysis. As we discussed earlier, it is nearly impossible to eliminate wall catalysis completely due to the finite surface-to-volume
0.00
700
800
900
1000
1100
Furnace temperature, Tf-1 (K)
1200
Fig. 12. CH4 and CO2 mole fractions measured during Run 5 as a function of the
furnace temperature of Zone I, Tf-1. The wall of the reactor tube used for Run 5 was
contaminated by wall coating of condensed-phase material from Pd(THD)2
decomposition.
430
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
5. . . .. . .,
6. Pdi + Pdj ? Pdi+j.
To examine the influence of catalysis on the reactor wall independently, an experiment (Run 5) was carried out without injecting the Pd(THD)2 precursor but using a quartz tube with a visible
coating of catalyst from a prior experiment. Fig. 12 shows the
CO2 and CH4 profiles as a function of the furnace temperature. It
is seen that the ‘‘contaminated” tube caused the methane mole
fraction to drop, starting at around Tf-1 = 730 K. It reached a plateau
at around Tf-1 = 1000 K. Interestingly, the ignition occurs at
Tf-1 1180 K (Table 2), about 30 K higher than experiments conducted in clean, unused tubes (Runs 1 and 2).
We interpret the above observation as follows. In Run 5, wall
catalysis indeed occurs, leading to the formation of the plateaus
in the concentration profiles. Ignition at a furnace temperature
1180 K is caused by the homogeneous process. A higher ignition
temperature is caused by a reduced methane concentration due to
catalytic conversion on the wall. Hence, it is clear that in this case
wall catalysis actually retarded the ignition process. More importantly, the catalytic ignition studies in Runs 3 and 4 (Fig. 6) show
constant CH4 concentration over the entire temperature range up
to the point of ignition. The absence of any significant drop in
CH4 concentration prior to ignition indicates that wall catalysis
was indeed suppressed.
This coalescence process leads to the formation of Pd nanoparticles, as observed in Figs. 7–11. The surface of the Pd particles becomes oxidized, especially for temperatures above 973 K, leading
to enhanced catalytic activity, which ignites the gas mixture at reduced temperature.
What remains unclear is why the oxide does not form at an
early stage of cluster and particle formation. The reason may be
thermodynamic and/or kinetic in nature. It is possible that for
small particles the formation of palladium oxides is not favorable
thermodynamically. A more plausible cause for the lack of bulk
PdO formation is the finite time required to mix the reactants
and diffuse oxygen to the surface. As the aerosol is injected into
the reactor and undergoes evaporation and decomposition, this
flow stream is not yet mixed locally with oxygen in the sheath.
Hence the initial particle nucleation process takes place in spatial
regions depleted of oxygen. If oxygen is present locally, it would
have been depleted by toluene decomposition and oxidation, as
evidenced by the early rise of CO2 in Fig. 6. The burst of nucleation
followed by coagulation causes the Pd(v) to condense onto existing
particle surface rapidly, before the particles have an opportunity to
mix by diffusion with the outer sheath of oxygen. The result is a
particle with a Pd core that oxidizes slowly at low temperatures,
and forms a real oxide layer only near or above 973 K, as confirmed by TEM and XPS results. Although additional work at the
elementary level is needed to confirm the above analysis, we are
now in a position to advance a chemical model to explain the ignition phenomenon observed.
4.6. Mechanism of particle formation and catalytic activity
Summarizing the above results, we may now advance a phenomenological description of the particle formation mechanism
and its catalytic activities. The kinetic processes of catalyst particle
formation may be described by the following processes:
1. 1–10 lm-diameter toluene droplets containing Pd(THD)2
vaporize, and the Pd(THD)2 decomposes, generating Pd vapor
or Pdi clusters,
2. Pd(v) + Pd(v) ? Pd2,
3. Pd(v) + Pd2 ? Pd3,
4. Pd2 + Pd2 ? Pd4,
4.7. Detailed kinetic modeling
To facilitate model comparison, Runs 6 and 7 were made, in
which the furnace temperature was held constant. These runs were
necessary because we were able to measure the temperature pro-
dN/d logDp (#/cm3)
2x109
Ignition
1x109
0
20
15
s)
e (m
Tim
10
5
100
0
10
<D p >
diameter,
Particle
(nm)
Fig. 13. Selected particle size distribution functions at various testing times measured at the exit of the flow reactor during Run 7. Furnace temperature Tf was kept at 973 K
while the catalyst concentration was gradually increased, leading to an increase in the surface-to-volume ratio f. The mixture ignited 20 min after the start of the run.
