Modeling and Numerical Simulation of Partially Premixed Flames
Ilona Zimmermann
Universitat¨ der Bundeswehr Munchen¨ Fakultat¨ fur¨ Luft- und Raumfahrttechnik
Thema der Dissertation: Modeling and Numerical Simulation of Partially Premixed Flames
Verfasser: Ilona Zimmermann
Promotionsausschuss:
Vorsitzender: Prof. Dr. sc. techn. J¨orn Sesterhenn 1. Berichterstatter: Prof. Dr. rer. nat. Michael Pfitzner 2. Berichterstatter: Prof. Wolfgang Polifke, Ph.D.
Tag der Pr¨ufung: 21.07.2009
Mit der Promotion erlangter akademischer Grad: Doktor der Ingenieurwissenschaften (Dr. Ing.)
Neubiberg, den 16.11.2009
Acknowledgements
This work was carried out at the University of Armed Forces Munich, Institute for Thermodynamics. I would like to express my sincere thank to Professor M. Pfitzner for his guidance and encouraging supervision during the entire project. He offered me the possibility to gain an insight to many different subjects. I am greatful to Professor Polifke who agreed to assist my work as co-examiner. I would also like to thank my colleagues of the Institute for Thermodynamics for interesting discussions, and my colleagues at ANSYS for the fantastic working atmo- sphere. Furthermore I would like to thank my friends who supported all the time.
Danksagung
Ein ganz besonderer Dank gilt nat¨urlich meiner Familie f¨urdie stetige Unterst¨utzung auf meinem Lebensweg und die Liebe die sie mir schenken. Vor allem mein Mann Mario gab mir stets den notwendigen R¨uckhalt und hat mich immer wieder auf meinem Weg best¨arkt.
Abstract
The demands on motor vehicles are steadily increasing. They have to be powerful and competitive but also environmentally friendly. This is not only due to the tight pol- lutant emission regulations, but also due to public interests. Although alternatives to internal combustion engines like hybrid or electro vehicles are under investigation and are already used for certain applications, it will take several decades until a wider use is possible. Thus conventional internal combustion engines have to be improved further. There is still potential like downsizing with supercharger, variable valve stroke or vari- able compression. But to be able to further improve the internal combustion engines, a deeper understanding of the combustion process is necessary. As experiments give only limited information, computational fluid dynamics (CFD) is a very powerful tool to gain more knowledge of the processes occurring within the engine and to improve the processes. In industry, the Eddy Dissipation Model is still a commonly used combustion model. As this is a very simple model, it is not successful to describe all the complex processes occurring in internal combustion engines. So this work focuses on the development of a suitable combustion model for direct injection engines in industrial applications. In such applications, combustion occurs under partially premixed conditions. The model presented is based on a flamelet concept with an additional transport equation for the reaction progress variable to account for the partially premixed regime. Different approaches are compared to close the source term of the reaction progress variable transport equation. The main effort has been spent for the correct description of the laminar flame speed, which changes with pressure, temperature and mixture composi- tion, which strongly vary in internal combustion engines. As the laminar flame speeds are derived in the context of premixed combustion, the validity range of the correlations is not wide enough for diffusion flames. So they are extended to the flammability limits, for which correlations are derived within this work to account for the pressure and temperature dependence. To also account for the tur- bulence effects, an effective laminar flame speed is introduced, used in pre-integrated tables, which also depends on the mean mixture fraction and mixture fraction fluctu- ation. The developed model is validated on a hydrogen jet flame, on a piloted methane- and heptane-air flame, and on a methane-air swirl burner. Especially for the hydrogen jet flame, and the piloted methane flame, a detailed set of experimentally obtained variables is available, like species mass fractions and temperature distributions. In comparison to the Eddy Dissipation Model, which is based on the assumption of a single step reaction, it could be seen that the Eddy Dissipation Model is not reliable enough to account for the complex processes occurring in internal combustion engines, whereas the flamelet model together with the reaction progress variable and the effective laminar flame speed shows promising and reliable results.
Contents
1 Introduction 1
2 Fundamentals of Turbulent Reactive Flows 5 2.1 Characteristic Variables ...... 5 2.2 Flame Types ...... 7 2.3 Laminar Flame Speed ...... 13 2.4 Flammability Limits ...... 14
3 Basic Equations for Modeling of Non Reactive Flows 17 3.1 Instantaneous Equations ...... 17 3.2 Turbulent Flows ...... 19 3.3 Turbulence Models ...... 22
4 Modeling of Turbulent Reactive Flows 27 4.1 Eddy Dissipation Model ...... 31 4.2 The Laminar Flamelet Concept ...... 32 4.3 The Reaction Progress Variable ...... 38 4.4 Partially Premixed Flames ...... 41
5 H2 Jet Flame 44 5.1 Computational Domain and Boundary Conditions ...... 45 5.2 Influence of Chemistry Table ...... 47 5.3 Choice of Turbulence Model ...... 51 5.4 Implementation of Laminar Flame Speed ...... 52 5.5 Results and Discussion ...... 59
6 Sandia D Flame 69 6.1 Computational Domain and Boundary Conditions ...... 70 6.2 Influence of Chemistry Table ...... 71 6.3 Choice of Turbulence Model ...... 75 6.4 Implementation of Laminar Flame Speed ...... 75 6.5 Results and Discussion ...... 84
7 Low Swirl Burner 97 7.1 Computational Domain and Boundary Conditions ...... 98 7.2 Results and Discussion ...... 99
8 Conclusion and Outlook 110 8.1 Conclusion ...... 110 8.2 Outlook ...... 111 A The Presumed β-pdf Approach 113
B Flammability Limits 118 B.1 Hydrogen ...... 118 B.2 Methane ...... 120
C Combustion Chemistry 122 C.1 Methane Mechanism ...... 122 C.2 Hydrogen Mechanism ...... 131
D Additional Figures 132 D.1 H2 Jet Flame ...... 132 D.2 Sandia D Flame ...... 134
E Material Properties of Hydrogen 135 Nomenclature
Latin Symbols
A flame front surface area m2 c reaction progress variable – cp heat capacity at constant pressure J/(kg K) D diffusion coefficient m2/s G stretch factor – g gravity m/s2 H total enthalpy J h specific enthalpy J/kg k turbulent kinetic energy m2/s2 l length m m mass kg P probability density function p pressure P a R specific gas constant J/(kg K) s stoichiometric mass ratio – sl laminar flame speed m/s st turbulent flame speed m/s T temperature K t time s U diffusion velocity m/s u velocity m/s W molecular weight kg/mol X mole fraction – x space coordinate m Y mass fraction – Z mixture fraction – Greek Symbols
χ scalar dissipation rate 1/s ε turbulent dissipation rate m2/s3 λ heat conductivity W/(Km) λ air excess ratio – µ dynamic viscosity kg/(ms)
µt turbulent viscosity kg/(ms) ν kinematic viscosity m2/s ν0, ν00 stoichiometric coefficients; forward, backward reac- – tion ω source term or turbulent frequency φ equivalence ratio – ρ density kg/m3
τc chemical time scale s
τf flow time scale s
τfl flame time scale s
τη Kolmogorov time scale s 2 τij stress tensor N/m
Abbreviations
CFD Computational Fluid Dynamics CPU Central Processing Unit DNS Direct Numerical Simulation JANAF Joint Army, Navy, and Air Force (USA) LES Large Eddy Simulation RANS Reynolds Averaged Navier Stokes Equation
Dimensionless Numbers
Da Damk¨ohlernumber Ka Karlovitz numer Le Lewis number Re Reynolds number Indices c chemical F fuel i species or space direction j species k species
O2 oxidant q quenching st stoichiometric t turbulent u unburnt
Chapter 1 Introduction
Combustion is the oldest and most important energy conversion method worldwide. In 2006, the total world primary energy demand was 11,741 millions of metric tons of oil equivalent (Mtoe). Figure 1.1 shows the development of the world energy demand recorded by the International Energy Agency (IEA) [168]. Since start of recording in 1971 the energy demand has already doubled, and is still steadily increasing. 81 % of the energy conversion for domestic heating, power generation and transportation is due to the combustion of either oil, coal or gas. Energy sources like nuclear, hydro, solar or wind energy still play a minor role. Thus combustion of fossil fuels also remains a key technology in the foreseeable future. Besides the generation of heat for conversion to secondary energy like thermal, mechanical or electric energy, the combustion also produces pollutants like oxides of nitrogen (NOx) or sulfur, which are responsible for acid rain and effects on health. Pollutants like carbon dioxide are in the discussion to be responsible for global warming.
Figure 1.1: World energy demand since 1971 [168]; **Other: geothermal, solar, wind, heat, etc
Therefore, especially in the transportation sector, climate protection and saving resources has moved increasingly into public interests. Future automotive engines for example have to fulfill several requirements demanded by political conditions, social trends as well as the wish for competitive products in the global surroundings. Public discussions about climate change and lobbyists lead to increased public consciousness concerning costs and environment protection. Additionally the energy industry is de- pendent on the availability of fuel and thus is dependent on the few countries producing oil. In the foreseeable future the resources will decrease and the mineral oil output will reduce. The legislator also influences the public opinion and decisions as well as the requirements for automotive engines by introducing tax benefits, driving limitations, 2 CHAPTER 1. INTRODUCTION consumption and emission regulations. The development of the emission regulations for pollutants is given in table 1.1 for petrol and diesel [169].
Emission EURO 1 EURO 2 EURO 3 EURO 4 EURO 5 EURO 6 [g/km] 1993 1997 2000 2005 2010 2014 Petrol CO 3.16 2.2 2.3 1 1 1 HC+NOx 1.13 0.5 HC 0.2 0.1 0.1 0.1 NOx 0.15 0.08 0.06 0.06 Diesel CO 3.16 1 0.64 0.5 0.5 0.5 HC+NOx 1.13 0.7 0.56 0.3 0.23 0.17 NOx 0.5 0.25 0.18 0.8 Particles 0.18 0.08 0.05 0.025 0.005 0.005
Table 1.1: Limitations for pollutant emissions according to EU exhaust gas regulations
These tight regulations and the continued demand for powerful cars which are also environmentally friendly is the main challenge of today’s design of internal combus- tion engines. Combustion engines will also be in the future of great importance as the development of alternative driving concepts did not yet reach all the necessary requirements. For examples hybrid vehicles are very heavy and are too expensive. Vehicles with electric drive are also very heavy and additionally have a short range as the storage batteries are still in a development status. Furthermore the infrastruc- ture for these alternatives is missing. The use of these concepts is only possible in combination with commonly used internal combustion engines. So in order to reduce emissions and save resources, the efficiency of combustion processes has to be improved. As the development for internal combustion engines is already very mature, it is necessary to get a deeper understanding of these combustion processes. To be able to fulfill the limitations, complex systems are necessary with complex injection processes. This has resulted in the introduction of new variable fueling technology. Instead of a single injection in Diesel engines, multiple injections per cycle are now used. Variable valve timing technology, variable turbocharging systems or even systems that allow varying engine compression ratios are also options to control the combustion process. As fossil fuels face some disadvantages, like limited power dynamics and unsatisfactory power-weight ratios, the use of hydrogen for internal combustion (IC) engines offers promising solutions [60]. One benefit of this fuel is the reduced carbon dioxide emis- sions and negligible NOx formation, as the main combustion product is water vapor. Combustion anomalies such as pre-ignition, backfiring and knocking can be avoided, which allows compression ratios in the range of diesel engines, high turbocharging and and downsizing measures. Furthermore it can be produced from water and a renew- able energy source like solar or wind power. The cryogenic storage and supercharging systems provide the ability of a high power density of spark-ignition engines. In order to reduce development time, engine manufacturers are continuously search- ing improvements of the development process. Since in experimental configurations the optical access is limited, efficient modeling and simulation becomes more and more im- portant. Simulations offer the advantage of supplying information that is not available from experiments due to e.g. missing optical access. Very complex laser diagnostics 3 are necessary to obtain detailed information. The simulation time improvements due to good parallelization and reliable model development together with increasing com- puter power, leads to its use in an early stage of the engine design process. Having such detailed information at this state already avoids the number of prototype parts. The simulation results can also be used to calibrate simpler low-dimensional models that in turn can be used for defining control algorithms. This work focuses on combustion model development for partially premixed flames. In order to benefit from the advantageous features of both, the premixed combustion where fuel and oxidizer are completely mixed before entering the combustion chamber, and the non premixed combustion where fuel and oxidizer enter separately, most tech- nical applications work under partially premixed conditions. If fuel and oxidizer enter separately the combustion chamber, and mix due to turbulence such that a stratified mixture exist at the moment when it is ignited, it is called partially premixed combus- tion. In automotive applications, turbulent flame propagation in a stratified mixture oc- curs for example in diesel engines. Liquid fuel is injected directly into the cylinder in several injections, early enough in the cycle so that the liquid fuel is able to vaporize and partially mix with the air before the mixture is auto-ignited. In modern direct injection spark ignition engines also an inhomogeneous mixture is ignited. The charge is stratified such that the flame initiated at the spark propagates through a partially premixed inhomogeneous mixture. A further example is the hydrogen direct injection. It is injected into the cylinder when the valves are closed in the beginning or later in the compression stroke, where it starts mixing with the surrounding air until it is ig- nited. It may also still be injected during the combustion process by multiple injections. In all of the examples the fuel-air mixture in the cylinder is partially premixed at the time of ignition. The fuel in partially premixed regions is rapidly consumed then combustion takes place under non-premixed conditions. Partial premixing thus leads to situations in which flames have characteristics of both, premixed and non-premixed flames. In the case of local extinction of a diffusion flame, intermixing of unreacted fuel and oxidizer is possible. Thus these two combustion regimes of partially premixed and non-premixed flames have to be covered by the developed combustion model. Due to the new variable fueling technology, the engine got more flexibility. Com- bined with electronic control, the engine design and calibration process has become more difficult. In order to achieve even small improvements, the combustion process has to be determined more precisely. A simple model, like the Eddy Dissipation model which assumes one-step chemistry and fast chemistry, will no longer be able to ful- fil the demands in the future. Nevertheless, simulations can not require too much computational time in order to remain competitive. So the use of Direct Numerical Simulation (DNS) or Large Eddy Simulation (LES) tools, especially in the combination with combustion, is limited due to the computational cost and thus is still not suitable for the simulation of internal combustion engines. That is why this work focuses on the optimization of combustion models for direct injection internal combustion engines using the Reynolds Averaged Navier Stokes (RANS) approach. In this work, the Com- putational Fluid Dynamic (CFD) tool ANSYS CFX is used. The goal of this work is the development and implementation of the improved com- 4 CHAPTER 1. INTRODUCTION bustion model with respect to partially premixed flames. Furthermore, it is validated with experimental data and compared to existing models commonly used in industry. The validation also shows the advantages and disadvantages of the model. The work is structured as follows: Chapter 2 gives an overview of the fundamentals of turbulent reactive flows, where some characteristic variables and combustion regimes for non-premixed and partially premixed flames are presented. The variables influencing the laminar flame speed and flammability limits, necessary for the developed model, are also discussed. Chapter 3 presents the basic equations for turbulent flows. Due to limited computer performance, it is necessary to apply an averaging procedure to obtain the Reynolds averaged Navier Stokes equations. This chapter also addresses the turbulence models used within this work. The modeling of turbulent reactive flows is discussed in chapter 4, where a review of combustion models for turbulent non-premixed and premixed flames is given, and the used and newly introduced combustion models are briefly discussed. Within chapters 5, 6 and 7, three different flames are examined, a hydrogen jet flame, a methane and heptane jet flame and a methane low swirl burner. A detailed description of the experimental setup is given. Different sources for the laminar flame speed and flammability limits for hydrogen, methane and heptane are compared and correlations for the temperature and pressure dependent flammability limits are de- rived. Chapter 8 gives a conclusion and an overview of the main findings of this work and provides an outlook with recommendations for further work. Chapter 2 Fundamentals of Turbulent Reactive Flows
Turbulent reactive flows involve a wide range of different processes related to ther- modynamics, chemical kinetics, fluid mechanics and transport processes. Concerning combustion, several chemical species react in multiple chemical reactions. In the follow- ing variables characterizing the reacting mixture are presented. Chapter 2.2 gives an overview of the different regimes that may occur in different applications. As within this work, special attention is paid on the laminar flame speed as well as on the flammability limits, these two parameters are described in separate chapters 2.3 and 2.4, respectively.
