LES/PDF for premixed combustion in the DNS limit

We investigated the behaviour of the composition probability density function (PDF) model equations used in a large-eddy simulation (LES) of turbulent combustion in the direct numerical simulation (DNS) limit; that is, in the limit of the LES resolution length scale Δ (and the numerical mesh spacing h) being small compared to the smallest flow length scale, so that the resolution is sufficient to perform a DNS. The correct behaviour of a PDF model in the DNS limit is that the resolved composition fields satisfy the DNS equations, and there are no residual fluctuations (i.e. the PDF is everywhere a delta function). In the DNS limit, the treatment of molecular diffusion in the PDF equations is crucial, and both the ‘random-walk’ and ‘mean-drift’ models for molecular diffusion are investigated. Two test cases are considered, both of premixed laminar flames (of thickness δ_L). We examine the solutions of the model PDF equations for these test cases as functions of Δ/δ_L and h/δ_L. Each of the two PDF models has advantages and disadvantages. The mean-drift model behaves correctly in the DNS limit, but it is more difficult to implement and computationally more expensive. The random-walk model does not have the correct behaviour in the DNS limit in that it produces non-zero residual fluctuations. However, if the specified mixing rate Ω normalised by the reaction timescale τ_c is sufficiently large (Ωτ_c ? 1), then the residual fluctuations are less than 10% and the observed flame speed and thickness are close to their laminar values. Away from the DNS limit (i.e. h/δ_L ? 1), the observed flame thickness scales with the mesh spacing h, and the flame speed scales with Ωh. For this case it is possible to construct a non-general specification of the mixing rate Ω such that the flame speed matches the laminar flame speed.     

Experimental measurements of geometric properties of turbulent stratified flames

Combustion under stratified conditions is common in many systems. However, relatively little is known about the structure and dynamics of turbulent stratified flames. Two-dimensional imaging diagnostics are applied to premixed and stratified V-flames at a mean equivalence ratio of 0.77, and low turbulent intensity, within the corrugated flame range. The present results show that stratification affects the mean turbulent flame speed, structure and geometric properties. Stratification increases the flame surface density above the premixed flame levels in all cases, with a maximum reached at intermediate levels of stratification. The flame surface density (FSD) of stratified flames is higher than that of premixed flames at the same mean equivalence ratio. Under the present conditions, the FSD peaks at a stratification ratio around 3.0. The FSD curves for stratified flames are further skewed towards the product side. The distribution of flame curvature in stratified flames is broader and more symmetric relative to premixed flames, indicating an additional mechanism of curvature generation, which is not necessarily due to cusping. These experiments indicate that flame stratification affects the intrinsic behaviour of turbulent flames and suggest that models may need to be revised in the light of the current evidence.     

Modelling of turbulent lifted jet flames using flamelets: a priori assessment and a posteriori validation

This study focuses on the modelling of turbulent lifted jet flames using flamelets and a presumed Probability Density Function (PDF) approach with interest in both flame lift-off height and flame brush structure. First, flamelet models used to capture contributions from premixed and non-premixed modes of the partially premixed combustion in the lifted jet flame are assessed using a Direct Numerical Simulation (DNS) data for a turbulent lifted hydrogen jet flame. The joint PDFs of mixture fraction Z and progress variable c, including their statistical correlation, are obtained using a copula method, which is also validated using the DNS data. The statistically independent PDFs are found to be generally inadequate to represent the joint PDFs from the DNS data. The effects of Z–c correlation and the contribution from the non-premixed combustion mode on the flame lift-off height are studied systematically by including one effect at a time in the simulations used for a posteriori validation. A simple model including the effects of chemical kinetics and scalar dissipation rate is suggested and used for non-premixed combustion contributions. The results clearly show that both Z–c correlation and non-premixed combustion effects are required in the premixed flamelets approach to get good agreement with the measured flame lift-off heights as a function of jet velocity. The flame brush structure reported in earlier experimental studies is also captured reasonably well for various axial positions. It seems that flame stabilisation is influenced by both premixed and non-premixed combustion modes, and their mutual influences.     