431
Tf = 973 K
20
0.02
15
0.01
10
Ignition
0
5
10
15
Area-to-volume ratio, ζ (cm-1)
Median diameter, <Dp> (nm)
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
0.00
25
20
Test time (min)
Fig. 14. Particle median diameter hDpi and area-to-volume ratio f measured at the
exit of the flow reactor during Run 7. Round and rectangular symbols represent hDpi
and f (STP), respectively.
Mole Fraction (PPM)
files along the reactor only under the condition of steady reactor
operation. Under the condition of transient furnace temperature,
the rise in the gas temperature lags further behind as compared
to the temperature profiles obtained under steady operating condition and shown in Figs. 3 and 4. For this reason, while the results of
Runs 3 and 4 are useful for examining the response of methane
ignition to furnace temperature, they cannot be modeled with
certainty.
For background, noncatalytic ignition of methane (Run 6), we
increase the furnace temperature in 5 K steps. The reactor was allowed to equilibrate for 30 min at each furnace temperature, until
20000
15000
CH4
CH4
CH4
10000
5000
CO
CO
CO
CO2
CO2
0
Mole Fraction (PPM)
ignition is observed. For the mixture composition given as F-1 in
Table 1, the ignition was observed at Tf = 1113 K with an uncertainty of ±5 K. This temperature is about 35 K lower than that of
Runs 1 and 2, which is reasonable considering that the furnace
temperature is transient in those runs. In run 7, the furnace temperature was held fixed at 973 K. The Pd(THD)2 loading was increased gradually while the Pd particle size distribution was
measured continuously. Fig. 13 shows selected PSDFs data at various times over the course of the run. The median diameter observed continued to increase, as seen in Fig. 14, while the total
number density remained nearly the same over the 20-min experiment period. When the catalytic loading reached 1.3 102 cm1,
i.e., the highest loading shown in Fig. 13, ignition occurred. At that
point, the atomic Pd-to-CH4 ratio was 2 104 and the surface
area-to-gas volume ratio f was 0.013 cm1, as shown in Table 2.
As before, the PSDFs are trimodal and the main mode is log-normal. This set of the experiments was reproducible, and was used
as the basis of modeling studies in what follows.
To verify the accuracy of USC Mech II for gas-phase methane
oxidation under conditions comparable to the current flow reactor
experiments, we first report the simulation results and compare
them to the turbulent flow reactor data of Hunter et al. [50]. In
those experiments, methane was highly diluted and oxidized in
an air-like mixture at pressures of 6 and 10 atm. The temperature
along the reactor axis was nearly constant and ranged from 950 to
975 K. As shown in Fig. 15, major and minor species concentrations
were predicted by USC Mech II accurately.
Additional computational cases are summarized in Table 4.
Simu-1 and -2 correspond to Run 6 and Simu-3–6 represent Run
7. In these simulations, the measured background temperatures
(Figs. 1 and 2) were used as a model input, i.e., the first term on
the right-hand side of Eq (3). Heat release is accounted for in the
second term of that equation. Of course, this description is not
CO2
1000
800
600
CH2O
CH2O
CH2O
400
200
C2H6
C2H6
C2H6
C2H4
0
C2H4
C2H4
Mole Fraction (PPM)
150
100
H2
H2
H2
50
CH3OH
CH3OH
CH3OH
0
0
40
80
120
160
200
Time (ms)
240
280
320
0
50
100
Time (ms)
150
200
0
50
100
150
200
250
Time (ms)
Fig. 15. Experimental (symbols [50]) and computed species profiles during homogeneous, non-catalytic oxidation of methane in a turbulent flow reactor. Left panels:
0.02CH4–0.205O2–0.775N2 at 6 atm and 953 K; middle panels: 0.02CH4–0.205O2–0.775N2 at 10 atm and 953 K; right panels: 0.021CH4–0.205O2–0.774N2 at 6 atm and
974 K.
432
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
Table 4
Summary of numerical simulations.
