Droplet characteristics and local equivalence ratio of reacting mixture in spray counterflow flames. Experimental Thermal and Fluid Science

Transcription

1 Accepted Manuscript Droplet characteristics and local equivalence ratio of reacting mixture in spray counterflow flames M. Orain, Y. Hardalupas PII: S (14) DOI: Reference: ETF 8225 To appear in: Experimental Thermal and Fluid Science Received Date: 25 February 2014 Revised Date: 16 May 2014 Accepted Date: 16 May 2014 Please cite this article as: M. Orain, Y. Hardalupas, Droplet characteristics and local equivalence ratio of reacting mixture in spray counterflow flames, Experimental Thermal and Fluid Science (2014), doi: /j.expthermflusci This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

2 Droplet characteristics and local equivalence ratio of reacting mixture in spray counterflow flames M. ORAIN 1, Y. HARDALUPAS 2 1 ONERA The French Aerospace Lab, F Palaiseau France 2 Imperial College London, Exhibition road, SW7 2AZ London UK Full-Length Article STATEMENT The content of this manuscript submitted to Experimental Thermal and Fluid Science is unpublished material that is not being submitted for publication elsewhere. Shortened Running Title Local equivalence ratio in spray counterflow flames Corresponding author: Dr. M. Orain, Physics Instrumentation and Sensor Department ONERA Tel: Fax: address: mikael.orain@onera.fr

3 Abstract Spray combustion was studied by injecting monodisperse 125 or 200 µm ethanol droplets in a premixed natural gas fuel flame flowing against an opposed heated air jet. Phase Doppler anemometry and chemiluminescence measurements allowed to characterise both droplet parameters and the local reacting mixture. Both types of droplets crossed the flow stagnation plane and entered the opposite jet. However, 125 µm droplets reversed their motion and oscillated around the stagnation plane, leading to increased droplet residence time in hot regions. The fuel vapour released by droplets close to the stagnation plane, mixed with the surrounding air and led to the ignition of a second fuel vapour flame below the natural gas flame. For 125 µm droplets, mean local equivalence ratio of the fuel vapour flame was about 0.8, suggesting lean-premixed combustion; whereas 200 µm droplets led to stoichiometric combustion. Group Combustion number was estimated from measurements and suggested that 125 µm and 200 µm droplets burned in different Group Combustion regimes. Keywords : Spray counterflow flames, droplet combustion, local equivalence ratio, droplet size, chemiluminescence. 2

4 1. Introduction Early studies on liquid fuel combustion commonly associated droplet burning with non-premixed reaction mode [1], which leads to high flame temperature and, therefore, high levels of NOx emissions. However, low-nox liquid-fuelled burners do exist and the NOx reduction must be associated with some kind of partially-premixed combustion [2]. Indeed, droplet dispersion around the reaction zone in liquid-fuelled burners can affect droplet evaporation and some premixing of fuel vapour with air may occur prior to the ignition of a partially-premixed flame. Nonetheless, the relative contribution of premixed and non-premixed combustion in low-nox liquid-fuelled burners is usually unknown, as well as the equivalence ratio of the reacting mixture and both quantities can contribute to reduced NOx emissions. Additionally, in an industrialtype combustor, it is difficult to determine the respective influence of each of these parameters. Consequently, in the literature, experiments used to better understand the fundamentals of spray combustion are often performed on idealised spray flames in order to minimize the coupling between the different effects and provide parametric results. Spray counterflow flames offer a stable reaction zone close to the flow stagnation plane with constant strain rate and good optical access for non-intrusive measurements, and have been extensively used to study droplet combustion, either experimentally [3-7] or numerically [8-12]. They can be divided in two types: 1) all droplets evaporate upstream from the reaction zone or 2) some droplets cross the flame front and burn beyond it. The first case results in fully single-phase combustion [13-15], which is of little interest for industrial burners where most of the droplets do not evaporate completely before reaching the flame front. The latter case is more representative of real burners because it deals with twophase flames. Indeed, this arrangement can simulate droplet dispersion and fuel vapour mixture formation around the reaction zone close to the stagnation region of a liquid-fuelled combustor stabilised by a recirculation zone. Two-stage spray counterflow flames were for example reported experimentally in [3, 16, 17] where single droplet burning beyond the main flame was observed with n-heptane or methanol, which are high-volatility fuels. More recently, Mikami et al. [18] could identify combustion of individual droplets or droplet clusters beyond the main flame in a counterflow burner operating with n-decane (i.e. a lowvolatility fuel of interest for model gas turbines) and proposed a mechanism to explain transition from isolated droplet to cluster combustion. Two-stage counterflow flames (operating with different liquid fuels) were also reported in computational studies [9, 10, 19] and Continillo and Sirignano [9] have indicated that both premixed-like and diffusion-like combustion can occur in low-strained n-octane spray counterflow 3