2.1 Characteristic Variables
The mole fraction Xi is defined as the number of moles of species i in relation to the total number of moles in the mixture. It can be related as
W Xi = Yi (2.1) Wi
where W is the mean molar mass, Wi is the molar mass of species i and the mass fraction Yi is defined as the ratio of the mass m of species i to the whole mass of the mixture m
m Y = i (2.2) i m
The total sum of all species mass fractions, respectively mole fractions, is equal to one. Species diffusion can be caused by concentration gradients, temperature gradi- ents, also called Soret effect, or pressure gradients. In general, the latter two are neglected [155]. So the diffusion velocities Ui can be expressed by the Stefan-Maxwell equation [51]
Ns ∂Xi X XiXj = (2.3) ∂x D (U − U ) i j=1 ij j i
where the binary diffusion coefficient Dij is independent of the mixture composition. It can be seen in equation 2.3 that the diffusion velocity of a species depends on the 6 CHAPTER 2. FUNDAMENTALS concentration gradients of all other species, which makes the evaluation of this equa- tion very CPU-hour-intensive. Later on in this work, this model is referred to under the name ‘complex’. To reduce the computational cost, the diffusion velocities can also be modeled by a Fick-like expression [99]
∂Yi UiYi = −Dim (2.4) ∂xi
with the mixture averaged diffusion coefficient Dim describing the diffusion of species i in the mixture. With the definition of the Lewis number
λ Lei = (2.5) ρDimcp the species diffusion fluxes can also be expressed as
λ ∂Yi ρUiYi = − (2.6) Leicp ∂xi
where λ is the thermal conductivity of the mixture, ρ the density, and cp is the specific heat capacity at constant pressure of the mixture
N Xs cp = Yicpi (2.7) i=1
the heat capacities cpi are tabulated in polynomial form for various species. For more detailed information about solving equations for the diffusion velocities, please refer to [115]. If fuel and oxidizer are completely consumed, and only combustion products are formed, the combustion occurs under stoichiometric conditions. If there is still fuel available after the combustion process, it is called rich and in the case that there is still oxidizer it is called lean. The global stoichiometric reaction describing the com- bustion of any hydrocarbon fuel can be expressed as
0 0 00 00 νF CmHn + νO2 O2 → νCO2 CO2 + νH2O H2O (2.8)
0 0 0 00 00 where νF = 1, νO2 = (m + n/4)ν , νCO2 = m and νH2O = n/2 are the forward and backward stoichiometric coefficients. 2.2. FLAME TYPES 7
The stoichiometric mass ratio s, which is defined as the minimum mass of oxidant per unit mass of fuel needed for complete combustion, can then be calculated as
0 νO2 WO2 s = 0 (2.9) νF WF
The index F refers to fuel, and Wi is the corresponding molecular weight of species i. Additionally, the mixture of fuel and air can be described by the so called air excess ratio λ, which relates the fraction of fuel and air in a mixture to the stoichiometric mass ratio. Its reciprocal is called equivalence ratio φ. Both can be expressed as follows
1 Y Y 1 Y λ = = O2,u F,u = O2,u (2.10) φ (YO2,uYF,u)st s YF,u
The index u refers to the initial conditions in the unburnt mixture and the index st refers to stoichiometric conditions.
2.2 Flame Types
Depending on the provision of the reactants and on the combustion process, the flame types can be classified in two basic flame types, the premixed and the non-premixed flames. Premixed combustion occurs if fuel and oxidizer are mixed before they enter the reaction zone. In the non-premixed combustion regime fuel and oxidizer are in- jected separately from each other. In that case fuel and oxidizer are mixed during the combustion process. When flame-fronts occur in the transition of non-premixed and premixed combustion, the combustion process is called partially premixed.
Non-Premixed Flames In non-premixed flames, fuel and oxidizer are injected separately from each other into the combustion chamber. As non-premixed flames do not exhibit inherent propagation speeds they cannot flash back. That is why non-premixed injection is often used due to safety reasons. As diffusion is the rate controlling process in these flames, a com- mon expression is also diffusion flame. The flame fronts are more complex than for premixed flames, as they cover the whole range of equivalence ratio from 0 for pure air to ∞ for pure fuel. Combustion takes place in the vicinity of stoichiometric mixture composition. Non-premixed flames can be found in early diesel engines, aircraft gas turbines or H2-LOx rocket combustion engines to name a few. The time needed for convection and diffusion, both being responsible for turbu- lent mixing, is typically much larger than the time needed for most of the combustion reactions to occur. So turbulent mixing and chemical reactions are the rate limiting processes. They are related through the dimensionless Damk¨ohlernumber Da, where 8 CHAPTER 2. FUNDAMENTALS
the timescale for the dissipation of turbulent fluctuations τt is related to a characteristic timescale τc for the chemical reaction system.
τ Da = t (2.11) τc
Poinsot et al. [124] introduced a regime diagram for non-premixed flames, presented in figure 2.1, depending on the Damk¨ohler number and the turbulent Reynolds num- ber Ret, which describes the ratio of turbulent inertial forces to molecular viscous forces
u0 l Re = t (2.12) t ν
0 where u is the velocity fluctuation, lt the integral length scale and ν the flow kinematic viscosity.
Figure 2.1: Regime diagram for non-premixed combustion [124]
For large Damk¨ohlernumbers, i.e. for fast chemistry, the flame is very thin, and the reactive layer is thinner than the diffusion layer, which is assumed to be equal to the Kolmogorov size, which is the size of the smallest eddies. This region is defined as the Flamelet region as the flame occurs as laminar flame elements. The Flamelet region is bounded where the flame Damk¨ohlernumber Dafl is equal to the Damk¨ohler number of the steady laminar flamelet assumption DaLF A. The flame Damk¨ohlernumber compares the molecular diffusion and chemical time scales. It is only a function of the local flame structure and does not take vortex time scales into account τ Dafl = f (2.13) τc
where τf , the flow time scale, can be estimated with the scalar dissipation rateχ ˜st as 2.2. FLAME TYPES 9
−1 τf ≈ (˜χst) (2.14)
The Damk¨ohlernumber can be related to the flame Damk¨ohler number according to Poinsot [124] as follows
p fl Da ≈ 2 RetDa (2.15)
For a longer chemical time scale, the thickness of the reactive layer becomes equal to the Kolmogorov length scale. In this case, a departure from laminar flame structures and unsteady effects is expected. For large chemical scales and small Damk¨ohlernum- bers extinction occurs. The regime of unsteady effects and extinction is divided where the flame Damk¨ohler number is equal to the Damk¨ohler number at extinction Daext. As the diffusion flame thickness ld is assumed to be equal to the Kolmogorov length 3/4 scale lη it follows lt/ld ≈ Ret . The regime of turbulent non-premixed flames has been discussed in several papers [18][33][92][118]. Compared to premixed flames it is difficult to classify a diffusion flame because the local flame scales, like local flame thickness and speed, depend on the local flow conditions such as local strain rates. So a second well established regime diagram derived by Peters [121] is presented in figure 2.2. He plots the ratio of the mixture p 002 fraction variance Zg and the diffusion thickness (4Z)F against the time scale ratio of the scalar dissipation rate at quenching χq to the conditional Favre mean scalar dissipation rateχ ˜st. The mixture fraction Z is a conserved scalar, which is 1 in the fuel stream, and 0 in the oxidizer stream and will be discussed briefly in chapter 4.2. The scalar dissipation rate χ can be interpreted as the inverse of a characteristic diffusion time. For χq > 1, three regions can be detected, the separated flamelet region, the con- χst nected flame zones and the connected reaction zones. p 002 The line Zg/(4Z)F = 1 separates two regimes, the separated flamelet and the connected flame zones. In the separated flamelet regime, the mixture fraction fluc- tuations are so large that they extend to sufficiently lean and rich mixtures. So the p 002 diffusion layers and thus the reaction zones are broken up. For Zg/(4Z)F < 1, the mixture fraction variance is too small to be able to break up the flame zones. In that case there might be intense mixing or partial premixing. p 002 Figure 2.2 shows an additional line with Zg/(4Z)R = 1, where the mixture fraction fluctuations are equal to the thickness of the reaction zone, which separates the connected flame zones and the connected reaction zones. Within the connected reaction zones regime, the fluctuations are smaller than the reaction zone thickness and are not able to disconnect the reaction zone. For values of χq/χ˜st < 1 the diagram shows the regime of flame extinction. 10 CHAPTER 2. FUNDAMENTALS
Figure 2.2: Regime diagram for non-premixed combustion [121]
Premixed Flames
In premixed flames, fuel and oxygen are already completely mixed before combustion takes place. Compared to diffusion flames, high temperatures occur and less soot is produced. The flame front separating the unburnt and the burnt gas state propagates due to heat and radical diffusion into the unburnt mixture. A characteristic quantity of premixed flames is the laminar flame speed sl, described in chapter 2.3. For example spark ignition engines or household burners operate under premixed conditions. Several researchers [3][17][125][120] introduced regime diagrams for turbulent pre- mixed combustion, also. The classical regime diagram for premixed combustion has been introduced already in 1985 by Borghi [17]. A modification of this diagram has been published by Peters [120] in 1999. Since they are both very similar, only the regime diagram of Peters is shown in figure 2.3.
The abscissa marks the ratio lt/lfl as a measure of the size of the turbulent eddies 0 that interact with the flame. The ordinate depicts u /sl as a measure for the turbulence intensity of the flow. Under the condition that the diffusivities for all species are equal, and unity Schmidt number Sc = ν/D = 1, where D is the diffusion coefficient, the flame thickness lfl and the flame time τfl can be expressed as follows
D D lfl = , τfl = 2 (2.16) sl sl
Thus equation 2.12 and 2.11 for the turbulent Reynolds number and Damk¨ohler number respecitvely, can be rewritten to 2.2. FLAME TYPES 11
Figure 2.3: Regime diagram for premixed combustion [120]
0 u lt Ret = (2.17) sl lfl
sl lt Da = 0 (2.18) u lfl
Furthermore, two turbulent Karlovitz numbers are distinguishing features in the regime diagram for premixed combustion (figure 2.3). The first one is defined as the ratio of the flame time scale τfl to the Kolmogorov time scale
2 2 τfl lfl uη Ka = = = 2 (2.19) τη lη sl
where τη, lη and uη are the Kolmogorov time, length and velocity scales.
According to [121], the second Karlovitz number relates the inner layer thickness lδ to the Kolmogorov length scale
2 lδ 2 Kaδ = 2 = δ Ka (2.20) lη with δ being the nondimensional thickness of the inner layer. 12 CHAPTER 2. FUNDAMENTALS
The regime diagram can now be separated in five regions:
• Laminar flames: In this region, only laminar, flat flames occur. It is separated from the turbulent region by the line Ret = 1
• Wrinkled flamelets: 0 The wrinkled flamelet regime lies in the region of u /sl < 1 and Ret > 1, where the laminar burning velocity still dominates the turbulent velocity fluctuations. Therefore the turbulent eddies within the flow field are not able to deform the flame front and only small wrinkling of the flame front can be observed. Since in industrial applications a higher turbulence intensity is required to enhance the combustion process, this regime is not of practical interest.
• Corrugated flamelets: 0 u /sl ≥ 1 and Ka < 1 limit the region of corrugated flamelets. The smallest eddies of size lη are larger than the flame front thickness, and are thus able to corrugate the flame front. The flame is not perturbed by the turbulent fluctua- tions, and can be locally considered as quasi laminar.
• Thin reaction zones: V V The thin reaction zones regime is defined by Ret > 1 Ka > 1 Kaδ < 1. The condition Ka > 1 implies, that the smallest eddies of the Kolmogorov size lη are smaller than the laminar flame thickness lfl. So they are able to penetrate into the reactive-diffusive flame structure, but they are still larger than the inner layer thickness lδ, which results from the criterion Kaδ < 1. The turbulent eddies, penetrating into the preheat zone, enhance the transport of chemical species and temperature. As they are not able to penetrate into the inner layer, the chemical reactions are essentially not influenced by turbulence.