No.
Ignition type
Background
Temperature input
Flow condition
in Table 1
Computed furnace
temperature at ignition Tf-ig (K)
Difference from
measureda (K)
Simu-1
Simu-2
Simu-3
Simu-4
Simu-5
Simu-6
Non-Catalytic
Non-Catalytic
Catalyticd
Catalyticd
Catalyticd
Catalyticd
Tb,center
Tb,wall
Tb,center
Tb,wall
Tb,center
Tb,wall
F-1
F-1
F-2b
F-2b
F-2c
F-2c
1127
1039
956
880
960
885
+14
74
17
93
13
88
a
b
c
d
Run 6 and Run 7 in Table 2 are used for comparisons with non-catalytic and catalytic cases, respectively.
Toluene was converted to CO2 and H2O, i.e. cold gas velocity = 52.8 cm/s, mole fractions (CH4/O2/CO2/H2O/N2) = 0.083/0.356/0.047/0.027/0.486, and / = 0.47.
Toluene was included as a reactant.
hDpi = 19 nm and f = 0.013 cm1 (STP), the ignition conditions in Run 7 of Table 2.
mathematically rigorous since local heat release would affect the
first term because of sensitivity of heat conduction and convection
towards local temperature. This uncertainty, however, is expected
to be smaller than the non-uniform temperature in the reactor. In
the simulations, we used both the measured centerline and nearwall temperatures as model input and treated them as the two limiting cases.
Table 4 shows that the predicted furnace temperature at which
the ignition occurs is only 14 K higher than the experimental value
when the centerline temperature was used as a model input (also,
see Fig. 16). Considering all of the experimental and model uncertainties, the close agreement is perhaps fortuitous, but encouraging
nonetheless. If the near-wall temperature was used in the simulation, the furnace temperature of ignition was predicted to be 74 K
lower than the observation. The results computed with the two
limiting temperatures bracket the experimental value, as expected.
Considering that the wall quenching of the free radicals was not
considered in either computational case, the current results are
considered as more than acceptable. In what follows, only computations made with the centerline temperature will be discussed,
though computational results using the near-wall temperatures
are reported in Table 4 for all cases.
Fig. 16 shows the CH4 and CO2 mole fraction profiles computed
as a function of the furnace temperature, comparing the non-cata-
0.14
CH4
0.12
CO2
4.8. Mechanism of methane ignition catalyzed by Pd particles
Mole fraction
0.10
0.04
0.02
0.00
800
Non-Catalytic
Simu-1
Catalytic
Simu-5
0.08
0.06
lytic case (Simu-1) with the catalytic cases (Simu-3 and -5). Also
shown in the figure are the corresponding experimental values.
As seen, the computed temperatures are within 20 K of the experimental values. Under comparable conditions, the model predicts a
reduction of 171 K in the furnace temperature at a catalyst loading
of 2 104 (atomic Pd-to-CH4 ratio) and a corresponding surface
area-to-gas volume ratio of 0.013 cm1, whereas the experimental
values give 140 K. The simulation results are expected to be sensitive to the local temperature inside the flow reactor, but since this
temperature is not a constant, it was not easy to quantify this
sensitivity.
Because toluene is used as the solvent for Pd(THD)2, its impact
on the gas-phase reaction kinetics is considered in the simulation.