5 flames. Coexistence of both burning regimes was also observed numerically by Akamatsu and co-workers [11, 12] in laminar n-decane spray counterflow flames and the influence of strain rate, overall equivalence ratio and droplet size was investigated. An additional interesting feature of two-stage spray counterflow flames concerns the possible occurrence of droplet oscillations around the burner stagnation plane which were first reported experimentally in [20] and later in [4, 21]. Indeed, the authors mentioned that droplet oscillations could be associated with increased droplet residence time in the burner, which could yield increased release of fuel vapour in the flow and, subsequently, modify the local equivalence ratio. Later on, in highly-strained counterflow spray flames (fuelled with methanol, ethanol and n-heptane), extensive numerical studies by Gutheil and co-workers [10, 22-25] have predicted the presence of a second maximum in the profiles of fuel vapour concentration and temperature along the burner axis, coinciding with the location where droplets reversed their motion. The authors have shown that the occurrence of multiple reaction zones depends on the flow strain rate, as well as on droplet size, albeit simulations were performed with monodisperse or bidisperse droplet sprays only. These numerical findings were a great source of inspiration for the present experiments. It is noted that the experimental studies mentioned above quantified mainly droplet characteristics (i.e. size and velocity), but not flame properties (i.e. local equivalence ratio and degree of premixing). The purpose of the current work was, therefore, to perform experiments in spray counterflow flames with monodisperse droplets penetrating the flame in order to quantify the characteristics of the local reacting mixture and correlate them with droplet size. The remaining paper comprises three parts. The first part describes the experimental facility, the optical techniques used for the measurement of spray characteristics and local reacting mixture. The second part presents the results and discusses the findings in terms of Group Combustion theory. Finally, the paper ends with a summary of the findings. 2. Experimental set-up 2.1. Test facility The counterflow burner used for this work (Fig. 1) was the same as that of [26] and similar to that of [27, 28]. It comprised two opposed brass pipes with inner diameter D=30 mm, separated by a distance H=25 mm. Premixed gaseous fuel (natural gas with 94 % methane) and air were injected from the upper duct, while 4

6 only preheated air at 420 K was injected from the lower duct. Coaxial jets of N 2 could be operated on both sides, in order to prevent diffusion flames from appearing at the edges of the flat flames. The equivalence ratio of the premixed natural gas was kept constant at Φ 0 =0.7 and the bulk gas velocity was fixed to V 0 =2 m/s (V 0 is the area-averaged velocity at the ducts exit). This corresponds to bulk flow strain rate of 160 s -1 and turbulent strain rate of 320 s -1. The bulk flow strain rate was classically evaluated as S=2V 0 /H. The turbulent strain rate was obtained from measured local velocity fluctuations and an estimate of the turbulent lengthscale of the flow [26], following the definition of Mastorakos [27]: S t 3 / 2 3 / 2 C q V0 = (1) νl t where C q is the normalised turbulence intensity (C q ~ 0.085), L t the turbulent lengthscale (L t ~ 3 mm) and ν the kinematic viscosity of air (ν=1.59e-5 m 2 /s at 300 K). The injection of ethanol (99 % purity) droplets in the counterflow burner was realised using a custom-made monodisperse droplet generator [29], located on top of the upper duct of the counterflow burner. The droplet generator was of similar type to that of Berglund and Liu [30]. It consisted of two piezoelectric transducers which imposed perpendicular vibrations onto a liquid jet passing through a pinhole (diameter: 50 or 100 µm). These vibrations yielded jet oscillations, which ultimately caused its break-up and led to a stream of monodisperse droplets which size was imposed by the pinhole diameter and the vibrations frequency. The droplet injection frequency was 20 and 10 khz, respectively for the 50 and 100 µm pinholes, leading respectively to droplets with diameter 125 µm (velocity 8 m/s) and 200 µm (velocity 12 m/s). Liquid flowrate was 1.2 and 2.5 cc/min for 125 and 200 µm droplets respectively. A single stream of monodisperse droplets was injected along the symmetry axis of the natural gas / air jet, at a distance of 120 mm from the stagnation plane of the counterflow burner. The droplet generator was water-cooled in order to maintain ethanol fuel at a constant temperature (298 K). Droplet sizes were selected, so that droplet evaporation upstream of the natural gas reaction zone was negligible. The coordinate system used in the present experiments has its origin on the burner vertical axis at the stagnation plane, Fig. 1. Results will be presented as a function of the radial distance r normalised by the radius of the counterflow burner R=15 mm and the axial distance z normalised by the distance between the 5

7 two ducts of the counterflow burner H=25 mm. Therefore, the value z/h=0 corresponds to the flow stagnation plane, the upper duct of the burner is located at z/h=0.5, while the lower duct is at z/h=-0.5. The value r/r=0 corresponds to the burner vertical axis, the edges of the burner are located at r/r=-1 and r/r= Optical measurement techniques In the literature, chemiluminescence from OH*, CH* and C 2 * radicals has often been used to determine the flame front structure [31, 32] or to monitor heat release [33, 34]. Additionally, correlation between the ratio of chemiluminescent intensity from those radicals and flame equivalence ratio has been considered in a few studies [35-39], although findings often depended on combustor geometry and fuel type. Extensive investigation of the influence of parameters such as equivalence ratio, strain rate and type of fuel on chemiluminescent intensity from OH*, CH* and C 2 * radicals has been reported in [40, 41]. Results have shown that the intensity ratio OH*/CH* has the potential to measure equivalence ratio of the reacting mixture in various flames. These two previous articles provide the full technical details and data processing of the chemiluminescence technique for time-dependent local measurement of equivalence ratio, which was used in the present study. Therefore, only the main features of the technique are repeated here. Simultaneous local measurements of the chemiluminescent intensity from OH*, CH* and C 2 * radicals in the ethanol droplet flame were achieved using the Cassegrain optical system of Akamatsu et al. [42] connected to a custom-made spectroscopic unit [40], Fig. 2. The Cassegrain optics made use of reflective mirrors, rather than lenses, to focus the collected light from the flame onto a pinhole in front of an optical fiber, which transmitted the light to the spectroscopic unit. The reflective mirrors eliminated achromatic aberrations and, as a consequence, the focal point for OH*, CH* and C 2 * chemiluminescence coincided. The Cassegrain optics had a nominal probe volume with diameter 200 µm and length 1.6 mm, which can resolve local flame characteristics. Limitations of the spatial resolution of the Cassegrain optics have been evaluated both experimentally and numerically by Hardalupas et al. [43-45] and shall be briefly mentioned here. These authors have shown that if the notional axis of the Cassegrain optics is placed tangentially to a cylindrical flame, the spatial resolution of the sensor is appropriate for local chemiluminescence measurements in flames with diameter smaller than around 60 mm (which is the case of the current flames). This was also verified in model swirl-stabilised burners [45, 46]. Furthermore, Hardalupas et al. [45] have shown that with conical flames (which is the case of the droplet flame in the present experiments), the spatial resolution of 6

