te Health & Safety Executive OFFSHORE TECHNOLOGY | REPORT - OTO 1999 011 CFD Calculation of Impinging Gas Jet Flames Date of Issue :April 1999 Health & Safety ExecutiveAgreement No MaT'SU/8872/3574 CED Calculation of Impinging Gas Jet Fames Authors: A.D. Johnson, L.C. Shirvill, A, Ungut Shell Research & Technology Centre, Thornton, Summary: ‘This report contains the results of Computational Fluid Dynamics calculations of 5 open air and impinging natural gas and propane gas jet flames performed for the UK Health and Safety Executive under contract agreement no MaTSU/8872/3574. The objective of the work was to provide information about flame temperatures, velocities, dynamic pressures and convective aud radiative heat transfer to flame-impinged surfaces and thereby to assist in the interpretation of experimental data and the development of jet fire tests of passive fire protection materials. The combustion, soot formation and radiative heat transfer models were developed by HSE Solutions and the numerical calculations were performed using the CFX software from AEA. Technology. Validation of the calculations where possible was made against experimental measurements taken at tests sites owned by British Gas Plc, SINTEF and South West Research Institute. Keywords: Hazards, CFD, Heat transfer, Jet Flame ImpingementEXECUTIVE SUMMARY Passive fire protection material (PFP) is used to place an insulating batvier between the heat loading from a fire and vulnerable objects such as vessels and supporting structures. ‘Traditionally, such materials have been tested using heat loading from a furnace operating under time-temporature conditions defined by a fire curve. However, fumsce tests are wnable to replicate the severe erosive forces, thermal shock or the balance between convective and radiative heat loading resulting from a impinging jet firo from a high prossure gas leak. ‘Therefore, a working group was formed from interested industry parties and regulatory authorities, with the objective of defining and validating a jet fire resistance test that could be performed in a reproducible manner at a number of test laboratories, and that replicated the key conditions of large-scale jet fires. ‘This report describes computational fluid dynamics (CFD) calculations carried out by Shell as part of the validation process, under commission from the Health and Safety Executive. The objective of the numerical modelling was to provide information about difficult-to-measure properties such as flame temperatures, velocities, dynamic pressures and convective and radiative heat transfer to flame-impinged surfaces. Calculations have been performed for two free, and three impinging natural gas and propane gas jet flames. Submodels have been developed to simulate the initial expansion of the high-pressure gas jet, turbulent gas-phase combustion, soot evolution and radiative and convective heat transfer. Numerical solutions were oblained using the CFX software fiom AEA Technology, Validation of the caloulations where possible was made against experimental measurements made by BG Technology plc, SINTEF Energy and South West Research Institute, The following conclusions can be drawn from the work. 1, The 0.3 ke/s propane gas jet flame used in the jet fire test does replicate the convective and radiative heat loading typical of a large-scale natural gas jet flame when it is fired into the test arrangement for planar and structural specimens. This supports the results found with PFP coated specimens tested at both scales. 2, Premixed flamelet modelling is requited in order to obtain an accurate prediction of the combustion behaviour of the lift-of region of the 0.3 ky/s propane gas jet flame. However, the diffusion flamelet model for natural gas combustion provides sufficiently accurate predictions of flame properties in the bulk of the flame to enable comparison of difficult to measure properties. 3. The Jet Fite Test for Tubular Specimens also appears to reproduce the convective and radiative heat loading that would be found inside a large-scale flame, However, although the peak dynamic pressures on the tubular are equivalent to those found in a large-scale flame, the region of high dynamic pressures is quite small and corresponds to a region of relatively low flame gas temperatures. 4. The work has also demonstrated that CFD is a useful tool for the analysis of the behaviour of impacting flames in complex geometries.CONTENTS EXECUTIVE SUMMARY CONTENTS L INTRODUCTION 1.1 Background to Project 1.2 Scope of Work MATHEMATICAL MODELS 2.1 Characteristics of Jet Flames to be Modeled 2.2 Equation Set PROPANE OPEN AIR JET FLAME 3.1 Brief Description of the Flame 3.2 Shock Structure Calculation 3.3 Flame Lift-Off Calculation for Laminar Diffusion Flamelet Modelling 3.4 Laminar Diffusion Flamelet Caloulation 3.5 Premixed Flamelet Calculation 3.6 Comparison Between Experiment and Premixed and Laminar Diffusion Flamelet Models 3.7 Conclusions with Regard to Improvements of the Premixed Flamelet Model over the Diffusion Flamelet Model NATURAL GAS OPEN AIR JET FLAME 4.1 Brief Description of the Flame 4.2 Shock Structure Calculation 4.3 Laminar Diffusion Flamelet Calculation of Open-air Flame 4.4 Comparison Between Laminar Diffusion Flametet Model and Experiment 4.5 Conclusions PROPANE JET FLAME IMPINGING ON EMPTY JET FIRE RESISTANCE TEST RIG 3.1 Brief Description of the Flame 3,2 3-D Fiame Calculation Using Diffusion Flamelet Model 5.3 Comparison Between Calculation and Experimental Measurements PROPANE JET FLAME IMPINGING ON JET FIRE RESISTANCE TEST RIG FOR TUBULAR SPECIMENS 6.1 Brief Description of the Flame 6.2 3-D Flame Calculation Using Premixed Flamelet Model 6.3 Comparison Between Calculation and Experimental Measurements 6.4 Conclusions Page iii 2 21 21 26 31 35 42 43 B 46 38 39 39 61 oo B 19 86,7, NATURAL GAS FLAME IMPINGING ON 2 I-BEAMS. 7.1 Brief Description of the Flame 7.2 3-D Flame Calculation Using Diffusion Flamelet Model 7.3 Comparison Between Calculation and Experimental Measurements 7.4 Conclusions REFERENCES 87 87 87 94 95CFD Calculation of Impinging Gas Jet Flames 1. INTRODUCTION 1.1. BACKGROUND TO PROJECT ‘The Piper Alpha disaster in 1988 highlighted the tisks of hydrocarbon jet fires on offshore installations, Jet fires may also present a hazard in onshore petrochemical plant, the main difference being that in the confines of a remote offshore platform the consequences are more likely to resuit in the loss of life, Lord Cullen's report [1] on the Piper Alpha disaster tecommended that Safety Cases should be prepared for all offshore installations; these would identify the hazards and where necessary the measures required to mitigate them. ‘To achieve this the hazards must be understood and in the specific case of jet fires this has driven the study of the hazards they pose. Lord Cullen also specifically recommended that realistic tests should be developed to indicate the resistance of passive fire protection materials under jet fire conditions. Traditional famace tests generate total heat fluxes similar to that within a fire however they are unable to reproduce the wider range of fire variables, such as the balance of radiative and convective heat loading, high gas velocities and thermal shock, These are major factors with regard to the performance of passive fire protection in actual fires and in particular, jet fires resulting from high pressure gas leaks, Tho first tests of passive fire protection in a large high-pressure natural gas jet fire were carried out in the Shell Offshore Flame Impingement Protection Programme (SOFIPP) [2]. Subsequently, many other such full scale demonstrations have been carried out, Although reliable, large scale testing requires the use of unique facilities and is too expensive to be recommended as a standard test method. Early work by