USC Mech II predicts the laminar flame speed and ignition delay of
toluene-oxidizer mixtures reasonably well [51]. However, the
model includes a limited number of reactions for toluene oxidation
under the current experimental conditions. Hence, the simulation
was conducted by assuming all of the carbon in toluene was converted to CO2 (Simu-3) and by assuming all of the toluene remains
as the reactant at the onset of the reaction (Simu-5). Unlike the
experimental observation, Simu-5 did not yield noticeable toluene
consumption prior to ignition, as seen in Fig. 16. The reason, of
course, is gas-phase reactions and/or catalytic processes not accounted for by the model. On the other hand, the computed ignition temperature is not at all sensitive to the toluene chemistry,
because Simu-3 and 5 yielded basically the same results. Overall,
the combined gas-phase and gas-surface model predicts the experimental results well.
Catalytic
Simu-3
Non-Catalytic
Exp. Run 6
Catalytic
Exp. Run 7
900
1000
1100
Furnace temperature, Tf (K)
1200
Fig. 16. Computational results of CH4 and CO2 mole fractions at the reactor exit as a
function of furnace temperature Tf for both non-catalytic (Simu-1) and catalytic
(Simu-3 and -5) cases. The model used the centerline temperature Tb,center as the
input. For Simu-3, toluene was assumed to be completely oxidized to CO2 and H2O
before ignition, whereas for Simu-5 toluene was included as a reactant. For both
catalytic cases, Pd particles with hDpi = 19 nm and f = 0.013 cm1 (STP) were
assumed.
Analysis of the computational results indicates that the ignition
process involves both thermal and radical runaway. Fig. 17 presents the evolution of gas-phase species concentration and catalyst
surface coverage just before ignition (Simu-3) against the axial position of the flow reactor and the reaction time. The model predicts
the surface to be nearly saturated by O(S), in agreement with the
XPS results at 973 K. For the reactor conditions used in Simu-3,
the concentration of O(S) was found to be independent of the initial surface composition. Simulations by assuming the initial Pd
surface being fully Pd(S) or fully saturated by O(S) yielded virtually
identical ignition temperatures. Additional tests were made by
introducing the Pd particles at different reaction times. The results
show that as long as the particles were ‘‘injected” computationally
in the first half of the reactor, the computed ignition temperature
remains the same.
As seen in Fig. 17, the initial rise in temperature is caused by
heating of the gas by the furnace until about 50 cm from the inlet.
The subsequent rise in temperature is largely the result of heat release due to catalytic reactions. This may be seen in the top panel
433
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
Residence time, t (sec)
0.00
0.15
0.27
0.36
0.44
0.51
0.58
0.64
1200
0.4
Mole fraction
1000
T center
0.3
800
Tb ,center
0.2
600
0.1
CH 4
400
Gas temperature, T (K)
O2
CO 2
H2O
0.0
200
Mole fraction
10 -4
CH 2 O
10 -6
10 -8
HO 2
10 -10
H2
OH
CO
O
H
10 -12
10 -1
10 -3
0.99
OH(S)
C(S)
10 -5
0.98
CO(S)
CH 3 (S)
10 -7
0.97
Pd(S)
H 2 O(S)
10 -9
0
10
20
30
O(S) site fraction, θΟ
Surface site fraction, θ
1.00
O(S)
H(S)
40
50
60
70
0.96
Distance from reactor inlet, x (cm)
Fig. 17. Gas-phase species mole fraction, surface site fraction h, reacting centerline temperature Tcenter and non-reacting, background centerline Tb,center as a function of the
distance from reactor inlet and residence time, computed for the furnace temperature Tf = 954.7 K (Simu-3: hDpi = 19 nm and f = 0.013 cm1 (STP)).
of Fig. 17 by comparing the centerline temperature profiles between the background and catalytic oxidation. From 50 to 60 cm,
almost all CH4 consumption and CO2 production is the result of
catalysis. At the same time, the reaction becomes autocatalytic,
in that the temperature rise causes an increase in the catalytic
reaction rate. At about 70 cm from the inlet, the temperature is
brought to 1000 K, at which point the increased desorption of
oxygen leads to a rapid rise in other surface sites and the overall
reaction rate. The temperature now increases rapidly. Above
1000 K, gas-phase reaction starts to play a role, as evidenced by
the rapid rise in the gas-phase H, O, OH, and HO2 concentrations.