Premixed flame response to oscillatory pressure waves

We numerically investigate the interaction of sinusoidal pressure waves and slightly corrugated premixed flames. Work up to now has demonstrated the well-known Rayleigh–Taylor instability by imposing a single pressure ramp function onto a corrugated premixed flame. This paper considers sinusoidal oscillations of pressure. Such inputs are important since observations of large-scale experiments suggest that the presence of acoustic waves might be expected to have a significant influence on the propagation of the flame. The numerical experiments reported in this paper show that oscillatory pressure waves of the order of 800 Hz can have a magnifying effect on the wrinkling of the flame, due to the effect of Rayleigh–Taylor instabilities, hence increasing the overall mass burning rate, and that a sinusoidal pressure wave interacting with a premixed flame causes the flame to increase its wrinkling with every cycle in pressure. The result of great interest is that this growing process reaches a maximum after a few cycles, creating a dynamic equilibrium in which the final time-averaged mass burning flux is larger compared to that of the initial flame. It is also shown that the final mass burning flux increases with amplitude of the pressure wave, and increases with increasing frequency over the range explored.     

An experimental study of partially premixed flame sound

The occurrence of oscillating combustion and combustion instability has led to resurgence of interest in the causes, mechanisms, suppression, and control of combustion noise. Noise generated by enclosed flames is of greater practical interest but is more complicated than that by open flames, which itself is not clearly understood. Studies have shown that different modes of combustion, premixed and non-premixed, differ in their sound generation characteristics. However, there is lack of understanding of the region bridging these two combustion modes. This study investigates sound generation by partially premixed flames. Starting from a non-premixed flame, air was gradually added to achieve partial premixing while maintaining the fuel flow rate constant. Methane, ethylene, and ethane partially premixed flames were studied with hydrogen added for flame stabilization. The sound pressure generated by methane partially premixed flames scales with M⁵ compared to M³ for turbulent non-premixed methane flames. Also, the sound pressure generated by partially premixed flames of ethane and ethylene scales as M^4.5. With progressive partial premixing, spectra level increases at all frequencies with a greater increase in the high-frequency region compared to the low-frequency region; flames develop a peak and later a constant level plateau in the low frequency region. The partially premixed flames of methane, ethylene, and ethane generate a similar SPL as a function of equivalence ratio when the fuel volume flow rate is matched. However, when fuel mass flow rate is matched, the ethane and ethylene flames produce a similar SPL, which is lower than that produced by the methane flame.     

Problems of predicting turbulent burning rates

Different approaches to the modelling of turbulent combustion first are reviewed briefly. A unified, stretched flamelet approach then is presented. With Reynolds stress modelling and a generalized probability density function (PDF) of strain rate, it enables a source term, in the form of a probability of burning function, P_b, to be expressed as a function of Markstein numbers and the Karlovitz stretch factor. When P_b is combined with some turbulent flame fractal considerations, an expression is obtained for the turbulent burning velocity. When it is combined with the profile of the unstretched laminar flame volumetric heat release rate plotted against the reaction progress variable and the PDF of the latter, an expression is obtained for the mean volumetric turbulent heat release rate. Through these relationships experimental values of turbulent burning velocity might be used to evaluate P_b and hence the CFD source term, the mean volumetric heat release rate.

Different theoretical expressions for the turbulent burning velocity, including the present one, are compared with experimental measurements. The differences between these are discussed and this is followed by a review of CFD applications of these flamelet concepts to premixed and non-premixed combustion. The various assumptions made in the course of the analyses are scrutinized in the light of recent direct numerical simulations of turbulent flames and the applications to the flames of laser diagnostics. Remaining problem areas include a sufficiently general combination of strain rate and flame curvature PDFs to give a single PDF of flame stretch rate, the nature of flame quenching under positive and negative stretch rates, flame responses to changing stretch rates and the effects of flame instabilities.     