SINTEF NBL in Norway and Shell Research {3] showed that key conditions of heat flux and velocity, typical of large fires, could also be produced at a much smaller scale, thereby offering the opportunity for a manageable standard jet fire test. ‘The tost is basod on an ignited jet of gaseous propane with a mass flow rate of 0.3 kg/s ic, an order of magnitude small than that used in the SOFIPP tests described in the previous section, ‘This jet fire develops a high velocity combined with a high heat flux through a recirculating flame caused by firing the jet into a 1.5 m square open front test box with 0.5 m sides, Following this initiative the UK Health and Safety Executive (UK HSB), in conjunction with the Norwegian Petroleum Directorate (NPD), set up a working group to examine jet fire testing and to work with industry to develop a standard jet fire test procedure for passive fire protection materials. Current membership of the Jet Fire Test Working Group includes the HSE, the NPD, UKOOA, Shell, British Gas, SINTEF NBL, the Health and Safety Laboratorios, the Southwest Research Institute in the USA, Lloyds Register and Exxon. ‘Through the Jet Fire Test Working Group the early work evolved into a test procedure, the Interim Jet Fire Test for Determining the Effectiveness of Passive Fire Protection Materials [4]. In developing this procedure it was recognised that validation would be required to both prove reliable uniformity of results between test laboratories, and to compare its results with those from large-scale tests, ‘The challenge in developing the Interim Jet Fire Test has been to reproduce the key conditions typical of large-scale jot fires that determine balance andlevel of convective and radiative heat loading. However, it is extremely difficult to make measurements of gas temperatures, velocities and convective and radiative heat transfer in a given flame in order to determine whether these conditions are being replicated. One way of making such a comparison is to use computational fluid dynamics (CFD) models that incorporate descriptions of the turbulent combustion and heat transfer. Shell Research has been developing such a model for a number of years [5-7]. Once such models have been validated, they are capable of providing detailed information about flow and heat transfer properties at any point in the computation domain. The UK HSE have therefore contracted Shell Research Limited to perform a CFD modelling study of five different turbulent gas jet flames with the objective of providing information that will be of use in interpreting results of experimental measurements and improving the design of jot fire tests for passive fire protection materials (Agreement No. MaTSU/8872/3574), This report contains the results of these calculations, 1.2. SCOPE OF WORK Calculations of the following five flames have been performed. 1, A.0.3 kg/s open air sonie propane gas jet flame from a 17.8 mm orifice. 2. A3 kg/s open air sonic natural gas jet flame from 20 min orifiee. 3. A0.3 kp/s sonic propane gas flame from 17.8 mm orifice, impinging on empty rig for Jet Fite Resistance Test (planar and structural specimens), 4, A.0.3 kp/s sonic propane gas jet flame from 17.8 min orifice, impinging on rig for Jet Fire Resistance Test (tubular specimens). 5. A3 kg/s sonic natural gas jet flame from 20 mm orifice impinging on two }-beams. ‘The mathematical models describing the physical sub-processes and the results of the simulations of each flame have been validated where possible against experimental measurements, Comparisons are made between the different flames for the properties of interest in the testing of passive fire protection materials viz, flame temperatures, velocities, dynamic pressures and convective and radiative heat transfer to flame-impinged surfaces,2. MATHEMATICAL MODELS 24 CHARACTERISTICS OF JET FLAMES TO BE MODELLED Figure 1 highlights the essential features of a gaseous jet fire that 2 CFD calculation must mod rh) Source. If the stagnation drive pressure, P, > P,4( 42) (typically about 2 bar), where 7 is the ratio of specific heat capacities, the exit velocity is sonic and the orifice exit pressure is greater than atmospheric. On exiting, the jet rapidly expands and accelerates to supersonic speeds, Subsequent deceleration results in a normal shock a few source diameters downstream of the orifice, followed by a further series of expanding and reflected shock waves forming a diamond shock pattern. Buoyancy dominates flow Cold soot escapes End of combustion Thermal plume and smoke ‘Ambiont flow => Flame lift-off point Radllative heat transfor +f Jot expands via series Turbulent combustion of shock waves Large eddies entrain air ‘ir entrainment Gombustion occurs rapidly without combustion inthin sheets “flametets” Boundary layer Flame intially blue (no soot) Convactive heat transfor ‘Amount of soot increases Radiative heat transfor along flame Figare 1 Schomatic of an impinging jet fire Flame Lift-off. No combustion can occur in the initial part of the jet because the turbulent strain rate is too high in the flammable parts of the shear layers at the edge of the jet. Further downstream the strain rate decreases sufficiently for a flame to exist. In most lifted hydrocarbon gas jet flames the first flame existence appears as a bluish premixed flame annulus around the outside of the central core of unburned fuel. Flame lift-off greatly increases air entrainment into the initial part of the jet and must therefore be predicted accurately. Turbulent Combustion. Downstream of the flame lift-off point, the turbulent strain rate falls sufficiently combustion to exist. The combustion timescale is fast compared to the turbulent mixing timescale and the flame instantaneously resembles an ensemble of reacting sheets or flamelets. The combustion in the flamelets can be taken either to be premixed (fuel and oxidant mixed before reaction) or diffusion (fuel and oxidant come together in the flame)depending on the degree of mixing in the region upstream of the flame lift-off point, A turbulent combustion model is required in order to determine temperatures, species concentrations, densities, and soot formation precursors. The combustion process is govemed by non-equilibrium chemistry. However, turbulent strain and temperature and concentration fluctuations make it infeasible to implement a large number of highly nonlinear chemical kinetics reactions directly into the turbulent flow calculations, Therefore Shell Research has Tesorted to the “conserved-scalar assumed probability density function (PDF)” approach coupled with offline calculation of laminar flame properties using detailed chemistry and ‘transport properties. Fuel atoms are neither destroyed nor created in chemical reactions, therefore, a conserved scalar can be defined, the mixture fraction /, representing the mass fraction of fuel atoms in any given mixture. / varies between 0 (pure air) and 1 (pure fuel). The instantaneous combustion properties are thus determined as unique functions of mixture fraction and local strain rate (and in the case of premixed combustion a reaction progress variable}, from laminar flamelet look-up tables. The effects of turbulent fluctuations in mixture fraction on the combustion properties are incorporated by assuming that the statistical variation in the mixture fraction can be modelled using a probability density function (PDF) of a presumed shape. A beta PDF is commonly used, which can be fully prescribed at each location in terms of the mixture fraction mean 7° and its variance g. Turbulent transport equations are solved for these two variables. This approach has the advantage that there are no difficult-o-solve chemical reaction source terms in the transport equations. The effects of turbulent fluctuations in strain rate and reaction progress variable are