Sie sind auf Seite 1von 20

This article was downloaded by: [Matiej]

On: 6 March 2011


Access details: Sample Issue Voucher: Combustion Science and TechnologyAccess Details: [subscription number
934500386]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-
41 Mortimer Street, London W1T 3JH, UK

Combustion Science and Technology


Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713456315

An Instability of Diluted Lean Methane/Air Combustion: Modeling and


Control
T. Wadaa; F. Jarmolowitzb; D. Abelb; N. Petersa
a
Institute of Combustion Technology RWTH Aachen University, Aachen, Germany b Institute of
Automatic Control RWTH Aachen University, Aachen, Germany

Online publication date: 24 November 2010

To cite this Article Wada, T. , Jarmolowitz, F. , Abel, D. and Peters, N.(2011) 'An Instability of Diluted Lean Methane/Air
Combustion: Modeling and Control', Combustion Science and Technology, 183: 1, 1 — 19
To link to this Article: DOI: 10.1080/00102201003789147
URL: http://dx.doi.org/10.1080/00102201003789147

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
Combust. Sci. and Tech., 183: 1–19, 2011
Copyright # Taylor & Francis Group, LLC
ISSN: 0010-2202 print=1563-521X online
DOI: 10.1080/00102201003789147

AN INSTABILITY OF DILUTED LEAN METHANE/AIR


COMBUSTION: MODELING AND CONTROL

T. Wada,1 F. Jarmolowitz,2 D. Abel,2 and N. Peters1


1
Institute of Combustion Technology, RWTH Aachen University, Aachen,
Germany
2
Institute of Automatic Control, RWTH Aachen University, Aachen,
Germany

In order to study possibilities of Model Predictive Control (MPC) to low temperature


combustion, the authors performed a numerical study of combustion with a lean highly
diluted methane/air mixture in a perfectly stirred reactor using a detailed chemical kinetic
model. Chosen conditions are the following: equivalence ratio 0.6, residence time of mixture
Downloaded By: [Matiej] At: 11:54 6 March 2011

in reactor 0.5 s; molar fraction of Nitrogen 0.9, temperature of incoming unburned gases
and reactor 1100 K, and heat loss coefficient 2  103 cal/(cm2Ks). At these conditions
strong oscillations are predicted in agreement with previous experimental findings of
M. de Joannon et al. (2004). It is found that, at low temperatures (<1300 K), reactions
forming CH3O and CH3O2 from CH3 had a strong influence on the oscillations. Once these
reactions were subtracted from the kinetic model, no oscillations were observed. A virtual
MPC suppressed the oscillations at low temperatures successfully and demonstrated
extremely low NO and N2O emissions (<0.1 ppm).

Keywords: Combustion instability; Heat loss; Limit cycle; Model reduction; NOx; Predictive optimization

1 INTRODUCTION
For practical combustors such as internal combustion (IC) engines and gas
turbines, improving the efficiency of fuel consumption and reducing combustion
emissions are important topics for research and development (Glarborg et al.,
1994, 1995; Rutar et al., 1996; Sullivan et al., 2002). Lean combustion fuel, in gen-
eral, decreases flame temperatures and combustion emissions. However, combustion
phenomena become unstable at low temperature conditions, and such instability
hampers the application of these technologies in many areas, such as in the homo-
geneous charge compression ignition engines (Anderlohr et al., 2009; Yao et al.,
2009; Zhao et al., 2009). In order to simply the phenomena and identify those diffi-
culties, a Jet Stirred Reactor (JSR) has been used for a fundamental study. The main
advantage of the JSR is the focus on influence caused by chemical reactions under
various conditions such as inlet temperature, pressure, or dilution. The JSR has been

Received 26 November 2009; revised 26 February 2010; accepted 18 March 2010.


Address correspondence to T. Wada, Institute of Combustion Technology, RWTH Aachen
University, Templergraben 64, 52056 Aachen, Germany. E-mail: t.wada@itv.rwth-aachen.de

1
2 T. WADA ET AL.

used in a wide range of temperatures for various fuels in order to derive chemical
kinetic mechanisms (Dagaut and Cathonnet, 2006; Dayma et al., 2009; Healy
et al., 2008a, 2008b).
We are facing three problems in understanding the results from those experi-
mental and numerical investigations. First, our understanding of their oxidation
processes is not sufficient enough to describe the influence of all hydrocarbon spe-
cies. At high temperature conditions (>1300 K), the kinetics have been well studied
(GRI-mech) and can be described using systematical reduction (Smith et al., 2006).
At low temperatures (<1300 K), thermal decomposition of the hydrocarbon species
plays an important role during its oxidation processes (Peters et al., 2002; de Vries
and Petersen, 2007). However, the description of the role of chemical reactions at
low temperatures (low-temperature chemistry) is still uncertain and is one of ongoing
topic in combustion research (Petersen et al., 2007; Dagaut and Togbe, 2009).
Second, a huge number of species and elementary reactions in a chemical kinetic
model causes critical problems in the computational load. For example, the
chemical kinetic model for n-heptane (n-C7H16) from Lawrence Livermore National
Laboratory (LLNL) involves 561 species and 2,539 elementary reactions. In order to
reduce the computational load, a time-scale-adapted calculation method was
Downloaded By: [Matiej] At: 11:54 6 March 2011