• Broken reaction zones: The last region is the broken reaction zone. It is only bounded by Kaδ ≥ 1. Eddies, smaller than the inner reaction zone, penetrate into this zone. Due to the enhanced heat loss towards the preheat zone there occurs local extinction of the flame. Fuel and oxidizer are able to mix without chemical reaction.
Partially Premixed Flames In technical applications usually neither pure premixed nor pure non-premixed com- bustion occurs. In the main applications it is common practice to make use of the advantages of both and avoid their drawback, resulting in partially premixed combus- tion. They can avoid for example the high pollutant emissions of diffusion flames, while still operating under safe conditions. Partially premixed flames occur for ex- ample, when fuel and oxygen enter the combustion separately, and partially mix by turbulence such that combustion takes place in a stratified medium when it is ignited. Examples for partially premixed flames are aircraft gas turbines or direct injection in- ternal combustion engines. This work focuses on the combustion model development 2.3. LAMINAR FLAME SPEED 13 for partially premixed flames. For example in the case of DI internal combustion en- gines, fuel enters the combustion chamber during the compression stroke and starts mixing with the air. When ignition occurs near top dead center, the fuel has partially mixed with the air and the flame propagates through a stratified mixture.
2.3 Laminar Flame Speed
As already mentioned in chapter 2.2, the laminar flame speed sl is the most important parameter governing the regime of premixed combustion. For partially premixed flames it is also of growing interest, as can be seen within this work. The laminar flame speed depends on the initial conditions of the mixture, its composition, its pressure and its temperature and is defined as the velocity at which the flame propagates normal to itself and relative to the flow into the unburnt mixture. The experimentally obtained laminar flame speeds of hydrogen, methane and heptane are shown in figures 5.8, 6.7 and 6.8 respectively, as well as the correlations derived from the experiments or chemi- cal kinetics calculations. It can be seen that the values of the laminar flame speed show a substantial scatter. This can be attributed to the fact that no experimental set-up exists which is able to measure a one-dimensional, planar, adiabatic, steady, unstrained laminar flame for which sl is defined. Additionally some published experimental results did not take stretch effects into account (e.g. [75] or [107]). Stretch effects caused by curvature of strain rate have an influence on the flame temperature and thus on the laminar flame speed. For small to moderate stretch rates, this can be expressed as follows [43]
sl,s = sl,u − LK (2.21) where the indices s and u refer to the stretched and unstretched laminar flame speed sl respectively. L is the Markstein length and K the stretch rate. Depending on the sign of L, the laminar flame speed can be increased or decreased. If the flame is sta- ble, the Markstein length is positive and the laminar flame speed is decreased. For a negative Markstein length the flame is unstable and quickly develops cellular structures. Stretch-free laminar flame speeds for hydrogen are presented by Verhelst [150]. He also showed that the stretched laminar hydrogen flame at ambient conditions has a higher laminar flame speed, which he addressed to thermodiffusively unstable flames or flame acceleration due to cellularity. A further difficulty is to obtain measurements at strain rates low enough, that a linear extrapolation to a strain rate of zero leads to accurate results. Tien et al. [145] showed that the linear extrapolation of the nonlinear dependence of the laminar flame speed on the strain rate may lead to over-predictions of sl by 5% to 15%. A further experimental uncertainty to mention is the flame cooling. In order to fix the flame at a certain position, the flame is often stabilized on a burner, which is only possible when the flow velocity on the unburnt side is smaller than the laminar flame speed. This is due to heat loss of the flame to the burner and does not fulfill the adiabatic condition any more. Additionally, depending on the measurement technique, researchers use different definitions of the laminar flame speed. It can be 14 CHAPTER 2. FUNDAMENTALS based on the entrainment velocity of the unburned mixture as follows
1 dmu sl = − (2.22) Aρu dt
where A is the flame front area, ρu the density of the unburnt mixture and the fraction dmu/dt is the rate of entrainment of the unburnt mixture into the flame front. The laminar flame speed can be also expressed by the rate of production of the reacted gas as
1 dmb sl = (2.23) Aρu dt
Although both equations 2.22 and 2.23 lead to the same result for the ideal case of a one-dimensional, stretch free, planar flame, they are different for non-planar flames due to the finite flame thickness. Within the recent years the calculation of 1D planar flames has also become a pop- ular method to obtain the laminar flame speed. Calculated laminar flame speeds for hydrogen, methanol and hydrocarbon fuels have been presented by Peters et al. [122]. The influence of heat loss due to radiation has been addressed by Kennel et al. [83]. Recent publications by Verhelst [150] for hydrogen, or by Hermanns [71] for methane- hydrogen-air mixtures directly compared measurements and simulations.
2.4 Flammability Limits
The flammability limits are defined as the concentration limits, where a homogeneous mixture of a combustible gas and oxidizer is still able to self support a flame. Different experimental setups are presented in literature to measure the flammability limits. A summary of different combustible gases and liquids measured with different methods is given by Coward et al. [47]. Measurements of flammability limits are performed either in a spherical vessel or bomb, or in a tube which can be mounted horizontally or vertically. In the case of vertical mounting, the mixture in the tube can be ignited at the top or at the bottom. For a mixture with an upward propagating flame ignited at the bottom, the flamma- bility limits are wider than those of a downward propagating flame ignited at the top. Jarosinski et al. [77] determined experimentally the lean limit extinction mechanism for upward and downward propagating flames, which is also described in detail by Van den Schoor [147]. In the case of downward propagating flames, the flame is extin- guished near the walls due to heat loss. Natural convection causes the cold combustion products to be carried ahead of the flame front until the mixture is diluted so much by the combustion products that the flame extinguishes. Due to gravity, the upward propagating flame is faster and has a convex shape towards the unburned mixture, than the downward propagating flame, which has a nearly flat shape. Therefore the heat loss is larger for a downward propagating flame causing earlier extinction which narrows the flammable range. For a horizontal mounting of the tube, the flammability 2.4. FLAMMABILITY LIMITS 15 limits lie in between the ones detected with the upward and downward propagating flame. The influence of the different experimental setups can be seen in Appendix B where the flammability limits for hydrogen, methane and heptane in air are listed for different setups. The flammability limits are not only influenced by the direction of flame propaga- tion. If the size of the vessel becomes too small, for example when the radius of the tube is too small, the limits are narrowed as the walls reduce the flame temperature [141]. This phenomenon can also be seen in appendix B for the flammability limits of a horizontal propagating methane/air flame. Additionally the ignition energy must be high enough to initiate a flame. If it is too high, the flame propagates faster as a result of the heat provided by the ignition source and not the heat liberated by the flame itself. Thus the flammability limits are broadened. A further influence on the flammability limit is the definition of the criteria at which a flame is able to propagate or not. If the flame has to propagate over a distance of 0.1 m, the flammability limits are wider than for a flame which has to propagate over a distance of 1.5 m. An increase in temperature broadens the flammable range, as less heat has to be supplied from the burning layer. The flammability limits vary nearly linearly with temperature. Van den Schoor [147] determined three mechanisms that are responsible for flame extinction and four mechanisms of flame propagation which have an influence on the flammability limits.
Mechanisms that lead to flame extinction:
• Chemical kinetics: When the mixture fraction of a combustible mixture reaches the flammability limits, the flame temperature decreases, which has an effect on the chain branch- ing and chain terminating reactions, whereas the chain terminating reactions are less sensitive on temperature. Thus the concentration of flame carriers is reduced which leads to flame extinction.
• Heat loss: When the heat loss to the surroundings becomes too large compared to the heat production within the reaction zone, the flame might extinguish.
• Flame stretch: Increasing the stretch rate leads to a thinner flame. The residence time reduces and only incomplete reaction is achieved. This lowers the flame temperature which in turn leads to extinction.
Mechanisms of flame propagation:
• Flame stretch: The flame stretch has a strong effect on the laminar burning velocity and on the flame temperature, which in turn has an influence on the flammability limits. 16 CHAPTER 2. FUNDAMENTALS
• Natural convection: A flame kernel will be created, when a near-limit mixture in quiescent atmosphere is ignited. During the kernel growth, it is also moving upwards which causes flow strain. With increasing flow strain, the probability of flame extinction also increases. The effect of pressure on the flow strain is negligibly small, whereas with increasing temperature the flow strain is decreased.
• Flame front instabilities: For flames close to the flammability limits, the most important flame front in- stability is the diffusive-thermal instability, which leads to cellular flame fronts. These cellular structures have a smaller radius of curvature which in turn leads to a higher flame stretch, reducing the flammable range.
• Preferential diffusion: The effects of preferential diffusion becomes particularly important for cellular flames which occur close to the flammability limits. The reactant with the higher diffusivity reaches the cells more quickly, so the cells are enriched with the higher diffusive species which causes a shift in composition.
Up to now numerical methods still fail to predict the flammability limits correctly [149][147]. Experimental data from literature are used within this work to derive cor- relations for the pressure and temperature dependent flammability limits. Chapter 3 Basic Equations for Modeling of Non Reactive Flows
The governing equations describing the flow of a mixture are based on the assumption that the fluid is regarded as continuum. This means that the smallest volume element is large compared the mean molecular distance, and the mean free path is small com- pared to the characteristic length of the flow field.
3.1 Instantaneous Equations
The flow field of a mixture can be described by partial differential equations for the con- servation of mass, momentum (also known as Navier-Stokes equations), species mass fraction of species k and energy. With Einsteins’s summation convention they can be written in Cartesian coordinates.
• Continuity equation ∂ρ ∂ + (ρuj) = 0 (3.1) ∂t ∂xj
• Momentum equation
∂ ∂ ∂p ∂τij (ρui) + (ρujui) = − + + ρgi (3.2) ∂t ∂xj ∂xj ∂xj
If the flow is non-buoyant, the term ρgi = 0.
• Species mass fraction ∂ ∂ (ρYk) + (ρ(uj + Vk,j)Yk) =ω ˙ k (3.3) ∂t ∂xj
The source term due to chemical reaction is given here for the sake of complete- ness. It is Zero for non-reactive flows, and will be discussed in detail in chapter 4.
• Energy equation
∂ ∂ ∂p ∂qj ∂ (ρH) + (ρujH) = − + (ujτij) + SE (3.4) ∂t ∂xj ∂t ∂xj ∂xj H is the total enthalpy, which is defined as the sum of static enthalpy h and kinetic energy of the fluid. qj the molecular transport of energy, and SE is a source or sink term for example due to radiative heat losses. 18 CHAPTER 3. BASIC EQUATIONS
In the above equations, ρ is the fluid density, ui is the velocity in direction xi, p is the static pressure and τij the viscous stress tensor, which can be expressed for Newtonian fluid as
∂ui ∂uj 2 ∂ul τij = µ + − δij (3.5) ∂xj ∂xi 3 ∂xl where µ is the laminar dynamic viscosity and δ denotes the Kronecker delta. The molecular transport of energy is defined by:
K ∂T X ∂Yk qj = −λ + ρDkhk (3.6) ∂xj ∂xj k=1
λ is the conductivity, Dk the diffusion coefficient and hk the enthalpy of species k.