A LES/PDF simulator on block-structured meshes

A block-structured mesh large-eddy simulation (LES)/probability density function (PDF) simulator is developed within the OpenFOAM framework for computational modelling of complex turbulent reacting flows. The LES/PDF solver is a hybrid solution methodology consisting of (i) a finite-volume (FV) method for solving the filtered mass and momentum equations (LES solver), and (ii) a Lagrangian particle-based Monte Carlo algorithm (PDF solver) for solving the modelled transport equation of the filtered joint PDF of compositions. Both the LES and the PDF methods are developed and combined to form a hybrid LES/PDF simulator entirely within the OpenFOAM framework. The in situ adaptive tabulation method [S.B. Pope, Computationally efficient implementation of combustion chemistry using in situ adaptive tabulation, Combust. Theory Model. 1 (1997), pp. 41–63; L. Lu, S.R. Lantz, Z. Ren, and B.S. Pope, Computationally efficient implementation of combustion chemistry in parallel PDF calculations, J. Comput. Phys. 228 (2009), pp. 5490–5525] is incorporated into the new LES/PDF solver for efficient computations of combustion chemistry with detailed reaction kinetics. The method is designed to utilise a block-structured mesh and can readily be extended to unstructured grids. The three-stage velocity interpolation method of Zhang and Haworth [A general mass consistency algorithm for hybrid particle/finite-volume PDF methods, J. Comput. Phys. 194 (2004), pp. 156–193] is adapted to interpolate the LES velocity field onto particle locations accurately and to enforce the consistency between LES and PDF fields at the numerical solution level. The hybrid algorithm is fully parallelised using the conventional domain decomposition approach. A detailed examination of the effects of each stage and the overall performance of the velocity interpolation algorithm is performed. Accurate coupling of the LES and PDF solvers is demonstrated using the one-way coupling methodology. Then the fully two-way coupled LES/PDF solver is successfully applied to simulate the Sandia Flame-D, and a turbulent non-swirling premixed flame and a turbulent swirling stratified flame from the Cambridge turbulent stratified flame series [M.S. Sweeney, S. Hochgreb, M.J. Dunn, and R.S. Barlow, The structure of turbulent stratified and premixed methane/air flames I: Non-swirling flows, Combust. Flame 159 (2012), pp. 2896–2911; M.S. Sweeney, S. Hochgreb, M.J. Dunn, and R.S. Barlow, The structure of turbulent stratified and premixed methane/air flames II: Swirling flows, Combust. Flame 159 (2012), pp. 2912–2929]. It is found that the LES/PDF method is very robust and the results are in good agreement with the experimental data for both flames.     

Modeling subfilter soot-turbulence interactions in Large Eddy Simulation: An a priori study

A new LES model for subfilter soot-turbulence interactions is developed based on an a priori analysis using large-scale DNS data of temporally evolving non premixed n-heptane jet flames at a jet Reynolds number of 15,000. In this work, soot formation is modeled in LES by solving explicit transport equations for soot moments, and the unclosed filtered soot moment source terms are closed by a presumed PDF approach. Due to the strong intermittency of soot fields, a previous modeling approach assumes the presumed PDF to be bimodal accounting for sooting and non-sooting subfilter regions but neglects any sub-structure of the soot distribution. In this work, the modeling framework is improved by a new presumed PDF model that explicitly accounts for the sub-structure of the sooting mode, which is modeled by a log-normal distribution. The previous and new models are assessed by means of their prediction of the filtered source terms and the filtered intermittency, and the log-normal distribution is found to significantly reduce modeling errors, in particular, for the coagulation source term. Introducing a log-normal distribution for the PDF of the sooting mode involves a large amount of additional model parameters, such as the width of the distribution and correlation coefficients among different soot moments, so model assumptions to reduce the number of model parameters are discussed by means of the DNS data. The conclusions are found to be robust with respect to a variation in the global Damköhler number in the DNS datasets. The final model formulation only requires solving two additional transport equations in LES compared to previous models, while significantly improved model predictions are obtained for the coagulation source term which is import for predicting the number of soot particles.     

Modelling of turbulent scalar flux in turbulent premixed flames based on DNS databases

A transport equation for scalar flux in turbulent premixed flames was modelled on the basis of DNS databases. Fully developed turbulent premixed flames were obtained for three different density ratios of flames with a single-step irreversible reaction, while the turbulent intensity was comparable to the laminar burning velocity. These DNS databases showed that the countergradient diffusion was dominant in the flame region. Analyses of the Favre-averaged transport equation for turbulent scalar flux proved that the pressure related terms and the velocity–reaction rate correlation term played important roles on the countergradient diffusion, while the mean velocity gradient term, the mean progress variable gradient term and dissipation terms suppressed it. Based on these analyses, modelling of the combustion-related terms was discussed. The mean pressure gradient term and the fluctuating pressure term were modelled by scaling, and these models were in good agreement with DNS databases. The dissipation terms and the velocity–reaction rate correlation term were also modelled, and these models mimicked DNS well.     

