simulation of a gas turbine combustor test rig using a reactor...
DESCRIPTION
Use of gas turbines as one of the most effective power generation technologies has ecological concerns caused by polluting combustion products. To reduce emissions different fuel compositions are being constantly investigated and gas turbines are developed by means of experiments or less expensive numerical simulations. Combustion processes can be modeled in computational fluid dynamics (CFD) with a good accuracy but it is time consuming and rather complicated in case of detailed chemistry. To overcome this issue a processing of CFD solution can be applied for a further building of equivalent chemical reactor networks (CRN) that allow to reduce calculation times and take minor species into account. The aim of this work is to choose a proper technique of CRN set-up and apply it for engineering tasks with the software tool 'LOGEsoft ReactorNetwork'. The first part of the thesis is devoted to investigation of existing CRN approaches, CFD processing instruments and testing and improvement of the 'LOGEsoft ReactorNetwork'. That software is successfully examined on the Sandia Flame D and a parameter study of the reactor network is carried out. The second part involves mechanism validation for methane/hydrogen mixtures and development of an equivalent reactor network for the Siemens atmospheric combustion test rig that serves as an experimental facility for enhancement of the 3rd generation dry low emission burner. The obtained CRN is validated against experimental data of NOx measurements and it showed reasonable results with deviations. A parameter study and mechanism sensitivity of the model is also conducted and some ways for the future improvement are suggested.TRANSCRIPT
Simulation of a Gas Turbine Combustor Test Rig using a Reactor
Network Approach with Detailed Chemistry
by
Oleg Bosyi
Master's Thesis in Power Engineering
Industrial supervisor:
Dipl.Ing. Lars Seidel
LOGE GmbH, Cottbus, Germany
Academic supervisor:
Prof. Dr.Ing. F. Mauß
Chair of Thermodynamics and Thermal Process Engineering,
Brandenburg University of Technology CottbusSenftenberg, Cottbus, Germany
In collaboration with:
Thommie Nilsson
LOGE AB, Lund, Sweden
Cottbus, July 2014
Abstract
Use of gas turbines as one of the most effective power generation technologies has
ecological concerns caused by polluting combustion products. To reduce emissions
different fuel compositions are being constantly investigated and gas turbines are
developed by means of experiments or less expensive numerical simulations. Combustion
processes can be modeled in computational fluid dynamics (CFD) with a good accuracy
but it is time consuming and rather complicated in case of detailed chemistry. To
overcome this issue a processing of CFD solution can be applied for a further building of
equivalent chemical reactor networks (CRN) that allow to reduce calculation times and
take minor species into account.
The aim of this work is to choose a proper technique of CRN setup and apply it for
engineering tasks with the software tool 'LOGEsoft ReactorNetwork'.
The first part of the thesis is devoted to investigation of existing CRN approaches,
CFD processing instruments and testing and improvement of the 'LOGEsoft
ReactorNetwork'. That software is successfully examined on the Sandia Flame D and a
parameter study of the reactor network is carried out.
The second part involves mechanism validation for methane/hydrogen mixtures and
development of an equivalent reactor network for the Siemens atmospheric combustion
test rig that serves as an experimental facility for enhancement of the 3 rd generation dry
low emission burner. The obtained CRN is validated against experimental data of NOx
measurements and it showed reasonable results with deviations. A parameter study and
mechanism sensitivity of the model is also conducted and some ways for the future
improvement are suggested.
Keywords: gas turbine, combustion, emissions, NOx, CFD, chemical reactor, reactor
network, CRN, ERN, hydrogenenriched.
2
Acknowledgment
This thesis is a final part of the Master program of Power Engineering at
Brandenburg University of Technology CottbusSenftenberg. The work is carried out at
LOGE GmbH (Cottbus, Germany) in cooperation with LOGE AB (Lund, Sweden) and
Siemens Industrial Turbomachinery AB (SIT) (Finspång, Sweden).
I would like to thank my supervisors Dipl.Ing. Lars Seidel and Prof. Dr.Ing. F.
Mauß for giving me such a pleasant experience and support. I also want to thank M.Sc.
Andrea Matrisciano at LOGE GmbH and M.Sc. Cathleen Perlman and Thommie Nilsson at
LOGE AB for a great help during the work.
In addition, I would like to express my appreciation of all the encourage from Dr.
Daniel Lörstad at SIT for provision of necessary data for the investigation.
Cottbus, July 2014
Oleg Bosyi
3
Statement of Authentication
The author of this paper declares that he/she has prepared the submitted work
himself/ herself, unassisted and without using any other resources other than those
indicated. All the direct or indirect cited information from other sources (including
electronic sources) is duly acknowledged without exceptions. The material, in this or
similar form, has not been previously submitted, either in full or in part, for other exams
at this or any other academic institution.
Place, Date Signature
4
Contents
Abstract..............................................................................................................................2
Acknowledgment................................................................................................................3
Statement of Authentication...............................................................................................4
Contents.............................................................................................................................5
List of Figures.....................................................................................................................7
List of Tables......................................................................................................................8
Nomenclature.....................................................................................................................9
Abbreviations...................................................................................................................10
1. Introduction.................................................................................................................11
1.1 Background...........................................................................................................11
1.2 Objectives of this work..........................................................................................12
1.3 LOGE AB................................................................................................................12
1.4 Siemens Industrial Turbomachinery AB.................................................................12
2. Theory..........................................................................................................................13
2.1 Gas turbine principles............................................................................................13
2.2 Combustion...........................................................................................................14
2.3 Governing equations of fluid flow and heat transfer..............................................14
2.4 Flame basic definitions..........................................................................................16
2.5 Emissions...............................................................................................................17
2.5.1 Carbon dioxide...............................................................................................18
2.5.2 Carbon Monoxide..........................................................................................18
2.5.3 Nitric Oxides..................................................................................................18
2.6 Reactor models......................................................................................................20
2.6.1 PSR Perfectly Stirred Reactor.......................................................................20
2.6.2 PFR PlugFlow Reactor................................................................................21
2.6.3 PaSR – Partially Stirred Reactor.....................................................................22
2.7 Reactor network building......................................................................................23
3. Software overview........................................................................................................26
3.1 Interface and features............................................................................................26
3.2 Testing and improvement......................................................................................28
3.3 Modules data and settings.....................................................................................29
4. Sandia Flame D............................................................................................................30
4.1 Reactor network setup. Approach 1......................................................................32
4.2 Reactor network setup. Approach 2......................................................................36
4.2.1 Parameter study..................................................................................................39
5. Siemens atmospheric combustion test rig.....................................................................44
5
5.1 Mechanism validation............................................................................................45
5.2 Reactor network setup..........................................................................................46
5.3 Parameter study.....................................................................................................49
6. Discussion and conclusions..........................................................................................52
7. Future works................................................................................................................54
References........................................................................................................................55
6
List of Figures
Figure 1. Gas turbine structure.........................................................................................13
Figure 2. Theoretical PSR.................................................................................................19