To close the system of differential equations, the ideal gas law relates the density to the pressure and temperature T :
K X Yk p = ρRT = ρ K X h = Ykhk (3.8) k=1 The specific enthalpy of a species k can be defined as the sum of chemical (heat of formation given at a reference temperature T0) and sensible enthalpy Z T 0 hk = ∆hf,k + cp,k dT (3.9) | {z } T0 chemical | {z } sensible cp is the heat capacity at constant pressure. In the case of an ideal gas, it is only a function of the temperature. 3.2. TURBULENT FLOWS 19 3.2 Turbulent Flows Turbulence is a three-dimensional, time-dependent, chaotic motion of vortices. Thus, a turbulent flow is characterized by a large spectrum of time and length scales. As turbulence occurs at high Reynolds numbers nearly all kinds of flows in nature and in technical applications are turbulent. Turbulence is irregular, dissipative and mixing intensive so that the transport of scalar properties is much faster than the molecular transport. The turbulent flow consists of movements with different lengthscales, in- teracting with each other. For the flow dynamic, the so called eddies represented by different dimensions, are important. The dimension of these eddies have an upper and a lower bound. The upper bound is given by the geometrical dimensions, the lower bound is given by the viscosity of the fluid. The viscosity of a fluid is able to dissipate vortices and velocity fluctuations only at very small lengthscales. Thus it avoids the formation of infinitely small eddies as the energy of very small eddies is transformed to heat. The smallest eddies to occur in a flow field are the so called Kolmogorov eddies with a length scale lη, a timescale τη and a velocity scale uη. With the viscosity ν and the dissipation rate ε they are defined as ν3 1/4 l = (3.10) η ε ν 1/2 τ = (3.11) η ε 1/4 uη = (νε) (3.12) For turbulence simulations, the Direct Numerical Simulation (DNS) approach is the most accurate, as the instantaneous transport equations (3.1 - 3.4) are solved directly without any modelling or averaging. Due to computer storage capacities and perfor- mance limitations, the application of DNS is restricted to small Reynolds numbers. The cells of the numerical grid have to have a size comparable to the Kolmogorov length scale of the flow, to be able to resolve the smallest eddies. Introducing ε ≈ u3/l into equation 3.10, the ratio of the length scale can be estimated as: 3/4 3/4 l/lη ∼ (ul/ν) = Re (3.13) Additionally, for turbulent reactive flows, the mesh has even to be fine enough to re- solve the inner structure of the flame. Nowadays, DNS studies are only possible on small geometries and flows with a reasonably small Reynolds number, but it is a valuable tool for the understanding of turbulent flames as its resolution allows a deep analysis of isolated crucial combustion phenomena. An alternative approach is the Large Eddy Simulation (LES). Three dimensional, unsteady turbulent motions are directly simulated down to a certain dimension and the small eddies are modeled. Because the small turbulent structures are modeled, a 20 CHAPTER 3. BASIC EQUATIONS coarser mesh may be used which reduces the computational cost compared to DNS. Al- though the importance of this method concerning combustion is growing in the field of research, especially for cases with large scale unsteadiness like bluff body flows, strong swirl or unsteady sound generation, the numerical costs are still too high for practical applications. As the focus of this work is the simulation of direct injection, internal combustion engines, the Reynolds averaged Navier Stokes equations (RANS) is the most suitable method. All the turbulent structures are modeled, which allows an even coarser grid than for the LES approach. In most applications, the turbulent fluctuations are of less interest than the mean values of the flow field. With RANS it is possible to simulate statistically stationary flows. The main idea of this approach goes back to the aver- aging of Osborn Reynolds [133]. He suggested splitting all variables ψ into a mean ψ¯ and a fluctuating ψ0 value ψ(x, t) = ψ¯(x, t) + ψ0(x, t) (3.14) Forming the mean on the left and right side of this equation, the mean value of the fluctuation disappears ψ0 = 0. Further rules for the Reynolds averaging method are a¯ =a ¯ a + b =a ¯ + ¯b ab¯ =a ¯¯b (3.15) ab =a ¯¯b + a0b0 ∂a ∂ = a¯ ∂x ∂x The Reynolds averaging has the disadvantage that it results in an unclosed quantity 0 0 0 of the form ρ ujui. Additionally, in several processes like combustion, large density fluctuation occur, which makes the use of Favre averages, which is a density weighted averaging procedure, more suitable. It can be written as ψ = ψe + ψ00 (3.16) Again the mean value of the fluctuation disappears ψf00, and the mean value is expressed with ρψ ψe = (3.17) ρ¯ 3.2. TURBULENT FLOWS 21 Using the Favre strategy to average the flow variables, the instantaneous balance equa- tions 3.1 - 3.4 can be derived as follows: • Continuity equation ∂ρ¯ ∂ = (¯ρu˜j) = 0 (3.18) ∂t ∂xj • Momentum equation ∂ ∂ ∂p¯ ∂τ¯ij ∂ 00 00 (¯ρu˜i) + (¯ρu˜ju˜i) = − + − ρ¯u]i uj +ρg ¯ i (3.19) ∂t ∂xj ∂xi ∂xj ∂xj • Species mass fraction ∂ ∂ ∂ 00 00 (¯ρYek) + (¯ρu˜jYek) = − Vk,jYkρ¯u]j Yk + ω˙ k (3.20) ∂t ∂xj ∂xj • Energy equation ∂ ∂p¯ ∂ (¯ρHe) − + (¯ρu˜jHe) = ∂t ∂t ∂xj K ! (3.21) ∂ ∂Te X ∂Yek 00 00 ∂ λ + ρD¯ khk − ρ¯u]j h + (ujτij) + SE ∂xj ∂xj ∂xj ∂xj k=1 On the right hand side of these equations, there occur unclosed terms due to the aver- aging procedure. The unknown quantities are the: 00 00 Reynolds Stresses u]i uj The Reynolds stresses arise from the non-linear convective term in the non-averaged equations. They are closed by the applied turbulence model (chapter 3.3), which ranges from a zero-equation Prandtl mixing length model to Reynolds Stress Models where transport equations for all components of the Reynolds stress tensor and the dissipa- tion rate are solved. 00 00 00 00 Species u]j Yk and Enthalpy u]j h Turbulent Fluxes The turbulent fluxes also result from the non-linear convective term of the un- averaged equation. The most common approach to close the turbulent fluxes is the gradient transport assumption for the species: 00 00 µt ∂Yek ρ¯u]j Yk = − (3.22) Sct,k ∂xi and for the enthalpy: ˜ 00 00 µt ∂h ρ¯u]j h = − (3.23) Sct ∂xi 22 CHAPTER 3. BASIC EQUATIONS where the turbulent viscosity µt is estimated from the turbulence model and Sct(,k) is the turbulent Schmidt number (of species k). This assumption is simple to implement and has the advantage to increases the stability of CFD codes. The diffusive contribu- tion is added to the laminar diffusion term. In a weak turbulent flame, counter-gradient transport may occur, where this assumption is not valid any more. Species Chemical Reaction Rate ω¯˙ k The modeling of the mean reaction rate is discussed in chapter 4. It is the main goal of most turbulent flame studies. The laminar diffusive fluxes for species or enthalpy as well as the pressure-velocity correlation are in general neglected. 3.3 Turbulence Models In order to close the Reynolds stress terms of the averaged equations, a large number of turbulence models have been developed, where some selected models, especially the models used within this work, are presented in the following. They can be distinguished between eddy viscosity models and Reynolds stress models Eddy Viscosity Models Within the eddy viscosity models, the Reynolds stresses are related to the mean veloc- ity gradient using the Boussinesq hypothesis 00 00 ∂u˜i ∂u˜j 2 ∂u˜l 2 ρ¯u]i uj = −µt + − δij + ρk¯ (3.24) ∂xj ∂xi 3 ∂xl 3 where k denotes the turbulent kinetic energy, which is defined as the variance of fluc- tuations in velocity: 3 1 X k = u]00u00 (3.25) 2 i i i=1 The turbulent viscosity (or eddy viscosity) µt requires further modeling. Zero- and one-equation models The Prandtl mixing length model [131] represents a zero-equation model. The tur- bulent viscosity, expressed by an algebraic equation, contains the velocity gradient. Therefore no additional transport equation has to be solved. 3.3. TURBULENCE MODELS 23 For the Prandtl-Kolmogorov one-equation model, a balance equation for the tur- bulent kinetic energy is solved. The Spalart-Allmaras [171] model solves a transport equation for the kinematic eddy viscosity. For the Prandtl mixing length model, and for the Prandtl-Kolmogorov model, em- pirical relations have to be given. They have no physical foundation and therefore they can not be applied universally. Two-equation models The most widely used turbulence models are the two equation models, as they are a good compromise between computational accuracy and numerical effort. For these models, two additional transport equations are solved. k − ε model The most popular two-equation model, which is still commonly used, is the k − ε model introduced by Jones and Launder [80]. The relation of turbulent kinetic energy k and dissipation ε to turbulent viscosity is given by the equation k2 µ = C ρ¯ (3.26) t µ ε where Cµ = 0.09 is a model constant. For the turbulent kinetic energy and dissipa- tion rate, two additional transport equations are solved with the assumption of local isotropy in the flow: ∂ ∂ ∂ µt ∂k (¯ρk) + (¯ρu˜ik) = µ + + Pk − ρε¯ (3.27) ∂t ∂xi ∂xi σk ∂xi 2 ∂ ∂ ∂ µt ∂ε ε ε (¯ρε) + (¯ρu˜iε) = µ + + Cε1 Pk − Cε2ρ¯ (3.28) ∂t ∂xi ∂xi σε ∂xi k k where Cε1 = 1.44, Cε2 = 1.92, σk = 1.0 and σε = 1.3 are the default model constants. The turbulent production Pk due to viscous and buoyant forces is modeled as 00 00 ∂u˜i Pk = −ρ¯u]i uj (3.29) ∂xj The k − ε model with the default model constants behaves fairly well for a wide range of wall-bounded and free shear flows. In addition it is very cheap, so it is still today a very popular model. But it is also well known to over-predict the spreading rate of axi-symmetrical jet flows. Flow separation is predicted too late and the amount of separation is underpredicted as a result. 24 CHAPTER 3. BASIC EQUATIONS k − ω model A further popular two-equation model is the k − ω model. Compared to the k − ε model, it does not involve the complex non-linear damping functions near the wall. Thus it is more accurate and more robust, especially concerning the prediction of flow separation. With this approach, the turbulent viscosity is expressed by the turbulent kinetic energy and the turbulent frequency ω as k µ = ρ (3.30) t ω Analogously to the k−ε model, two transport equations are solved for the turbulent kinetic energy and the turbulent frequency according to Wilcox [157] ∂ ∂ ∂ µt ∂k (¯ρk) + (¯ρu˜ik) = µ + + Pk − ρC¯ µkω (3.31) ∂t ∂xi ∂xi σk ∂xi ∂ ∂ ∂ µt ∂ω ω 2 (¯ρω) + (¯ρu˜iω) = µ + + Cω1 Pk − Cω2ρω¯ (3.32) ∂t ∂xi ∂xi σω ∂xi k where the default model constants are σk = 2.0, σω = 2.9, Cω1 = 5/9 and Cω2 = 0.075. The calculation of the source term Pk is given in equation 3.29. Although this model behaves quite well for flow separation, the dissipation in free shear flows is underpredicted. SST model To overcome the deficiencies of both, the k − ε model, which behaves well for free shear flows, and the k − ω model, which shows good performance in the near wall region, Menter introduced the SST model [100]. This model blends between the k − ω model near the wall and the k − ε model in the free stream flow. To combine these two models, the differential equations of the k − ε model are transformed to a k − ω formulation and a blending function F1 is introduced, where the Wilcox formulation is multiplied by F1 and the transformed k − ε formulation by (1 − F1). The modified transport equations are [170] ∂ ∂ ∂ µt ∂k (¯ρk) + (¯ρu˜ik) = µ + + Pk − ρC¯ µkω (3.33) ∂t ∂xi ∂xi σk3 ∂xi 3.3. TURBULENCE MODELS 25 ∂ ∂ (¯ρω) + (¯ρu˜iω) = ∂t ∂xi ∂ µt ∂ω 1 ∂k ∂ω ω 2 µ + + (1 − F1)2¯ρ + CSST 1 Pk − CSST 2ρω¯ ∂xi σω3 ∂xi σω2ω ∂xi ∂xi k (3.34) The coefficients of this model are linear combinations of the corresponding coefficients of k − ω model and the transformed k − ε model Φ = F1Φk−ω + (1 − F1)Φtrans.k−ε (3.35) To accounted for the transport of turbulent shear stress from which the overpredic- tion of eddy-viscosity results, a limiter is introduced to the eddy-viscosity a1k µt = ρ (3.36) max (a1ω, SF2) where F2 is the second blending function restricting the limiter to the wall boundary layer, S is an invariant measure of the strain rate and a1 = 0.31 is a model constant. Reynolds Stress Model The main disadvantage of the Boussinesque hypothesis is the assumption of isotropic turbulence. To overcome this drawback, a second category of turbulence models, the Reynolds stress models, was introduced. Either algebraic expressions or transport 00 00 equations are solved for the individual stress componentsρ ¯u]i uj and for the dissipa- tion rate. Thus seven additional equations have to be solved in a 3D flow simulation. Although the exact turbulence production term and the inherent modeling of stress anisotropies should make the Reynolds stress models more suited for complex flows where sudden changes in mean strain rate occur, practice shows that they do not often give better predictions than the eddy viscosity transport equation and have the disad- vantage of additional numerical cost and are less robust as more transport equations have to be solved. For the low swirl burner test case examined in chapter 7, the Baseline (BSL) Reynolds stress model is used. Similar to the SST model it blends between an ε- formulation in the outer region and an ω-formulation close to the wall. The equations for the Reynolds stresses can be written as follows: ∂ ∂ 2 0 ∂ µt ∂τij (ρτij) + (ulρτij) = −ρPij + β ρωkδij − ρΠij + µ + (3.37) ∂t ∂xl 3 ∂xl σ∗ ∂xl 26 CHAPTER 3. BASIC EQUATIONS The ε based transformed transport equation results to ∂ ∂ (ρω) + (ρu˜iω) = ∂t ∂xi (3.38) ω 2 ∂ µt ∂ω 1 ∂k ∂ω α3 Pk − β3ρω + µ + + (1 − F1)2ρ k ∂xi σω3 ∂xi σ2ω ∂xi ∂xi Here the coefficients are also blends between the ε and ω based models. A detailed description of the models used within this work can be found in [170], and an overview of several turbulence models is given by Wilcox [157]. Chapter 4 Modeling of Turbulent Reactive Flows Although it is possible to describe reacting flows with exact governing equations, it could already be seen in chapter 3 that even modeling of nonreactive flows with DNS is not possible for the simulation of reacting flows within practical devices. In addition to the transport equations for nonreactive flows, the transport equation for each species needs to be solved. As can be seen in figure 4.1 a wide range of turbulent and chemi- cal timescales is involved in combustion modeling. The chemical processes occur on a molecular level and occur on a wide range of time scales, which span several orders of magnitude. Fast chemical processes occurring on a time scale of nano seconds corre- spond to reactions which reach equilibrium conditions, whereas NO-formation happens very slowly. So the chemical timescales span a much larger range than the time scales of the fluid flow like turbulent mixing time, diffusion or heat transfer, which depends on the flow conditions and geometrical dimensions. As a result, the coupling of turbulence and reaction makes the system of partial differential equations very stiff. Figure 4.1: Physical and chemical timescales in turbulent combustion To be able to simulate combustion in industrial applications, the Reynolds aver- aged Navier-Stokes equations are still the best choice. In these equations, the chemical source term appears in unclosed form and needs to be modeled. Depending on the way how the fuel and oxidizer are injected into the combustion chamber, several combustion models have been developed for nonpremixed and premixed flames. It is possible to include effects