8 the sensor is even better than with cylindrical flames because the amount of light collected from outside the probe volume is reduced, due to a smaller intersection volume between the flame and the solid angle of collection of the sensor. The left hand side of Fig. 3 depicts the way the Cassegrain optics was oriented with respect to the droplet flame, and how the sensor was traversed to measure radial profiles of chemiluminescence intensity, which corresponds to the point measurements traverse direction in the article of Hardalupas et al. [45], hence the most appropriate way to benefit from the high spatial resolution of the Cassegrain optics. On the right hand side of Fig. 3, a horizontal cross-section of the intersection between a flame (16 mm diameter) and the collection angle of the sensor is represented to scale. As can be seen, the intersection is limited, and it is even further reduced for flames with a smaller diameter. Additionally, the reaction zone is rather thin compared to the cross-section area between the droplet flame and the light collection angle of the sensor. Together with the fact that the Cassegrain optics collects light from its probe volume 420 times more efficiently than outside its probe volume (see Akamatsu et al. [42]), this suggests that contributions from outside the probe volume are marginal compared to the chemiluminescent signal from the probe volume, in the present experiments. Finally, placing a mask (with height 10 mm) at the front of the collection optics further reduced potential contributions from chemiluminescence emissions of the natural gas flat flame which are located outside the probe volume. The spectroscopic unit comprised two dichroic mirrors (Optical Coatings Japan, with efficiencies above 90 %) to split up the collected light into three parts. Each part was directed on appropriate interference filters (Optical Coatings Japan) specific to the radical considered. The filters were centred respectively at nm, nm, nm with corresponding efficiencies of 22 %, 45.3 %, 71.7 % and full width at half maximum (FWHM) of 13.5 nm, 1.9 nm, 2.1 nm for OH*, CH* and C 2 * radicals respectively. The collected light intensities were transformed into electrical signals by three photomultiplier tubes (Hamamatsu R269). The temporal signals of OH*, CH* and C 2 * were then amplified and digitised simultaneously using a 16 bit A/D card (DT2827, Data Translation) and recorded onto a personal computer. The sampling frequency was 50 khz for each chemiluminescence signal and the duration of each measurement record was 1.64 s per radical, at each position of the measurements. This duration was long enough to obtain statistics over more than 1000 turnover timescales of flow fluctuations, which had a typical timescale less than 1 ms, based on the ratio between the integral lengthscale and the rms of the flow axial velocity at the outlet of the upper duct of the counterflow burner [26]. 7

9 Although the spectroscopic unit also monitored C 2 * chemiluminescence, results are reported only for OH* and CH* radicals here. The time-dependent signals of chemiluminescent intensity from OH* and CH* radicals and the intensity ratio OH*/CH* were analysed using the method Processing 1 [40, 41] which will be briefly recalled now. First, a threshold was selected for the raw intensity CH* signal (threshold = 25 % of the maximum intensity); which allows the detection of signals above the background noise and also to numerically filter out the contributions of the chemiluminescence signal that could originate from outside the probe volume of the Cassegrain optics. The probability of flame presence was obtained from the proportion of time that the chemiluminescent intensity of CH* radicals is higher than the threshold level. Second, samples above the threshold for the CH* signal were identified; these were considered to arise from chemical reaction in the optical probe volume. The corresponding points in time in the OH* intensity signal were identified to evaluate the instantaneous OH*/CH* ratio. Within the interval of the signal record being above the threshold, we identified the maximum values of the CH* and OH* signals. If the maxima of the intensities coincided within a tolerance of ±2 sampling points around the CH* intensity maximum, we calculated the instantaneous value of the OH*/CH* ratio from these two maxima. This approach was selected because it was considered that each time window for which the CH* intensity was above the threshold represented a laminar flamelet in the current flames. However, we also evaluated the results when the ratios for all the intensities above the threshold were considered and there was little influence on the final measurements. This was in agreement with the reported findings in counterflow flames of Hardalupas and Orain [40] for the methods Processing 2 and 3, which also considered the ratios for all the data points above the selected threshold. Third, The mean and rms of the fluctuations of the OH*/CH* ratio were calculated from the instantaneous OH*/CH* ratio, taking into account the different efficiencies of the interference filters, dichroic mirrors and the spectral response of the photomultiplier tubes in the spectroscopic unit. Figure 4 represents the calibration curve describing the dependence of OH*/CH* intensity ratio upon equivalence ratio for V 0 =2 m/s (which corresponds to the bulk gas velocity in the current experiments), as measured by Orain and Hardalupas [41] in ethanol-fuelled premixed opposed flames. This curve was used as the calibration for the conversion of the chemiluminescent measurements into equivalence ratio. OH*/CH* intensity ratio exhibits a monotonic behaviour with flame equivalence ratio Φ 0 which can be approximated by a linear fit (represented by the dash line in Fig. 4) in the form: OH CH * * = Φ (2) 0 8