Figure 3. Theoretical PFR.................................................................................................21
Figure 4. Conceptual diagram of PaSR reactor.................................................................23
Figure 5. Flame Zone Mapping Based on CFD Result with Swirl Angle of 450 .................24
Figure 6. Schematic Layout of 6Element CRN Model......................................................25
Figure 7. LOGEsoft ReactorNetwork main interface.........................................................26
Figure 8. Available modules.............................................................................................26
Figure 9. Example of a reactor network...........................................................................27
Figure 10. Reactor data....................................................................................................27
Figure 11. Calculation and output parameters.................................................................28
Figure 12. Reactor output................................................................................................28
Figure 13. Flame D.. .......................................................................................................30
Figure 14. Sandia Flame D validation..............................................................................31
Figure 15a. Sandia Flame D Approach 1. Principle..........................................................32
Figure 15b. Sandia Flame D Approach 1. CRN.................................................................33
Figure 16. (a): temperature profile; (b): CH4 concentration, linear; (c): CH4 conc.Logarithmic; (d): OH conc.; (e): CO2 conc.; (f): H2O conc...............................................34Figure 17. (a): CH4 concentration, linear; (b): CH4 conc. Logarithmic; (c): OH conc.; (d):CO2 conc.; (e): H2O conc..................................................................................................35
Figure 18a. Sandia Flame D Approach 2. Principle..........................................................36
Figure 18b. Sandia Flame D Approach 2. CRN.................................................................37
Figure 19. (a): temperature profile; (b): CH4 concentration, linear; (c): CH4 conc.Logarithmic; (d): OH conc.; (e): CO2 conc., (f): H2O conc...............................................38
Figure 20. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2
conc.; (e): H2O conc.........................................................................................................40
Figure 21. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2
conc.; (e): H2O conc.........................................................................................................41
Figure 22. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2
conc.; (e): H2O conc.........................................................................................................42
Figure 23. Schematic layout of the combustion test rig....................................................44
Figure 24. 2D model of the test rig combustion chamber.................................................45
Figure 25. (a): CH4,; (b): CO2; (c): CO; (d): H2O concentrations vs hydrogen content inthe mixture.......................................................................................................................46
Figure 26. Velocity magnitude of the flow and schematic zones mapping........................47
Figure 27. Combustion rig chemical reactor network.......................................................47
Figure 28. Thermal NOx formation trend along the central axis......................................48
7
Figure 29. (a): NOx vs hydrogen content; (b): Temperature vs hydrogen content...........48
Figure 30. (a): NOx/Temp. vs inlet mass flow rate; (b): NOx/Temp. vs inlet temperatureof the mixture..................................................................................................................50
Figure 31. (a): NOx and (b): temperature sensitivity to different reaction schemes.........51
List of Tables
Table1. Sandia Flame D flow conditions..........................................................................30
Table 2. CPU times in case of two schemes in the first approach.....................................36
Table 3. Time step size influence (Output results for reactor Nr 9).................................39
Table 4. Mechanisms CPU times......................................................................................43
Table 5. Time step size influence (Output results for outlet)...........................................49
Table 6. CPU times..........................................................................................................50
8
Nomenclature
ρ – mixture mass density
u – flow velocity vector
t – time
x – spatial coordinate
g – gravitational acceleration field
p – thermodynamic pressure
I – unit tensor
σ – stress tensor
cp – specific heat
T – temperature
λ – thermal conductivity
Υi – mass fraction
ω – molar net change rate
M – molar mass
W – molecular weight
Lei – Lewis number of species I
m – mass flow rate
τ – residence time
Ax – crosssectional flow area
C p – mean heat capacity per unit mass of gas
Qe – heat flux from the surroundings
ae – area per length unit
c – concentration
τc – chemical reaction time
τmix – micromixing time
V – reactor volume
Acell flow – cell surface area through which the flow goes
∆t – time step size
9
Abbreviations
CFD – computational fluid dynamics
CRN – chemical reactor network
ERN – equivalent reactor network
SIT – Siemens Industrial Turbomachinery
PSR – perfectly stirred reactor
PFR – plug flow reactor
PaSR – partially stirred reactor
RANS – Reynoldsaveraged Navier–Stokes
URANS – unsteady Reynoldsaveraged Navier–Stokes
CPU – central processing unit
mf – mole fraction
DLE – dry low emission
10
1. Introduction
1.1 Background
One of the most important power generation technologies is utilization of gas
turbines – facilities that burn gaseous fuels and convert chemical energy into electricity
and heat. However, in terms of environment safety different combustion products can be
undesirable pollutants such as COx and NOx. Gas turbines mostly operate on natural gas
as a fuel which composition can vary depending on the location it has been transported
from. Also it is essential to use additives or diluents in its composition to improve
combustion properties and reduce polluting emissions, for example, by blending natural
gas with hydrogen. When burning there is a very complex process of interaction between
the flow and flame, the chemical reactions depend on many factors. That leads to a need
of adequate estimation and study of the combustion process.
An effective way to avoid expensive and time consuming fullscale experiments is
numerical simulation. Nowadays it is possible to predict the most important flame
properties such as heat release, velocity and main species concentration by means of
computational fluid dynamics (CFD) with a good accuracy. However, in order to predict
minor species such as CO, NOx, SOx and so forth it is necessary to have a more detailed
view of fuel chemical composition and combustion reaction chemistry. CFD has
computational limits such as a huge time consumption, so there is a need of an
alternative approach. Use of a chemical reactor network (CRN) can be such a method. Its
concept is based on extraction of an ‘‘equivalent’’ network of ideal chemical reactors as a
simple flow model from CFD simulation that previously was performed using a simplified
kinetics mechanism on a fine grid. The resulting CRN significantly reduces computational
times to calculate minor species concentrations using detailed chemical reaction schemes.
11
1.2 Objectives of this work
1. Investigate different existing approaches of building up equivalent reactor
networks based on CFD solution.
2. Test the software 'LOGEsoft ReactorNetwork' by LOGE AB and improve it.
3. Set up a CRN for the Siemens atmospheric combustion test rig, perform parameter
studies and evaluate sensitivity to different reaction schemes.
1.3 LOGE AB
LOGE, Lund Combustion Engineering, is a software development company based in
the cities of Lund, Sweden and Cottbus, Germany. The main competence of the company
is the development of software tools used for simulating chemical processes, such as the
combustion in technical devices (engines, furnaces), or chemical processes on surfaces
(christal growth, catalysts). The company's main simulation tool is its own software suite
LOGEsoft, which is a comprehensive solution for detailed chemical kinetics modeling of
engineering applications. [15]
1.4 Siemens Industrial Turbomachinery AB
Siemens Industrial Turbomachinery AB (SIT) develops, manufactures, markets and
services individual gas turbines all the way to complete power plants on a global market.
The business comprise of approximately 2800 employees with an annual turnover of
more than 1 billion EUR. The head office is located to Finspång.
SIT supply customers all over the world with various gas turbine solutions. The
turbines are characterized by low environmental impact and high efficiency. Within the
Siemens group, SIT is responsible for industrial steam turbines with power of 60180 MW
and for gas turbines with power levels of 1550 MW.
SIT gas turbine offering consists mainly of five lines: SGT500, SGT600, SGT700,
SGT750 and SGT800. [1]
12
2. Theory
Combustion is a complex physical and chemical process of converting raw materials
into combustion products during exothermic reactions, accompanied by a heat release.
The chemical energy stored in the components of the original mixture can be released in
form of heat and light radiation. The luminous zone is called flame front or flame.
2.1 Gas turbine principles
Figure 1. Gas turbine structure. [2]
A gas turbine is an engine of continuous action in which scapular device energy of
compressed and/or warm gas is transformed to mechanical work on its shaft. Burning of
fuel can occur both out of the turbine, and inside of the turbine. Basic elements of its
design are: a rotor (the working shovels fixed on disks) and a stator made in form of the
leveling device.
Gas turbines are used as parts of gasturbine engines, stationary gas turbine units
and steamgas turbine units. A gas turbine unit consists of two main parts: a power
turbine and a generator. The stream of gas of high temperature moves the shovels of the
power turbine. Heat used by a heat exchanger or a copperutilizator provides increase in
the general efficiency of the whole plant. Gas turbines can work both with liquid and with
13
gaseous fuel: in a usual operating mode — natural gas, and in reserve (emergency) mode
— it is automatically switched to a diesel fuel.