from both sides into a model for partially premixed flames. In the following, an overview of existing models is given for non-premixed and pre- mixed flames. Furthermore, a more detailed description of the combustion models still commonly used in industry is given. These models are also used within this work for 28 CHAPTER 4. TURBULENT REACTIVE FLOWS comparison purposes. The proposed combustion model developed especially for par- tially premixed internal combustion engines is presented in section 4.4. Models for Non-Premixed Combustion During the last decades, several non-premixed combustion models have been de- veloped. A fundamental basis has been introduced by Burke and Schumann [35] in 1928. It is an analytical solution of the reacting structure of a diffusion flame, which is often called ’flame sheet model’. The assumption of this solution is infinitely fast and irreversible chemistry. The complete combustion process is described by a single step reaction. The flame front thickness is assumed to be so small that it may be treated as a geometrical surface. The reaction occurs when fuel and air are in stoichiometric proportion (mixed-is-burned). If there are no heat losses, the chemical equilibrium state is a function of mixture fraction. In this model infinitely fast and reversible chemistry is assumed. Especially at stoichiometry this assumption is different to the Burke-Schumann limit as dissociation of the products is accounted for. Therefore fuel and oxidizer can coexist, provided that the temperature is high enough. Most algorithms to calculate the chemical equilib- rium are based on the minimization of Gibbs free energy, which is used to compute the species mole fraction. Examples for the equilibrium calculation of an H2/Air system can be found in [119] and [137]. The assumption of fast chemistry (infinite Damk¨ohlernumber) is no longer valid with the Laminar Chemistry model, which is also called Finite Rate Chemistry model. Finite rate chemistry becomes important for finite Damk¨ohlernumber. The reaction zone broadens around Z = Zst, the stoichiometric mixture fraction, which results in a more complex flame structure. In the Laminar Chemistry model, all conservation equations have to be considered. If the mixing time is much faster than the chemi- cal reactions, the reactants can be considered as well mixed and the effect of turbu- lent fluctuations can be ignored. The finite chemical effects of a non-premixed coflow hydrogen-air diffusion flame have been examined by Magre et al. [98]. The Eddy Dissipation Model derived by Magnussen and Hjertager [97] is based on a single step reaction and fast chemistry assumption. As it is still industrial standard, it is also used within this work for comparison reasons. Therefore a detailed description can be found in chapter 4.1. The models mentioned up to now have the disadvantage that they do not prop- erly take the turbulence-chemistry-interaction into account. Thus several models have been introduced based on the presumed pdf (probability density function) approach of conserved scalars. The Flamelet Concept, described in detail in chapter 4.2, is based on this approach. A variety of models have been developed which are based on a presumed pdf for the mixture fraction and a transport equation for a reaction progress variable. The first to mention is the ”Mod`eleIntermittent Lagrangien” (MIL). This model is supposed to be valid in the thickened wrinkled flame regime and was first introduced by Borghi et al. [19] and improved by Gonzalez et al. [64] and Obounou et al. [114]. Borghi tried to simulate the interaction between turbulence and chemical reaction, particularly where chemical and turbulent timescales are in the same order of magnitude. For this, he takes the whole spectrum of turbulent timescales into account, instead of a single mean 29 timescale. This model assumes a high activation energy, which is applicable in combus- tion. The local instantaneous consumption rate is determined by the fuel mass fraction on the lean side, and by the oxidizer mass fraction on the rich side of the mixture. As mixture fraction pdf, a β-function is used, and for the pdf distribution of timescales an exponential function is assumed. A further model, which uses the presumed pdf for the mixture fraction and reaction progress variable models is the PEUL (probabilistic Euler Lagrangian) model [18][153]. The turbulent flow field is calculated with an Eulerian balance equation for the mean velocity, turbulent kinetic energy and dissipation rate. The Lagrangian tracking of a fluid particle only in composition space and not in physical space describes the chemi- cal reaction. Schlatter [137] applied this method for hydrogen-air and methane-air jet flames. The approach of Conditional Moment Closure was independently proposed by Kli- menko [84] and Bilger [14]. Both give a detailed review of this model in [85]. The assumption behind this model is that most of the fluctuations in the scalar quantities can be associated with the fluctuations of a key quantity. Instead of using conventional averages, the reactive scalars are conditioned on the mixture fraction. For premixed combustion a conditioning on the progress variable is proposed [85]. Means or av- erages are the first moments, second moments are variances and co-variances of the fluctuations about the averages, and third moments are triple correlations between the fluctuations. Due to the conditioned description of the transport equations, the unclosed terms can be closed by modeling the higher level moments in terms of the lower moments. Roomina et al. [135] and Bradley et al. [26] applied this method to a methane-air jet flame. Pope [129] discusses in detail different pdf transport equation models, whereas M¨obus[108] and M¨obus et al. [109] examined the scalar and the joint pdf trans- port equation methods for a fast hydrogen jet flame. Lindstedt et al.[93] examined a methane-air jet flame with the joint pdf approach. Within these models, a transport equation for the pdf itself is solved. Compared to the presumed pdf method, the shape of the pdf is not restricted, and the reaction term is closed as it depends only on local quantities. Closure is needed for molecular mixing and turbulent convection terms into the pdf. The joint pdf approach includes the flow parameters and thermochemical parameters. In the composition pdf approach only the pdf of the thermochemical vari- ables is solved for, the flow field is computed with RANS, using a turbulence model for closure. Models for Premixed Combustion As for the non-premixed flames, several models have been developed for premixed combustion. The Eddy Break Up (EBU) model proposed by Spalding [142], closes the chemical source term of the species transport equation. It assumes a one-step reaction with infinitely fast chemistry, where turbulent mixing is the rate determining process, and an infinitely thin flame. The Bray-Moss-Libby (BML) model, introduced by Bray and Moss [32] and further improved by Bray et al. [29][31] is based on the assumption of a one-step irreversible reaction between fresh gases and combustion products. Further assumptions of this model are incompressible flow, constant chemical properties and unity Lewis number. 30 CHAPTER 4. TURBULENT REACTIVE FLOWS Thus, the reaction progress can be described with a single scalar variable c, which is 0 for fresh gases and 1 for burnt gases. The pdf of the progress variable is assumed to consist of two delta functions, one at c = 0 and at c = 1. Additionally a Favre aver- aged transport equation for the mean progress variable is solved, neglecting molecular diffusion. Among several existing flame surface density models, namely the Cant, Pope and Bray (CPB) model [37], the Mantel and Borghi (MB) model [101], the Cheng and Diringer (CD) model [41] or the Choi and Huh (CH) model [42], the Coherent Flame Model (CFM) based on the ideas of Marble and Broadwell [102] is the most popular one. Under the flamelet assumption based on flame surface area, the main idea of all these models is to obtain the mean reaction rate from a product of the flame surface density and the local consumption rate per unit of flame area. Furthermore, a trans- port equation for the flame surface density is solved. The CFM model has been further improved by Trouv´eet al. [146] and Colin et al. [45] to the Extended Coherent Flame Model (ECFM). An adaption of the ECFM model to hydrogen internal combustion engines has been presented by Knop et al. [87]. The mean chemical source term de- pends only on the turbulent time scale and thus operates in the condition of infinitely fast chemistry. The so called G-equation model describes the propagation of the flame front. The G-equation was first introduced by Williams [158]. This flamelet model is a kinematic approach based on a level-set formalism for the nonreactive scalar G instead of the reaction progress variable c. An iso-scalar surface which divides the flow field in two regions is defined as G(x, t) = G0. If the point x is located in the burnt gas, then: G > G0, if it is located in the fresh gas G < G0. Although the choice of G0 is arbitrary, it is fixed for a certain combustion simulation [140]. For RANS modeling, a transport equation for the Favre mean and the variance of G is solved. This model for pre- mixed combustion is equivalent to the Flamelet concept for non-premixed combustion described in chapter 4.2, where a tansport equation for the mean and variance of the mixture fraction is solved. The G-equation model is described in detail in [121]. As well as for the non-premixed case, the basic ideas for pdf transport equation combustion modeling are the same for premixed combustion. This equation is solved using a Lagrangian particle tracking method called Monte-Carlo method. Although the Monte-Carlo method decreases the complexity of this approach, still a large num- ber of particles per grid cell is needed, which makes it difficult to apply to industrial applications as it is very time consuming. A new Monte-Carlo joint PDF approach has been introduced by Brandt et al. [23][24][25], where the representative ensemble of particles is generated in a preprocessing step with a biased random number generator. A more detailed summary of non-premixed and premixed combustion models with respect to internal hydrogen combustion engines is given in [70]. 4.1. EDDY DISSIPATION MODEL 31 4.1 Eddy Dissipation Model Still a popular combustion model in industrial applications is the Eddy Dissipation Model (EDM) derived already more than 30 years ago by Magnussen and Hjertager [97]. It is an extension to non-premixed combustion of the Eddy Break Up (EBU) model proposed by Spalding [142] and is based on the assumption of a single step reaction and fast chemistry. In addition to the common conservation and transport equations for cold flows, a transport equation for each species has to be solved [124] ∂ (ρYj) ∂ (ρuiYj) ∂ ∂Yj + = ρDj +ω ˙ j (4.1) ∂t ∂xi ∂xi ∂xi where ui is the velocity component in i direction, Dj is the diffusion coefficient, Yj the species mass fraction andω ˙ j is the reaction rate of species j. During the mixing process, the turbulent eddies follow a cascade process from the largest eddies down to molecular scales. The Eddy Dissipation Model assumes that fuel and oxygen are in separate eddies and the cascade process controls the rate of combustion. So the process which determines the reaction rate is the mixing process on the smallest length scale. As fuel and oxygen occur as fluctuating intermittent quan- tities, the fluctuations and the mean concentrations of the species can be related. The local rate of combustion is determined by the equation that gives the lowest reaction rate [124] ! ! 1 YeO YeP ε YeO YeP ρω˙ F = Cmagρ min YeF , , β ≈ Cmagρ min YeF , , β (4.2) τt ν 1 + ν k ν 1 + ν ω˙ F fuel mean burning rate τt turbulent mixing time k turbulent kinetic energy ε dissipation rate Cmag, β model constants ν stoichiometric oxygen to fuel mass ratio The indices F , O and P of the mass fraction Y denote fuel, oxidizer and products respectively. The rate of combustion is related to the turbulent mixing time scale and thus with the turbulent kinetic energy and the dissipation rate. Additionally the EDM limits the rate of combustion to the minimum mean concentration of reacting species. This model is still commonly used as it has the advantage that it can be implemented easily, needs low computational resources and is very robust, which is especially im- portant in industrial applications with complicated geometries, complicated flow fields and large grids. But this model does not include any effects of chemical kinetics and it overestimates the reaction rate where the flow field is highly strained. Additionally, the results depend strongly on the model constants, which have to be fitted to each case and cannot easily be extrapolated to new operating conditions. 32 CHAPTER 4. TURBULENT REACTIVE FLOWS 4.2 The Laminar Flamelet Concept In order to uncouple chemical kinetics from turbulence, Peters [116][117] introduced the laminar flamelet concept. It is based on the assumption that the flame structure of the local mixture fraction is determining the main characteristics of the combustion process. Thus a diffusion flame is considered to consist of an ensemble of laminar diffusion flamelets, which are stretched and wrinkled by the flow field. Due to this stretching and wrinkling, the molecular transport in the flamelet and thus the chemi- cal reaction is affected. The reaction is assumed to take place within these flamelets in a thin reactive-diffusive layer. As long as this thin sheet, also called fuel consumption or inner layer, is smaller than the Kolmogorov eddies, it can be regarded as laminar. If the Kolmogorv eddies become smaller than the inner layer, they are able to destroy the structure. The location of the reaction layer is defined by the fuel consumption layer, which for simplicity is located at stoichiometric condition, so that the flame surface is defined as the surface of stoichiometric mixture fraction Z(x, t) = Zst. The mixture fraction Z is a conserved scalar. For a system with an oxidizer stream and a fuel stream, it can be defined by the mass fractions of fuel (index 1) and oxidizer (index2) Y Z = 1 (4.3) Y1 + Y2 In the fuel stream, the mixture fraction is equal to 1, and in the oxidizer stream it is equal to 0. The stoichiometric mixture fraction is defined by −1 ν YF,1 Zst = 1 + (4.4) YO,2 YF,1 is the mass fraction of the fuel in the fuel stream, whereas YO,2 denotes the mass fraction of oxidizer in the oxidizer stream. The stoichiometric oxygen-to-fuel mass ratio is given by ν. The stoichiometric mixture fraction for H2/air is 0.0284, for methane/air 0.055 and for C3H8/air flames it is 0.0601. According to [121], the mixture fraction can be related to the well known equiva- lence ratio φ as follows Z 1 − Z φ = st (4.5) 1 − Z Zst In order to take non equilibrium effects in the combustion model into account, Peters [116] introduced the flamelet equations, which reduces the three dimensional problem in physical space to a one dimensional in mixture fraction space. The detailed chemistry calculations done in a preprocessing step thus do not need to be coupled with the turbulent flame calculation. 