10 Unfortunately, the counterflow burner used by Orain and Hardalupas [41] could not be operated in the nonpremixed mode with ethanol fuel, which would probably produce soot anyway, leading to a strong emission over a wide range of the flame spectrum which could interfere with the emission bands of CH* and C 2 * radicals centred at 431 and 516 nm respectively. Therefore, following the idea of Panoutsos et al. [47], chemiluminescence measurements were complemented with numerical simulations of stoichiometric premixed and non-premixed ethanol-fuelled counterflow flames. Computations were performed using the one-dimensional flame code OPPDIFF [48] and CHEMKIN software [49]. The chemical kinetics mechanism used for ethanol combustion was that of Marinov [50], while the mechanism proposed by Panoutsos et al. [47] was selected for the chemiluminescence reactions of OH* and CH* radicals. It is noted that Benvenutti et al. [51] proposed a chemical kinetics mechanism for ethanol combustion which includes excited species like OH*, CH*, C 2 *, CH 2 O* and HCO* radicals. However, it is a reduced mechanism with a limited number of reactions for the chemiluminescent species, which may lead to large uncertainties on the formation and destruction of those excited species. Figure 4 shows that experiments and simulations agree quantitatively within 30 % for stoichiometric premixed flames, which may be associated with the general limitations of chemiluminescence sub-mechanisms modelling as pointed out by Panoutsos et al. [47]. Nonetheless, an important outcome of computations concerns the large difference between the values of the OH*/CH* intensity ratio in diffusion and premixed ethanol-fuelled counterflow flames, which is similar to results in CH 4 /air counterflow flames [47]. Indeed, non-premixed flames exhibits an OH*/CH* ratio four times lower than premixed flames for the present flow conditions. This feature provides a criterion to discriminate without any doubt both types of flames, which is a crucial issue when a chemiluminescence-based technique is used to measure the equivalence ratio of the reacting mixture in a partially-premixed flame. Finally, the degree of non-stoichiometric reaction was quantified by considering the probability of values of local equivalence ratio being outside the range [0.95, 1.05], which was considered to represent stoichiometric reaction, owing to the accuracy of the measurement technique (i.e. ±0.05 units of equivalence ratio [40, 41]). The interest to determine the degree of non-stoichiometric reaction is related to the maximum flame temperature associated with stoichiometric combustion that leads to maximum NO X emissions. Droplet characteristics (size, velocity, flux and number density) were measured with a custom-made phase Doppler anemometer (PDA) [52]. The Ar + laser used for PDA measurements was operating at a wavelength 9

11 of nm, with a power of 100 mw. The transmission optics had a focal length of 600 mm, which resulted in a beam crossing angle of 3.08º and a probe volume with dimensions μm 3. The frequency shift between the two beams was provided by a rotating grating and was fixed to 6 MHz. The light scattered by the droplets was collected at 30º off-axis from the forward direction by the receiving optics (500 mm focal length) and focused onto a spatial filter of 300 µm width, which limited the nominal length of the probe volume. The collected light was subsequently directed to three photomultipliers and the detected signals were processed by a zero-crossing counter, which was interfaced to a PC. The selected optical arrangement allowed to measure droplet size between 2 μm and 280 μm, which covered all droplet diameters encountered in the present experiments. The droplet size validation criteria for the PDA were based on the approach of Hardalupas and Taylor [53]. Briefly, the method consists of using three photodetectors (similarly to commercial PDA systems by Artium, Dantec or TSI), which improves the sizing accuracy by rejecting measurements from signals with phase ratio values lying outside the expected value for spherical scatterers and which would otherwise have been validated by a two photodetector system. The measured droplet diameter was based on the number density of liquid droplets through the probe volume, instead of the droplet number flux, using the approach of Hardalupas and Taylor [54] which was discussed further by Hardalupas et al. [55]. Quickly, the method consists of measuring the residence time of particles in the probe volume and the number density of each size class, which improves the measurement accuracy of the PDA technique compared to conventional early PDA approaches based on dividing the mass flux by the mean velocity, for each size class. The experimental uncertainties for the droplet size and velocity were less than 2 % and the droplet number density and droplet flux have uncertainties of the order of 15 %. 3. Results This section is divided in two parts. The first sub-section describes the dispersion of droplets injected in the counterflow burner, and presents results on droplet diameter and velocity, liquid flux and number density based on PDA measurements. In the second sub-section, visualisation shows that droplets can ignite after crossing the counterflow flame front, which leads to a second fuel vapour flame ignited below the natural gas flame. Chemiluminescent intensity measurements are used to determine the mean local equivalence ratio of the second flame due to droplets, and the corresponding degree of non-stoichiometric reaction. 10

12 3.1. Droplet behaviour in the counterflow burner Influence of droplet diameter on droplet trajectory in the burner at isothermal conditions Figure 5 shows photographs of droplet trajectories in the counterflow burner at isothermal conditions (i.e. no combustion), for two droplet diameters (125 and 200 µm). As can be seen, 200 μm droplets go straight into the lower duct where the liquid fuel impacts the walls. By contrast, 125 µm droplets cross the stagnation plane, reach the lower duct wherein they reverse their motion, return to the stagnation plane and oscillate around it. This different behaviour is attributed to the inertia of the droplets, as well as their initial axial velocity (8 m/s for 125 µm droplets vs. 12 m/s for 200 µm droplets). Figures 6(a) and 6(b) show the evolution of the axial velocity probability density function (PDF) of 125 and 200 µm droplets, measured by PDA, at three axial locations in the counterflow burner: z/h=+0.44 (close to the upper duct), z/h=0 (at the stagnation plane) and z/h=-0.44 (close to the lower duct) on the burner axis (r/r=0). Each PDF is normalised by its total number of events (50000 droplets). The axial velocity is defined as negative for droplets travelling downwards and positive for droplets moving upwards. Hence, following this sign definition, the initial axial velocity is -8 m/s and -12 m/s, for 125 and 200 µm droplets respectively. The velocity PDF is unimodal, with negative values, for 200 μm droplets, which shows that these droplets move only in the downward direction. By contrast, the velocity PDF of 125 μm droplets exhibits a bimodal shape with negative and positive values, which indicates that these droplets cross the stagnation plane and about 22 % of them reverse their motion to oscillate around the stagnation plane. Figure 6(c) shows the evolution of the axial velocity PDF of 125 µm droplets at four radial locations in the counterflow burner: r/r=0 (on the burner axis), r/r=0.33, r/r=1 and r/r=1.47 on the stagnation plane (z/h=0). At the first three locations, the PDF is bimodal, which indicates the presence of droplets moving downwards as well as upwards (the percentage of droplets moving upwards increases with radial distance). At the last position (r/r=1.47), the PDF becomes monomodal and symmetrical to 0 m/s, which shows that droplets are also moving both downwards and upwards at that location, albeit with reduced axial velocity. Bimodal PDFs at different radial locations suggest occurrence of multiple successive crossings of the stagnation plane by the droplets. The behaviour of 125 µm droplets is clearly more interesting than the trend observed with 200 µm droplets, because droplet oscillations around the stagnation plane lead to an increased droplet residence time in the burner. For example, the average residence time of a 125 µm droplet in the gap between the stagnation plane and the exit plane of the lower duct can be estimated as around 3 ms for droplets moving downwards 11