incorporated using other PDFs. Soot evolution, Incandescent soot is the dominant source of thermal radiation in most hydrocarbon jet flames because the soot emits radiation over the whole infrared spectrum, whereas emission from gaseous sources, such as water vapour and carbon dioxide, is confined to narrow spectral bands. ‘Thus, to obtain good predictions of the radiative heat transfer, accurate calculations of the soot volume fraction are required. Furthermore, the bright yellow soot is often used as a marker for the position of the flame, Timescales for soot evolution reactions are relatively slow compared with turbulent fluctuations, therefore, finite rate chemistry has to be included in transport equations describing the soot evolution Radiation, Radiative heat transfer plays an integral part in the modelling of a combusting system, It shares an equal role with convection for the transfer of heat within a turbulent diffusion flame, but is entirely responsible for the heat emanating from the flame to external objects. At typical flame temperatures of 1500-1700K, the peak in the Planck spectrum overlaps strong absorption/radiation bands of water and carbon dioxide in the infra-red. These bands, together with the continuous radiation from aty soot, chiefly contribute to the radiative heat transfer within the flame and to external objects. The volume fraction of soot in the natural gas and propane gas flames simulated in this project is insufficient for the spectral radiation to saturate locally, hence the radiative emission is not that of a black body and an absorption coefficient has to be determined. 2.2 EQUATION SET The flow equations are given in a co-ordinate free tensor notation as described in [8]. The main combusting flow is solved as a steady state problem, However, the high speed, highly compressible jet source calculations liad to be performed as time dependent calculations with ‘ime steps as small as | ps to ensure numerical stability,2.2.1 Continuity Ve(at) =0 22.2 Momentum Ve(SU@U)-Ve[uo{VO +(viy')} ve +(P)-3)g=0 and, neglecting the bulk viscosity . 1 puveu™ kis the turbulence kinetic energy Jk"): & = 72 ui the fluctuating component of the velocity vector Horr is the effective viscosity; Myr =]+ Hy, the dynamic laminar viscosity = 1.8 x 10% ke/ (ms), Be e ¢ is the dissipation rate of the turbulence kinetic energy (I kg" s", @is the tensor product, and VU is a rank two tensor, 2 is the gravitational acceleration along the vector x, His the eddy viscosity w, = C,p Note that gravity is neglected in the jot source calculation 2.2.3. Turbulence kinetic energy ¥-(p08)-v«{(u+#2)ve] = Prt Py +P, - Be . ‘Where the shear production tenn is: a a es 2 P= (v (¥ "| ay 7, = Hy VU «(VO + (VE)')-5 V6 and the production due to buoyancy is: Hay Por and an additional source term is used in the premixed flamelet combustion model to take into account mean density and pressure gradients created due to the heat released by the combustion process [9]: 9 bh ~ ea {Spivens tt vpev?)} Hy Voit + at) fy-— Eee2.2.4 Turbulence dissipation rate Velots)-v4[(ns#}pe] ace +Cma(F,.0)+P)-CB- where an additional production term is used in the premixed flamelet combustion model to take into account mean density and pressure gradients created due to the heat released by the combustion process (9] p,=—“# (vp ewe) Pp Where C;, C;, and C; are constants. is an assigned turbulent Prandtl number. 2.2.5 Scalar parameters ‘The turbulent transport equation for a scalar in general form is written as ve(p00)-v4l(n, +#}vo) =P, Se. In this model turbulent diffusivity term aa is assumed to be much greater than the ® molecular diffusivity tem Dq@ so that the numerical values of D@ need to be only approximately accurate, Pd is the source term for scalar . Additional scalar variables satisfying the turbulence transport equations are solved for the combustion invariant fuel mass fraction, the variance of mass fraction, soot mass fraction, soot number density, acetylone mass fraction, and oxygen mass fraction. The molecular diffusivity tetm, is taken to be 2.59x10° all scalar parameters respectively, The units of the source term Pep, and the scalar parameter ® should be defined as (scalar-unit m* s'), and (scalar-unit kg-mixture®) respectively, for example enthalpy is expressed in J/kg. ‘The source term for the Fayre averaged mixture fraction f is defined as zero: P, = 0 ‘The Favre averaged fluctuation of the mixture fraction g is defined as: pxf"xf" so and the source term for the scalar transport equation for g is given as (9: 2g 5°. P, a WI )-GB 782.2.6 Enthalpy equation Jn CFX-F3D, the enthalpy for the mixture is normally defined as: a lo HHH, 000 +k ‘Where the Favre averaged thermat enthalpy (/7,,) may be expressed as: A, =7,F-G, aTog where ° and Tre is the reference temperature (normally taken as 298 K). ‘The definition of enthalpy is modified in the laminar diffusion flamelet model and the variable used represents the loss of thermal enthalpy from its adiabatic value. The kinetic energy terms in the enthalpy scalar are only used for the highly compressible jet shock stricture calculation. In general these terms are negligible in jet flame combustion calculations and the temperature field calculated will be correct, but for the flow conditions close to the nozzle where velocity is high this may lower the temperature of the mixture artificially (approximately 10-50 K). This may be a slight drawback for simulating the lift- off region of the premixed flamelet model where the local temperature is used to define a Teaction progress factor. However, detailed test calculations with the full enthalpy representation showed that the turbulent stretch effects are dominant in this region, and the mixture temperature is affected mostly by the air entrainment. ‘The diffusivity used in the enthalpy equation is: Dy =7Z- + Where 2.is the thermal conductivity (taken as 2.8 x 107, W/mK) and Cp is the A Ce fluid specific heat capacity, "The source term in the enthalpy equation Py is different for the jot shock structure and laminar diffusion flamelet and premixed flamelet combustion models 2.2.7 Turbulence constants Where possible, the turbulence constants used by the authors who originally developed the different models for the jet source calculations and the laminar diffusion flamelet or premixed flamelet calculations, In all calculations, the turbulent Prandtl mumbers for scalars is taken as 0.7, 6 I, and the constants C,, C; and C'yare taken as 1.44, 0 and 2 respectively.‘Table 1 lists the values of the constants that changed from model-to-model, Table 1 Turbulence constants used in the various model calculations “Model GG; ~_—Referonce_ Tet source 0.06 192 13 10 Natural gas lift-off 0.06 1.92 1,22 35 Propane lift-off conn 19 422 6 Diffusion flamelet 0.06 192 12 6 Premixed flamelet 0.07272 19 1.22 11,12 2.2.8 Jet shock structure model ‘The simulation of the jet shock structure is performed as a time dependent calculation, Thus, Ok > time derivatives #8. are added to the loft hand side of each equation (where = ©, H, and scalar respectively) Buoyancy is neglected (g = 0) The source term in the enthalpy equation is the time derivative of the pressure = The calculations are performed using an axisymmetric grid and the calculation domain is modelled as a hole in an adiabatic wall with the jet released into open air, Uniform flow is assnmed at the jet inlet, The jet inlet velocity, temperature and pressure are calculated using isentropic flow equations. The turbulence kinetic energy and dissipation rate are determined ‘rom correlations for fully developed pipe flow [13]. Constant pressure boundary conditions are applied on the other sides of the computational domain. ‘The mean density {7 is given by the ideal gas equation for each of the two fluid species “fuel” and “air” . _ PM, Pan RF and the mixture fraction, 7 . Thus, FOP) Real Mya My P ‘The calculation domain for the 3 kg/s natural gas release is 0.4 m (20xD) in the radial direction and 2 m in the axial direction. The calculation domain for the 0.3 ke/s propane gas release is 0.089 m (5xD) in the radial direction and 0.89 m in the axial direction, Cartesian numerical grids, hybrid differencing schemes for all flow variables, central differencing for density and timesteps of the order of 5 ys were found to give the best numerical stability and predictions of the jet structure downstream of the shocks. 