proposed (Ju et al., 2009). This method applies different discretizations for species,
which are categorized based on their time scale. Third, elementary reactions are
evaluated differently based on different validation data. Unless the rate constants
of a few elementary reactions are measured directly, most of the values of other
reactions have to be estimated. Therefore, differences in the order of reaction rates
can be observed frequently. In addition, a set of the same elementary reactions might
give different results because of different rate constants. In order to observe the
influence of elementary reactions, the same elementary reactions have to be used.
For instance de Joannon et al. (2005) studied oscillations with diluted methane=air
mixtures under a wide range of temperatures and fuel concentrations. They identified
decomposition of the acetyl radical CH3CO and oxidation of the vinyl radical C2H3
as keys to modulate oscillations at low temperatures. However, they pointed out that
different oscillatory conditions, such as composition of the mixture, were shown by
the use of different chemical kinetic models. In other words, this difference implies
that influence of specific reaction path may be not valid with another chemical
kinetic model and cannot be examined using another hydrocarbon fuel. In order
to solve these problems, a reference model has to be defined for various fuels and
well validated over a wide temperature range. In addition, a reduction method for
this reference model is needed to reduce computational load.
Because instabilities prevent the application of low-temperature combustion
with its advantage of low emissions measures have to be taken to stabilize it. There-
fore, active control is studied under low-temperature conditions to avoid combustion
instabilities. As a control methodology, a Model Predictive Control (MPC) is inves-
tigated as a strategy utilizing an explicit kinetic model of the combustion chemistry.

2 MODEL DESCRIPTION
There is a constant mass flow rate into the constant-volume Perfectly Stirred
Reactor (PSR). The reactor is at atmospheric pressure, with an inflow of unburned
DILUTED LEAN METHANE/AIR COMBUSTION 3

mixture and an outflow of burned gas. The inflow temperature and mass fraction of
the jth species of the unburned and burned mixtures are denoted by T and Yj ,
respectively, where the superscript  denotes inlet conditions. We assume constant
inlet and low-Mach conditions, as well as constant pressure in the reactor. Under
these assumptions, the outlet mixture can be assumed to be the same as the one in
the reactor. In addition, the wall temperature is assumed to be the same as that of
the inlet. A heat loss occurs when the reactor temperature differs from that of the
wall. Due to rapid mixing, the heat loss can be modeled as homogeneous sink term.
There is no additional heat flux from the wall. The governing equation for species
conservations is

dYj 1  m
_ chem;j
¼ ðYj  Yj Þ þ ð1Þ
dt s q

Here, t, q, m
_ chem, and s denote time, density, the change of mass due to the chemical
reactions, and residence time of the mixture in the reactor (s  qVR =m _ where
VR stands volume of the reactor and m _ for the mass flow rate), respectively. The
governing equation of energy conservation is
Downloaded By: [Matiej] At: 11:54 6 March 2011

dT 1 X   X hj m
_ chem;j aAðT  T  Þ
cp ¼ Yj ðhj  hj Þ   ð2Þ
dt s j j
q qVR

where cp, h, a, and A denote specific heat, enthalpy, the heat loss coefficient, and
surface area of the reactor, respectively. The first terms in the right-hand side of
the Eqs. (1–2) represents incoming and outgoing flows. In this study, VR and A,
which depends on experimental apparatus, are taken from the experimental study
by de Joannon et al. (2005) and a part of the subroutine of Cantera on Matlab
was used to solve the governing equations.

3 RESULTS AND DISCUSSIONS


3.1 Model Reduction
In our study, a chemical mechanism from Ranzi’s group (Milan, Italy) was
used as a reference mechanism (Ranzi et al., 2005). This mechanism has been vali-
dated under a wide temperature range. It is suitable for various fuels, but with a
smaller number of species and reactions using the lumping method (Ranzi et al.,
2001). Only 341 species are involved in this mechanism for hydrocarbon species
up to 16 carbon atoms, instead of 561 species in one for species up to 7 carbon atoms
from LLNL. There are two restrictions to a reduction process. First, the calculated
maximum reactor temperature has to remain constant. A lower or higher flame tem-
perature caused by the reduction process indicates a significant change of reaction
mechanism. Second, the frequency of the oscillations, time intervals between auto-
ignitions, has to remain constant. The mean residence time of mixture in combustors
is determined by inlet conditions and is independent from combustion phenomena.
Therefore, a different frequency of the oscillations indicates that the reduction
process modifies the chemistry significantly or the mechanism. Reductions consist
4 T. WADA ET AL.

of a step-by-step processes; a) carbon atom screening (CAS), b) production rate


screening (PRS), and c) rate of progress screening (RPS), including the previously
mentioned restrictions.

3.1.1 Carbon atom screening (CAS). CAS is a reduction process to sub-


tract unnecessary hydrocarbon species from a chemical kinetic model. A chemical
kinetic model consists of a hierarchical structure based on the number of carbon
atoms within the species (Warnatz et al., 1996). The highest class of the hierarchy
describes H2 oxidation and has to be considered for all combustion phenomena.
On the other hand, lower classes that describe the other reactions with CO, CH3OH,
or heavier hydrocarbons species have to be considered, depending on the tested fuel.
These classes can be subtracted from the chemical kinetic model if there is no influ-
ence on the entire combustion process. Figure 1 shows temperature profiles using
intermediate models of CAS with CH4-air mixtures at atmospheric pressure. Con-

ditions are the following equivalence ratio (/  nYCH 4
=YO 2 where n denotes stoi-
chiometric mass ratio), 0.6, T ¼ 1100 K, XN2 ¼ 0.9, and s ¼ 0.5 s (this mixture
composition and the inlet conditions are used for rest of this study). C5 indicates
a chemical kinetic model that contains only hydrocarbon species up to five carbon
atoms to be considered. Species that contain more than six carbon atoms have been
Downloaded By: [Matiej] At: 11:54 6 March 2011

removed. This results show that temperature profiles using models C2–C5 are
practically identical. On the other hand, using C1 chemistry only shows significant
differences. Therefore, hydrocarbon species containing up to two carbon atoms play
an important role for the lean CH4=air combustion under the tested conditions.
Therefore, the C2 model is the best chemical kinetic model for our purpose.