13 (based on an average droplet velocity ~ -3.8 m/s), whereas it is about 9 ms for droplets moving upwards (based on an average droplet velocity ~ +1.4 m/s). Nonetheless, the residence time of a 125 µm droplet oscillating around the stagnation plane (i.e. after the first and second crossings) is difficult to determine, because it depends on the droplet trajectory and the associated droplet velocity that varies in space. Such information may be obtained by particle tracking velocimetry (PTV), which was, however, not available for the current experiment. Therefore, the droplet residence time was evaluated from numerical simulations and was found to be of the order of a few tens of ms [26]. Similarly, the total residence time of a 125 µm droplet in the burner (i.e. when a droplet is between the exit planes of the two ducts, as well as when it is inside the lower duct) is hard to estimate, because we do not know experimentally how far droplets travel inside the lower duct before they reverse their motion. Again, from numerical simulations, the total residence time inside the burner was evaluated to be around 160 ms for a 125 μm droplet, with the droplet being about 100 ms below the stagnation plane (i.e. in the region of negative z values, including inside the lower duct). Therefore, in a reacting counterflow burner, it is anticipated that 125 μm droplets should remain a long time in hot regions where evaporation is enhanced, which may lead to release of fuel vapour below the natural gas flat flame Influence of the counterflow flame on droplet diameter As indicated in the section describing the experimental set-up, combustion experiments were performed with the equivalence ratio of the fully-premixed natural gas flame Φ 0 =0.7. Although not shown in the paper, PDFs of droplet axial velocity similar to those presented in Fig. 6 were observed in the counterflow flame, with a bimodal shape for 125 µm droplets and a unimodal shape for 200 µm droplets. Figures 7(a) and 7(b) present the droplet size distributions conditional on 125 and 200 µm droplets moving downwards (negative axial velocity) at different axial locations in the reacting counterflow burner. For 125 µm droplets, a slight shift towards small diameters is observed in the size distributions, due to flame crossing and increased droplet residence time in regions where the local temperature is above ethanol boiling point (351 K). By contrast, for 200 µm droplets, no shift of the size distribution is observed between the different locations along the vertical axis of the burner. Figure 7(c) presents the size distributions conditional on 125 µm droplets moving upwards (positive axial velocity) at different axial locations in the burner. The size distribution at the stagnation plane (z/h=0) is narrower than that at the exit plane of the lower duct (z/h=-0.44), which shows 12

14 that small droplets evaporate as they return to the flame. In addition, intermediate droplet sizes, smaller than the originally-injected 125 µm droplets, appear in the size distribution at z/h=-0.44, which suggests strong evaporation. In Fig. 7, each PDF is normalised by its total number of events. Table 1 indicates that, for 125 µm droplets, the droplet arithmetic mean diameter D 10 decreases by around 5 % of the initial diameter, when droplets move from the exit of the top to the opposed duct. More interesting, the mean diameter is reduced by 10 % between the moment a droplet crosses the flame moving downwards for the first time, and the moment it returns to the flame and crosses it moving upwards. This demonstrates that some volume of liquid fuel is evaporated below the stagnation plane, which can modify the local equivalence ratio due to fuel vapour mixing with the surrounding air. By contrast, Table 2 shows that, for 200 µm droplets, D 10 remains almost constant from the upper to the lower duct, suggesting very limited liquid fuel evaporation between the stagnation plane and the exit plane of the lower duct. The amount of vaporised liquid fuel below the natural gas flame is evaluated from the radial evolution of the droplet volume mean diameter D 30 which is represented in Fig. 8, for the injection of 125 µm droplets. Indeed, as opposed to D 10 which is a population-based mean diameter, D 30 is a volume-related mean diameter. Therefore, D 30 is more suitable to evaluate a change of droplet volume (or mass), which is directly related to the volume of fuel vapour generated by the evaporation process. A proper evaluation of the amount of vaporised liquid requires to separate the D 30 diameter of droplets moving downwards ( D droplets moving upwards ( D + 30 ). Figure 8 shows that 30 ) from that of D 30 is 125 µm (i.e. the size of the originally-injected droplets) close to the centreline, whereas it reduces as radial distance increases, due to droplets that have reversed their motion at least twice. The evolution of + D 30 indicates that droplets which move upwards after reversing their motion are consistently smaller (around 110 µm) than the originally-injected 125 µm droplets, which confirms liquid fuel evaporation below the stagnation plane. Figure 9 presents the radial profiles of liquid volume flux at the stagnation plane of the counterflow burner (z/h=0) for droplets moving downwards and upwards respectively, for the injection of 125 μm droplets. The liquid flux, which represents the volume of liquid per unit of area per unit of time that passes through the PDA probe volume, was quantified according to the method of [54]. In Fig. 9, at each radial location, the + volume flux of droplets moving downwards ( G, counted as negative), respectively upwards ( G, counted 13