4.2. THE LAMINAR FLAMELET CONCEPT 33 Figure 4.2: Surface of a stoichiometric mixture in a turbulent jet diffusion flame [121] As can be seen in figure 4.2, to obtain the flamelet equations, a coordinate transfor- mation of Crocco type has to be performed using the mixture fraction as independent coordinate. At a definite time at least one coordinate, e.g. x1, has to be normal to the surface of stoichiometric mixture. The origin of the coordinate system can move together with the surface or independently, and has no influence on the results. For the coordinate transformation, x1 is replaced by the mixture fraction Z, whereas the other coordinates do not change t, x1, x2, x2 → τ, Z, Z2,Z3 (4.6) with the transformation rules follows [116] ∂ ∂ ∂Z ∂ = + ∂t ∂τ ∂t ∂Z ∂ ∂Z ∂ = (4.7) ∂x1 ∂x1 ∂Z ∂ ∂ ∂Z ∂ = + (k = 2, 3) ∂xk ∂Zk ∂xk ∂Z When x1 can be uniquely expressed by Z, the transformation is valid. Then laminar flamelets exist in the turbulent flow field. So the analysis can be performed within an infinite region around the reaction zone. The coordinate transformation results in the flamelet equations, where the species mass fraction Yi and temperature T are transformed into mixture fraction space. The flamelet equation for the species mass fraction can be written as [170] 2 ∂Yi 1 χ ∂ Yi ρ − ρ 2 =ω ˙ i (4.8) ∂t 2 Lei ∂Z 34 CHAPTER 4. TURBULENT REACTIVE FLOWS The convective derivatives have disappeared. The flamelet equations for the tempera- ture follows to 5 3 1 2 4 z }| { z}|{ z }| { z }|2 { z }| { N ∂T 1 ∂p χ ∂ T χ ∂T ∂cp X 1 χ cp,i ∂Yi ∂T ρ − − ρ − ρ − ρ ∂t c ∂T 2 ∂Z2 2c ∂Z ∂Z 2 Le c ∂Z ∂Z p p i i p (4.9) N ˙ 1 X q˙R hchem + m˙ h − = c i i c c p i p p | {z } |{z} | {z } 6 7 8 The individual terms of this equation have the following meaning: 1. instationary temperature change 2. instationary pressure change 3. heat diffusion with scalar dissipation rate as diffusion coefficient 4. convection in mixture fraction space 5. convection in mixture fraction space 6. chemical source term 7. radiation 8. enthalpy per unit time The terms 4 and 5 in equation 4.9 describe the convection in mixture fraction space, which results from differential diffusion in physical space between temperature and mixture fraction. N is the number of chemical species, ρ the density,q ˙R the radiative heat loss, cp,i the heat capacities of species i at constant pressure,ω ˙ i the chemical production rates hi the enthalpies and Lei the Lewis number of species i. The Lewis number is defined as λ Lei = (4.10) ρcpDim with the heat conductivity of the mixture λ and the molecular diffusion coefficient Dim. The instantaneous scalar dissipation rate χ [1/sec], which is an essentially non- equilibrium parameter and can be interpreted as the inverse of a characteristic diffusion time, couples the flamelet equations with the flow field in physical space 4.2. THE LAMINAR FLAMELET CONCEPT 35 ∂Z 2 χ = 2DZ (4.11) ∂xj with the thermal diffusivity DZ . The influence of convection and diffusion, normal to the surface of stoichiometric mixture fraction due to the transformation, is also in- cluded in the scalar dissipation rate. As can be seen in appendix D, where the temperature distribution for H2/air and methane/air flames as function of the mixture fraction is shown for different scalar dissipation rates, the maximum flame temperature reaches the adiabatic flame tem- perature for χ → 0. When the scalar dissipation rate increases, the solution is more and more influenced by finite rate chemistry and the increasing heat loss across the flame sheet. In general when flamelet models are used in combustion simulations, the pre- integrated results are stored as a pre processing step in flamelet libraries. Therefore, they are not dependent on the flow field, and the scalar dissipation rate χ(Z) needs modeling before its use within the flamelet equations. According to [116] it can be approximated as a function of mixture fraction diffusivity a χ = exp{−2[erfc−1(2Z)]2} (4.12) π with the assumption of constant density and diffusivity. erfc is the complementary er- ror function. As CFD codes provide only an unconditioned, averaged scalar dissipation rate, the above equation can also be expressed as follows exp{−2[erfc−1(2Z)]2} f(Z) χ(Z) = χ(Z ) ≡ χ(Z ) (4.13) st −1 2 st exp{−2[erfc (2Zst)] } f(Zst) For this equation the definition of the mean scalar dissipation rate at stoichiometric mixture χst is needed, and can be calculated as ε˜ 002 Cχ Zg f(Zst) χ(Z ) = k˜ (4.14) e st R 1 ˜ 0 f(Z)P (Z) dZ ˜ where Cχ = 2 and P (Z) is the Favre averaged probability density function. A further approach to obtain flamelet tables for a combustion simulation is to eval- uate a laminar, counterflow diffusion flame (see figure 4.3). With this geometrical setup it is possible to obtain a one-dimensional diffusion flame structure, which can also be mapped into mixture fraction space. Within this work, the code CHEM1D [172] is used for this purpose. In the counterflow configuration, the flame is stabilized near the stagnation plane and can be at first considered as two-dimensional planar or axially symmetric. In the 36 CHAPTER 4. TURBULENT REACTIVE FLOWS Figure 4.3: Counterflow diffusion flame following u is the velocity in x or radial direction, and v is the velocity in y or axial direction. With the ansatz u = Ux, the velocity u is related to the gradient U of the velocity u and the direction x. Therefore, the governing equations can be expressed as follows: Continuity d(ρv) + (j + 1)ρU = 0 (4.15) dy Momentum dU d dU ρv = −ρU 2 + ρ a2 + µ (4.16) dy ∞ dy dy Mixture fraction dZ d dZ ρv = ρD (4.17) dy dy dy Reactive scalars dψ d dψ ρv i = ρD i + ω (4.18) dy dy i dy i For planar configurations j = 0, and for axially symmetric configurations j = 1. a is the strain rate. The scalar dissipation rate is already given in equation 4.11. With the assumption of sufficiently large velocities, far away from the stagnation plane, a potential flow may be assumed. Thus the boundary conditions for the oxidizer stream (y → ∞) are v∞ = −ay, U∞ = a and Z = 0. For the fuel side (y → −∞) 1/2 the boundary conditions are v−∞ = −(ρ∞/ρ−∞) ay, U−∞ = aρ∞/ρ−∞ and Z = 1. A 4.2. THE LAMINAR FLAMELET CONCEPT 37 detailed description of the laminar counterflow flame can also be found in [121]. Within the flamelet generation code implemented in ANSYS CFX 11 [170], only a single value for the Lewis number of the single species may be provided by the user, which leads in most cases to the assumption of unity Lewis number for all species. The CHEM1D code has the advantage, that it also accounts for the species differential dif- fusion. Three different models are available. A Lewis number can be provided for each species according to ANSYS CFX 11. The ’Mixtureaveraged’ model and the ’Complex’ model calculate the diffusion velocities based on assumption described in chapter 2. The Favre mean species mass fraction can then be calculated with a probability density function P˜ Z 1 ˜ ˜ Yi = Yi(Z)P (Z)dZ (4.19) 0 Commonly a β-pdf shape is used for P˜(Z, x, t). The pre-integration of flamelet ta- bles is done with CFX-RIF [170] automatically. The tables which are generated with CHEM1D [172] still have to be pre-integrated, which is done within this work with the calculation of the β-pdf function described in detail in appendix A. Within the framework of the flamelet model, instead of solving the transport equa- tion for each species, only the transport equations for the Favre mean mixture fraction and its variance are solved. Assuming that all species have equal diffusivities, the transport equation for the Favre mean mixture fraction can be written as [170] " # ∂(¯ρZe) ∂(¯ρu˜jZe) ∂ µt ∂Ze + = µ¯ + (4.20) ∂t ∂xj ∂xj σZ ∂xj where µ is the molecular dynamic viscosity, µt the turbulent viscosity and σZ is a model coefficient. As the mixture fraction is a conserved scalar, its transport equation contains no source term. The transport equation for the mixture fraction variance Zg002 can be modeled as follows [170] 2 002 002 " 002 # ! ∂(¯ρZg) ∂(¯ρu˜jZg) ∂ µt ∂Zg µt ∂Ze + = µ¯ + + 2 − ρ¯χ (4.21) ∂t ∂x ∂x σ ∂x σ ∂x e j j Zg002 j Z j where σ and C are model coefficients. The first term on the right hand side rep- Zg002 χ resents the production of variance, whereas the last term models its dissipation. The instantaneous scalar dissipation rate χe is modeled as ε˜ 002 χ = Cχ Zg (4.22) e k˜ 38 CHAPTER 4. TURBULENT REACTIVE FLOWS withε ˜ and k˜ being the dissipation rate and the turbulent kinetic energy, respectively. ˜ As the species mass fractions Yi are previously stored in a flamelet table, they can be obtained within the CFD simulation by coupling the CFD code with the flamelet libraries as shown in figure 4.4, adapted from [121] Figure 4.4: Code structure of the flamelet model The properties Favre mean mixture fraction, mixture fraction variance and scalar dissipation rate are calculated from the CFD code. From these values, the pre- integrated mean values for the species mass fractions are then obtained by look-up in the flamelet library. 4.3 The Reaction Progress Variable In addition to the conservation equations together with the equations of state, the transport equation for the Favre mean mixture fractions and its variance, the trans- port equation for the Favre mean progress variablec ˜ has to be solved for the improved combustion model presented in chapter 4.4. Similar ideas have, for example, been used by Domingo et al. [52] for partially premixed flamelets in a Large Eddy Simulation (LES) for non-premixed lifted flames and a similar model where the flamelet model is coupled with a pdf transport equation model has been used by Demiraydin et al. [50] for turbulent non-premixed methane/air flames. The progress variable is defined as the normalized mass fraction of products and describes the progress of reaction within a flamelet. By definition c = 1 for completely burned mixtures and c = 0 for unburned mixtures. ˜ The mean species mass fraction Yi can be approximated as Yei = (1 − c˜) Yei,u +c ˜Yei,b (4.23) where the index u and b denote the unburned and burned values of the species mass fraction. The species mass fraction of the burned state Yei,b is derived from the flamelet model in CFX. The mass fraction for the unburned state can be obtained by linear blending of fuel and oxidizer composition. 4.3. THE REACTION PROGRESS VARIABLE 39 The transport equation of the reaction progress variable has the following form [28] ∂ (¯ρc˜) ∂ (¯ρu˜jc˜) ∂ 00 00 + = − ρ¯ugj c + w˙c (4.24) ∂t¯ ∂xj ∂xj 00 00 whereas the turbulent scalar transport termρ ¯ugj c is modeled with a gradient transport assumption: ∂ (¯ρc˜) ∂ (¯ρu˜jc˜) ∂ νt ∂c˜ + = − ρ¯ · + w˙c (4.25) ∂t¯ ∂xj ∂xj Sct ∂xj Sct = 0.9 is the turbulent Schmidt number and νt is the turbulent viscosity. To close the source term w˙c in equation 4.25, several models are examined within this work, which are presented in the following. Zimont The turbulent flame speed closure model was first introduced by Zimont [164] and further improved by Zimont et al. [166][165]. The model assumes that the combustion front consumes fuel at a turbulent flame speed st. The flame brush width of the com- bustion front increases in time and space. This is the so called Intermediate Steady Propagation (ISP) combustion regime. The source termw ˙c of the transport equation for the reaction progress variable (equation 4.25) is modeled by w˙c = ρust|∇c˜| (4.26) where ρu is the density of the unburned, unreacted mixture. The turbulent flame speed is then defined according to [165] as 03/4 1/2 −1/4 1/4 st = AGu sl au lt (4.27) where A is a model constant (for hydrogen combustion it is equal to 0.61) and u0 is the turbulent velocity intensity. The turbulent integral length scale lt can be obtained 3/2 from the turbulent kinetic energy and its dissipation rate lt = k /ε. au is the thermal diffusivity of the unreacted mixture. Wrinkling and straining of the flame front caused by the turbulent velocity field are included in the expression for the turbulent flame speed, which is also dependent on changes in operating pressure, fuel concentration, composition and temperature in a physically meaningful way. The stretch factor G models the extinction and accounts for reduction of the flame velocity due to large strain rate. It can be obtained by integrating the log-normal distribution of the turbulent dissipation rate [164] 40 CHAPTER 4. TURBULENT REACTIVE FLOWS ( ) 1 r 1 ε σ G = erfc − ln cr + (4.28) 2 2σ εe 2 In this equation σ is the standard deviation of the turbulent dissipation rate ε lt σ = µstrln (4.29) lη where the default value of the model constant µstr is 0.28. The critical dissipation rate εcr can be obtained from the kinematic viscosity ν and the critical velocity gradient gcr 2 εcr = 15νgcr (4.30) If the velocity gradient is smaller than this critical value, differently strained flamelets are reduced to unstrained ones, where no quenching occurs. If it is larger, the flamelets are highly strained. For ε > εcr extinction of the flamelets occurs. The model has been successfully applied to a hydrogen/air jet diffusion flame by Zimmermann et al. [163][154]. Bradley As second correlation for the turbulent flame speed, the correlation of Bradley et al. [27] was examined. It is based on a large number of experiments performed to measure the turbulent flame speed for different fuels and conditions, which are presented by Abdel-Gayed et al. [4]. These measurements were performed using the double-kernel technique, where two kernels are ignited at the same time and meet each other in homogeneous, isotropic stationary turbulence. A Schlieren technique is used to detect the flame kernels. The turbulent flame speed is derived from the time span until the kernels collide. Bradley et al.[27] validated their correlation in the range of 0.01 < Ka Le < 0.63, where Ka is the Karlovitz stretch factor, for which they obtained the following equation st −0.3 0 = 0.88(Ka Le) (4.31) uk 0 uk is the effective cold mixture root mean square (r.m.s.) turbulent velocity acting on a flame. 4.4. PARTIALLY PREMIXED FLAMES 41 For the Karlovitz number, the definition derived by Abdel-Gayed et al. [1][2] is used 0 2 u −0.5 Ka = 0.157 Ret (4.32) sl When time approaches infinity t → ∞, the effective cold mixture r.m.s. turbulent velocity can be approximated as u0 [95], which results in the following expression for the turbulent flame speed −0.3 00.55 0.6 0.15 −0.15 st = C Le u sl lt ν (4.33) with C = 0.88 · 0.157−0.3. Lindstedt-Vaos´ Lindstedt and V´aos [94] propose to model the source term of equation 4.25, the trans- port equation for the reaction progress variable, using a correlation for the flame surface density Σ, based on an assumption of fractal dimension of the flame surface 3/4 sl ε˜ w˙c = ρustΣ = CRρu c˜(1 − c˜) (4.34) ν1/4 k˜ Depending on the experimental configuration, the model constant CR lies between 1.25 and 1.95. For example, the model constant with a value of 1.25 is adapted to the database of Abdel-Gayed [2], whereas the value of 1.95 is correlated to the database of Gulati et al. [68]. Lindstedt and V´aos [94] tuned the constant to 2.6. Bray et al. [30] investigated a methane flame with promising results although a 0 higher value of CR was needed to fit the data at high u . Brandl et al. [24] also suc- cessfully applied this model at higher pressures, where he modeled the turbulent scalar 00 00 transport termρ ¯ugj c with a first order gradient transport assumption. 