Pressure wave excitation of natural flame frequencies

Theoretical investigations by McIntosh predicted the existence of a high-frequency natural resonance in premixed flames. In this work we examine this phenomenon using numerical techniques and predict the resonance frequency for commonly used hydrocarbon flames. It is found that resonance occurs when the width of the pressure pulse is similar to the the width of the flame region. The resonance frequency is found to increase with the burning velocity and decrease with the burnt temperature.     

Turbulent flame visualization using direct numerical simulation

Combustion phenomena are of high scientific and technological interest, in particular for energy generation and transportation systems. Direct Numerical Simulations (DNS) have become an essential and well established research tool to investigate the structure of turbulent flames, since they do not rely on any approximate turbulence models. In this work two complementary DNS codes are employed to investigate different types of fuels and flame configurations. The code is π³ is a 3-dimensional DNS code using a low-Mach number approximation. Chemistry is described through a tabulation, using two coordinates to enter a database constructed for example with 29 species and 141 reactions for methane combustion. It is used here to investigate the growth of a turbulent premixed flame in a methane-air mixture (Case 1). The second code,Sider is an explicit three-dimensional DNS code solving the fully compressible reactive Navier-Stokes equations, where the chemical processes are computed using a complete reaction scheme, taking into account accurate diffusion properties. It is used here to compute a hydrogen/air turbulent diffusion flame (Case 2), considering 9 chemical species and 38 chemical reactions.     

Large-eddy simulation of a bluff-body stabilised turbulent premixed flame using the transported flame surface density approach

A premixed propane–air flame stabilised on a triangular bluff body in a model jet-engine afterburner configuration is investigated using large-eddy simulation (LES). The reaction rate source term for turbulent premixed combustion is closed using the transported flame surface density (TFSD) model. In this approach, there is no need to assume local equilibrium between the generation and destruction of subgrid FSD, as commonly done in simple algebraic closure models. Instead, the key processes that create and destroy FSD are accounted for explicitly. This allows the model to capture large-scale unsteady flame propagation in the presence of combustion instabilities, or in situations where the flame encounters progressive wrinkling with time. In this study, comprehensive validation of the numerical method is carried out. For the non-reacting flow, good agreement for both the time-averaged and root-mean-square velocity fields are obtained, and the Karman type vortex shedding behaviour seen in the experiment is well represented. For the reacting flow, two mesh configurations are used to investigate the sensitivity of the LES results to the numerical resolution. Profiles for the velocity and temperature fields exhibit good agreement with the experimental data for both the coarse and dense mesh. This demonstrates the capability of LES coupled with the TFSD approach in representing the highly unsteady premixed combustion observed in this configuration. The instantaneous flow pattern and turbulent flame behaviour are discussed, and the differences between the non-reacting and reacting flow are described through visualisation of vortical structures and their interaction with the flame. Lastly, the generation and destruction of FSD are evaluated by examining the individual terms in the FSD transport equation. Localised regions where straining, curvature and propagation are each dominant are observed, highlighting the importance of non-equilibrium effects of FSD generation and destruction in the model afterburner.     

湍流分层燃烧的直接数值模拟和小火焰模型研究

湍流分层燃烧广泛应用于工业燃烧装置,但是目前还比较缺乏适用于湍流分层燃烧的高精度数值模型。本文利用直接数值模拟数据库,对高Karlovitz数分层射流火焰的小火焰模型表现进行了先验性评估。考虑了两种小火焰模型,一种是基于自由传播层流预混火焰的小火焰模型M1,另一种是基于分层对冲小火焰的小火焰模型M2。研究发现M1和M2在c-Z空间的结果与直接数值模拟在定性上是一致的。在物理空间,M2对过程变量反应速率脉动值的预测结果要优于M1.     

A DNS assessment of linear relations between filtered reaction rate,flame surface density,and scalar dissipation rate in a weakly turbulent premixed flame