2.2.9 Laminar diffusion flamelet combustion model ‘The correct determination of the flame lift-off point is important because air entrainment into the jet is greatly enhanced when there is no combustion,‘The mean Eulerian strain rate Sis given in terms of the ms turbulent velocity w’ and the Taylor lengthscale A aw ( . ae %) In the region of the flame lift-off point 3 is much higher (xxx for natural gas and yyy for propaae) than the extinction strain rate for laminar natural gas (~ 500 s*) or propane flames (-990 s") . Therefore, the combustion at the flame lift-of point is likely to be premixed and the flame lift-off position must be determined by a premixed combustion model. ‘The method for determining flame lift-off position for the natural gas jet flames is described in reference [5]. It is based on a simplification of premixed laminar flamelet combustion modelling work by Gu [11] and Bradley et al {14]. In Gu’s model, the average turbulent heat release rate is given by _ ih 1 a= f ANfa@.Nee.naa im 8 where © is a non-dimensional reaction progress variable, which varies from 0 to { through the flame, q,(0, f) is the laminar unstretched premixed flame heat release rate, p(®,f) is the joint pdf, P,(/) is the probability that the local mean turbulent strain rate is sufficiently low to permit flamelets to bum and f,, and fing, are the mixture fractions corresponding to the upper and lower flammability limits. The flame lift-off position was taken to be the point at which the mean turbulent heat release rate exceeded a certain critical value (chosen to be 3 MW/m* for natural gas after calculation of various lifted flames). Shell Research has not yet developed a premixed laminar flamelet combustion model for natural gas jet flames. ‘Therefore, a simplification of the model was derived, which enabled the estimation of the lift- off position from the turbulent flow properties alone. Consideration of a number of natural gas flames calculated by Gu suggested that the flame lift-off point coincided with the location where , first became greater than 0.3 on the stoichiometric contour. P, is determined from the turbulent kinetic energy dissipation rate ¢, the fuel Lewis number, the laminar buming velocity and the kinematic viscosity of air. ‘The method for determining the flame lift-off position for the propane gas jet flames is described in reference [6] and uses the same premixed laminar flamelet formulation as Gu [LL]. However, the critical mean turbulent heat release rate corresponding to the flame lift-off point was found to be 10 MW/m’ for propane flames, Isothermal mixing is assumed to occur upstream of the lift-off point. ‘The laminar diffusion flamelet combustion model is an adaptation of a methodology originally published by Fairweather et al [15] and first implemented in Shell Research by Barker et al [5]. Look-up tables of chemical species mass fractions, temperature, Cp and density as functions of instantancons mixture fraction are obtained from calculations of planar counterflow adiabatic strained laminar diffusion flames using the computer program RUN-IDL [16] from Cambridge University Engineering Department. Full chemical kinetics, detailed transport andthermodynamic properties are provided by the CHEMKIN-II code from Sandia Laboratories 07), ‘The chemical mechanism used for the calculation of the reaction rates for counterflow methane air diffusion flames incorporates 78 elementary reactions for 27 species in order to predict the formation of C,H, a soot precursor which has successfully been used in the modelling of soot formation in turbulent natural gas and propane diffusion flames. The mechanism used for propane air flames has 31 species and 97 reactions. ‘The mixture faction formulation is given by Bilger [18] for fuels containing carbon and hydrogen only, at where J; is the mass fraction of the element J, M, is the atomic weight of element / and the indices © and F indicates that the mass fiaction comesponds to the fuel and oxidant respectively. Counterflow laminar methane-air diffusion flamelets have been calculated in a planar geometry for strain rates ranging from 2s“! to the extinction strain rate of 504 5"! Counterflow propane-air diffusion flamelcts have been calculated for strain rates from 10 =“! up to an extinction strain rate of 990 s“!. One effect of strain is to reduce the maximum adiabatic flame temperature, For the methane flames, the maximum temperature reduces from 2202 K at 2.5 s*l to 1787 K at 504 s“l, whilst the maximum temperature of the propane flames reduces from 2101.98 at a strain rate of 10s"! to 1651.37 K at 990 s*!, Another effect of strain is to modify species concentrations. Figure 2 shows the variation in the mass fraction of the soot precursor acetylene in the methane flamelets. oot z 5 om —25 (ist ‘5 0008 —~20(1/s) 5 —50(18) go 3 008 —— 150(1/s)] E 0008 —— 450 (1s) g 0.002 | 500(t/s) zo 0 02 04 068 OB 1 Mixture Fraction Figure 2 Mass fraction of C7Hz profiles at six different strain rates - Methane flame 10For a given strain rate, s, the adiabatic flamelet temperature T,y(f, 5) is adjusted off-line to account for radiative heat loss using the approach of Crauford at al. 19}, POF S)= Tea S SHV XTaa f 8)! Te (3) J] Taq" (s) is the maximum adiabatic temperature at that strain rate and % parameterises the effect of radiative heat loss. Initially, calculations were performed using a look-up tables from single flamelet and a single valuc of %, throughout the turbulent flame (In reference [5] a flamelst with strain rate of 100 s“] and y= 0.15 was used for a 3 kg/s sonic natural gas flame, whilst in reference {6], two calculations of a subsonic natural gas jet flame used strain rates of 60 s“! and 500 s-! with y= 0.15 and a calculation of a sonic propane jet flame used a flamelet with a strain rate of 280 s“! with x= 0.21. ‘The flamelet density is adjusted using the ideal gas law, Pan SS) Tal $8) 2S Tia) ‘No adjustment is made to the gascous species concentrations in the flamelet library which retain their adiabatic values, In the most recent adaptation of the model, the local flame properties are determined by selecting a flamelet according to the mean Eulerian strain rate Fand % is calculated from the loss in enthalpy due to radiative onergy exchange. Favre average flame properties are determined using a Beta function PDF whose form is given by the mean and variance of the mixture fraction. Thus, for example the local Favre temperature is given by \ F=[ry.nPne ‘and the mean density is given by 1fol = Bi ® J 0lf,.5) ne The local value of is determined by converting the standard CFX-F3D enthalpy equation into an equation for the gain/loss in the total enthalpy from its adiabatic value, fh, It is assumed that the total enthalpy loss does not alter the product composition from the adiabatic one and therefore is totally accounted for by a loss in thermal enthalpy, = cam _ fhe ine = ett = J,“ Cp a. If the specific heat capacity of the mixture Cp, is taken to be its adiabatic value Cp,,, then, Fins = Pag Fag P)giving, Ping x= Gas) It can be shown that Fines = bang + hyo F)hiyag The laminar flamelet libraries are constructed with the zero point thermal enthalpy and total enthalpy equalling the chemical enthalpy at 298.1SK. The thermal enthalpies then follow