3.1.2 Production rate screening (PRS) and rate of progress screening


(RPS). Further screenings consider also the reactor temperature during the oscilla-
tions and the oscillation frequency. These values strongly depend on the heat loss
due to the walls. Figure 2 shows the influence of the heat loss on the temperature

Figure 1 Comparison of temperature profiles between C1–C5 models under atmospheric pressure (CH4=
air, / ¼ 0.6, XN2 ¼ 0:9, T ¼ 1100 K, and s ¼ 0.5 s). C2–C5 models are perfectly overlapped.
DILUTED LEAN METHANE/AIR COMBUSTION 5

Figure 2 Influence of heat loss due to walls on the temperature profiles. (a heat loss coefficient [cal=
(cm2Ks)]).
Downloaded By: [Matiej] At: 11:54 6 March 2011

profiles by changing the heat loss coefficient (a[cal=(cm2Ks)]) using the C2 model.
From the figure it can be observed that the temperature profiles strongly depend
on a. In addition, the heat loss affects the profiles in two ways. First, the heat loss
decreases the maximum temperature. In all cases, it was observed that the tempera-
ture increased rapidly within 1 s until it was close to the adiabatic flame temperature
(Tad ¼ 1619 K). The peak temperature values were inversely proportional to the heat
loss and were significantly influenced by it. Second, the heat loss led to both steady
and unsteady conditions after the temperature reached the maximum value. In the
case of low values of a (a ¼ 0 and 1  104), the temperature decreased gradually
and reached steady conditions. In contrast, unsteady conditions were observed in
the case of high values of a (a ¼ 4  104 and 1  103). Figure 3 shows a comparison

Figure 3 Change of oscillation frequency by change of heat loss coefficients.


6 T. WADA ET AL.

between experimentally (de Joannon et al., 2004) and numerically determined


oscillation frequency. a depends on the experimental apparatus and may be a unique
value. In order to numerically reproduce the experimentally observed oscillations, we
need to investigate the a. The numerically determined frequency decreases as a
increases and agrees with reference values at a ¼ 2  103; this a is used for rest of
this study.
PRS and RPS are processes to remove from the chemical kinetic model
negligible species and elementary reactions, respectively. For these processes, the
chemical mass production rates are addressed. The rate of change of mass
production in Eqs. (1–2) can be written as

_ chem;j ¼ Wj x_ j
m
¼ Wj ðx_ Pj  x_ Cj Þ ð3Þ
X  
F R
¼ Wj n j;i qi  qi
_ _ ð4Þ
i

Here, W, x_ P , x_ C , n, q_ F , and q_ R denote molecular weight, production rate, consump-


Downloaded By: [Matiej] At: 11:54 6 March 2011

tion rate, stoichiometric coefficient, forward reaction rate, and reverse reaction rate,
where subscript i denotes, ith elementary reaction. For the PRS, Eq. (3) is used to
evaluate importance of species. Negligible species have more complicated structures
and extremely smaller x_ P and x_ C than the others. This process specifies a reaction
network. The network consists of connected entire species of a certain importance
and describes the base of reaction processes from fuel to product. After this process,
the number of species is reduced subject to the two constrains mentioned previously.
Equation (4) is used to evaluate the importance of elementary reactions for RPS with
the same concept as PRS. The negligible elementary reactions have extremely small
q_ F and q_ R . The influence of these negligible elementary reactions is, therefore, also
small. This reduction process has to satisfy the following restrictions: constant flame
temperature, frequency of the oscillations, and number of species. After all these
reduction processes, a skeletal kinetic model is established (see Appendix). Table 1
shows the results of all screenings. The number of species and reactions decreases
dramatically, but the peak temperature and oscillation frequency do not change.
Figure 4 shows a comparison between C2 and the Skeletal models. The result shows
that the different lines perfectly overlap and the reduction process has been
accomplished successfully. This skeletal model is for lean CH4=air mixture with
elementary reactions taken from Ranzi’s reference mechanism. As a result, 341 spe-
cies and 9173 reactions from the original chemical kinetic model from Ranzi’s
research group are reduced to 32 species and 192 reactions as a Skeletal model.

Table 1 Reduction processes from the reference kinetic model (341 species and 9173 elementary reactions)

Num. Species Num. Reactions Peak Temp. [K] Oscillation Freq. [Hz]

CAS 56 592 1481 0.551


PRS 32 280 1486 0.548
RPS 32 192 1486 0.546
DILUTED LEAN METHANE/AIR COMBUSTION 7

Figure 4 Comparison of temperature profiles between those with Skeletal and C2 models.

3.2 Oscillations in PSR


Downloaded By: [Matiej] At: 11:54 6 March 2011

3.2.1 Typical oscillations. Figures 5a and 5b show the profile of tempera-


ture and the major species, O2, H2O, CO, CO2, and CH4, for one cycle of the oscil-
lation. The time and temperature are normalized by use of time at the minimum
temperatures (beginning of the cycle), period of the cycle, and the T . According
to an experimental investigation done by de Joannon et al. (2004), this pattern is
classified as cusp-shape oscillations. After every peak, the temperature decreases
until it is close to Tin, thereby implying the occurrence of flame quenching. On the
other hand, the peak temperature does not reach Tad even if temperature increases
rapidly and shows similar behavior to that for autoignition. As shown in
Figure 5b, no CH4 remains in the reactor when the temperature reaches high values.
As CH4 is consumed, O2 is also consumed until no CH4 remains in the reactor, and
the other products, H2O and CO2, and the intermediate CO are formed. Meanwhile,
the incoming unburned mixture is mixed with the burned gas and further dilution
occurs after the first peak. Therefore, the peak temperature of the following peaks
cannot reach the same temperature as one of the first peak (see Figure 4). In
addition, the hot residual gas enhances CH4 consumption, which triggers autoigni-
tion and complete consumption of CH4 at around 6.8 s. Figure 5c shows the NOx
profile using the Skeletal model with a detailed NOx kinetic model (Faravelli et al.,
2003; Frassoldati et al., 2003) added. NO and N2O are formed when the temperature
increases rapidly and approximately the same amount of NO and N2O are observed.
These results are caused by two effects. One is that the Prompt NO mechanism does
not play an important role. Under lean conditions NO formation involving hydro-
carbon radicals are suppressed (Steele et al., 1995). On the other hand, the thermal
NO mechanism plays an important role when the temperature is sufficiently high. In
addition, the nitrous oxide mechanism also plays an important role at low tempera-
tures even for a lean mixtures (Correa, 1992; Malte and Pratt, 1974). The second
effect is that N2O destruction depends on temperature and its influence can be
neglected below 1300 K, (Hulgaard and Dam-Johansen, 1993). Once N2O is formed
8 T. WADA ET AL.
Downloaded By: [Matiej] At: 11:54 6 March 2011