2.2 Combustion
Three factors necessary for combustion:
fuel
oxygen
temperature
Flammable mixture: fuel combined with a sufficient amount of oxygen is ignited at a
certain temperature. The main combustible components in the fuel are: carbon (C),
hydrogen (H2) and mixtures formed during the combustion.
Complete combustion, also known as stoichiometric combustion, in theory it is an
ideal combustion process as a result of which the fuel burns completely. An example of
complete burning can be expressed as:
CH4(g) + 2O2(g) CO→ 2(g) + 2H2O(g)
Chemical reaction schemes, or mechanisms, describe in a stepwise manner the exact
collisions and events that are required for the conversion of reactants into products.
Mechanisms achieve that goal by breaking up the overall balanced chemical equation into
a series of elementary steps. An elementary step is written to mean a single collision or
molecular vibration that results in a chemical reaction. [3]
2.3 Governing equations of fluid flow and heat transfer
If a chemically reacting flow is considered, the system at each point in space and
time is completely described by specification of pressure, density, temperature, velocity of
the flow, and concentration of each species. These properties can be changing in time and
space. The changes are the result of fluid flow (called convection), chemical reaction,
molecular transport (e.g., heat conduction, diffusion, and viscosity), and radiation. A
mathematical description of flames therefore has to account for each of these processes.
Some properties in reacting flows are characterized by the fact that they are
conserved. Such properties are the energy, the mass, and the momentum. Summation
over all the processes that change the conserved properties leads to the conservation
equations, which describe the changes in reacting flow; accordingly, these equations are
often called the equations of change. These equations of change (an extended set of the
14
socalled NavierStokes Equations) are the general starting point for mathematical
descriptions of chemically reacting flows. Because all systems are described by the
conservation equations, the main difference from one system to another are the boundary
conditions and physicochemical conditions. [4]
Mass conservation:
∂ ρ∂ t
+∂ ( ρu )
∂ x=0
(2.1)
Momentum conservation:
∂ ρu∂ t
+∂ ( ρuu )
∂ x=pg−
∂P∂ x , (2.2)
where ρ is the mixture mass density, u the flow velocity vector, t the time, x the
spatial coordinate, g the gravitational acceleration field, P = pIσ and is the stress
tensor with p as thermodynamic pressure, I the unit tensor and σ is the stress tensor
which is calculated using Fick's Law of friction for a compressible mixture.
Mass conservation of species:
∂ (ρY i )∂ t
+∂ ( ρuY i )
∂ x= ∂
∂ x ( λLe ic p
∂Y i
∂ x )+ Ri (2.3)
Energy conservation of species:
∂ (ρc pT )∂ t
+∂ (ρuc pT )
∂ x= ∂
∂ x ( λ ∂T∂ x )+ ∂
∂ x ( λc p∑i=1
N s
( 1Lei
−1)c p,iTY i
∂ x )−∑
i=1
N s
( hi ωiMW i
ρ+ρug+σ (∇ u )+(∂ p∂ t
+u∇ p))(2.4)
In these equations, cp is the specific heat, T is the temperature and λ is the thermal
conductivity of the mixture, Υi = ρi/ρ is the mass fraction, h the enthalpy, ω the molar
net change rate, M the molar mass, W the molecular weight, and Lei is the Lewis number
of species i.
Lewis number is a dimensionless number which expresses the ratio of thermal
diffusivity ( /cλ p) to species mass diffusivity ( Dρ m,i) as:
Lei=λ
ρDm,i c p (2.5)
where Dm,i is the mixtureaveraged diffusion coefficient. Effects of Lewis number
become substantial when the thermal and mass diffusivity of the fuel differ and Le 1≠ . [5]
15
2.4 Flame basic definitions
In combustion processes, fuel and oxidizer (typically air) are mixed and burned. It is
useful to identify several combustion categories based upon whether the fuel and oxidizer
is mixed first and burned later (premixed) or whether combustion and mixing occur
simultaneously (nonpremixed). Each of these categories is further subdivided based on
whether the fluid flow is laminar or turbulent.
Laminar Premixed Flames: In laminar premixed flames, fuel and oxidizer are
premixed before combustion and the flow is laminar.
A premixed flame is said to be stoichiometric, if fuel (e.g., a hydrocarbon) and
oxidizer (e.g., oxygen O2) consume each other completely, forming only carbon dioxide
(CO2) and water (H2O). If there is an excess of fuel, the system is called fuelrich, and if
there is an excess of oxygen, it is called fuellean.
Premixtures of fuel and air are characterized by the air equivalence ratio, ,Ф
(sometimes air number) or the reciprocal value, the fuel equivalence ratio [4]
Ф=(Fuel /Air )actual
(Fuel / Air )stoichiometric (2.6)
Accordingly, premixed combustion processes can now be divided into three groups,
rich combustion: Ф>1
stoichiometric combustion: Ф=1
lean combustion: Ф<1
The burning of freely burning premixed laminar flat flames can be characterized by
the laminar burning velocity vL (e.g., in m/s); other names in the literature are flame
velocity or flame speed. It depends only on the mixture composition, the pressure and the
initial temperature.
Sometimes premixed flame fronts burn and propagate into a turbulent fluid flow. If
the turbulence intensity is not too high, curved laminar premixed flame fronts are
formed. The turbulent flame can then be viewed as an ensemble of many premixed
laminar flames.
The advantage of premixed combustion is that much greater control of the
combustion is possible. By lean premixing (Ф<1), high temperatures are obtained and
hence combustion with low production of soot and unburned hydrocarbons is
accomplished. However, due to higher heat more NOx is formed.
16
Despite the advantages, premixed combustion is not widely used because of the
potential for accidental collection of large volumes of premixed reactants, which could
burn in an uncontrolled explosion.
Laminar Nonpremixed Flames: In laminar nonpremixed flames (laminar diffusion
flames), fuel and oxidizer are mixed during the combustion process itself. The flow is
laminar.
Turbulent Nonpremixed Flames: In this case nonpremixed flames burn in a turbulent
flow field, and for low turbulence intensities the socalled flamelet concept can be used.
Nonpremixed flames are mostly used in industrial furnaces and burners. Unless very
sophisticated mixing techniques are used, nonpremixed flames show a yellow
luminescence, caused by glowing soot particles formed by fuelrich chemical reactions in
the rich domains of the nonpremixed flames. [4]
Adiabatic Flame Temperature is the flame temperature under constant pressure, with
no heat exchange and when the combustion is complete. This is the highest possible
flame temperature.
17
2.5 Emissions
One of the driving factors in modern gas turbine design is reducing emissions, and
the combustor is the primary contributor to a gas turbine's emissions. Generally speaking,
there are five major types of emissions from gas turbine engines: smoke/soot, carbon
dioxide (CO2), carbon monoxide (CO), unburned hydrocarbons (UHC), and nitrogen
oxides (NOx , which is a sum of NO and NO2) .
2.5.1 Carbon dioxide
It is a harmful pollutant and takes place in the global warming process, it is a product
of the combustion process and is primarily mitigated by reducing fuel usage. On average,
1 kg of jet fuel burned produces 3.2 kg of CO2. Carbon dioxide emissions will continue to
drop as manufacturers make gas turbine engines more efficient.