as, , ; = [erat and iyo = fepar as vhs In this work the inlet temperature of the fuel and air is Z, = 298.0K,, with the result that the contribution of the thermal enthalpies to A,,, are a factor of a hundred smaller than typical values of hys4. Thus, with negligible loss in accuracy, Prose = ~hyaa Now within CFX-F3D, the thermal enthalpy /,,,is normally related to the temperature by finsa= Cp T- Cag hey ‘The modified use of the enthalpy as an enthalpy loss is ensured by setting t= haga Cai with the initial condition that i,,, = 0 The specific heat capacity is kept positive by setting CPrep = 10,000 I/kg*K, Tuy = 1000K whilst setting 4=0279W / m K maintains a realistic value of the laminar thermal diffusivity. ‘Upstream of the flame lift-off point, the instantaneous fluid density is determined as a function of mixture fraction using a look-up table created from the jet inlet profile derived from the jet shock stmcture calculation, The fluid temperature is set to the ambient temperature, 2.2.10 Premixed flamelet combustion model In addition to the work being carried out for the HSE, Shell Research has its own programme to develop and validate improved CED models. Since starting the work on the HISE contract a new “Premixed flamelet” combustion model has been developed for high pressure propane ‘gas jet flames. This improved combustion model has been used under the separate Shell Research model development project to calculate two of the jet flame cases in the HSE contract. The laminar diffusion flamelet mode! described in chapter 2.2.9 is not so good at predicting the structure of a jet flame near the flame lift-off point. The propane jet flames under study in this contract are impacting on objects at a point close to the flame Lif-offregion. Therefore, in order to provide the best possible information to the HSE, the results of calculations using the new model are being made available in this report. A brief description of the model is given in this report, Full details of the model will be the subject of future publication, The combustion model proposed by Bradley [12] is adopted, and our implementation follows closely the details published by Gu [10]. ‘The non-premixed turbulent jet fire is supposed to consist of an assembly of premixed laminar flamelets, The extent to which the ‘combustion reaction is completed in individual flamelets is modelled with assigned separate Favre probability functions of the combustion invariant mixture fraction (f), the turbulent strain rate (s), and a reaction progress factor (@). The PDF of @ is assumed to be conditional to the local mixture fraction, Turbulent heat release rate ‘The source term in the enthalpy equation is the average turbulence heat release rate 7, which is defined locally as: fmx 4ooL = - 5 0/P a= fu} Ja@.F.9P OR y>® ca where P(/) is the Favre PDF of the mixture fraction, %) is the Favre PDF of the reaction progress variable conditional on the fuel mass fraction £, and P(s) the Favre PDF of turbulent strain rte (s). 4,(0,f,5) is the heat release rate calculated for the premixed Jaminar flamelet subjected to a strain rate of s. The reaction progress variable 6 is defined as: T-T, TA)-% The Favre average of @ for a given instantaneous value of the mixture fraction fis defined as 8 T-7, ‘Where T,, is the calculated adiabatic flame temperature of the fuel mass fraction £ and T,, is the unburned reference temperature. Probability density finctions of Favre averaged parameters are beta distribution fimctions defined as: BE) roige‘in which Where C() is the gamma finction, and &"? is the Favre average of the fluctuating component. In turbulent flow, laminar flamelets are better represented by a geometry where the unbumed premixed gases counterflows against the bumed combustion products. Numerical solutions show that for this condition there is not a clearly defined flame extinction caused by flame stretch [20], and the laminar heat release rate reduces gradually with inercasing strain rate s Bradley et al [9, 11, 12, 21] proposed an empirical expression for turbulent combustion probability P, based on a large number of experiments with various fuels at Leeds University. ‘This approach assumes that the heat release rate is unaffected by the strain rate, and provides an empirical uniquely defined extinction strain rate sq*best fitted to the experimental turbulent heat release rate data (J; Jao. It is further assumed that the negative strain does not affect the laminar heat release rate which is set to the unstretched value, n= 18@) a= ~ Ja,@)F@)a0 0 whore q(0) is the laminar heat release rate calculated for unstretched premixed combustion. 3q° indicates the positive empirical quenching strain rate which is a function of the mixture fiaction f‘The turbulent heat release rate can be rewritten as: Sf max. 1 — - as 50/P H= [ANP ao.NPY tae fii 3 In this analysis flamolets are not quenched by negative strain rate, and any possible increase of turbulent heat release rate because of negative Markstein numbers for rich propane mixtures is neglected. ‘The Favre averaged fluctuating component of 0 is estimated from a correlation based on computed values as reported by Gu [11] 6 =Ge(1-6) 2.707K x, wae Gy -( aBradley[12] simplified the integration of the turbulent heat release rate by fitting the premixed laminar heat release rate by a function of the following type : (ass) 0.F) = dane rage 0" (1-0)" where a, and b ate curve fit constants to give the best fit to the computed strain free heat release curves. ‘The turbulent average heat release rate may then be expressed as: rar 42,0, 1 XT, HEP) T) (eb, Aa in which, (a+B) Ta+p)t(a+a)r(b+p) Aart) oe Te) Br (a+b+0+p) ara atb+a+B Aa,b,f)= Combustion Probability ‘The effect of turbulence stretch on premixed flamelets is modelled by assigning a combustion PDF (P,) which depends on a turbulent Kariovitz stretch factor (K), and the Lewis number (Le). This approach was originally proposed by Bradley et a. [9, 11]. ‘The turbulent Karlovitz. stretch factor is defined as the ratio of the mean turbulent strain rate ‘based on the Taylor length scale and a laminar flamelet strain rate as: wv, Vv, | & KTP li where the mean Bulerian strain rate (u'/h) is defined as au’ fe » ” Visv ‘Where 2 is the Taylor length scale, w’ is the turbulent velocity, v, is the unburned viscosity, uy, is the laminar buming velocity, and s is the dissipation rate of turbulence kinetic energy.‘The combustion probability function is given [11] by: KL, =0166-0.014KL, > - 00132 KL, KL, 08em(— and KL, a1 +032exp(— 2132 Fig. 3 shows the variation of P, as a function of KL,, As observed experimentally [23] combustion probability reduces to a very low value (0.2) for Kle greater that 1.5 Figure 3 Probability of combustion as a function of turbulent Karlovitz stretch factor. Mean density. In the premixed flamelet model the mean fluid density is given by BMAP), 5 oe ear) where P, is the ambient pressure and F-1 BA-1,2.2.11 Soot evolution The soot evolution model follows the approach of Lindstedt et al. [15, 23, 24], Turbulent transport cquations are solved for soot mass faction 7 (kg-soov/kg-mixture) and soot aumber density N (particles-soot/kg-mnixture), ‘The source term for turbulent transport equation for the soot mass fraction is given by : Pag FM, AF MEMEO" MBER ges!) ‘Where M, is the soot molar mass, (2.011 kg/kmol. ‘The instantaneous soot nucleation rate _r,is described by the pyrolysis of acetylene, and given by: 1 eval - evn] (kmol m* s") Where [isthe molar concentration in. kmol m, and [C, /¥, ] M, ‘Where Yiu is the mass fraction of acetylene, and Mju is the molecular weight, Soot mass growth is assumed to proceed by the adsorption of acetylene molecules at the surface of each particle, and is presumed to be first order in acetylene concentration. To account for the experimental observation that soot particle ageing reduces surface reactivity it is assumed that the number of active sites present locally in the flame is proportional to the square root of the total area available, The instantaneous surface reaction source rate is given n=C, enol - *\c.mWA 2 6M,)> where A = 7 (%} ™P, where p, is the soot density taken as 2000 kg m*. In this model it is assumed that all soot particles are perfectly spherical, and uniform in size. “The instantaneous soot oxidation rate follows the expression formulated by Leo et al.