Linear relations between (i) filtered reaction rate and filtered flame surface density (FSD) and (ii) filtered reaction rate and filtered scalar dissipation rate (SDR), which are widely used in Large Eddy Simulation (LES) research into premixed turbulent combustion, are examined by processing DNS data obtained from a statistically 1D planar flame under weakly turbulent conditions that are most favourable for the two approaches (flamelet combustion regime, single-step chemistry, equidiffusive mixture, adiabatic burner, and low Mach number). The analysis well supports the former approach provided that the filtered reaction rate is combined with filtered molecular transport term. In such a case, both the RANS and LES FSD approaches are based on local relations valid within weakly perturbed flamelets. Accordingly, simply recasting RANS expressions to a filtered form works well. On the contrary, while the FSD and SDR approaches appear to be basically similar at first glance, the analysis does not support the latter one, but shows that a ratio of the filtered reaction rate to the filtered SDR is strongly scattered within the studied flame brush, with its conditionally mean value varying significantly with Favre-filtered combustion progress variable. As argued in the paper, these limitations of the LES SDR approach stem from the fact that it is based on a relation valid after integration over weakly perturbed flamelets, but this relation does not hold locally within such flamelets. Consequently, when a sufficiently small filter is applied to instantaneous fields, the filter may contain only a part of the local flamelet, whereas the linear relation holds solely for the entire flamelet and may not hold within the filtered flamelet volume. Thus, the present study implies that straightforwardly recasting well-established RANS equations to a filtered form is a flawed approach if the equations are based on integral features of local burning.     

Effects of Lewis number,density ratio and gravity on burning velocity and conditional statistics in stagnating turbulent premixed flames

DNS is performed to analyse the effects of Lewis number (Le), density ratio and gravity in stagnating turbulent premixed flames. The results show good agreement with those of Lee and Huh (Combustion and Flame, Vol. 159, 2012, pp. 1576–1591) with respect to the turbulent burning velocity, S_T, in terms of turbulent diffusivity, flamelet thickness, mean curvature and displacement speed at the leading edge. In all four stagnating flames studied, a mean tangential strain rate resulting in a mean flamelet thickness smaller than the unstretched laminar flame thickness leads to an increase in S_T. A flame cusp of positive curvature involves a superadiabatic burned gas temperature due to diffusive–thermal instability for an Le less than unity. Wrinkling tends to be suppressed at a larger density ratio, not enhanced by hydrodynamic instability, in the stagnating flow configuration. Turbulence is produced, resulting in highly anisotropic turbulence with heavier unburned gas accelerating through a flame brush by Rayleigh–Taylor instability. Results are also provided on brush thickness, flame surface density and conditional velocities in burned and unburned gas and on flame surfaces to represent the internal brush structures for all four test flames.     

How fast can we burn, 2.0

We review the state of the art in measurements and simulations of the behavior of premixed laminar and turbulent flames, subject to differential diffusion, stretch and curvature. The first part of the paper reviews the behavior of premixed laminar flames subject to flow stretch, and how it affects the accuracy of measurements of unstrained laminar flame speeds in stretched and spherically propagating flames. We then examine how flow field stretch and differential diffusion interact with flame propagation, promoting or suppressing the onset of thermodiffusive instabilities. Secondly, we survey the methodology for and results of measurements of turbulent flame speeds in the light of theory, and identify issues of consistency in the definition of mean flame speeds, and their corresponding mean areas. Data for methane at a single operating condition are compared for a range of turbulent conditions, showing that fundamental issues that have yet to be resolved for Bunsen and spherically propagating flames. Finally, we consider how the laminar flame scale response of flames to flow perturbations interacting with differential diffusion leads to very different outcomes to the overall sensitivity of the burning rate to turbulence, according to numerical simulations (DNS). The paper concludes with opportunities for future measurements and model development, including the perennial recommendation for robust archival databases of experimental and DNS results for future testing of models.     

Laser-induced fluorescence determination of flame temperatures in comparison with CARS measurements

Temperature profiles in several premixed low pressure H₂/O₂/N₂ flames and in an atmospheric pressure CH₄/air flame were determined by laser-induced fluorescence (LIF) and by CARS experiments. In the LIF study, temperatures were derived from OH excitation spectra, CARS temperatures were deduced from N₂ Q-branch spectra. The present study is the first quantitative comparison of these two methods for temperature determination in flames burning at pressures up to 1 bar. The resulting temperatures showed good agreement.     