To understand the impact of surface reactions on ignition, we
carried out sensitivity analysis for the simulation discussed above.
The sensitivity coefficient defined as DTf-ig/hTf-igi, where DTf-ig is
the change in the furnace temperature of ignition resulting from
perturbing the pre-factors by a factor of 2, and hTf-igi is an average
before and after perturbation. The results are shown in Fig. 18. It is
seen that the key steps are adsorption and desorption of O2 (R1)
and the oxidative dissociation of CH4 on a vacant site (R11). These
results are in agreement with what Fujimoto et al. [13] reported for
CH4 oxidation from 500 to 800 K over PdO surfaces. The overall catalytic reaction rate appears to be limited by this reversible adsorption and desorption process, which controls the density of vacant
Pd sites needed for CH4 oxidative dissociation on catalyst surfaces.
Water or OH poisoning is predicted to play a minor role under the
conditions studied. Around 900 K, the major pathway for water
formation is the surface metathesis reaction between two surface
OH sites.
The numerical results reported in Fig. 17 are complicated by the
variable centerline temperature. To better understand the ignition
mechanism, we carried out calculations for stoichiometric methane oxidation in air under the adiabatic and isobaric conditions
at an initial temperature of 900 K and 1 atm pressure. The computed temperature–time histories are shown in Fig. 19. The calcu-
434
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
103
O2+2Pd(S)
R1b
2O(S)
H2O+Pd(S)
R2b
H2O(S)
2OH(S)
H2O(S)
H2O+Pd(S)
H2O(S)+O(S)
R5b H2O(S)+O(S)
R11
p = 1 atm
O2+2Pd(S)
R2f
R5f
2O(S)
Ignition delay time, τ (sec)
R1f
2OH(S)
CH4+Pd(S)+O(S)
102
101
Non-Catalytic
100
10-1
10-2
Catalytic
CH3(S)+OH(S)
-0.06
-0.04 -0.02
0.00
0.02
0.04
Sensitivity coefficient, - ΔTf-ig/<Tf-ig>
0.06
Fig. 18. Sensitivity coefficients, DTf-ig/hTf-igi, computed for Simu-3. A positive
sensitivity coefficient indicates lowered ignition temperature. Reactions with noiselevel sensitivities are omitted.
lation for catalytic oxidation used typical Pd loading and size, i.e.,
hDpi = 19 nm and the surface-to-volume ratio equal to 0.013 cm1
(STP). Under this Pd loading, the ignition delay time is shortened
by about two orders of magnitude compared to homogeneous,
non-catalytic methane oxidation. For the catalytic case, analysis
shows that the initial temperature rise at around 102 s is entirely
caused by heat release from catalytic reactions. Interestingly the
temperature levels off at around 1400 K just before it takes off
again some 102 s later. This behavior is explored in detail, as
shown in the inset of Fig. 19. Without considering gas-phase reactions, the predicted temperature plateaus at around 1400 K, far below the adiabatic flame temperature. In other words, the catalytic
reaction shuts itself off at that temperature because surface
desorption of O2 becomes so rapid that the catalytic activity is lost.
Yet because the catalytic reactions had brought the gas temperature to 1400 K, close to the cross-over temperature of H + O2
2200
Catalytic
1800
1800
1600
T (K)
Temperature, T (K)
2000
Ignition
due to
gas-phase
oxidation
1400
1200
1600
1000
0.02
1400
Non-Catalytic
Heat release due to
catalytic oxidation
1000
800
10-4
10-3
10-2
Time, t (sec)
600
800
1000
1200
1400
1600
Initial temperature, T0 (K)
Fig. 20. Ignition delay time computed as a function of initial temperature of a
stoichiometric methane/air mixture under adiabatic and isobaric conditions. The
catalytic case used hDpi = 19 nm and f = 0.013 cm1 (STP).
chain branching and termination under that pressure, homogeneous gas-phase reactions become active. Eventually ignition occurs because of both gas-phase thermal and radical runaways.
Hence, the present analysis suggests a two-step process to ignition.