15 as positive), is normalised by the total volume flux of droplets. Since the initial monodisperse droplet trajectories extend radially over a few millimetres from the centreline, it is necessary to identify a radial location r 0 where droplets cross the stagnation plane moving downwards for the first time. It is also required to define a section (of radius r 2 -r 1 ) where droplets cross the stagnation plane moving upwards for the first time. Figure 10 schematically represents droplets trajectory between the two opposed ducts and summarises the definition of r 0, r 1 and r 2. The radial variations of D 30 and liquid flux at the stagnation plane in Figs. 8 and 9 can define the corresponding radial values of the two sections as r 0 /R=0.33, r 1 /R=0, r 2 /R=1. r 0 /R is selected as 0.33, because up to this radius the volume mean diameter size of injected droplets) and the volume flux D 30 of Fig. 8 is close to 125 µm (the original G of Fig. 9 is nearly constant. Visualisation of droplet trajectory also confirms that the initial droplets can disperse up to this radial position. Please note that the presence of downwards liquid flux for r/r>0.33 is due to droplets that cross the stagnation plane for the second time downwards, after they have reversed their motion twice. In Fig. 9, the upwards liquid flux + G is different from zero on the centreline (please note that, theoretically, droplets can oscillate around the stagnation plane, on the centreline, if they are injected exactly at r/r=0 with no radial velocity) and increases with radial distance, therefore, r 1 /R=0 and r 2 /R=1 were selected accordingly. From the values of r 0, r 1 and r 2, the spatially-averaged volume mean diameter D 30 of droplets moving downwards or upwards can now be quantified by integrating the radial profiles between the appropriate radial boundaries. The expression is given by: D r = 2πrD r πrG ( r) G ( r) dr ( r) dr for droplets moving downwards D r2 + r1 30 = r2 r1 + 2πrD30 ( r) G 2πrG + + ( r) dr ( r) dr for droplets moving upwards (3) Using formulae in equation (3) with the appropriate profiles of Figs. 8 and 9, 14 D 30 and + D 30 are evaluated as 124 μm and 112 μm respectively. As a consequence, the volume of liquid fuel vaporised below the stagnation plane is represented by the change of the averaged volume mean diameter, which indicates that 26 % of the initial volume of liquid fuel injected in the counterflow burner evaporates below the stagnation plane and modifies the local equivalent ratio. The effect of this vapour release on the counterflow flame will be examined in the next section.

16 The same analysis cannot be performed for 200 μm droplets because PDA measurements show that droplet evaporation is limited between z/h=0 and z/h= As mentioned previously, 200 µm droplets go straight into the lower duct where they impact the walls and ultimately evaporate, and all the fuel vapour mixes with the air supplied to the lower duct. However, it is difficult to evaluate the amount of fuel vapour burnt in the second flame (close to the burner axis). Indeed, it is not possible to assume that all the vapour generated in the lower duct is used to form the second flame close to the centreline, as the vapour probably spreads all over the section of the lower duct and may also burn at the outer edges of the natural gas flat flame. Nevertheless, it is expected that the amount of fuel vapour released by 200 µm droplets in the lower jet is higher than that released by 125 µm droplets during oscillatory motion around the stagnation plane Flame behaviour with droplets Influence of droplets on the counterflow flame Figure 11 shows photographs of the counterflow flame, without and with droplet injection. As can be seen, a second flame is present below the natural gas flame when droplets are injected. The fuel vapour generated by droplet evaporation below the stagnation plane mixes with the surrounding air and leads to ignition of this second flame which is elongated along the centreline, but also extends radially and parallel to the gas flame. Figure 12 shows two images of the flame recorded at different times by an intensified CCD camera (exposure time: 10 µs, acquisition rate: 25 Hz) which suggest that the second flame is not always present below the main reaction zone. This flame intermittency is probably due to the temporal variation of local equivalence ratio below the stagnation plane which is, therefore, not always enough to ignite and sustain a flame. From statistics over 400 CCD images, the probability of presence of the second flame close to the centreline was evaluated as 70 % for 125 μm droplets and increased to 90 % for 200 μm droplets. This finding may have its origin on the larger amount of fuel vapour released by the 200 μm droplets after they enter the lower jet. This presumably leads to a higher equivalence ratio of the fuel/air mixture in the second flame and, therefore, higher flame stability. This will be quantified thereafter. 15

17 3.2.2 Characteristics of the second flame formed by droplets The chemiluminescent intensity signals from OH* and CH* radicals were processed as mentioned in section 2.2 and provide the following quantities for the second flame related to the fuel vapour released by the evaporating droplets. (i) Probability of flame presence at different locations in the flow, defined as the time that the CH* intensity is higher than a selected intensity threshold; (ii) mean local equivalence ratio of the reacting mixture, based on the OH*/CH* ratio of the chemiluminescent intensities; (iii) Probability Density Function of the local equivalence ratio of the reacting mixture, as obtained from the time-dependent analysis of the chemiluminescent intensities; and (iv) degree of non-stoichiometric reaction. (a) Injection of 125 µm droplets Figure 13(a) shows that the probability of flame presence on the burner axis decreases from 75 % to 11 % with increase of axial distance from the stagnation plane. It is maximum close to the centreline, where most of the droplets are present and decreases in the radial direction, where droplet number density reduces and the released fuel vapour is lower; which explains that no flame is present at large radial locations. These probability values indicate that the fuel vapour flame is intermittent, because of the inability to ignite a fuel/air mixture, probably when the local equivalence ratio is lower than a critical value for ignition related to the local strain rate. The probability of flame presence suggests that the time-averaged fuel vapour flame has a conical shape around the centreline of the burner and the mean flame dimensions are estimated as 9 mm along the z axis of the burner with a radius at the base close to the stagnation plane of around 9 mm. These values are close to those observed from long exposure photographs. Figure 13(b) indicates that the time-averaged mean equivalence ratio of the fuel vapour flame is less than stoichiometric at all locations, which suggests lean-premixed combustion. Close to the centreline, it decreases from 0.8 to 0.5 with increase of axial distance from the stagnation plane. This probably comes from the increased droplet number density close to the stagnation plane, due to contribution of droplets reversing their motion in the opposed hot air jet and incoming droplets from the upper jet. As a consequence, the amount of vapour released close to the stagnation plane is higher than in the hot air stream of the lower jet due to enhanced droplet vaporisation; which justifies the increased equivalence ratio close to the stagnation plane. In addition, velocity measurements with PDA showed that droplets decelerate close to the stagnation plane; which increases their residence time in that flow region and, therefore, the release of fuel vapour leading to locally richer mixtures. 16