Figure 5 Temperature, major species, and NOx profiles within an cycle from Figure 1.

and temperature decreases, N2O is not converted to NO and remains in the reactor.
It should be noted that NO2 can be neglected because its formation mechanism is not
important under the present condition (Dagaut and Dayma, 2006). This result
implies that keeping reactor condition at low temperatures (i.e., avoiding rapid
increase in temperature) is the most effective way to reduce NOx emissions.

3.2.2 Influence of specific reactions. Figures 6 show the net production


rates of major products from CH3 (x_ N CH3 where N denotes net values). The major
products are CH2O, CH3OH, CH3O, and CH3O2. These values were taken at the
minimum temperature after the temperature peak (the beginning of the cycle, Tmin;
1100 K), 1% temperature increase (T1%; 1111 K), and inflection point of temperature
profile during the temperature increase (Tinflec; 1324 K) using the Skeletal model.
These figures shows the temperature dependency of the CH4 oxidation. The CH3
produced mainly CH3O and CH3O2 at Tmin and T1% (see Figure 6a and 6b), though
CH3 produced CH2O and CH3OH at Tinflec instead of those species (see Figure 6c).
The CH3OH is known as an intermediate which has considerable importance at high
temperatures (Fernandes et al., 2006; Krasnoperov and Michael, 2004; Pilling, 1996;
Walch, 1993). Moreover, the Tinflec is close to the inner-layer temperature which is
the minimum temperature to sustain the premixed flame (Goettgens et al., 1992,
Seshadri, 1996). Therefore, these results clearly show that the CH3O and CH3O2
DILUTED LEAN METHANE/AIR COMBUSTION 9
Downloaded By: [Matiej] At: 11:54 6 March 2011

Figure 6 Comparison of CH3 consumption rates to lead CH2O, CH3OH, CH3O, and CH3O2 using
skeletal model under the same condition as that in Figure 4. Figure 6a is at the minimum temperature Tmin
on a cycle (beginning of the cycle); Figure 6b is at 1% increase temperatures; Figure 6c is at inflection point
of the temperature profile during autoignition.

are the major products from CH3 at low temperatures. In other words, these results
imply that reactions forming CH3O and CH3O2 from CH3 at low temperatures (i.e.,
the time period between the beginning of the cycle and before autoignition) plays an
important role for the oscillations.
In order to study roles of the CH3O and CH3O2 at Tmin and T1%, the net rate of
progress (q_ N
i ) of CH3O and CH3O2 related reactions using the Skeletal models
are compared. We considered CH3O related reactions (R1–R4) and CH3O2 related
reactions (R5–R7) to be the following:

R1: CH3O (þM) ¼ CH2O þ H (þM)


R2: HO2 þ CH3 ¼ CH3O þ OH
R3: O2 þ CH3O HO2 þ CH2O
R4: O2 þ CH3 ¼ CH3O þ O
R5: O2 þCH3 ¼ CH3O2
10 T. WADA ET AL.
Downloaded By: [Matiej] At: 11:54 6 March 2011

Figure 7 Comparison of net rate of progress (q_ N


i ) of major CH3O and CH3O2 related reactions using
Skeletal at Tmin and T1%. Figures 7a and 7b are at Tmin and Figure 7c and 7d are at T1%.

R6: CH3O2 ¼ CH2O þ OH


R7: CH3O2 þ CH3 ¼ CH3O þ CH3O

Figure 7 shows q_ N
i of R1–R7 at Tmin and T1% using the Skeletal model. These
results clearly show reaction paths of CH3O and CH3O2 between CH3 and CH2O. In
Figures 7c and 7a, the values of R1 and R2 are are remarkably larger than the values
of R3 and R4. Therefore, the CH3O related reaction path between CH3 and CH2O is
CH3!CH3O!CH2O. The R2 has a larger value than R4, which is a well-known
reaction at high temperatures (Herbon et al., 2005). This means that the HO2 radical
is important to form CH3O from CH3 at low temperatures. The CH3O2 related reac-
tion path between CH3 and CH2O is shown in Figures 7b and 7d. R5 and R6 have
larger values and describe it as CH3!CH3O2!CH2O. Moreover, this result also
shows different importance in elementary reactions by comparison between R4
and R5. A relatively large value of R7 is shown in Figure 7d. This reaction describes
the connection between CH3O and CH3O2 using CH3. As a result, the important
reaction paths of the CH4 oxidation processes at low temperatures are reactions
forming CH2O from CH3O and CH3O2, and connecting between CH3O and CH3O2.
In order to evaluate those influence on the oscillations, the corresponding
elementary reactions are removed from the Skeletal model. Table 2 shows the models
DILUTED LEAN METHANE/AIR COMBUSTION 11