[25], and assumes that all soot are exclusively oxidised by molecular oxygen to form carbon monoxide at typical flame temperatures.Ma = ovFea{- Boa where [0,]= » Yox is the mass fraction of oxygen, and Mox is the molecular weight. The source term for turbulent transport equation for the soot number density N (particles kg-l) is given by : us 52 pls avs (particles ar3 sl) Where Na is the Avogadro number, Cinin is the number of carbon atoms in the incipient carbon particle assumed to be constant, Na=6.022x1026 particles/kmol. The instantaneous agglomeration rate is given as: 6M, )* [6xT ml) Vp, ‘Where k is the Boltzman constant, 1.381x10-23 JK. ‘The model constants used in the laminar diffusion flamelet model for natural gas and propane are the same as used by Fairweather et al (15, 24] and are given in Table 2. Table 2 Constants used in the soot evolution equations ‘Natural Gas Propane Cl 136x106 10000 c 500 12000 C3 1.78 x 104 358 TiLK 21000 21000 TK 12100 12100 13K 19680 19680 Ca 3 9 Coin 90000 100 The application of the soot evolution equations is slightly different in the premixed flamelet model. Turbulent transport equations are solved for Favre averages of acetylene and oxygen mass finctions. Favre averaged values of every variable ate then substituted into the instantaneous equations. ¥y Finally soot volume fraction (fy) used in radiation calculations is defined as: 7, = 2 182.2.12 Radiative heat transfer Radiation heat transfer is calculated using the CEX-RADIATION|26] solver that simulates volume heating, and cooling from the cafculated flow grid by the CEX-F3D[8] solver. CFX- RADIATION provides two methods for the radiative heat transfor calculation, the Discrete transfer model [27], which calculates energy transfer along pre-defined rays fired from surfaces, and the Monte Carlo model which traces the histories of simulated packets of photons. Unless special pseudo-surfaces are specified, the discrete transfer method propagates rays from solid surfaces and the boundaties of the computational domain, The CFX-F3D domain for open-air flames is large compared with the flame volume, thus a large proportion of the rays from the boundaries would be wasted since they do not pass through the flame. The Monte Carlo method is more efficient for a free flame problem, and is the one used in this work, because the photon packets originate from areas of high luminosity, id est., inside the flame. They immediately sample the absorption/emission characteristics of the flame and then, if they survive, travel to outer boundaries where they contribute to external radiation. Radiative properties are assumed to be constant within each CFX-RADIATION zone which usually contain many CFX-F3D cells, Total cooling fram each zone is calculated as: Hogs 24% XV KET" ‘Where o is Stefan’s constant, V is the volume, and T is the temperature of each radiation zone. The temperature weighed average absorption coefficient for the radiation zone is provided by a user programme defined as: x DY, «4i(l. 50, Pago Toh) xT Dyer where N is the number of CFX-F3D cells in each CEX-RADIATION zone, V; is the volume, T; is the local temperature, and aj is the absorption coefficient of each CFX-F3D cell. The absorption coefficient is a function of species concentrations ,soot volume fraction (fy), local temperature, and the pathlenght (Ij) defined as: 3 fi _ 364, A, j is the surface area of the cell. 4 where ‘Temperature of the radiation zone is defined as: 1 x day,’ r=|2 v 19Using these temperature, and absorption coefficient fields CEX-RADIATION computes a total heating for each zone (Aheat-z W/m3) which can be related to the source term for the enthalpy equation in cach CFX-F3D cell ( Gay); (W/m3) as: ¥, (na) Llane XGEH XV), -40°* TE xa, x, erg Soy + Tn the application of the laminar diffusion flamelot model Zj'=(p#yand the absorption/emission coefficient for each CFX-F3D cell is calculated from the CO, and H,O concentrations and the soot volume fraction using the Edward's wide band model [28]. More details of this mode! are given in reference [5]. The air outside of the flame is not in radiative equilibrium and therefore it is not strictly correct to assign it equal absorption and emission coefficients. Therefore, the absorption coefficient is set to zero for cell temperature less than. 500 K and calculation of radiative heat fluxes to objects outside the flame is made by estimating the atmospheric absorption along an average pathlength from the flame to the receiver using an atmospheric transmissivity model derived by Wayne [29]. Tn the application of the premixed flamelet model 7;' =(7)‘a natrow band spectral radiation method [30] is used to calculate a lookup table for estimating the local effective absorption coefficients in a five-dimensional data matrix of local temperature, soot volume fraction, COz and Hj0 mass fractions, and the pathlength. To avoid problems with the reaction progress variable, which is defined in terms of an adiabatic flame temperature, the radiative heat transfer source term in the enthalpy equation is switched on only for the last few iterations of a calculation whilst the turbulent heat release rate source term is frozen. 2.2.13 Convective heat transfer In CFX-F3D convective heat transfer is derived from the boundary condition for the enthalpy equation where : * Jy is the wall heat flux in W/m2 * 1 is the wall shear stress and equal to ty=pCy,/2k * Ht is the distance from the wall to the centre of the neighbouring cell, normalised by the enthalpy linear sublayer thickness * Hquid and Hyyal} are the enthalpies at the fluid coll centre next fo the wall and at the wall respectively (see reference [8] for a full description of the madel). Note that no account is taken of the cooling effect of a lower temperature wall on the temperature of the gas flowing over it, Note that, in the application of the laminar diffusion flamelet combustion model, care has to be taken in defining properties to calculate the convective heat transfer because the enthalpy used is not the thermal enthalpy but rather the enthalpy loss from the adiabatic enthalpy. “Thus, the convective heat flux to walls is determined by caleulating thermal enthalpies in the wall cells and neighbouring cells from the fluid properties as a post process. Finally, for the comparison of properties of open-air flames a correlation for the average convective heat flux to a cylinder, derived from work by Churchill et al [33], was used. 203. PROPANE OPEN AIR JET FLAME 3.1 BRIEF DESCRIPTION OF THE FLAME The flame is the same as that specified for the Interim Jet fire Test [4]. Measurements from this flame were carried out at Southwest Research Institute [31]. Propane is released horizontally from a 17.8 mm orifice with a mass flowrate of 0.3 kg/s. Stagnation pressure and temperature are approximately 4,5 bara, and 323 K respectively. The Jet flow is choked with an exit pressure of 2.6 bara, and highly under-expanded close to the nozzle and expands non-isentropically to the ambient pressure through one or more barrel- shaped, and normal shock structures. Because of very high flow velocities and restricted fuel and air mixing in this region the combustion can only take place further downstream where flow is expanded to atmospheric pressure. As shown in Fig 4, the location of the flame lift- off point is marked by a blue-white combusting annulus of flamo approximately 0.6-0.7 m fiom the nozzle, It is, therefore, assumed that the under-expanded region of the jet is unaffected by the downstream combustion, and simulated separately as a non-combusting axisyinmetric fully compressible flow. The first highly luminous yellow flame appears at about 1.6 m downstream of the nozzle, The end of the luminous yellow flame is about 7-8 m horizontally from the release point. Figured Video image of 0.3 kp/s open air propane gas jet flame 3.2 SHOCK STRUCTURE CALCULATION ‘The calculation domain for the 0.3 kg/s propane gas release is 0.089 m (5 times the inlet diameter) in the radial ditection and 0,89 m in the axial direction, The numerical grid consists of 200 cells in the axial direction and 60 cells in the radial direction. 