That is, catalytic reactions result in initial heat release and raise the
gas temperature. Although the temperature rise would shut off the
catalytic activity at some point, gas-phase radical chain branching
takes over eventually and completes the ignition process.
To assess the influence of initial temperature on the ignition delay time, we extend the above calculation to a range of 700–
1500 K. Fig. 20 shows that under the level of Pd loading discussed
earlier, the ignition delay time is shortened by two orders of magnitude from 700 to about 1200 K. When the initial temperature is
at least 1400 K, the catalytic and noncatalytic ignition delay is
identical because the catalyst activity is suppressed for reasons already discussed and also because gas-phase reaction kinetics tend
to dominate at high temperatures. Over the range of temperature
shown in Fig. 20 and up to 40 atm pressure (results not shown
here), the ignition mechanism remains the same. That is, we observe a sequential, two-step process in which catalytic reaction
dominates initially, but gas-phase reactions take over after catalytic processes are deactivated by the rising temperature.
5. Conclusions
0.03
t
1200
10-3
10-1
100
Fig. 19. Temperature–time histories computed for a stoichiometric methane/air
mixture under adiabatic and isobaric conditions, showing the difference in ignition
onset between catalytic and noncatalytic ignition of the mixture with initial
temperature T0 = 900 K and pressure p = 1 atm. The catalytic case used a median
particle diameter hDpi = 19 nm and the surface-to-volume ratio f = 0.013 cm1
(STP). The inset also shows the temperature history computed without considering
gas-phase reactions.
Freely suspended palladium nanoparticles were generated
in situ in a laminar flow reactor from an aerosol composed of
Pd(THD)2 diluted in toluene. Mixing the aerosol with a stream of
methane, oxygen, and nitrogen (equivalence ratio / 0.5) under
a heated condition can lead to a notable reduction in the ignition
temperature of the unburned mixture, compared to homogeneous
methane oxidation without palladium particles. Measurement of
the particle size distribution shows that the median particle diameter is around 20 nm, and the surface-to-gas volume fraction is
0.02 cm1 (STP). The observed size distribution is consistent with
a particle formation mechanism dominated by homogeneous
nucleation and particle size growth by coagulation. TEM/XPS analysis shows that the particles have a crystalline palladium core and
a few atomic layers of PdO on the particle surface for particles observed near or above 973 K furnace temperature, but very little
oxide was formed at 773 K. Coupled with the observation that cat-
T. Shimizu et al. / Combustion and Flame 157 (2010) 421–435
alytic ignition occurred near 973 K, Pd oxidation kinetics must play
a critical role in the catalytic process.
Numerical simulation using a detailed, combined gas-phase and
gas-surface reaction model shows that the experimental results are
well captured by the model. Simulation results suggested that the
surfaces of Pd nanoparticles give rise to catalytic reactions at a
temperature which would otherwise not lead to substantial gasphase reactions. The heat release resulting from catalysis eventually brings the gas to a temperature where the catalytic reaction
is overtaken by the gas-phase reactions, which eventually leads
to thermal and radical runaway and fuel ignition.
The key surface reaction steps were found to be the reversible
oxygen adsorption/desorption, which governed the vacant Pd site
density and the overall rate of the catalytic reaction. Methane oxidative dissociation on a vacant Pd site is the rate-limiting step for
catalytic conversion of methane to CO and hence the heat release.
Acknowledgements
The Utah and USC groups gratefully acknowledge support for
this work from the Air Force Office of Scientific Research through
the MURI program (FA9550-08-1-0400). The Utah, USC, TDA, and
Reaction Systems efforts were also supported by AFOSR STTR contracts (FA8650-06-C-2673 and FA9550-07-0106) to TDA, with subcontracts to Utah and USC. In addition, the authors thank Mr. Brian
Windecker of TDA Research for design and assembly of the catalyst
test apparatus.
References
[1] M.J. Lewis, J. Propul. Power 17 (2001) 1214–1221.