2.5.2 Carbon Monoxide
Carbon monoxide is dangerous for human health and environment in terms of
poisoning. Carbon monoxide is often formed in fuelrich conditions, because of the lack of
adequate oxygen to produce CO2. Although, in stoichiometric or fairly lean conditions, a
high amount of CO is found due to the dissociation of CO2. In reality, CO emissions tend
to be the highest at lowload conditions. The reason might be due to:
• Deficient burning rates due to a too small equivalence ratio, or too low residence
time.
• Nonuniformity in equivalence ratio which creates spots with too lean or too rich
mixtures.
Combustion efficiency and thus CO emissions are highly influenced by engine and
combustor inlet temperatures, combustion pressure and primaryzone fuel to air ratio (or
equivalence ratio). [5]
2.5.3 Nitric Oxides
This is a pollutant that participates in a chain reaction removing ozone from the
stratosphere, that lead to increase of ultraviolet radiation on the earth's surface.
18
NOx is a common term to describe NO, NO2 and N2O. Gas turbine combustor mostly
contains NO. There are four general mechanisms of NOx formation: Thermal NOx, Prompt
NOx, N2O and Fuel NOx.
Thermal NOx (Zeldovich NOx): it usually has a large concentration at temperatures
more than 1750 K. For this type there are three reactions of formation:
O + N2 NO + N→
N + O2 NO + N→
N + OH NO + H→
Formation of thermal NOx is largely controlled by flame temperature. As flame
temperature rises, NOx production boosts up. Nevertheless, although temperatures are
higher at the rich side, thermal NO peaks at the lean side. This is because of the
competition between fuel and nitrogen for the available oxygen. On the lean side, there is
an excess of oxygen which can be consumed by nitrogen. On the rich side, however,
oxygen is mainly used by the fuel. [7] When the air is preheated thermal NOx is the first
NOx formation mechanism.
Prompt NOx (Fenimore NOx): in temperature lower than 1800 K, HCN is oxidized to
NO in the flame front and prompt NOx is formed:
CH + N2 HCN + N→ 2
HCN + O2 NO + ...→
Prompt NOx is important in fuel rich condition and is formed in relatively low
temperature (about 1000 K). [8]
Fuel NOx: this type of NOx is associated with the presence of N2 in the fuel and in
general is not important for gas turbines because of low nitrogen presence in natural gas
and other common fuels.
Nitrous NOx (from N2O): N2O is important in high pressure and high temperature
conditions. Emission of N2O is not significant, but it can serve as an intermediate to NOx
emissions. [9]
19
2.6 Reactor models
2.6.1 PSR Perfectly Stirred Reactor
Figure 2. Theoretical PSR. [10]
The perfectly stirred reactor is an ideal chemical reactor in which perfect mixing is
achieved inside the control volume. This means that the composition within the reactor is
everywhere the same [11]. The gas from the inlet is instantly mixed with the existing
reactor content and the composition in the outlet will be the same as in the reactor, so
called back mixing [12]. The mixture in a perfectly stirred reactor is blended with the
combustion products and heated so quickly that in passing through the chamber with a
sampling probe it is difficult to locate any regions with composition or temperature
different than in other regions. Due to intense recirculation in the chamber, no defined
directions of the flow can be distinguished either [13].
The basic equations in the PSR model are the following.
Mass conservation:
min−mout=0(2.7)
with m indicating the mass flow rate and indexes in and out representing the inlet
and outlet flows from the control volume, respectively.
Considering [accumulation] = [generation] + [in] − [out], the above equation
becomes:
ωiMW iV−m (Y in,i−Y out,i)=0(2.8)
for i=1, 2, …, Ns,
where V is the volume of the reactor.
This creates Ns equations with Ns + 1 unknowns. The additional equation is obtained
from the energy balance.
Energy conservation of species:
20
dTdt
=1c p [ 1τ ∑i=1
N s
Y in,i (hin,i−hi )−∑i=1
N s
( hi ωiMW i
ρ−q 'lossρu )]
(2.9)
where
τ=ρu /m (2.10)
is the residence time and
ρ=pMWmix /RT (2.11)
is the mixture density, and
MW mix=1/∑i=1
N s
(Υ i /MW i ) (2.12)
is the heat loss flux. [14]
2.6.2 PFR PlugFlow Reactor
Figure 3. Theoretical PFR.
In the ideal plugflow reactor or tube reactor no recirculation occurs and the flow is
homogeneous with respect to velocity and species concentration in a radial direction from
the axis of symmetry. The physical interpretation is that the flow is trapped in between
two membranes/pistons that move with the same velocity as the flow along the tube, thus
inhibiting any recirculation/mixing in the axial direction of the tube. [16]
There is a clear analogy to an ideal perfectly mixed mixing tube of a gas turbine
burner. For the plugflow reactor some assumptions can be made for ignition delay
applications:
• Steady state, onedimensional flow
• Only gasphase reactions
• Ideal gas behavior
21
• Ideal frictionless flow
Applying the above simplifications the mass conservation equation can be written as:
ρudA x
dx+ρA x
dudx+uA x
dρdx
=0(2.13)
Ax is the crosssectional flow area. The conservation of species can be formulated as
ρuA x
dY i
dx=W i ω i Ax
(2.14)
The energy equation is formulated as
ρuA x(∑i=1
N s
hidY i
dx+C p
dTdx
+ududx )=aeQe
(2.15)
C p is the mean heat capacity per unit mass of gas, Qe the heat flux from the
surroundings to the outer wall of the tube whose area per length unit is ae. [17]
2.6.3 PaSR – Partially Stirred Reactor
In many practical combustion devices, e.g., gas turbines, the characteristic time
scales for mixing are of the same order of magnitude as the time scales for chemical
kinetics. When modeling such practical combustion devices it is important to account for
both effects. However in order to include detailed reaction chemistry, simplifying
assumptions regarding the fluid flow description are necessary to avoid excessive
computational and storage expenses. The partially stirred reactor (PaSR) is one such
model based on the probability density function (PDF) transport equation of the physical
quantities, assuming statistical spatial homogeneity. The model accounts for mixing and
is computationally efficient for large coupled chemical reaction mechanisms involving
many chemical species.
The PaSR model can be derived from the one point joint scalar PDF. The PDF
equation is solved numerically using a Monte Carlo particle method with time splitting
techniques. This method involves approximating the PDF by an ensemble of stochastic
particles, and has been successfully exploited for solving high dimensional PDF equations.
[18]
22
Figure 4. Conceptual diagram of PaSR reactor (the reaction zone is painted). [19]
In the PaSR approach, a computational cell is split into two different zones: in one
zone all reactions occur, while in the other one there are no reactions (Fig. 4). Therefore,
the composition changes due to mass exchange with the reacting zone. In addition. the
reaction zone is treated as a PSR, in which all reactants are assumed to be perfectly
mixed with each other. This allows to neglect any fluctuations when calculating the
chemical source terms. Three average concentrations are presented in the reactor, the
mean mixture concentration of the feed c0, the mixture concentration in the reaction zone
c, the mixture concentration at the exit of the reactor c1.
The whole combustion process is regarded as two processes. In the first process
initial concentration in the reaction zone changes from c0 to c, in the second process the
reacted mixture (with concentration c) is mixed with the unreacted mixture (with
concentration c0 by turbulence), the results is the averaged concentration c1. The reaction
rate of this computational cell is determined by the fraction of the reactor in this cell. It
seems quite clear that it should be proportional to the ratio of the chemical reaction time
τc to the total conversion time in the reactor, i.e. the sum of the micromixing time τmix
and reaction time τc:
κ i=τc
τ c+τmix (2.16)
The micromixing time τmix characterizes the exchange process between reactant
mixture and unburnt mixture. The overall reaction rate ω and the homogeneous
reaction rate ω of this computational cell have the following relationship: [20]
ci1−c i
0
dt=ωi=k i ωi
(2.17)
23
2.7 Reactor network building
In this section different techniques of CFD data extraction with further equivalent
chemical reactor networks development are described. In principle, a CRN is developed
by analyzing a flow field obtained with CFD. Then regions with certain characteristics are
identified and modeled by chemical reactors, as a rule by PSRs and PFRs.