Table 2 Subtracted models

Model Subtracted elementary reaction

S1 CH3O2 þ CH3 ¼ CH3O þ CH3O


S2 HO2 þ CH3 ¼ CH3O þ OH
S3 O2 þ CH3 ¼ CH3O2

(S1, S2, and S3) with the corresponding removed elementary reactions. Figure 8
shows the comparison of oscillation frequency using skeletal and models S1–S3. A
significantly small frequency is observed using S1 and no oscillations are observed
using S2 and S3. This difference that their importance at time period between begin-
ning of the cycle and before autoignition. It also proves that the CH4 oxidation pro-
cess at low temperatures is different than that at high temperatures. Therefore, the
following hypothesis can be established: heat release caused by CH4 oxidation with
intermediates, CH3O and CH3O2, increases the temperature gradually from close to
T and activate reaction path for the high temperatures instead of those for low tem-
peratures. Figure 9 shows comparison of ignition delay times (signs) at 1100 K using
the Skeletal Model and model S1–S3. These sign show required time to activate reac-
Downloaded By: [Matiej] At: 11:54 6 March 2011

tion paths at high temperatures and significant influence caused by subtracting


important elementary reactions at low temperatures. S3 especially shows the longest
sign and the removed reactions is the most remarkable reaction path of CH4
oxidation processes at low temperatures. In addition, the oscillation frequency and
sign are were inversely proportional. For a fixed s, the longer sign indicates that large
amount of unburned fuel flows out before it produces products. Therefore, no or
slow oscillations are observed by using a model that has long longer sign than one
using skeletal model.
Switching between reaction paths at high and low temperatures plays an
important role for the oscillations. The temperature decreases from the peak tem-
perature due to the heat loss. If the heat loss is large enough, the reaction path of

Figure 8 Comparison of oscillation frequency using the skeletal, S1, S2, and S3 models. Subtracted
reactions from skeletal model are shown in Table 2.
12 T. WADA ET AL.

Figure 9 Comparison of the sign using the C2, skeletal, S1, S2, and S3 models in CHR model under
atmospheric pressure and adiabatic condition. Subtracted elementary reactions of S1, S2, and S3 are
shown in Table 2.
Downloaded By: [Matiej] At: 11:54 6 March 2011

CH4 oxidation process changes. Reaction paths for high temperatures are deacti-
vated and those for low temperatures are activated. Under the present conditions,
this switching occurred at 1221 K. Then, this reaction path keeps oxidizing CH4
and releases heat. Once temperature increases and activates the reaction paths for
high temperatures, autoignition occurs and temperature reaches the peak tempera-
ture again. This periodic heat release causes the continuous autoignition. Therefore,
heat release rate, heat loss, and residence time of the mixture are the key factors of
the combustion instability in the PSR. On the other hand, this periodic heat release
does not allow the reactor to be at a stable condition such as mild combustion
(Cavaliere and de Joannon, 2004) at low temperatures. Once the temperature is
sufficient to activate reaction paths for high temperatures, NOx (i.e., thermal NO)
is produced even if the inlet temperature is at low temperatures (<1300 K). In order
to extend the operation range of combustor with low emission, a closed-loop
controller is required.

3.3 Control of the PSR


The main goal of the control is to damp the oscillations. The oscillations
represent periodic trajectories in the state space. This behavior is called a nonlinear
limit cycle (Khalil, 2002). The equilibrium point shows our desired conditions, repre-
senting steady combustion at low temperatures. Therefore, damping the oscillations
is equivalent to making the equilibrium point asymptotically stable. There are two
choices to achieve the goal. One is the use of the nonlinear model and the other is
to use a linearization. The latter is used in the present study to reduce computational
load and achieve real-time capability for the proposed control strategy.
3.3.1 System analysis. To reduce computational load for the control
algorithm and for system analysis, two steps are taken. The first step is to find the
equilibrium point. The equilibrium point can be found by use of Differential
DILUTED LEAN METHANE/AIR COMBUSTION 13

Algebraic Equation (DAE) control (Wissel, 1996), which is not applicable in real-
time for this set-up. The next step is to linearize the system around the equilibrium
point using finite differences. Because / has a strong influence on the system it is
chosen as an input variable. YCH4 is chosen as an output variable. The linearized
discrete system with input and output variables can be written as

xkþ1 ¼ Axk þ B/k ð5Þ

YCH4;k ¼ Cxk ð6Þ

where x, A, B, and C represent the state vector consisting of temperature and Y of all
species in the system, the system matrix, input and output vectors, respectively. The
subscript k denotes the discrete time step. This system is used for control.
3.3.2 Model predictive control. As a control method model predictive con-
trol (MPC) is used. Figure 10 shows the MPC scheme. MPC is a control technique
which minimizes the difference between predicted and reference outputs by changes
of the system input. In order to obtain an optimal input variable a cost function (J) is
minimized using the linear Eqs. (5) and (6) for prediction. For our purpose, the input
Downloaded By: [Matiej] At: 11:54 6 March 2011

and output variables are / and YCH4 . J can be written as

Hp
X X
H u 1
ref
J¼ QðYCH4;k  YCH 4;k
Þ2 þ RD/2
k ð7Þ
k¼Hw k¼0

where Q, R, represent tuning parameters and superscript ref represents the reference
value. The outputs are compared in the prediction horizon Hw  k  Hp. The changes
of the inlet are determined in the control horizon 0  k  Hh. The control strategy
relies on finding the minimum of J yielding the input changes D/k . Only the
first input change D/0 is applied to the system (Eqs. [1] and [2]). The MPC is com-
bined with a Luenberger state-observer (for details see Maciejowski, 2001; Boyd
and Vandenberghe, 2004). In summary, the MPC calculates an input signal by

Figure 10 Schematic of the Model Predictive Control scheme.