[2] D. Ciuparu, M.R. Lyubovsky, E. Altman, L.D. Pfefferle, A. Datye, Catal. Rev. 44
(2002) 593–649.
[3] D. Ciuparu, L. Pfefferle, Appl. Catal. A 209 (2001) 415–428.
[4] K. Persson, L.D. Pfefferle, W. Schwartz, A. Ersson, S.G. Jaras, Appl. Catal. B 74
(2007) 242–250.
[5] Y.K. Rao, Stoichiometry and Thermodynamics of Metallurgical Processes,
Cambridge Press, Cambridge, 1985.
[6] G. Zheng, E.I. Altman, Surf. Sci. 462 (2000) 151–168.
[7] T. Schalow, B. Brandt, M. Laurin, S. Schauermann, S. Guimond, H. Kuhlenbeck, J.
Libuda, H.-J. Freund, Surf. Sci. 600 (2006) 2528–2542.
[8] A.K. Datye, J. Bravo, T.R. Nelson, P. Atanasova, M. Lyubovsky, L. Pfefferle, Appl.
Catal. A 198 (2000) 179–196.
[9] M.M. Wolf, H.Y. Zhu, W.H. Green, G.S. Jackson, Appl. Catal. A: Gen. 244 (2003)
323–340.
[10] Å. Johansson, M. Försth, A. Rosén, Surf. Sci. 529 (2003) 247–266.
[11] J.C.G. Andrae, A. Johansson, P. Bjornbom, A. Rosen, Surf. Sci. 563 (2004) 145–
158.
[12] S.L. Tait, Z. Dohnalek, C.T. Campbell, B.D. Kay, Surf. Sci. 591 (2005) 90–107.
[13] K. Fujimoto, F.H. Ribeiro, M. Avalos-Borja, E. Iglesia, J. Catal. 179 (1998) 431–
442.
[14] K. Sekizawa, M. Machida, K. Eguchi, H. Arai, J. Catal. 142 (1993) 655–663.
[15] R.W. Sidwell, H.Y. Zhu, R.J. Kee, D.T. Wickham, C. Schell, G.S. Jackson, Proc.
Combust. Inst. 29 (2002) 1013–1020.
435
[16] H. Wang, X. You, A.V. Joshi, S.G. Davis, A. Laskin, F. Egolfopoulos, C.K. Law,
High-Temperature Combustion Reaction Model of H2/CO/C1–C4 Compounds,
2007, <http://ignis.usc.edu/USC_Mech_II.htm>.
[17] M. Lashdaf, H. Hatanpaa, M. Tiitta, J. Therm. Anal. Calorim. 64 (2001) 1171–
1182.
[18] B. Van Devener, S.L. Anderson, T. Shimizu, H. Wang, J. Nabity, J. Engel, J. Yu, D.
Wickham, S. Williams, J. Phys. Chem. A, submitted for publication.
[19] M.E. Coltrin, R.J. Kee, F.M. Rupley, Int. J. Chem. Kinet. 23 (2004) 1111–1128.
[20] S.G. Davis, C.K. Law, H. Wang, Combust. Flame 119 (1999) 375–399.
[21] A. Laskin, H. Wang, Chem. Phys. Lett. 303 (1999) 43–49.
[22] A. Laskin, H. Wang, C.K. Law, Int. J. Chem. Kinet. 32 (2000) 589–614.
[23] Z. Qin, V. Lissianski, H. Yang, W.C. Gardiner Jr., S.G. Davis, H. Wang, Proc.
Combust. Inst. 28 (2000) 1663–1669.
[24] H. Wang, Int. J. Chem. Kinet. 33 (2001) 698–706.
[25] C.K. Law, C.J. Sung, H. Wang, T.F. Lu, AIAA J. 41 (2003) 1629–1646.
[26] S.G. Davis, A.V. Joshi, H. Wang, F. Egolfopoulos, Proc. Combust. Inst. 30 (2005)
1283–1292.
[27] A.V. Joshi, H. Wang, Int. J. Chem. Kinet. 38 (2006) 57–73.