Method 1: On the most basic level, for very simple 2D geometries with no
recirculating flows, the streamlines are divided into several plug flow reactors. The
division is refined until no considerable change is observed in the results; or in other
words a sensitivity analysis is implemented via the number of reactors. [21]
However, this approach is not suitable for complex geometries where a combination
of recirculation zones play a big role and should be taken into account.
Method 2: Falcitelli et. al. [22] have divided the full combustor into many small
perfectly stirred reactors. In this method, the flow field is broken down into many small
regions based on temperature and composition parameters. The grouping is done
regardless of the geometrical properties. The recycling flow (the flow that enters one
reactor from another) is directly taken from the CFD calculations between the adjacent
cells. As every few cells are grouped into one reactor, the network's resolution would not
be far from the CFD's resolution therefore making it easier to obtain accurate results. [21]
In case of relatively simple flows this method makes sense, nevertheless, if the flow
field is complicated and consists of thousands of cells it seems to be impossible to build
up a network manually with several hundred reactors after a preliminary extraction of
mass exchange and other parameters for each cell. This approach definitely needs to be
automated by using a certain code coupled with computational resources.
Method 3: Thanh Hao et. al. [23] have suggested to analyze the flow field
information in CFD in order to determine combustion zones in the combustor. A gas
turbine combustor can be divided into zones based on the flow velocity, temperature and
species concentration. Figure 5 displays the flame shape and flow of gas from one zone to
another.
Figure 5. Flame Zone Mapping Based on CFD Result with Swirl Angle of 450. [23]
24
Afterwards, the CRN has been built up and schematically shown in the Figure 6.
Figure 6. Schematic Layout of 6Element CRN Model. [23]
The method allows to simplify complicated schemes and manually extract necessary
parameters. The authors have conducted validation of this approach on a variety of flame
modifications against experiments and a good agreement has been obtained.
25
3. Software overview
3.1 Interface and features
The software tool 'LOGEsoft ReactorNetwork' developed by LOGE AB aims for
creation of reactor networks. For all the further work in this thesis the version v1.00.006
was used. The Figure 7 shows its interface.
Figure 7. LOGEsoft ReactorNetwork main interface.
In the Figure 8 the modules available to be connected into a network are listed. The
Stochastic PSR has not been implemented yet at the time of this work.
Figure 8. Available modules.
26
Some modules in a bigger scale can be seen in the Figure 9. A fuelair mixture inlet is
connected with PSRs. The mass exchange between the reactors is expressed in kg/s and
given by the user. Also, for each module, except mixers and splitters, an initial gas
composition with its temperature and pressure need to be specified. Figure 10
demonstrates what inputs a reactor has. A PSR requires specification of either volume or
residence time. For a PFR the geometry is represented as length, diameter and surface
area.
Figure 9. Example of a reactor network.
Figure 10. Reactor data.
The user may choose what tolerance the calculation should have as well as the time
step size. The time step size does not serve for chemistry calculation but for the flow.
Output frequency defines time steps to be written in output files. These options are
shown in the Figure 11.
27
Figure 11. Calculation and output parameters.
Figure 12. Reactor output.
As shown in the Figure 12 the output for each reactor can be seen as a function of
time.
3.2 Testing and improvement
First of all, the tool was needed to be validated before its implementation in real
cases. The main purpose of the testing was to prove that it is working stably and all
balances are kept. An examination of code and execution of that code in various
environments and conditions has been conducted. After a debugging process some new
features have been included to enhance performance and extend possibilities of the
product.
28
3.3 Modules data and settings
All the chemical reactor networks in this thesis are built from homogeneous
perfectlystirred reactors using transient calculation. In terms of calculation simplification
it was decided to specify volumes since there was no option to take residence times
directly from the CFD.
Mass flow rates of a flow have been extracted from the CFD for each cell (and further
summarizing of the corresponding cells) with the formula:
m=ρVAcellflow (2.17)
where V is a reactor volume, Acell flow – cell surface area through which the flow goes.
Finally, the following parameters have been specified for the reactor networks:
• flow boundaries: inlet mixture composition, temperature and pressure (keep
constant).
• mass flow rates between modules
• PSR/PFR: volume/geometry, initial mixture composition (air was used),
temperature and pressure (change with time and mostly affect only on the convergence
speed).
In all schemes in this work a spark of 0.01 s long in the first PSR was applied for
ignition and a value of 1e6 for a steady state tolerance was used. The time step size of
1e3 seconds is treated as default and optimal for most cases but also investigated later
on.
29
4. Sandia Flame D
In order to investigate the process of building up equivalent reactor networks and
find their sensitivity to different parameters the Sandia Flame D [24] was used as a
reference. Briefly, it is a methane/air jet flame. Figure 13 shows photographs of it.
Figure 13. Flame D (left) with Nd:YAG laser beam and closeup of the pilot flame (right).
Flow properties Main inlet Pilot CoflowCH4 0.16 O2 0.2 0.05 0.23N2 0.65 0.73 0.77Mixture fraction 1.0 0.27 0.0H2O 0.1 CO2 0.12 Temperature, K 293.0 1880.0 293.0Pressure, atm 1 1 1Average inlet velocity, m/s(variable along radius)
55.87 13.44 1.01
Area, mm2 0.66 3.54 517.89
Table1. Sandia Flame D flow conditions.
The experimental setup has a main inlet, a pilot flow and a coflow. They are
described in the Table 1.
The flame with a corresponding set of boundary conditions had been simulated by
Nilsson [25] and described in his thesis being written in parallel to this work. All flow
calculations in the Nilsson's work use the time dependent form of the RANS equations
(also known as unsteady RANS or URANS). Flow equations are solved using the finite
volumebased CFD software STARCD. [25]
30
Figure 14. Sandia Flame D validation. [25]
Mechanism with 163 species is used and described by Shenk et. al. [26]. The
validation sample can be seen in the Figure 14. There is a good agreement between the
CFD solution and experiments, therefore the resulting flow field can serve as a trusted
reference for the further CRN development.
In order to model the flame with equivalent reactor network two different
approaches have been verified and are described in the following sections.
31
4.1 Reactor network setup. Approach 1
In the first and simplest approach the reactor network is based on a sequence of
volumes into which the flame has been cut. These volumes are used as volumes of
transient PSRs in the CRN. The reaction scheme used was the same as which had been
used in STARCD [26]. The schematic principle and equivalent reactor network can be
seen in the Figures 15a and 15b respectively.
Figure 15a. Sandia Flame D Approach 1. Principle.
32
Figure 15b. Sandia Flame D Approach 1. CRN.
Afterwards, the results from the CRN have been compared to those from the CFD
solution, where the values are taken as median of all cells in a certain corresponding part
of the flame. The comparison is shown in the Figures 16 (af).
33
(a) (b)
(c) (d)
(e) (f)
Figure 16. (a): temperature profile; (b): CH4 concentration, linear; (c): CH4 conc. logarithmic; (d): OH conc.;
(e): CO2 conc.; (f): H2O conc.
As one can see the combustion prediction in the CRN differs from CFD. From the
residence time formula it can be explained that combustion in the CRN directly depends
on volumes chosen for the reactors and will be different according to different choice of
the flame parts' size. In order to force the CRN combustion to go the way it goes in CFD,
it was decided to use isothermal PSRs with a temperature profile from CFD, which is seen
34
in the Figure 16a. In such reactors under a constant temperature mass fractions may vary.