14 T. WADA ET AL.

comparing the measured value YCH4;0 and the predicted values YCH4;k with a desired
ref
reference trajectory YCH 4;k
using an internal process model. In the present study the
input signal is updated every 10 ms, which is different from the time step used in the
solver of Eqs. (1) and (2). The optimization has to be done in every time step k,
which makes this a time critical part for real-time application.
3.3.3 Control strategy. Because the internal controller model is linear, the
nonlinear limit cycle cannot be predicted correctly. Thus, a special control strategy
to handle the nonlinear phenomenon is necessary. This strategy drives the system
near the equilibrium point before the MPC is activated. The MPC can stabilize
the system if the system is close enough to the equilibrium point (i.e., inside the
region of attraction). Once the MPC stabilizes the system, the influence from the
nonlinearity decreases and the capability of the MPC increases dramatically.
3.3.4 Control results. Figures 11 shows the T, / , YCH4 , and YCO2 profiles.
The first period, 0 < t < 2 s, demonstrates the oscillating behavior with inactive con-
troller. The observed peak in the temperature profile corresponds to results from the
previous section (see Figure 4). In the second period, 2 < t < 3.5 s, the / is lowered
to break the limit cycle (see Appendix). The system is driven into a stable condition
Downloaded By: [Matiej] At: 11:54 6 March 2011

using a set value of / ¼ 0.2. / is decreased gradually and reaches the set value at
t ¼ 2.8 s; meanwhile, the peak temperature decreases. The heat release decreases as
incoming fuel decreases. In addition, the heat release cannot increase the reactor
temperature to activate the reaction path at high temperatures mentioned previously.
Once residual gas from the previous period (0 < t < 2 s) is flown out, the system
reaches the equilibrium point. In the third period, 3.5 < t < 15 s, the MPC is activated
with a reference value of the original condition (/ ¼ 0.6). After / reaches the
reference value around 4 s, a flat temperature profile can be observed. During this

Figure 11 Influence of the MPC on temperature, / , fuel, and product profiles. The MPC was activated at
t ¼ 3.5 s and deactivated at t ¼ 15 s.
DILUTED LEAN METHANE/AIR COMBUSTION 15

Figure 12 Influence of the MPC on NO and N2O profiles. The time of activation and deactivation of the
Downloaded By: [Matiej] At: 11:54 6 March 2011

MPC is the same as Figure 11.

transition the nonlinear limit cycle is avoided, making the linearized model sufficient
for prediction. The result shows the significant effect of the MPC, which stabilizes
the system. The system remains at the equilibrium point and most of CH4 is con-
verted to the products. At the beginning of the third period, 15 < t < 20 s, the
MPC is deactivated. The system oscillates again. The oscillations represent the same
limit cycle as in the first period 0 < t < 2 s. This result proves that the system cannot
remain at low temperature without control. Figure 12 shows the influence of the
MPC on the NOx emissions during the same simulation shown in Figure 11. When
the MPC is activated, NO and N2O emissions are extremely small (<0.1 ppm). These
results show that the proposed control strategy using a linear model within the MPC
and a special starting trajectory is suitable to damp the oscillations and can reduce
the emissions significantly. In addition, the control strategy can be used for real-time
applications even if the combustion phenomenon is highly nonlinear.

4 CONCLUSIONS
In order to study possibilities of MPC on low temperature combustion, we
performed a numerical study of combustion with a lean highly diluted methane=
air mixture in a perfectly stirred reactor using a detailed chemical kinetic model in
a PSR with a virtual MPC. To represent the heat transfer to the walls in agreement
with experiments, we chose the following conditions: equivalence ratio (/) ¼ 0.6;
residence time of mixture in reactor (s) ¼ 0.5 s; molar fraction of N2 ¼ 0.9; tempera-
ture of incoming unburned gases and reactor, 1100 K; and heat loss coefficient
(a) ¼ 2  103 cal=(cm2Ks). The results lead to the following conclusions:

. A reduction in the chemical kinetic model was brought about by using step-by-
step reduction processes, CAS, PRS, and RPS, for number of carbon atoms,
16 T. WADA ET AL.

production rate, and rate of progress, individually. The obstained skeletal model
consisted of 32 species and 192 reactions and showed good agreement with the
result obtained using a reaction flow analysis with the reference model Ranzi
et al. (1994). By use of a reference mechanism for various fuels with this reduction
processes, characteristics of fuels could be emphasized with the same elementary
reactions.
. Continuous oscillation was observed when a ¼ 2  103 cal=(cm2Ks). This shape
was categorized as cusp-shaped oscillations and was in good agreement with
experimental investigation done by de Joannon et al. (2004). During the oscil-
lation periods, each peak reached a temperature that was relatively lower than
the adiabatic flame temperature even if CH4 was perfectly consumed during
periods of high temperature. This result can be attributed to the mixing of the
unburned mixture with the burned gas.
. The influence of each reaction path with activation energies set to infinity for spe-
cific reactions was equivalent to that of removing reactions from the model. Reac-
tions were specified using reaction flow analysis and the results showed that C1
reactions, which are reactions forming CH3O and CH3O2 from CH3, had a strong
influence on all the oscillations. Once only one of these reactions was subtracted
Downloaded By: [Matiej] At: 11:54 6 March 2011

from the kinetic mechanism, no oscillations were observed because of relatively


longer ignition delay time than s.
. A transition from steady to unsteady conditions in the temperature profile was
observed by using various values of the a. Above the temperature at which reac-
tion paths for low temperatures is not active, no oscillations were observed and
the MILD combustion mode (Cavaliere and de Joannon, 2004), was observed.
In contrast, once the temperature reached the temperature regime of reaction
paths for low temperatures, oscillations began because of the periodic heat pro-
duction caused by the reaction path at low temperatures. Therefore, this periodic
heat release governs the oscillations has to be suppressed to obtain the MILD
combustion mode at further low temperatures.
. An MPC strategy was applied to damp the oscillations (to stabilize the system). In
the controller the linearized system around the equilibrium point was used. To
handle the high nonlinearity a special starting trajectory was used. The / of the
incoming mixture was chosen as input variable and fuel concentration was chosen
as output variable. By applying the proposed control strategy the MILD combus-
tion mode was observed at low temperatures. During the MILD combustion
mode, the majority of the CH4 is converted to products and extremely low emis-
sions are observed. This shows that this MPC strategy is effective to damp the
oscillations and a method to extend the area of operations of low temperature
combustion.