4.4 Combustion Model for Partially Premixed Flames In a direct injection internal combustion engine for automotive applications, cold fuel is injected into the cylinder during the compression stroke, often in multiple injections. Finally the mixture is ignited. The compression causes a rise of pressure and tem- perature of the mixture, which results in a temperature difference between pure air, fuel/air mixture and injected (cold) hydrogen. Thus, the influence of temperature and pressure of the unreacted mixture is very important and has to be taken into account 42 CHAPTER 4. TURBULENT REACTIVE FLOWS in the flamelet tables. The effect on combustion products like H2O or CO2 or radicals like O2 and OH are demonstrated for a hydrogen/air flame in chapter 5.2, figure 5.4 and for a methane/air flame in chapter 6.2, figure 6.3. Although the mass fractions of the radicals are small, they have a strong influence on the pollutant prediction. In the models to close the source term of the reaction progress variable transport equation, the laminar flame speed sl occurs within all mentioned correlations (Zimont, Bradley and Lindstedt-V´aos). As could be seen in chapter 2.3, and can be seen for hydrogen, methane and heptane combustion in chapters 5.4 and 6.4, the laminar flame speed is also strongly dependent on temperature and pressure, as well as on the mixture fraction or equivalence ratio, respectively. As the laminar flame speed was especially used in the past for premixed combustion, only a constant laminar flame speed was generally applied. With the extended use for partially premixed flames, also the de- pendence on mixture fraction is applied successfully in [161][162][163][154][136]. To describe the combustion process in a direct injection gasoline engine, Kech et al. [81] used a similar approach as presented in this work. It is also based on a combination of mixture fraction and reaction progress variable. The laminar flame speed was calculated in dependence on the mean mixture fraction and its variance as follows: 2/3 s¯ = s P Z, Z002 C (4.35) l(Z˜) l,Zst Zst e g Zst where CZst = 0.9. This is only an approximation of the detailed description presented below. Polifke et al. [126] developed an extended turbulent flame speed closure model for inhomogeneously premixed combustion, where the influence of model parameters on mixture fraction is examined in detail. Although the laminar flame speed covered the whole mixture fraction range also occuring in non premixed combustion, it is only dependent on the mean mixture fraction, and not on its variance. In the framework of LES, Domingo et al. [52] showed that equation 4.25 neglects terms expressing fluxes in mixture fraction space. This approach was also picked up by Duwig et al. [55] and Durand [54]. The right hand side of equation 4.25 then contains two extra terms. The first term contains the scalar dissipation rate and represents the mixing effect. It can be neglected when the fuel mass fraction in the burned state is a linear function of the mixture fraction. In the model considered in this work, this is only the case near stoi- chiometry and thus can be neglected compared to the other terms. The second term is the cross-scalar dissipation rate defined between the progress variable and the mixture fraction. When the length scale of the mixture fraction variation is large compared to the flame thickness, this term can also be neglected, which was done in this work. The mixture fraction dependence of the laminar flame speed can be obtained by two options, either by previously calculating the laminar flame speed for different mix- ture fractions, temperatures and pressures with a chemistry code like CHEMKIN [82] or CHEM1D [172], or by using one of the several published correlations available for different kinds of fuel [107][86][111]. Using a chemistry code has the disadvantage that a large range of mixture fraction, pressure and temperature combinations has to be calculated in a previous step for internal combustion engines, which is very time con- suming. Due to the analytical expressions of the laminar flame speed correlations, this dependence can be accounted for very quickly. But all expressions have the dis- 4.4. PARTIALLY PREMIXED FLAMES 43 advantage, as they are developed for premixed combustion cases, that they are not valid in the whole mixture fraction range for diffusion flames with 0 ≤ Z ≤ 1. For non-premixed flames Z = 1 for fuels, and Z = 0 for the oxidizer. To extend the range of validity of the different correlations for laminar flame speed, within this work, the last valid points of the flame speed correlation on the lean or rich side, respectively, are connected linearly with the lower or upper flammability limits. Outside the flammability limits the laminar flame speed is so small that the flame is quenched immediately, as soon as a perturbation occurs. So a value of 0 m/s can be assumed for the laminar flame speed outside the flammable range. The previous optimizations widen the range of validity of any correlation for the laminar flame speed to the whole range occurring in diffusion flames. The correlations are dependent on pressure, temperature and equivalence ratio (or mixture fraction). As commercial CFD codes solve time-averaged equations, only the Favre averaged mean mixture fraction and its variance are known. The time-averaged source term in the transport equation for the reaction progress variable in partially premixed combustion contains a mixture fraction dependence within the laminar flame speed. Additionally to take the turbulence effects on the time averaged source term into account, a proba- bility density function (pdf) P is used. The resulting effective laminar flame speed eZ,e Zg002 used in the correlations is defined by: Z s˜ (Z, Z002) = s (Z)P (Z; x, t) dZ (4.36) l e g l eZ,e Zg002 Thus the effective laminar flame speed also depends on the Favre averaged mixture fraction and its variance. To be able to use the effective laminar flame speed efficiently within a CFD code, a table has to be generated for the laminar flame speed depending on pressure, temperature, Favre mean mixture fraction and mixture fraction variance. Examples of the laminar flame speed correlation extension and table generation are given in chapter 5 for hydrogen and in chapter 6 for methane and heptane combustion. It is common practice in non-premixed combustion simulations to use a β-pdf shape [76] Γ(a + b) P˜(Z; x, t) = Za−1(1 − Z)b−1 (4.37) Γ(a)Γ(b) where Γ is the gamma function, and α and β can be expressed by the mean mix- ture fraction and its variance. The implemented β-pdf approach within this work is explained in detail in appendix A. Chapter 5 H2 Jet Flame For evaluation purposes, the ’H2 Jet Flame’ of the TNF workshop [167] has been se- lected as a test case for the non-premixed hydrogen diffusion flame model. It offers the advantage of well defined boundary conditions and a relatively simple flow field. The hydrogen jet flame has been experimentally examined by Barlow and Carter [10][8][9] and Flury [62] with a helium dilution of 0 %, 20 % and 40 %, which offers a good data set for comparisons with numerical methods. Thus it has been used by several researchers for model evaluation in the framework of Large Eddy Simulations (LES) and RANS simulations. Near the injection nozzle, the flame is characterized by thin mixing and reaction layers, whereas downstream the burnt gases occupy a large region of the domain and the reaction layer is broad with near equilibrium regions at the flame tip. Due to the relatively fast combustion chemistry of hydrogen, there is no local extinction at the injection velocities considered. Additionally, there occurs no soot formation. Figure 5.1: Hydrogen Jet Flame The following chapters give an overview of the computational domain and bound- ary conditions derived from the experimental setup(chapter 5.1), followed by exami- nation of the influence of the Lewis number (chapter 5.2), the comparison and choice 5.1. BOUNDARY CONDITIONS 45 of a suitable turbulence model (chapter 5.3), and the influence of the laminar flame speed (chapter 5.4). The results, where also different closure models are compared, are discussed in chapter 5.5. The presented simulations are performed using the CFD (computational fluid dynamics) tool ANSYS CFX [170]. 5.1 Computational Domain and Boundary Conditions The hydrogen jet in both configurations is injected from a central tube with an inner diameter of 3.75 mm and a wall thickness of 0.55 mm. In the case of Barlow and Carter [9] the burner was centered at the exit of a vertical prismatic wind tunnel contraction with a cross sectional area of 0.3 m × 0.3 m. The air coflow at the experimental setup of Flury [62] is surrounded by a vertical wind tunnel with a hexagonal base and a di- ameter of 0.6 m. A comparison of both setups with LDV (Laser Doppler Velocimetry) measurements has shown no difference [62]. The radial positions where experimental data are available are marked in red in figure 5.1. Using the CFD program CFX, it is not possible to solve 2D flow fields. To be able to solve a ’quasi 2D’ case, a plane with the dimension 0-150 mm in radial direction and 0-800 mm in axial direction has been rotated by three degrees to create a 3D case. To be able to use a structured grid, the axis has been replaced by a small tube (diam- eter 0.02 mm). Calculations with more cells in the third direction showed negligible differences, so a width of one cell in circumferential direction is sufficient and reduces calculation time. In the domain with 156,000 nodes and 236,193 elements, the grid has been refined near the visible flame. A grid refinement study showed that the solution is grid independent. A symmetry boundary condition has been chosen for the rotated surfaces. At the “axis tube” and the radial boundary an adiabatic no slip wall has been placed. The air coflow has a velocity of 1 m/s initialized with a constant profile, where the mean velocity varies over the radius by about 1.3 % and the turbulence intensity is about 10 %. The inlet temperature is set to 298 K. Figure 5.2(a) shows the normalized velocity profile measured with 20 % and 40 % helium dilution. Using pure hydrogen, no measurements of the velocity profile exist, since it was not possible to seed the fuel due to security reasons [137]. The velocity was normalized in both cases with the maximum velocity Umax measured. In the case of a dilution of 20 % helium, the maximum jet velocity is 302.49 m/s, with a dilution of 40 % helium the maximum jet velocity is 263.21 m/s. The Reynolds number of the hydrogen jet is 10,000. Figure 5.2(a) shows an additional velocity inlet profile calcu- lated with the power law 1 r n u(r) = 1 − U (5.1) R max 1 A fully turbulent velocity profile can be obtained by the so called 7 -rule, where n = 7, but the profile representing the measurements is obtained with n = 12 [113], so the 46 CHAPTER 5. H2 JET FLAME (a) Normalized radial inlet velocity (b) Turbulent kinetic energy Figure 5.2: Profiles of velocity and turbulent kinetic energy for different dilutions of Helium, 2 20 % Helium dilution, 4 40 % Helium dilution, - power law with n = 12 measured profile does not represent fully developed turbulent tube flow. The measured inlet profile for the turbulent kinetic energy of the radial inlet profile with a dilution of 20 % and 40 % Helium is shown in figure 5.2(b). The turbulent kinetic energy can be expressed as 1 k ≡ u0 u0 (5.2) 2 i i For the dissipation rate ε no measurements are available. To determine the rela- tionship between Taylor and Kolmogorov scales, Pope [130] defines the lengthscale L characterizing the large eddies as k3/2 L ≡ (5.3) ε CFX [170] advises to use 30 % of the inner diameter of the tube as the choice of the length scale yielding L = 0.3 · 3.75mm. With this assumption, the dissipation rate ε can be evaluated. Calculations with the Eddy Dissipation Model and different length scales, ranging from 1L, 0.5L and 0.3L, for the calculation of the dissipation rate, used as inlet boundary condition, show nearly no difference in the behavior of the achieved results (see figure 5.3). This has also been seen by Schlatter [137] who performed several tests with different levels of k and ε and obtained only negligible sensitivity of the results. The inlet temperature of the fuel is 298 K. As the modification of hydrogen is not known, hydrogen is assumed to be normal hydrogen (see appendix E), which consists of 25% para and 75% ortho hydrogen. The CFD simulations were performed using the High Resolution discretization scheme and double precision. Depending on the local flow field, the High Resolu- tion discretization scheme blends between a first and a second order advection scheme. 5.2. INFLUENCE OF CHEMISTRY TABLE 47 Figure 5.3: Influence of different length scales on the simulation results The convergence criterion is 1e-6 for the residuum of all transport equations. Further boundary conditions like the choice of the turbulence model, the flamelet table and laminar flame speed are given in the following chapters 5.2-5.4 5.2 Influence of Chemistry Table Within a DI engine, the cold fuel is injected into the cylinder during the compression stroke. This is often done in multiple injections until the mixture is ignited. The com- pression causes a rise of pressure and temperature of the mixture, which results in a temperature difference between pure air, hydrogen/air mixture and injected (cold) hy- drogen. As described in chapter 4.4, the change in unburned temperature and pressure has an influence on the flamelet table. The laminar flamelet tables were generated with CHEM1D [172]. The β-pdf de- scribed in appendix A is used to generate flamelet tables which take the turbulence effects into account. They are needed for the Flamelet model, and the Flamelet + Re- action Progress Variable model. The Conaire hydrogen chemical reaction mechanism [46] used contains 10 species. It is listed in appendix C.2. For this flame, pressure and temperature variations as they occur in internal combustion engines are negligible. Therefore flamelet tables were generated with a pressure of 1 bar and a temperature of fuel and oxidizer of 298 K. 10 laminar flamelet tables ranging from a strain rate of a = 10 1/s to the maximum strain rate depending if differential diffusion effects are taking into account or not are used. 100 points are distributed non-uniformly along the mixture fraction axis with a refinement of resolution near stoichiometry. Temperature and Pressure Dependence Figure 5.4 shows the influence of pressure and temperature on the mass fractions of H2O, O and OH as a function of mixture fraction Z. Temperatures and pressures range from ambient conditions of 1 bar and 300 K to elevated conditions of 50 bar and 900 K. The laminar flamelet tables have been generated with a strain rate of 10 1/s. 48 CHAPTER 5. H2 JET FLAME (a) YH2O (b) YO (c) YOH Figure 5.4: Temperature and pressure dependence of species mass fractions of H2O, O and OH In addition, the stoichiometric mixture fraction is plotted. It can be seen that changes in temperature and pressure have little influence on the mass fraction of H2O in contrast to the strong effect on the mass fractions of the mi- nor species O and OH. The influence on H2O occurs mainly at stoichiometric mixture fraction. The maximum mass fraction drops with increasing temperature and increases with increasing pressure. It can also be expected that the influence on the global flow field is nearly negligible since the mass fraction of H2O represents the main combustion product of hydrogen, and so temperature and density are closely related to the H2O concentration. Figures 5.4(b) and 5.4(c) clearly show changes in the mass fraction of O and OH. With increasing temperature both mass fractions increase strongly and with increasing pressure the maximum decreases. Increasing temperature also broadens the mixture fraction range where the minor species occur. Increasing pressure decreases the mix- ture fraction range. As the mass fractions of O and OH are very small, they only exert little influence 5.2. INFLUENCE OF CHEMISTRY TABLE 49 on the main flow field, but as can be seen from the