[28] X.Q. You, H. Wang, E. Goos, C.J. Sung, S.J. Klippenstein, J. Phys. Chem. A 111
(2007) 4031–4042.
[29] S.G. Davis, C.K. Law, H. Wang, J. Phys. Chem. A 103 (1999) 5889–5899.
[30] S.-A. Seyed-Reihani, G.S. Jackson, Chem. Eng. Sci. 59 (2004) 5937–5948.
[31] J.F. Kramer, S.-A.S. Reihani, G.S. Jackson, Proc. Combust. Inst. 29 (2002) 989–
996.
[32] O. Deutschmann, R. Schmidt, F. Behrendt, J. Warnatz, Proc. Combust. Inst. 26
(1996) 1747–1754.
[33] K. Yagi, D. Sekiba, H. Fukutani, Surf. Sci. 442 (1999) 307–317.
[34] Y. Cao, Z.-X. Chen, Surf. Sci. 600 (2006) 4572–4583.
[35] E. Broclawik, J. Haber, A. Endou, A. Stirling, R. Yamauchi, M. Kubo, A. Miyamoto,
J. Mol. Catal. A 119 (1997) 35–44.
[36] E. Broclawik, R. Yamauchi, A. Endou, M. Kubo, A. Miyamoto, J. Chem. Phys. 104
(1996) 4098–4104.
[37] A.B. Mhadeshwar, H. Wang, D.G. Vlachos, J. Phys. Chem. B 107 (2003) 12721–
12733.
[38] Z.G. Li, H. Wang, Phys. Rev. E 68 (2003) 061206.
[39] Z.G. Li, H. Wang, Phys. Rev. E 68 (2003) 061207.
[40] H. Wang, Transport Properties of Small Spherical Particles, in: S. S. Sadhal,
(Ed.), Interdisciplinary Transport Phenomena V: Fluid, Thermal, Biological,
Materials & Space Sciences, 2008.
[41] B. Zhao, Z.W. Yang, Z.G. Li, M.V. Johnston, H. Wang, Proc. Combust. Inst. 30
(2005) 1441–1448.
[42] B. Zhao, Z. Yang, M.V. Johnston, H. Wang, A.S. Wexler, M. Balthasar, M. Kraft,
Combust. Flame 133 (2003) 173–188.
[43] B. Zhao, Z.W. Yang, J.J. Wang, M.V. Johnston, H. Wang, Aerosol. Sci. Technol. 37
(2003) 611–620.
[44] J. Singh, R.I.A. Patterson, M. Kraft, H. Wang, Combust. Flame 145 (2006) 117–
127.
[45] B. Zhao, K. Uchikawa, H. Wang, Proc. Combust. Inst. 31 (2007) 851–860.
[46] A.D. Abid, N. Heinz, E. Tolmachoff, D. Phares, C. Campbell, H. Wang, Combust.
Flame 154 (2008) 775–788.
[47] A.D. Abid, E.D. Tolmachoff, D.J. Phares, H. Wang, Y. Liu, A. Laskin, Proc.
Combust. Inst. 32 (2009) 681–688.
[48] H. Wang, A.D. Abid, Size distribution and chemical composition measurements
of nascent soot formed in premixed ethylene flames, in: H. Bockhorn, A.
D’Anna, A.F. Sarofim, H. Wang (Eds.), Combustion Generated Fine
Carbonaceous Particles – Proceedings of an International Workshop held in
Villa Orlandi, Anacapri (May 13–16, 2007), Karlsruhe University Press,
Karlsruhe, 2009, pp. 367–384.
[49] J.D. Landgrebe, S.E. Pratsinis, Ind. Eng. Chem. Res. 28 (1989) 1474–1481.
[50] T.B. Hunter, H. Wang, T.A. Litzinger, M. Frenklach, Combust. Flame 97 (1994)
201–224.
[51] Z.M. Djurisic, A.V. Joshi, H. Wang, Detailed kinetic modeling of benzene and
toluene combustion, in: The Second Joint Meeting of the US Sections of the
Combustion Institute, Berkeley, CA, March 2001 (paper 238).