18 It is noted that the decrease of the mean equivalence ratio with distance from the stagnation plane agrees with the observed reduction of the probability of flame presence in Fig. 13(a). Far from the burner axis, values of time-averaged mean equivalence ratio are lower than the lean flammability limit of ethanol which is 0.45 according to [56]. However, lean flammability limits in the literature are measured for cold fullypremixed environments and without assessing the influence of flow strain rate. Therefore, it may be that at large radial locations, preheating due to the presence of premixed natural gas flame and the incoming hot air stream from the lower duct allows ignition of fuel/air mixtures leaner than 0.45 at certain times. However, ignition of mixtures with low equivalence ratios is a rare event, as indicated by the low probability of flame presence (see Fig. 13(a)) and is, therefore, associated with rare flow events leading to favourable strain rates for ignition of lean mixtures. Figure 14 shows the probability density function (PDF) of equivalence ratio at different axial and radial locations in the fuel vapour flame. At both axial positions, each PDF is normalised by the total number of events of the corresponding PDF recorded at r/r=0.07. Close to the burner axis, increase of axial distance from the stagnation plane leads to the most probable value of equivalence ratio to shift towards lower values; which explains why local mean equivalence ratio decreases along the centreline. The shape of the PDFs suggests that combustion mainly occurs in lean-premixed mode, although the probability of stoichiometric equivalence ratio cannot be neglected near the centreline at z/h= The range of equivalence ratios of reacting mixtures broadens with reduction of z, possibly due to increased temporal fluctuations of fuel vapour generated by droplet evaporation. Additionally, the PDFs indicate some reaction of rich-premixed mixtures close to the stagnation plane which do not occur at z/h= For a given axial location, the PDFs slightly broaden with radial distance, suggesting larger inhomogeneities of the fuel vapour/air mixture. At z/h=-0.10, the shape of the PDFs is almost Gaussian close to the burner axis, but becomes skewed towards lower values of equivalence ratio at large radial distances. From the probability density functions of equivalence ratio, an estimate of the degree of non-stoichiometric reaction (η) is obtained by evaluating the percentage of reacting events with equivalence ratio lower than 0.95 (corresponding to lean-premixed mode) or higher than 1.05 (corresponding to rich-premixed mode). Figure 15 shows the degree of non-stoichiometric reaction of the fuel vapour flame at different axial and 17

19 radial locations, which increases with radial distance, so that combustion becomes fully non-stoichiometric. Close to the centreline, η is between 82 % and 100 %, suggesting that lean- or rich-premixed combustion dominates for this droplet size, despite some contribution of stoichiometric combustion is observed at z/h= This finding is supported by the observed blue colour of the second flame. Findings for 125 μm droplets suggest that droplets remaining in relatively hot regions of liquid-fuelled burners for a long time (e.g. in the recirculation zone) may release fuel vapour, which can mix with air before ignition and lead to partiallypremixed combustion and potential reduction of NOx emissions. (b) Injection of 200 µm droplets Figure 16(a) indicates that the probability of flame presence on the burner axis slightly decreases with increase of axial distance from the stagnation plane, but always remains between 80 and 93 %. It is maximum on the axis and decreases in the radial direction. This suggests that the fuel vapour flame is intermittent, but to a smaller extent than for 125 µm droplets where flame probability is lower at all locations. The time-averaged axial length of the fuel vapour flame is estimated with the eye as about 35 mm, hence significantly longer than for 125 µm droplets, as the flame penetrates inside the lower duct. The mean flame radius at the base close to the stagnation plane is around 15 mm, which is close to the value obtained with long exposure photographs. By contrast with observations for 125 µm droplets, time-averaged mean equivalence ratio of the fuel vapour flame remains close to stoichiometric values at all locations on the burner axis, Fig. 16(b). This may be attributed to the fact that 200 µm droplets do not reverse their motion, so that, close to the centreline, fuel vapour is generated mainly from droplets moving downwards. In addition, 200 µm droplets have constant large velocity between the two ducts (see Table 2), so that droplet residence time in that flow region is constant irrespective of the axial locations considered and, therefore, droplet evaporation does not vary significantly. Far from the burner axis, equivalence ratio is low, similarly to that observed for 125 µm droplets, and correlated with low probability of appearance of ignited mixtures. Similarly to Fig. 14, in Fig. 17, at both axial positions, each PDF is normalised by the total number of events of the corresponding PDF recorded at r/r=0.07. Near the burner axis, the most probable value of the equivalence ratio PDFs is close to 1.0, which justifies that mean local equivalence ratio remains constant with axial distance along the centreline. Additionally, the shape of the PDFs suggests that combustion occurs 18