The results are presented in the Figure 17 (ae).
(a) (b)
(c) (d)
(e)
Figure 17. (a): CH4 concentration, linear; (b): CH4 conc. logarithmic; (c): OH conc.; (d): CO2 conc.; (e): H2O
conc.
35
Scheme CPU time
Real User Sys
Default PSRs 16m 36.247s 16m 29.044s 0m 0.956s
Isothermal PSRs 3m 55.795s 3m 54.928s 0m 0.208s
Table 2. CPU times in case of two schemes in the first approach.
It is obvious, that in case of isothermal PSRs the convergence has been achieved four
times faster and the tendency of species concentrations also has become better but still
the scheme does not predict this flame well enough.
The probable reason was that the reactors had been created from nonhomogeneous
parts of the flame, so that nonsimilar zones had been mixed.
Thus, it was decided to try another approach to take homogeneity into account.
4.2 Reactor network setup. Approach 2
In terms of specification of reactor network parameters the second approach
resembles the first one. The only difference is in the principle of zones derivation which
were defined by the flame temperatures. That principle is demonstrated in the Figure
18a.
Figure 18a. Sandia Flame D Approach 2. Principle.
36
In the Figure 18b an equivalent reactor network is shown.
Figure 18b. Sandia Flame D Approach 2. CRN.
Then the results were compared to CFD and shown in the Figure 19 (af).
37
(a) (b)
(c) (d)
(e) (f)
Figure 19. (a): temperature profile; (b): CH4 concentration, linear; (c): CH4 conc. Logarithmic; (d): OH conc.;
(e): CO2 conc., (f): H2O conc.
The temperature profile obtained in the CRN is very close to the one from CFD, it
means the combustion went properly and caused an obvious relatively good agreement
for the species concentrations as well. The scheme has been also tested with use of
isothermal reactors but the results mostly did not change.
38
4.2.1 Parameter study
It was decided to carry out a parameter study for the reactor network made with the
second approach to see its stability and sensitivity to various factors. The influence of
inlet temperature of main inlet and coflow, time step size and mechanism sensitivity
have been investigated.
Time step size, t:Δ
Time step
size, s
Reactor Nr.8 prediction CPU time
CH4, mf OH, mf CO2, mf H2O, mf T, [K] real user sys
1.0E1 8.060055E06 3.275011E10 1.356732E02 1.110755E02 530.21 0m42.918s 0m41.876s 0m0.124s
1.0E2 8.060055E06 3.275011E10 1.356732E02 1.110755E02 530.21 1m5.175s 1m1.268s 0m0.308s
1.0E3 8.059971E06 3.210346E10 1.356732E02 1.110755E02 530.21 4m14.898s 4m3.336s 0m1.076s
5.0E4 8.060055E06 3.275011E10 1.356732E02 1.110755E02 530.21 8m6.396s 7m39.196s 0m1.492s
1.0E4 8.060055E06 3.275011E10 1.356732E02 1.110755E02 530.21 35m58.561s 35m36.804s 0m6.440s
1.0E5 8.060055E06 3.275011E10 1.356732E02 1.110755E02 530.21 248m27.451s 247m5.132s 0m43.692s
Table 3. Time step size influence (Output results for reactor Nr 9).
With time step size of 0.1 and 0.02 the initial spark time has been increased as
Δt×5 in order to ensure ignition of the mixture. Despite residence times of some
reactors are smaller and some are bigger than t, Δ as seen in the Table 3 for the current
scheme it does not have an influence on results and the change of time step size only
between 1e3 and 5e4 seconds gives different values. So, for the scheme with these
certain conditions the time step size of 5e4 seconds leads to the goal results while
consuming less computational time.
Inlet temperature:
Mixture temperature variety has been done in three cases and results for the main
inlet and the coflow are shown in the Figures 20 (ae) and 21 (ae) respectively. For the
main inlet the next cases are used: Case 1: Tmain inlet = 293 K; Case 2: Tmain inlet = 350 K;
Case 3: Tmain inlet = 410 K.
39
(a) (b)
(c) (d)
(e)
Figure 20. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2 conc.; (e): H2O conc.
No obvious difference for the combustion behavior can be seen. For the coflow
temperature the following cases are used: Case 1: Tcoflow = 293 K; Case 2: Tcoflow = 350 K;
Case 3: Tcoflow = 410 K;
40
(a) (b)
(c) (d)
(e)
Figure 21. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2 conc.; (e): H2O conc.
Increase of coflow temperature causes a higher final flame temperature that
logically reduces amount of methane and enlarges OH; concentration of the other species
did not change that much.
41
Mechanism sensitivity:
For this research the following mechanisms have been chosen for comparison with
the previous results: LOGEfuel C4 v1.0 with 235 species, Reduced Methane Mech. with
28 species and Optimized Methane Mech. with 20 species, all developed by LOGE AB.
Their performance can be seen in the Figure 22 (ae). Also, in the Table 4 computational
times of the mechanisms are summarized.
(a) (b)
(c) (d)
(e)
Figure 22. (a): temperature profile; (b): CH4 concentration; (c): OH conc.; (d): CO2 conc.; (e): H2O conc.
42
Mechanism CPU timereal user sys
Shenk et. al. [26] 4m 14.898s 4m 3.336s 0m 1.076sLOGEfuel C4 v1.0 6m 24.935s 6m 17.832s 0m 0.800s
Reduced Methane 0m 10.796s 0m 10.548s 0m 0.136s
Optimized Methane 0m 5.729s 0m 4.816s 0m 0.092s
Table 4. Mechanisms CPU times.
As we can see from the plots the Optimized mechanism is the most beneficial one in
terms of computational time and rather good prediction.
In conclusion, it can be stated that the second approach is sufficiently enough to
simulate the flame in terms of time costs and accuracy. However, in order to decrease the
deviation the zones should also be splitted into more reactors to take into account a
variety of mass flows within the flame, this process is just a matter of time. The current
scheme has demonstrated that the software tool performs properly giving reasonable
results, so it can be applied to other engineering problems.
43
5. Siemens atmospheric combustion test rig
Siemens Industrial Turbomachinery manufactures the SGT800 that is the third
generation dry low emission (DLE) gas turbine. For experimental studies of the burner an
atmospheric combustion test rig situated in Finspång, Sweden, has been built. A detailed
description of the rig can be found in [27]. A schematic setup is presented in the Figure
23.
Figure 23. Schematic layout of the combustion test rig. [28]
CFD computations on the burner have been performed by Nilsson [25] in STARCD
using two different 2D meshes provided by Siemens. These meshes do not include the
mixing chamber and the swirl cone upstream of the burner; instead the velocity and
turbulence profiles used at the inlet were taken from previous 3D calculations preformed
by Bruneflod [29], which included those features. The chemical composition and
temperature at the inlet are computed based on reported mass flow rates of fuel and air
into the mixing chamber. [25] The 2D model imported in STARCD is shown in the
Figure 24.
44
Figure 24. 2D model of the test rig combustion chamber.
5.1 Mechanism validation
Gas turbine simulation is a complicated engineering task and has to be treated
accordingly. First, in order to start modeling of combustion processes in the rig and trust
results, the most important component should be examined – a reaction scheme. Four
mechanisms were chosen to be tested:
• GRI3.0 with 53 species [30]
• Ranzi et. al. mechanism with 114 species, described in [31]
• Optimized mechanism.
• Reduced mechanism.