APPENDIX

. The reduction processes, PRS and RPS, strongly depend on their threshold values.
Figure 13 shows the influence of threshold values in the PRS on peak tempera-
tures. The dashed line denotes the peak temperatures after the CAS and symbols
denote the peak temperatures after the PRS with corresponding threshold values.
DILUTED LEAN METHANE/AIR COMBUSTION 17

Figure 13 Influence of threshold value for reduction processes.


Downloaded By: [Matiej] At: 11:54 6 March 2011

Table 3 Set value of / and the earliest time for MPC


activation (tMPC)

Set value of / tMPC [s]

0.180 2.58
0.215 2.62
0.230 2.64

This result shows that the peak temperature dramatically decreases when the
threshold value is larger than 4  103 kmol=m3s. Therefore, the optimal
threshold value is 3  103 kmol=m3s. In the same manner, the optimal threshold
value, 2  106 kmol=m3s, for the RPS is chosen.
. Under the examined conditions the limit cycle is finally broken after the appear-
ance of the second peak in Figure 11 at t ¼ 2.6 s. At this point all remaining fuel is
consumed. After this peak the MPC can be activated and successfully drives the
system to the original set point of / ¼ 0.6. Table 3 shows the earliest time possible
for successful MPC activation. At these times, the second peak appears and the
MPC drives the system to the original set point and stabilizes it there.

ACKNOWLEDGMENT
The authors acknowledge valuable discussions with Prof. Eliseo Ranzi
(Dipartimento di Chimica, Materiali e Ingegneria Chimica, Politecnico di Milano,
Italy) and Dr. Mara de Joannon (Istituto di Ricerche sulla Combustione, C.N.R.,
Napoli, Italy), and financial support from the Deutsche Forschungsgemeinschaft
(DFG) to Sonderforschungsbereich (SFB) 686 at RWTH-Aachen University.
18 T. WADA ET AL.

REFERENCES
Anderlohr, J.M., Piperel, A., Pires da Cruz, A., Bounaceur, R., Battin-Leclerc, F., Dagaut, P.,
and Montagne, X. 2009. Influence of egr compounds on the oxidation of an hcci-diesel
surrogate. Proc. Combust. Inst., 32, 2851.
Boyd, S., and Vandenberghe, L. 2004. Convex Optimization. Cambridge University Press,
Cambridge, England.
Cantera. 2007. Cantera Code site. Retrieved 27 July, 2010 from http://code.google.com/p/
cantera/
Cavaliere, A., and de Joannon, M. 2004. Mild combustion. Prog. Energy Combust. Sci., 30, 329.
Correa, S.M. 1992. A review of nox formation under gas-turbine combustion conditions.
Combust. Sci. Tech., 87, 329.
Dagaut, P., and Cathonnet, M. 2006. The ignition, oxidation, and combustion of kerosene: A
review of experimental and kinetic modeling. Prog. Ener. Combust. Sci., 32, 48.
Dagaut, P., and Dayma, G. 2006. Mutual sensitization of the oxidation of nitric oxide and a
natural gas blend in a jsr at elevated pressure: Experimental and detailed kinetic modeling
study. J. Phys. Chem. A, 110, 6608.
Dagaut, P., and Togbe, C. 2009. Experimental and modeling study of the kinetics of oxidation
of butanol-n-heptane mixtures in a jet-stirred reactor. Energy & Fuels, 23, 3527.
Dayma, G., Togbe, C., and Dagaut, P. 2009. Detailed kinetic mechanism for the oxidation
Downloaded By: [Matiej] At: 11:54 6 March 2011

of vegetable oil methyl esters: New evidence from methyl heptanoate. Energy & Fuels,
23, 4254.
de Joannon, M., Cavaliere, T., Faravelli, A., Ranzi, E., Sabia, P., and Tregrossi, A. 2005.
Analysis of process parameters for steady operations in methane mild combustion
technology. Proc. Combust. Inst., 30, 2605.
de Joannon, M., Sabia, P., A. Tregrossi, A., and Cavaliere, A. 2004. Dynamic behavior of
methane oxidation in premixed flow reactor. Combust. Sci. Tech., 176, 769.
de Vries, J., and Petersen, E.L. 2007. Autoignition of methane-based fuel blends under gas
turbine conditions. Proc. Combust. Inst., 31, 3163.
Faravelli, T., Frassoldati, A., and Ranzi, E. 2003. Kinetic modeling of mutual interactions in
no-hydrocarbon low temperature oxidation. Combust. Flame, 132, 188.
Fernandes, R.X., Luther, K., and Troe, J. 2006. Falloff curves for the reaction ch3 þ o2(þm)
yields ch3o2(þm) in the pressure range 2–1000 bar and the temperature range 300–700 k.
J. Phys. Chem. A, 110, 4442.
Frassoldati, A., Faravelli, T., and Ranzi, E. 2003. Kinetic modeling of mutual interactions in
no-hydrocarbon high temperature oxidation. Combust. Flame, 135, 97.
Glarborg, P., Dam-Johansen, K., and Miller, J.A. 1995. The reaction of ammonia with
nitrogen dioxide in a flow reactor: Implications for the nh2 þ no2 reaction. Int. J. Chem.
Kinet., 27, 1207.
Glarborg, P., Dam-Johansen, K., Miller, J.A., Kee, R., and Coltrin, M. 1994. Modeling
the thermal denox process in flow reactors, Surface effects and nitrous oxide formation.
Int. J. Chem. Kinet., 26, 421.
Goettgens, J., Mauss, F., and Peters, N. 1992. Analytic approximations of burning velocities
and flame thicknesses of lean hydrogen, methane, ethylene, ethane, acetylene, and
propane flames. Proc. Combust. Inst., 24, 129.
Healy, D., Curran, H.J., Dooley, S., Simmie, J.M., Kalitan, D.M., Petersen, E.L., and
Bourque, G. 2008a. Methane=propane mixture oxidation at high pressures and at
high, intermediate and low temperatures. Combust. Flame, 155, 445.
Healy, D., Curran, H.J., Simmie, J.M., Kalitan, D.M., Zinner, C.M., Barrett, A.B., Petersen,
E.L., and Bourque, G. 2008b. Methane=ethane=propane mixture oxidation at high pres-
sures and at high, intermediate and low temperatures. Combust. Flame, 155, 435.
DILUTED LEAN METHANE/AIR COMBUSTION 19