10 equal-sized cells are used across the radius of the jet inlet. This is the smallest, coll dimension used in the calculation domain. It is the same as the length of the first cell in the axial direction, Cell sizes increase gradually in the axial and radial directions (outside the radius of the jet), Hybrid differencing schemes for all flow variables, central differencing for density and timesteps of the order of 5 ts were found to give the best numerical stability and predictions of the jet structure downstream of the shocks. Convergence within each timestep is taken to have been achieved when the mass residual source sum is less than 10“ kg/s. a‘The net mass flow out through the pressure boundaries is 1.3 kg/s thus the mass conservation is accurate to within 0.01 %. The order of magnitude of other relevant absolute residual sums is given in Table 3. Table 3 Magnitude of absolute residual sums - shock structure (0.3 kg/s propane gas) Variable T Vv E € a F g Order of magnitude of 107 10° lo" 10? 10 10° 104 absolute residual sums Figure 5 shows calculated values of the U-velocity, The position of 3 diamond shocks can be seen clearly. U, mis, —— i | << 05m Figure 5 Axial velocity distribution of the underexpanded jet (0.3 kg/s propane gas) Figure 6 shows a comparison between measurements and predictions of dynamic pressure along the axis of the jet. 16 14 é 12 e —— CFD prediction 3 10 (Shock) 8 Experimental Results £08 * swan c g 06 04 & P02 00 02 04 06 og 10 Distance along jet centeline from nozzle, m Figure 6 Dynamic pressure along the axis of the underexpanded jet (0.3 kg/s propane gas)Radial profiles of jet properties are taken at an axial position (0.5 m) where the local pressure is close to ambient, but far enough upstream for the local turbulent strain rate to be too high to support combustion. These profiles are then used as inlet conditions for the CFD premixed flamelet calculation. The ambient, and the enthalpy reference temperature is set to 288 K, with 1.013 bara reference pressure. Figures 7-9 show the properties from the shock structure calculation that are used as input data for the flame lift-off and 3-dimensional flame structure calculations. 160 104 ‘0 —Dvvetocky 120 erty 12 @ 100 z Bw 1262 2 0 i 3 0 128 20 ° 148 oO 0.03 0.06 0.09 Radia} distance, m re7 Radial profiles of axial velocity and density at 0.5 m from nozzle (0.3 ke/s propane gas) oa aot 038 oot 03 fg gos ooo $ E oz cong 2 045 Wixture traction 0,004 2 ot = og | Metre faction ena - variance 3 bo 0 as ove 008 Fadia stance, m Figure 8 Radial profiles of mixture fraction and its variance at 0.5 m from nozzle (0.3 kg/s propane gas) 22000 ‘1490000 ‘1800 -—* ‘200000 1800 |—— ein 1400 ‘e00000 ‘1200 soom00 g = 1000 oo ecoo00 600 400000 400 1 20 ‘200000 0 To 9 0.03 0.06 0.08 Radial distance, m Figure 9 Radial profiles of lc and ¢ at 0.5 m from nozzle (0.3 kg/s propane gas) 3.3. FLAME LIFT-OFF CALCULATION FOR LAMINAR DIFFUSION FLAMELET MODELLING As has been mentioned in chapter 2.2.10, the flame lift-off position for the laminar diffusion flamelet model of combustion is determined for the propane flame by using an adiabatic premixed flamelet model. Figure 10 shows a schematic of the computation domain, Figure 11 shows shaded contours of adiabatic temperatures in the flame lift-off region, together with the 10 MW/m’ contour used to determine the flame lift-off position and shaded contours of speed from the jet shock structure calculation. CCCT differencing is used for all flow variables and upwind differencing is used for the density. The absolute residual sums are given in Table 4, The net mass flow out through the prossure boundaries is 3.98 kp/s Table 4 Magnitude of absolute residual sums - flame lift-off (0.3 kg/s propane gas) u Vv Mass E e H @ 15 Tx 6x10? 9x10! 7x1? 2x 108A x 107 Tx 10° cs pressure boundary 0.01 aus inlet 66 cols pressure | | 106 cots sons [sinal pe inlet profile / boundary 0 symmetry axis Deslis 36 oclls I ET 0 03 is ™ Figure 10 Schematic of computation domain - flame lift-off (0.3 kg/s propane gas) 24max temp. 1819 K >_ —_ Liftoff 0.725 m 10 Mim? Figure 11 Determination of flame lift-off position (0.3 kg/s propane gas) Figures 12 and 13 show measured and predicted dynamic pressures along the jet centreline and radially ata distance of 1 m from the nozzle. 16 14 & —=CFD prediction Lito M12 $10 © Experimental Resuls 3 (swen ge a ge 5 & 4 a 2 ° 04 06 os 10 12 14 16 18 20 Distance along jet centreline from nozzle, m Figure 12 Dynamic pressure along the jet axis - adiabatic lift-off model (0.3 kg/s propane gas)25 CFO prediction Liter g Caleulaton} & Experimental Resuls g (swe) 5 2 i 3 2 0.00 0.05 0.40 018 0.20 028 Distance from jet centretine, m Figure 13 Dynamic pressure at radial positions 1m from nozale - adiabatic lift-off model (0.3 kg/s propane gas) The predicted flame lift-of position of 0.725 m is applied in all calculations of the 0.3 ke/s propane gas flame using the laminar diffusion flamelet model. 3.4 LAMINAR DIFFUSION FLAMELET CALCULATION ‘The computation domain is shown in figures 14 and 15. It is divided into 6 rectangular blocks with a nonuniform Cartesian grid. Only hal of the flame is calculated as the plane Z=0 is a symmetry plane, The jet inlet is modelled as a rectangular inlet, 0.089 m wide by 0.178 m high, centred 1 m above the ground. The calculated jet profiles 0.5 m from the release point are mapped onto a uniform 8 x 16 cell grid covering the jet inlet, All cell centres on the plane X=0, not within a distance of 0.089 m from the jet contreline are given a boundary condition of an inflow of 1 m/s, with typical turbulence levels for open air flows. CCCT differencing is used for all flow variables and central differencing is used for the density. The absolute residual sums are given in Table 5. The net mass flow out through the pressure boundaries is 35.6 kp/s Table S Magnitude of absolute residual sums - open air laminar diffusion flamelet (0.3 kg/s propane gas) UG VW Mas ok «@ H f ¥ iF a ¥ 06 96 O1 4x10? 