As the corresponding experimental data the work of Le Cong et. al. [32] on
investigation of methane/hydrogen blends oxidation at atmospheric pressure in a PSR
was chosen. All the four mechanisms have been tested and validated as shown in the
Figures 25(ad).
45
(a) (b)
(c) (d)
Figure 25. (a): CH4,; (b): CO2; (c): CO; (d): H2O concentrations vs hydrogen content in the mixture.
As it seen from the plots above the mechanisms Reduced, Optimized and one by
Ranzi et. al. show a rather good agreement with experiments. The Optimized mechanism
is the most beneficial in terms of computational times and accurate prediction, whereas
the GRI3.0 has the worst trends. However, since there was no experimental data for NOx
measurements in that PSR case, all the four mechanisms are still interesting for
combustion and emission simulation under gas turbine conditions.
5.2 Reactor network setup
A reactor network for the rig has been developed according to the third method
described in the Section 2.7. Only a coldflow solution from [25] was available. The
processing of the CFD results was based on determination of recirculation zones which
are shown in the Figure 26.
46
Figure 26. Velocity magnitude of the flow and schematic zones mapping.
All data required for reactors has been extracted in a similar way it had been done
for the Sandia flame described in the Section 4.1. In the Figure 27 an equivalent CRN is
given. The reactors represent the flame zones as following: PSRs 1 – dome recirculation,
2 – main flame, 3 – immediate post flame, 4 – main recirculation. Reactors 5 and 6 are
PFRs and they model post flame zone and dilution zone respectively.
Figure 27. Combustion rig chemical reactor network.
The experiments had been carried out using natural gas as a fuel whereas in this
work it was replaced by methane, since the composition of corresponding natural gas was
not known. For the first simulation the mechanism Optimized M. with included thermal
NOx (Zeldovich reactions) chemistry. That NOx part had been taken from the GRI3.0
mechanism. Referring to the section 2.5.3 such a decision was made based on neglection
of a relatively small input of prompt NOx into the whole picture at high temperatures,
whereas thermal NOx plays the most important role. Moreover, prompt NOx strongly
depends on CHradicals and should better be ignored for simplification. The resulting
mechanism had 23 species. Outlet temperature and NOx concentration have been
compared to experimental data and plotted below. In the Figure 28 NOx concentration
47
from the CRN along the central horizontal axis of the rig starting from inlet is given. Due
to confidential reasons all the values are scaled.
Figure 28. Thermal NOx formation trend along the central axis.
The trend shows that the amount of NOx quickly increases in the flame zone, shortly
drops afterwards and slightly forms through the rest part of the combustion chamber.
According to the experimental data provided by Siemens the CRN has been tested
with corresponding conditions. The following Figures 29a and 29b compare NOx
concentrations and temperatures obtained in the reactor network and compared to
experimental values for five cases differed by hydrogen content in the fuel mixture.
(a) (b)
Figure 29. (a): NOx vs hydrogen content; (b): Temperature vs hydrogen content.
Based on the theory previously written and the last two plots above it can be stated
that the NOx formation process mainly depends on temperature. The combustion test rig
CRN captures a correct trend of such a dependency but not absolute values.
Temperatures reached in the reactor network with no hydrogen in the mixture are about
48
30 K lower than the experimental, therefore NOx values do not meet logically. Probable
reasons of those deviations will be discussed in the discussion part later on. Anyway, from
the plots we see that with higher temperatures the NOx agreement is better, which can be
explained by its formation mechanisms described in the Section 2.5.3.
5.3 Parameter study
The aim of this parameter study was to estimate prediction under different
conditions and also see the reactor network sensitivity to other reaction schemes.
Time step size:
Time stepsize, s
Outlet, scaled values [] CPU time
T CH4 CO CO2 OH NOx real user sys
1.00E01 104.10900 0.73720 0.96908 0.99308 1.05817 0.98963 0m 30.450s 0m 29.440s 0m0.060s
1.00E02 104.10900 0.73720 0.96908 0.99308 1.05817 0.98963 0m 20.433s 0m 19.792s 0m 0.048s
5.00E03 104.09306 1.12110 0.96627 0.99308 1.05638 1.00188 0m 21.554s 0m 19.608s 0m 0.056s
1.00E03 104.09306 1.12110 0.96627 0.99308 1.05638 1.00187 0m 48.125s 0m 46.964s 0m 0.064s
1.00E04 104.09306 1.12110 0.96627 0.99308 1.05638 1.00187 6m 52.969s 6m 21.004s 0m0.820s
1.00E05 104.09306 1.12110 0.96627 0.99308 1.05638 1.00187 54m 58.149s 54m 25.140s 0m2.210s
Table 5. Time step size influence (Output results for outlet).
Based on results given in the Table 5 for the current reactor network at time step size
of between 0.005 and 0.001 the values stop changing and further reduction of time step
does not improve the accuracy, thus for this scheme the optimal time step size is 5e3
seconds, since it takes less CPU time and leads to the goal results.
In the Figures 30a and 30b the dependency of NOx concentration and temperature of
the flow on the mixture inlet velocity and temperature is shown.
49
(a) (b)
Figure 30. (a): NOx/Temp. vs inlet mass flow rate; (b): NOx/Temp. vs inlet temperature of the mixture.
Apparently, the faster mixture comes to the burner, the lower values of temperature
and NOx amount the flow will have after combustion. Also, a higher temperature of
incoming mixture leads to higher flame temperatures and, as it was mentioned before,
leads to more intensive NOx production.
Mechanism sensitivity:
In the Figures 31a and 31b we can see sensitivity of this scheme to four chosen
mechanisms, using methane with no hydrogen content as a fuel. The reaction schemes
are: Optimized mechanism+Thermal NOx, Reduced+Thermal NOx (modified
analogically) with 31 species, Ranzi et. al. and GRI3.0. The Table 6 shows the
mechanisms computational time consumed for the simulation.
MechanismCPU time
real user sys
Optimized M. + T. NOx 0m 48.125s 0m 46.964s 0m 0.064s
Reduced M. + T. NOx 1m 33.210s 1m 28.300s 0m 0.184s
GRI3.0 [30] 3m 49.907s 3m 46.604s 0m 0.324s
Ranzi et. al. [31] 12m 44.989s 12m 42.288s 0m 0.260s
Table 6. CPU times.
50
(a) (b)
Figure 31. (a): NOx and (b): temperature sensitivity to different reaction schemes.
In comparison to bigger mechanisms, as shown before, the Optimized+Thermal NOx
mechaninism gives higher temperatures and lower NOx emissions and is still closer to the
measured data. Its performance is also better in terms of computational time. That all
proves its ability to be used as reaction scheme for the further CRN development of the
combustion test rig.
Also, it is worth to notice that the detailed reaction schemes do not influe so much on
the final temperature (<10 K difference), as on the NOx formation, so its prediction is
directly dependent on the NOx chemistry of a mechanism.
51
6. Discussion and conclusions
In this thesis different techniques of a chemical reactor network setup were
investigated. An inhouse software tool 'LOGEsoft Reactor Network' was tested, improved
and applied to two cases in order to see its performance for emission prediction in
parallel to low computational costs.
The first case was modeling of the Sandia Flame D based on its validated CFD
solution. The CRN development has been tried with two approaches: splitting the flame
by 'disks' in series and determination of homogeneous zones by temperature. The results
were far not accurate in the first approach since nonhomogeneous parts of the flame had
been represented by PSRs where a perfect mixing occurs and combustion went other way.
In order to force the combustion process resemble CFD, isothermal PSRs with the CFD
temperature field were used. The results became better and trends got more reasonable
but still the solution was incorrect. In the second approach a resulting CRN temperature
profile nearly repeated the one from CFD. It meant the combustion behavior went right
way, so the outcome for species concentrations was also in a good agreement with CFD.