Herbon, J.T., Hanson, R.K., Bowman, C.T., and Golden, D.M. 2005. The reaction of
ch3 þ o2: Experimental determination of the rate coefficients for the product channels
at high temperature. Proc. Combust. Inst., 30, 955.
Hulgaard, T., and Dam-Johansen, K. 1993. Homogeneous nitrous oxide formation and
destruction under combustion conditions. AIChE J., 39, 1342.
Ju, Y., Gou, X., Sun, W., and Chen, Z. 2009. An on-grid dynamic multi-timescale method
with path flux analysis for multi-physics detailed modeling of combustion. J. Combust.
Soc. Jpn., 51, 200.
Khalil, H.K. 2002. Nonlinear Systems. Prentice Hall, New York.
Krasnoperov, L., and Michael, J.V. 2004. High-temperature shock tube studies using multi-
pass absorption: Rate constant results for oh þch3, oh þch2, and the dissociation of
ch3oh. J. Phys. Chem A, 108, 8317.
Maciejowski, J.M. 2001. Predictive control with constraints. Pearson Hall.
Malte, P.C., and Pratt, D.T. 1974. Measurement of atomic oxygen and nitrogen oxides in
jet-stirred combustion. Proc. Combust. Inst., 15, 1061.
Peters, N., Paczko, G., Seiser, R., and Seshadri, K. 2002. Temperature crossover and
non-thermal runaway at two-stage ignition of n-heptane. Combust. Flame, 128, 38.
Petersen, E.L., Kalitan, D.M., Simmons, S., Bourque, G., Curran, H.J., and Simmie, J.M.
2007. Methane=propane oxidation at high pressures: Experimental and detailed chemical
kinetic modeling. Proc. Combust. Inst., 31, 447.
Downloaded By: [Matiej] At: 11:54 6 March 2011

Pilling, M. 1996. Radical-radical reactions. Annu. Rev. Phys. Chem., 47, 81.
Ranzi, E., Dente, M., Goldaniga, A., Bozzano, G., and Faravelli, T. 2001. Lumping
procedures in detailed kinetic modeling of gasification, pyrolysis, partial oxidation and
combustion of hydrocarbon mixtures. Prog. Ener. Combust. Sci., 27, 99.
Ranzi, E., Frassoldati, A., Granata, S., and Faravelli, T. 2005. Wide-range kinetic modeling
study of the pyrolysis, partial oxidation, and combustion of heavy n-alkanes. Ind. Eng.
Chem. Res., 44, 5170.
Ranzi, E., Sogaro, A., Gaffuri, P., Pennati, G., and Faravelli, T. 1994. A wide range modeling
study of methane oxidation. Combust. Sci. Tech., 96, 279.
Rutar, T., Kramlich, J.C., Malte, C, P., and Glarborg, P. 1996. Nitrous oxide emissions
control by reburning: Laboratory experiments. Combust. Flame, 107, 453.
Seshadri, K. 1996. Multistep asymptotic analyses of flame structures. Proc. Combust. Inst., 26, 831.
Smith, G.P., Golden, D.M., Frenklach, M., Moriarty, N.W., Eiteneer, B., Goldenberg, M.,
Bowman, C.T., Hanson, R.K., Song, S., Gardiner, J.W., Lissianski, V.V., and Qin, Z.
2006. Gri-mech 3.0. Retrieved 27 July 2010 from http://www.me.berkeley.edu/gri-mech/
version30/text30.html
Steele, R.C., Malte, P.C., Nicol, D.G., and Kramlich, J.C. 1995. Nox and n2o in
lean-premixed jet-stirred flames. Combust. Flame, 100, 440.
Sullivan, N., Jensen, A., Glarborg, P., Day, M.S., Grcar, J.F., Bell, J.B., Pope, C.J., and
Kee, R.J. 2002. Ammonia conversion and nox formation in laminar coflowing non-
premixed methane-air flames. Combust. Flame, 131, 285.
von Wissel, D. 1996. DAE control of dynamical systems: Example of a riderless bicycle. Ph.D.
thesis, Ecole nationale supérieure des Mindes de Paris, France.
Walch, S. 1993. Theoretical characterization of the reaction ch3 þ oh yields ch3oh yields
products: The lch2 þ h2o, h2 þ hcoh, and h2 þ h2co channels. J. Chem. Phys., 98, 3163.
Warnatz, J., Maas, U., and Dibble, R. 1996. Combustion. Springer, New York.
Yao, M., Zheng, Z., and Liu, H. 2009. Progress and recent trends in homogeneous charge
compression ignition (hcci) engines. Proc. Combust. Inst., 35, 398.
Zhao, F., Asmus, T.N., Assanis, D.N., Dec, J.E., Eng, J.A., and Najt, P.M. 2003. Homo-
geneous Charge Compression Ignition (HCCI) Engines. SAE International, Warrendale,
PA.

Das könnte Ihnen auch gefallen