Comparison of a spectral model for premixed turbulent flame propagation to DNS and experiments

A recently developed spectral model for premixed turbulent combustion in the flamelet regime (based on the EDQNM turbulence theory) has been compared with both direct numerical simulations (DNS) and experimental data. The 128³ DNS is performed at a Reynolds number of 223 based on the integral length scale. Good agreement is observed for both single- and two-point quantities (i.e. ratio of the turbulent to laminar burning velocities, scalar autocorrelation, dissipation and scalar-velocity cross correlation spectra) for the two different values of u′/s _L0 considered. The model also predicts the rapid transient behaviour of the flame at early times. An experimental set-up is then described for generating a lean methane-air flame and measuring two-point spatial correlations along the midpoint of the flame brush (i.e. along the C¯=0.5 contour). The experimental measurements in the flamelet regime take the form of a discontinuous or ‘telegraph’ signal. The EDQNM model, in contrast, describes an ‘ensemble’ of flames, and thus is based solely on continuous variables. A theoretical relationship between the correlation obtained from the EDQNM model and the equivalent correlation for a discontinuous (experimental) flame is derived. The relationship is used to enable a meaningful comparison between experimentally observed and model correlations. In general, the agreement is good for the three different cases considered in this study, with most of the error occurring at the lowest Reynolds number (Re _L =22). Furthermore, it is shown that considerably more error would result if no attempt is made to convert the ensemble representation in the model to an equivalent single-flame or ‘telegraph’ signal.     

A posteriori testing of algebraic flame surface density models for LES

In the application of Large Eddy Simulation (LES) to premixed combustion, the unknown filtered chemical source term can be modelled by the generalised flame surface density (FSD) using algebraic models for the wrinkling factor Ξ. The present study compares the behaviour of the various models by first examining the effect of sub-grid turbulent velocity fluctuation on Ξ through a one-dimensional analysis and by the LES of the ORACLES burner (Nguyen, Bruel, and Reichstadt, Flow, Turbulence and Combustion Vol. 82 [2009], pp. 155–183) and the Volvo Rig (Sjunnesson, Nelsson, and Max, Laser Anemometry, Vol. 3 [1991], pp. 83–90; Sjunnesson, Henrikson, and Löfström, AIAA Journal, Vol. 28 [1992], pp. AIAA–92–3650). Several sensitivity studies on parameters such as the turbulent viscosity and the grid resolution are also carried out. A statistically 1-D analysis of turbulent flame propagation reveals that counter gradient transport of the progress variable needs to be accounted for to obtain a realistic flame thickness from the simulations using algebraic FSD based closure. The two burner setups are found to operate mainly within the wrinkling/corrugated flamelet regime based on the premixed combustion diagram for LES (Pitsch and Duchamp de Lageneste, Proceedings of the Combustion Institute, Vol. 29 [2002], pp. 2001–2008) and this suggests that the models are operating within their ideal range. The performance of the algebraic models are then assessed by comparing velocity statistics, followed by a detailed error analysis for the ORACLES burner. Four of the tested models were found to perform reasonably well against experiments, and one of these four further excels in being the most grid-independent. For the Volvo Rig, more focus is placed upon the comparison of temperature data and identifying changes in flame structure amongst the different models. It is found that the few models which largely over-predict velocities in the ORACLES case and volume averaged in a previous a priori DNS analysis (Chakraborty and Klein, Physics of Fluids, Vol. 20 [2008], p. 085108), deliver satisfactory agreement with experimental observations in the Volvo Rig, whereas a few of the other models are only able to capture the experimental data of the Volvo Rig either quantitatively or qualitatively.     

Parametric study on a compound-drop spray flame

The introduction of compound-drop spray in a combustion system is a new concept. These droplets bear two gasification stages to cause an integral positive or negative effect on a premixed flame to raise or lower the local temperature of the gasification region. In this paper, we adopt a compound drop which contains a water core encased by a layer of shell fuel. A one-dimensional homogeneous lean or rich premixed flame with the dilute compound-drop spray was investigated by using large activation energy asymptotic analysis. The compound-drop spray burning mode was defined and divided into completely pre-vaporised burning (CPB), shell pre-vaporised burning (SPB) and shell partially pre-vaporised (SPP) burning modes by way of the gasification zones of the shell fuel and the core water relative to the flame position. The influences of the initial droplet radius, the shell-fuel mass fraction and the liquid loading of the compound-drop spray on the lean and rich flames were analysed. By means of the normalisation parameter of flame propagation mass flux (), enhancement, suppression or extinction of the compound-drop spray flame can be represented clearly. Furthermore, from the observation of extinction, the necessary conditions of extinction of a lean spray flame by the internal heat transfer are that the spray is a negative effect and causes a sufficient heat loss rate at flame sheet downstream side. For a rich spray flame, three extinction patterns were observed; they occur in SPP, SPB or at the critical SPB mode, but do not in CPB. The extinction maps of the compound-drop spray demarcate the patterns and also indicate the limitations and corresponding conditions of the flame extinction.