However, in order to decrease the deviation the zones should have been splitted into a
bigger number of reactors in order to take into account different mass flow rates which
vary along radius and height. The scheme has shown that the software tool performs
properly, the results are reasonable and it can be applied to other engineering problems.
The second case has pictured such a problem. As it was mentioned before gas
turbines are being constantly evolved in terms of emissions reduction and simulation
tools are in charge to avoid realscale experiments. An experimental atmospheric
combustion test rig by SIT, which serves for investigation of DLE burner used in the
SGT800 turbine, was modeled in CFD with a subsequent data extraction. The equivalent
reactor network has been built based on recirculation zones of the 'cold' flow. The
outcome captured the main trends but not absolute values. Firstly, it might be caused by
different heat release when burning methane whereas for the experiments natural gas
was used. Secondly, for the hydrogen cases during the experiments by adjusting mass
flow the temperature has been aimed to be the same, while in the CRN the flow does not
change and thereafter deviations of temperature are bigger. Thirdly, a subjective factor
that is expressed in manual data extraction with some essential deviations should be
noticed. Also, of course, the reactor network had been built upon a 'coldflow' solution
which may differ from a 'hot' one. In this work the 'coldflow' was a single available
picture of how the flame would look like and the corresponding ERN could not capture
the flame field for the hydrogen cases properly.
52
Nevertheless, the resulting values are reasonable in comparison to experimental data
that proves a suitability of such a scheme to be developed onward. A way of improvement
might be a more detailed zone mapping of the flame and accurate and full extraction of
mass flows and other interactions between the zones. It is rather complicated and time
consuming when this work is carried out by hand, that is why automation of the process
by use of computational sources is absolutely rational.
53
7. Future works
The current reactor network method needs to be used for further development, in
particular by means of two major features: using CFD 'hotflow' solutions (for hydrogen
cases also desirable) and implementation of automatic instruments for data extraction.
As an alternative approach partiallystirred (stochastic) reactors should be tested to
take turbulent effects on the flow into account for a better simulation of flames under gas
turbine conditions.
Despite the most frequently used above mechanism showed a good performance, its
NOx chemistry has to be also further improved.
In addition, more experimental investigations are highly desirable as well as
application of the reactor network to other similar gas turbine combustor modifications.
54
References
1. http://www.setatwork.eu/database/actors/A397.htm
2. geflexibility.com/static/globalmultimedia/flexibility/photos/howgtworklg.jpg
3. www.sparknotes.com/chemistry/kinetics/mechanisms/section1.html
4. Warnatz, J. et. al., Combustion: Physical and Chemical Fundamentals, Modeling and
Simulation, Experiments, Pollutant Formation. 4th ed. Springer Berlin Heidelberg New
York, 2006
5. Sepideh, S.M., Network Modeling Application to Laminar Flame Speed and NOx
Prediction in Industrial Gas Turbines. Master's thesis, Linköping University, 2013
6. http://en.wikipedia.org/wiki/Combustor
7. Lefebvre, A.H. and Ballal, D.R. Gas Turbine Combustion: Alternative Fuels and
Emissions. Third Edition. CRC Press, Taylor & Francis Group, 2010.
8. AlFawas, A.D., Dearden, L.M., Hedley, J.T., Missaghi, M., Pourkashanian, M.,
Williams, A. and Yap, L.T., NOx Formation in Geometrically Scaled Gas – Fired industrial
Burners. 25th Symposium (International) Combustion, The Combustion Institute,
Pittsburgh, 1994.
9. Hamedi, N., Numerical Study of NOx and Flame Shape of a DLE Burner. Master's
thesis, Linköping University, 2012
10. http://www.maidhof.com/img/psr.gif
11. Turns, S.R., An introduction to Combustion, 1993
12. Andersson, S., Förbränningsteknik, 1996
13. Chomiak, J., Combustion: A study in theory, fact and application, 1990
14. Tillmark, C., Kinetic study of combustion behavior in a gas turbine influence from
varying natural gas composition. Lund Institute of Technology, Lund University, 2006.
15. http://www.loge.se/Company/company.html
16. Schmidt, L.D., “The Engineering of Chemical Reactions”. New York: Oxford University
Press, 1998.
17. Daniel, J., Combustion Characteristics of MCV/LCV Fuel A Numerical Chemical
Kinetic Study at Gas Turbine Conditions. Master's thesis, Lund Institute of Technology,
Lund University, 2007.
18. Bhave, A. and Kraft, M., Partially Stirred Reactor Model: Analytical Solutions and
Numerical Convergence Study of PDF/Monte Carlo Method. Cambridge Centre for
Computational Chemical Engineering, University of Cambridge, 2002.
19. Hua, W., Yongchang, L., Mingrui, W., Yusheng, Z., “Multidimensional modeling of
Dimethyl Ether (DME) spray combustion in DI diesel engine”. Journal of Zhejiang
University SCIENCE, 2005.
55
20. Hallaji, M. and Mazaheri, K., Numerical simulation of turbulent nonpremixed
combustion in diluted hot coflow using PaSR combustion model. Combustion Institute,
MCS 7, 2011.
21. GhaziHesami, S., Cost Effective Emissions and Minor Species Predictions via
Coupling of Computational Fluid Dynamics and Chemical Reactor Network Analysis.
Master's thesis, Concordia University Montreal, 2009.
22. Falcitelli, M., Tognotti, L., Pasini, S., An Algorithm for Extracting Chemical Reactor
Network Models from CFD Simulation of Industrial Combustion Systems. Combustion
Science and Technology, 2002.
23. Thanh Hao, N. and Jungkyu, P., CRN Application to Predict the NOx Emissions for
Industrial Combustion Chamber. Asian Journal of Applied Science and Engineering, Volume
2, No 2, 2013.
24. http://www.sandia.gov/TNF/DataArch/FlameD.html
25. Nilsson, T., Development of a model for gas turbine combustion. Master's thesis,
Combustion physics, Lund University (to be published later).
26. Schenk, M., Moshammer, K., Oßwald, P., KohseHöinghaus, K., Leon, L., Seidel, L.,
Mauss, F., and Zeuch, T., Detailed mass spectrometric and modeling study of isomeric
butene flames. Combustion and Flame, 160(3), 2013.
27. Lindholm, A., Lörstad, D., Magnusson, P., Andersson, P. and Larsson, T., Combustion
stability and emissions in a lean premixed industrial gas turbine burner due to changes in
the fuel profile. ASME paper GT200959409, 2009.
28. Lörstad, D., Lindholm, A., Larfeldt, J., Lantz, A., Collin, R., Aldén, M., Investigation of
hydrogen enriched natural gas flames in a SGT700/800 burner using OH PLIF and
chemiluminescence imaging. Proceedings of ASME Turbo Expo, 2014.
29. Bruneflod, S., Flow simulations of an axisymmetric twodimensional 3:rd generation
DLE burner, Master's thesis. Department of Applied Physics and Mechanical Engineering
Division of Fluid Mechanics, Luleå University of Technology, 2010.
30. http://www.me.berkeley.edu/grimech/version30/text30.html
31.
http://creckmodeling.chem.polimi.it/index.php/currentversionnovember2013/c1c3me
chanism
32. Le Cong, T., Dagaut, P., Oxidation of H2/CO2 mixtures and effect of hydrogen initial
concentration on the combustion of CH4 and CH4/CO2 mixtures: Experiments and
modeling. Proceedings of the Combustion Institute 32, 2009.
56