Zel’dovich mechanism [160], these radicals have a large effect on the NO production and thus on the pollutant prediction. O + N2 ↔ NO + N N + O2 ↔ NO + O (5.4) N + OH ↔ NO + H Therefore if only the flow field is of interest, it is sufficient to use one flamelet table at one pressure and temperature, but as soon as an accurate prediction of NO is of interest, it is necessary to use pressure and temperature dependent flamelet tables. Differential Diffusion Flamelet tables for the hydrogen jet flame have been generated with the laminar coun- terflow flame of CHEM1D [172] at 298 K and 1 bar with a strain rate of 10 1/s. As reaction mechanism, the Conaire mechanism [46] has been selected. As described in chapter 2.1, there are three options available to describe the diffusion process: com- plex, mixture averaged and with user supplied Lewis numbers for all species. Figure 5.5 shows a comparison of these three methods where the Lewis number for all species is set to unity. Additionally the chemical equilibrium (dark blue dots) obtained with Cosi- lab [173] and the stoichiometric mixture fraction (red vertical dashed line) are shown. It can clearly be seen, that the simulations with the options ’complex’ and ’mixture averaged’ result with a maximum temperature of 2540 K in a higher value than the adiabatic flame temperature with 2390 K. As expected, the maximum temperature using unity Lewis numbers lies slightly below the adiabatic flame temperature. These super-equilibrium temperature levels have also been observed for turbulent H2/N2/air jet diffusion flames [106] and have been investigated on laminar counterflow flames and turbulent H2/N2/air flames with direct numerical simulations (DNS) [63]. Such a significant difference between unity and non-unity Lewis number has also a strong influence on the mixing process, which can already be seen in chapter 5.3 and 5.5. Gicquel et al. [63] examined in detail the effects which lead to the super-equilibrium temperature and found that the local mixture when taking differential diffusion effects into account is less strongly diluted with inert species than the mixture obtained with unity Lewis numbers. Whereas the amount of hydrogen and oxygen determines the heat release, the inert species act as heat ‘sink’, since the inert species have to be heated without providing energy themselves. Comparison with Temperature Scatter Plots Whether the consideration of differential diffusion effects is really important for this hydrogen jet flame can be estimated from experimental scatter plots. Figure 5.6 shows the experimental temperature scatter plot at y/L = 1/8 compared to the flamelet tables at different strain rates between 10 and 5000 1/s. It can be seen that flamelet 50 CHAPTER 5. H2 JET FLAME Figure 5.5: Differential diffusion effects tables with unity Lewis number fit better the experiment than the one which takes differential diffusion effects (Complex) into account. Thus, the flame is not influenced by this effect. Within the HyICE project, Colin [44] showed that for IC engines dif- ferential diffusion effects can be neglected already for Reynolds numbers above 100. With increasing strain rate a, the maximum temperature decreases. Simulations with CHEM1D [172] yield quenching at a strain rate of a = 11627 1/s. At this strain rate, the stoichiometric scalar dissipation rate is χst = 145 1/s. Thus, the experiment is far from quenching. A full set of scatter plots at all measured downstream positions is given in appendix D.1. Figure 5.6: Temperature distribution at y/L = 1/8 5.3. CHOICE OF TURBULENCE MODEL 51 5.3 Choice of Turbulence Model The spreading rate of round jets, known as the ’round-jet/plane-jet anomaly’ is over predicted by most turbulence models. It is defined as the value of radial distance to axial distance where the velocity reaches half of the centerline value [157]. Different approaches have been developed to improve the prediction [128][105][110]. Dally et al. [48] propose to use either the k-ε model or the Reynolds Stress model with a modifica- tion of Cε1 = 1.6 (standard value: Cε1 = 1.44). Due to the greater computational cost of the Reynolds Stress model, the k-ε model has been chosen here. Calculations have been performed with different values of Cε1 (1.44, 1.5 and 1.6). The value of 1.5 shows the best fit to the experimental data of the mixture fraction field, which is shown in figure 5.7. Figure 5.7(a) shows the mean mixture fraction plotted along axial direction normalized with the visible flame length Lvis = 675 mm. As the mixture fraction is a conserved scalar it is only influenced by the flow field. Therefore it is suited to com- pare different turbulence models. Figure 5.7(b) shows the isolines of the stoichiometric mixture fraction. The experimental stoichiometric flame length is Lst = 475 mm. Although the standard values of the k-ε model predict the stoichiometric flame length well, it is in poor agreement with the experiment close to the axis. Thus the following calculations were performed with Cε1 = 1.5. With this choice, one slightly underpre- dicts the mixture fraction at the first measurement point at y/Lvis = 0.125 (y/L = 1/8) and slightly overpredicts the next measurement point y/Lvis = 0.25 and the measure- ment points closer to the jet tip. (a) Mixture fraction over y/Lvis (b) Stoichiometric mixture frac- tion profiles Figure 5.7: Influence of different values of Cε1 52 CHAPTER 5. H2 JET FLAME 5.4 Implementation of Laminar Flame Speed Within the laminar flamelet plus reaction progress variable model presented here, three source term closure models for the reaction progress variable transport equa- tion are compared: The Zimont model [166][165], the Lindstedt-V´aos model [94] and the Bradley correlation [27]. Within all three models, the laminar flame speed has to be provided. Commonly, a constant value is used. When premixed flames are studied, the laminar flame speed is in general provided as a function of the mixture fraction or the equivalence ratio. Within this work, it is additionally provided as a function of the mean mixture fraction and mixture fraction variance to account for turbulence effects. Figure 5.8: Laminar flame speed of hydrogen/air mixture for different experiments and correlations at ambient temperature and pressure Figure 5.8 shows a comparison of different experiments and correlations for the laminar flame speed at ambient temperature and pressure. The symbols denote exper- iments (dark blue rhombe: Verhelst [150], red squares: Dowdy et al. [53], green delta: Koroll et al. [88], yellow circle: Egolfopoulos et al. [59], bright blue rhombe: Aung et al. [7], blue line: simulation with CHEM1D [172], green line: correlation of Iijima and Takeno [75], orange line: correlation of Knop et al. [86]). As can be seen, the maximum value of the laminar flame speed is not reached at stoichiometric conditions for a hydrogen flame, as is the case for many hydrocarbon fuels, but on the rich side for hydrogen-air combustion. The double kernel measurements of Koroll [88] give higher burning velocities than the stretch-free measurements by [150][53][59][7]. The simulation with CHEM1D and the chemistry of Conaire [46] shows good agreement especially with the experiment of Dowdy [53]. The correlation of Knop [86] agrees well with the experiments, especially at the lean side. As was demonstrated by Zimmermann et al. [161] the laminar flame speed correlation used has a noticeable influence on the results. In the following details of the correlations used are presented. 5.4. IMPLEMENTATION OF LAMINAR FLAME SPEED 53 Correlation of Milton and Keck [107] α β Tu p sl = Sl0 (5.5) T0 p0 with Sl0 = 2.17m/s, α = 1.26, β = 0.26, p0 = 1atm, T0 = 298K. The domain of validity is 0.5atm < p < 7atm, and 198K ≤ Tu ≤ 550K at a stoichiometric equivalence ratio φ = 1. Thus, this correlation does not take the changes in equivalence ratio into account. The laminar flame speed for this test case is sl = 2.163m/s. Correlation of Iijima and Takeno [75] α p Tu sl = Sl0 1 + β ln (5.6) p0 T0 with 2 3 Sl0 = 2.98 − (φ − 1.70) + 0.32(φ − 1.70) [m/s] α = 1.54 + 0.026(φ − 1) β = 0.43 + 0.003(φ − 1) p0 = 1atm, T0 = 291K The domain of validity is 0.5 ≤ φ ≤ 4.0, 0.5atm < p < 25atm and 291K ≤ Tu ≤ 500K Correlation of Knop et al. [86] α β Tu p sl = Sl0 (5.7) T0 p0 The correlation domain is separated in two regions • 0.25 ≤ φ ≤ 1.90 5 4 3 2 Sl0 = −1.7460φ + 10.8087φ − 25.5368φ + 26.9161φ − 9.4727φ + 1.1588 • 1.90 < φ φ − 1.90 S = S (φ = 1.90) + [1.08 − S (φ = 1.90)] l0 l0 l0 5.05 − 1.90 with β = −0.9791φ2 + 1.7829φ − 0.6323 α = 1.62 5 p0 = 1.013 · 10 P a T0 = 298 K 54 CHAPTER 5. H2 JET FLAME The validity domain of this correlation is 0.25 ≤ φ ≤ 5.0, 1 bar < p < 10 bar and 298 K ≤ Tu ≤ 550 K As under real engine conditions the ’unburned’ temperature Tu and pressure p are not constant, several researchers [65][151][150][152][34] have examined the influence on the laminar flame speed for hydrogen combustion. Figure 5.9 shows the results ob- tained by Verhelst [151] for lean mixtures. With increasing pressure, the laminar flame speed decreases, and with increasing temperature it increases. Figure 5.9: Temperature and pressure dependence of the laminar flame speed [151] Extension of the laminar flame speed Because all the correlations for the laminar flame speed have been derived in the con- text of premixed combustion, the validity domain does not cover the whole range of diffusion and partially premixed flames. In order to amend this problem, an approach is presented in chapter 4.4 to overcome this problem using the pressure and temper- ature dependent flammability limits. Since the mean laminar flame speed is not only dependent on the Favre mean mixture fraction, but also on its variance, the improve- ments possible using a pre integrated laminar flame speed table are demonstrated in chapter 5.5. The hydrogen limits of flammability in air are 4 vol. % and 75 vol. % [38] at ambi- ent temperature and pressure. That means that the flammable range lies in the range of fuel/air equivalence ratios of 0.099 ≤ φ ≤ 7.135 which corresponds to a mixture fraction range of 0.0029 ≤ Z ≤ 0.1726. As can be seen in figure 5.8 where the correla- tion of Iijima & Takeno [75] is plotted, the correlations, generally derived for premixed combustion, don’t cover the whole flammable range, and thus are not applicable in the whole range of diffusion flames, with 0 ≤ Z ≤ 1, usually producing unphysical results outside the given domain of validity (see figure 5.10). In diffusion flames Z = 1 for fuels, and Z = 0 for oxidizer, as both reactants are injected in non-premixed form. The 5.4. IMPLEMENTATION OF LAMINAR FLAME SPEED 55 Figure 5.10: Temperature and pressure dependence of the laminar flame speed [151] equivalence ratio φ can be expressed by the mixture fraction Z and its stoichiometric value Zst as Z 1 − Z φ = st (5.8) 1 − Z Zst The term involving Zst is constant for a given combination of fuel and oxidizer. For hydrogen-air combustion, the stoichiometric mixture fraction is equal to Zst = 0.0285. For pure fuel φ → ∞ and for pure oxidizer φ = 0. Figure 5.11 shows the temperature dependent flammability limits of hydrogen. The experiment (red dots) has been performed by White [156] with a downward propagating flame in a 25 mm in diameter tube. An increase of the oxidizer temperature broadens the flammability limits because less heat has to be supplied from the burning layer. The flammability limits vary nearly linearly with temperature. As the flammability limits are also influenced by the size of the vessel and the direction of flame propagation, appendix B shows the values for different vessel sizes, and in the following, correlations are derived for a downward propagating flame and an assumed upward propagating flame. The correlations for the upward propagation are assumed, as only experimental results at ambient temperature are available. The green lines in figure 5.11 represent the correlation derived for the lower flammability limit, and the blue line for the up- per flammability limit. With the assumption of an upward propagating flame, the flammability limits at ambient temperature and pressure are higher for the hydrogen lean side, and lower for the hydrogen rich side. Finally the flammability limits with a broader flammable range are used for the correlated curves for the laminar flame speed. The following expressions give the temperature dependence of the lower and upper flammability limit of hydrogen in air, where LF L0 and UFL0 denote the lower and up- per flammability limit of hydrogen at ambient temperature and pressure respectively. The correlation of flammability limits obtained from experimental results of a down- ward propagating flame [156] marked with dashed lines can be written as 56 CHAPTER 5. H2 JET FLAME Figure 5.11: Temperature dependent flammability limits of hydrogen T − T 0 LF L = LF L0[vol.%] 1 − (5.9) T,experiment 1200 [K] T − T 0 UFL = UFL0[vol.%] 1 + (5.10) T,experiment 2700 [K] The approximations for the lower and upper flammability limits assuming an upward propagating flame, marked with solid lines in figure 5.11, are T − T 0 LF L = LF L0[vol.%] 1 − (5.11) T,corrected 500 [K] T − T 0 UFL = UFL0[vol.%] 1 + (5.12) T,corrected 2800 [K] The index 0 refers to a reference state at T 0 = 293 K and p = 1 bar. The pressure dependent flammability limits are shown in figure 5.12. The grey solid line [13] and the grey solid line with filled dots [144] show the experimental re- sults obtained with a downward propagating flame, and the solid line with no filled dots [16] shows the experimental results obtained within a spherical vessel. The differ- ent results obtained from the measurements cannot be only attributed to experimental uncertainties. Moreover, they result from different interpretations. Bone et al. [16] (solid line with no filled dots) defined the criterion for the flammability limits to be 100 % combustion, whereas Berl et al. [13] defined the flammability limits for 80 % combustion. 5.4. IMPLEMENTATION OF LAMINAR FLAME SPEED 57 Figure 5.12: Pressure dependent flammability limits of hydrogen As can be seen, with increasing pressure, the flammability limits are at first nar- rowed and with further increasing pressure they are widened. Berl et al. [13] ascribe this phenomenon to the highly caloric ignition. When the silver wire fuses, heat quan- tity is supplied to the system. It has a stronger influence at low pressures, hence a low mass in the volume. The higher the pressure the smaller the influence of this phe- nomenon. The dashed lines show the correlations adapted to the experiment of Berl et al. [13], and the solid lines show again the approximations assuming an upward propagating flame. The green lines denote the lower flammability limits, and the blue lines the upper flammability limits. Due to the fact that the flammability limits are at first narrowed and with further increasing pressure widened, the correlations are split up to two ranges for the lower flammability limit as well as for the upper flammability limit. The experiments performed with a downward propagating flame can be approxi- mated by • Upper flammability limit – 0 atm ≤ p < 9.22 atm p − 285 UFL = (5.13) p1,experiment −4 – 9.22 atm ≤ p < 220 atm −3 UFLp2,experiment = 2.17017 lnp − 6.48263 · 10 p + 63.753 (5.14) 58 CHAPTER 5. H2 JET FLAME • Lower flammability limit – 0 atm ≤ p < 18.6844 atm p + 31 LF L = (5.15) p1,experiment 4.578 – 18.6844 atm ≤ p < 220 atm −2 −4 2 −7 3 LF Lp2,experiment = 11.5476−4.0747·10 p+1.8864·10 p −3.0605·10 p (5.16) The pressure and temperature dependent lower and upper flammability limits can be approximated as