2 30 2x10° 8x10t 5x10% 7x10* 1x10”Y a 7 pressure boundary 1 m/s inlet Beells pressure boundary ——~"| 1.089 - \ ! oot | stinlet . * cells o | Limsinte Wall 12 cells GO cells 05s os ™ x Figure 14 ‘Schematic of computation domain - open air (0.3 kg/s propane gas) Y m 7 pressure boundary Symmetry 33 cells plane 1.089 ‘ 0.911 ° cells | 12 cells 0 _, 8 cells 25 cells TTT 0 0.089 2™ Zz Figure 15 Schematic of computation domain - open air (0.3 ke/s propane gas) Pa‘The numerical grid used for the radiative heat transfer calculation is not the same as for the calculation of the flowfield. The CFX-RADIATION zones are constructed from planar partitions perpendicular to the X, Y and Z-axes. There are 30, 27 and 16 partitions in the X, Y and Z, directions respectively. Each zone contains a number of CFX-F3D cells and the boundary of each zone is chosen to coincide with the boundary of a CFX-F3D cell. The positions of the partitions were chosen in order to resolve sharp gradients of the key variables affecting radiative heat transfer, viz. temperature and species concentrations. ‘A comparison of visible flame shape calculated fom CFD computations is not trivial. On the ‘one hand, the video camera or eye is making an integration of the visible flame emissions along rays projected through the flame onto a 2-dimensional image plane, On the other hand the CFD computation produces a 3-dimensional field of temperatures and grey-gas absorption coefficients. A direct comparison can only be made if the emission in the visible spectrum is calculated and projected onto a similar image plane to that used by the video camera, and if this projected integrated emission is calibrated with the response of the camera to different intentions of visible radiation. Altematively a simple comparison can be made by looking at T* representing the presence of high radiative emission. Figure 16 shows a comparison between values of T*on the flame symmetry plane and a visible flame probability plot obtained from image analysis of video taken of the experiments carried out by SWRI. The predicted flame is slightly too buoyant. This is believed to be due to the unphysical sudden change in density and velocity at the flame lift-off plane (at x=0,725m from the exit nozzle plane). The predicted flame width at the lift-off point is also too wide. A comparison between measured and predicted profiles of dynamic pressure and temperature is given in ‘Chapter 3.6, S.858SE+12 4g817E+12 3.90708 +12 2.93226+12 1.08746 +12 ‘9.8288E+11 7.98626 +03 Visite Flame probability Above 50% 105% 510% 13% ot-1% Below 0.1% Figure 16 ‘Comparison of Video Analysis and predictions of 7* on flame symmetry plane - open air (0.3 kg/s propane gas)Figure 17 shows the calculated horizontal velocities in the release direction on the flame symmetry plane. This clearly shows the unphysical jump in velocity at the flame lift-off point, as the combustion mode! switches on. The plot also shows an unphysical expansion of the flame at the lift-off point as the calculation attempts to maintain both mass and momentum conservation following a sudden reduction in density duc to combustion. 4 U-velocity, m/s 3 p> 15000 +02 S000E +02 2 1.02506 +02 7.50006 +01 1 4.7500E+01 a 2.0000€+01 6 < 0.0000 +00 0.8 3.5 65 9.5 Figure 17 U-velocity on flame symmetry plane - open air (0.3 kg/s propane gas) ed mean mixture fractions between the flammable limits of 0.031 Figure 18 shows the calcula to0.11 4 Mixture fraction wg 1.1000E-01 9.9467E-02 8.4983E-02 7.0800~02 Bee 0.5 3.5 6.5 8.5 Figure 18 Mixture fraction on flame symmetry plane - open air (0.3 kg/s propane gas) Predicted Favre averaged temperatures on the flame symmetry plane are shown in Figure 19, 4 Favre Temperature, K wy 15445E+09 1.33596 +03 112726403 9.1860E+02 7.0998 +02 Bf sosx0e 2.020602 0.5 3.5 65 9.5 Figure 19 Favre temperatures on flame symmetry plane - open air (0.3 kg/s propane gas)Figure 20 shows calculated soot volume fraction together with contour lines of flame temperature and figure 21 shows the grey gas absorption coefficient. In the laminar diffusion flamelet model most of the soot is predicted to be in the first half of the flame, which is unrealistic. The grey gas absorption coefficient has its highest values in the region where there is the most soot and also demonstrates the unrealistic behaviour of the simulation near the start of the flame in that no cold unreacting core is predicted. ‘Temperature contour lines: 800, 1050, 1300 K Soot volume fraction 3 > 3.,0000E-07 2.6190E-07 2.0852E-07 1.S714E-07 1.0476E-07 52381E-08 ° ‘o.0000E +00 08 35 6.5 95 Figure 20 Soot volume fraction on flame symmetry plane - open air (0.3 kg/s propane gas) 4 Grey gas absorption coefficient 3 > 1.00006 +00 8,7302E-01 .9341E-01 8z381E~-01 3.4321E-01 1.74602-01 0900 +00 ° os 35 es 8.5 Figure 21 Grey gas absorption coefficient on flame symmetry plane - open air (0.3 kg/s propane gas) Finally, the radiation characteristics of a flame may be described by calculating the fraction of total energy radiated (F-number) [32]. The F-number calculated for the simulation of this flame using the laminar diffusion flamelet model is 0.24. Although there were no extemal thermal radiation measurements for this flame, comparison with results for other flames with similar properties suggests that this result is reasonably accurate. The overprediction of emission from the lift-off region appears to be matched by the underprediction of emission (due to lower flame temperatures) in the bulk of the flame. 303.5 PREMIXED FLAMELET CALCULATION ‘The computation domain and boundary conditions are the same as that used for the laminar diffusion flamelct calculation shown in figures 14 and 15, except that the Y and Z axes are interchanged so that vertically up is in the positive Z direction. CCCT differencing is used for all flow variables and central differencing is used for the density. The absolute residual sums are given in Table 6. The total mass flow through the domain is 35.7 kg/s Table 6 Magnitude of absolute residual sums - open air laminar premixed flamelet (0.3 kg/s propane gas) U VW Mas kb «© &# f T, ¥ 08 04 17 1x10" 1 15 2x10 2x10" _1x10% 9x10 2x10“ 4 Temperature, K Zm 1600 3 1475 2 1350 1 1225 1100 ° 6. 5 35 65 95 Xm Figure 22 ‘Comparison of measured flame shape shown as a black frustum shape with the calculated net volumetric thermal emissions plotted as an iso-emission surface of 450 kWm®. The temperature distribution is false colour mapped onto the iso-emission. surface and the colour key is shown Experimentally determined luminous flame shape is compared with CFD results in Figure 22 where the measured flame frustum is shown together with the calculated net volumetric thermal emissions plotted as an iso-emission surface of 430 kWm”. This figure is chosen to be approximately 20 % of the maximum value in order to be consistent with the luminosity threshold level used in the image analysis to discriminate flame regions (see figure 16). The temperature distribution is false-colour mapped on to the iso-emission surface, and the colour key is also shown. Low temperature areas of the iso-emission surface mark the high soot volume fraction regions, and the high temperature areas indicate lower soot concentrations. ‘There is good agreement between calculated and measured flame shapes. The effect of ‘buoyancy on the flame tilt angle is slightly more pronounced for the CFD flame. probably 31because of the effect of the enclosure on the laboratory jet fire. The CFD prediction of the position of the first luminous part of the flame is in good agreement with the experiment. Figures 23, 24, and 25 show the axial velocity field, combustion independent mixture fraction in the flammable range (0.031 to 0.11), and flame temperature plotted across the cross section of the jet fire. The results display a realistic simulation with overall features in agreement with experiments as discussed above and do not have the unphysical expansion of the jet at the flame lift-off point that appears in the laminar diffusion flamelet calculation (see figures 17-19). The flame temperatures in the bulk of the flame calculated using the premixed fiamelet model are approximately 400 K higher than the flame temperatures calculated using ‘the laminar diffusion flamelet model. A more detailed comparison between the two model’s predictions and measurements is made in Chapter 3.6. Axial Velocity 4 mst > 1.5000E+02 3 1.3000E+02 am 1.0250E+02 7.5000E+01 1 4.7500E+01 . 2,0000E+01 05 35 8.5 95 <9:00006+00 xm Figure 22 Calculated axial velocity field on the symmetry plane - 0.3 kg/s open air propane 4 Mixture Fraction 3 > 1.1000E-01 zm 9.9467E-02 2 8.4989E-02 7.0500E-02 ‘ 5.6017E-02 ° 4.1533E-02 05 35 65 9.5 ¥¢3.1000E-02 Xm Figure 23 Calculated mixture fraction on the symmetry plane - 0.3 kg/s open air propane, Mixture fractions representing the combustible range are shown. 32