20 in both stoichiometric and lean- or rich-premixed regimes, which differs from findings with 125 µm droplets where very few stoichiometric events were observed. Near the burner axis, the extent of the tail of the PDFs towards high equivalence ratios remains similar with distance from the stagnation plane (as opposed to 125 µm droplets), which is in line with the limited variation of local mean equivalence ratio with axial distance. The most probable value of the PDFs shifts towards lower values of equivalence ratio as radial distance increases, in agreement with the mean values observed in Fig. 16(b). The PDFs also broaden with radial distance, indicating more inhomogeneities in the fuel vapour/air mixture. Similarly to 125 µm droplets, the shape of the PDFs is nearly Gaussian close to the burner axis, but skewed towards low values of equivalence ratio at large radial distances. Figure 18 shows that the degree of non-stoichiometric reaction increases with radial distance, leading to fully non-stoichiometric combustion. Close to the centreline, η is between 47 % and 65 %, which is significantly lower than for 125 µm droplets. This suggests that 125 and 200 µm droplets lead to different combustion regimes, which will be discussed in the next section. This finding is also supported by some yellowish flames observed around the stream of droplets in the fuel vapour flame for 200 µm droplets Droplet Group Combustion regime This section assesses the correlation between the results in the second flame generated by the fuel vapour and droplet combustion regime for 125 and 200 µm droplets. Combustion of an isolated droplet is classically associated with diffusion regime, because the fuel vapour forms a region of stoichiometric reaction around each droplet. By contrast, droplets burning in a cloud may lead to partially-premixed combustion depending on air entrainment within the cloud and subsequent mixing with fuel vapour. The notion of droplets burning as a cloud rather than individually is known in the literature as Group Combustion which was first introduced by Suzuki and Chiu [57]. The generic term of Group Combustion comprises different types of droplet burning regimes, which are identified from the group combustion number G and the inter-droplet distance d defined as [58]: G n( Re ρd Sc ) 4πλr C 1 / 2 1 / 3 d 2 = R b (4) p 19

21 d r = 1 4π (5) 3 d 3 1/ 3 (n rd ) Where Re and Sc are the Reynolds and the Schmidt numbers, ρ the density (kg.m -3 ), λ the thermal conductivity (W.m -1.K -1 ), C p the heat capacity at constant pressure (J.kg -1.K -1 ) and D the diffusion coefficient (m 2.s) of the gas surrounding a droplet in the cloud. r d and R b are respectively the droplet radius and the radius of the droplet cloud, n is the droplet number density defined as: n b = (6) 4π 3 R b 3 N where N b is the number of droplets present in the cloud. Assuming Le=1 and neglecting convective effects on droplet evaporation, G reduces to: G 2 = 4π (7) nr d R b In the current experiments, the cloud radius R b is taken as half the length of the fuel vapour flame (determined from images recorded with a camera), which is considered as representative of droplet group combustion here (see Fig. 19). Considering L f as the length of the fuel vapour flame, the time for a droplet to travel vertically from the top to the bottom of the flame is defined by T t =L f /V d, where V d is the mean droplet velocity averaged over the flame length (obtained from PDA measurements). As a result, the number of droplets injected into the burner during this period of time is N t =N T t, where N is the number of droplets injected per second by the monodisperse droplet generator into the burner (N=20000 s -1 and N=10000 s -1 for 125 and 200 µm droplets respectively). 200 μm droplets only move downwards and therefore N b =N t. By contrast, 125 μm droplets reverse their motion and move both upwards and downwards in the burning cloud, which increases the number density of the cloud, as quantified by PDA measurements. As a result, N b contains contributions from droplets moving in both directions, and, therefore, N b >N t. Table 3 summarises values of the different parameters for both droplet sizes which exhibit a different group combustion number G. Following the terminology of Candel et al. [59], results suggest that 125 μm droplets burn in the Critical Group Combustion regime (CGC) and 200 μm droplets in the Internal Group Combustion regime (IGC), Fig. 20. IGC implies droplets burning in a cloud which is surrounded by single droplet burning. In the current experiments, combustion of isolated droplets may be responsible for some 20

22 yellowish flames observed around the stream of 200 μm droplets, albeit these are rare events. This might explain the reduced degree of non-stoichiometric reaction in the central region close to the main droplet stream. On the other hand, CGC implies only droplets burning in a cloud, which justifies the measurements associated with increased lean-premixed combustion and the blue colour of the second flame with 125 μm droplets. The inter-droplet distance d is smaller for 125 µm droplets than for 200 µm droplets, which also justifies increased droplet cloud combustion. Therefore, the values obtained for the droplet group combustion number are consistent with the measured degree of non-stoichiometric reaction and the local equivalence ratio of the reacting mixture. An implied suggestion from the group combustion analysis is that droplet evaporation in clouds may limit the ability of fuel vapour to ignite fast and form non-premixed stoichiometric reaction around individual droplets, which assists premixing of fuel vapour and air prior to ignition, leading to reduced NOx emissions. 4. Concluding remarks Monodisperse ethanol droplets with diameter of 125 and 200 µm were injected in a premixed counterflow burner operating with natural gas / air mixture on one side and a hot air stream on the other side. 200 µm droplets crossed the stagnation plane and entered the opposite hot air jet where they evaporated, whereas 125 µm droplets reversed their motion and oscillated around the stagnation plane, leading to increased droplet residence time in hot regions. The ethanol vapour generated below the stagnation plane mixed with the surrounding heated air and led to ignition of a fuel vapour flame below the natural gas flat flame for both droplet diameters. The findings are as follows: The flame was intermittent and the probability of flame presence was maximum close to the burner axis, where most of the droplets were present. Flame probability was higher with 200 µm droplets. 125 µm droplets led to ignition of a fuel vapour flame with equivalence ratio around 0.8, whereas 200 µm droplets were responsible for a stoichiometric flame. Near the burner axis, the degree of non-stoichiometric reaction of the fuel vapour flame was in excess of 80 % with 125 µm droplets, suggesting mainly lean-premixed combustion, whereas it was about 50 % for 200 µm droplets, indicating occurrence of stoichiometric reaction. 21

23 Group combustion analysis suggests that 125 µm droplets burned only in a cloud ( Critical Group Combustion regime), while cloud burning together with single droplet combustion could occur for 200 µm droplets ( Internal Group Combustion regime). 22

24 References [1] Williams A., Combustion of droplets of liquid fuels: a review, Combustion and Flame, Vol. 21, 1-31 (1973). [2] Correa S.M., A review of NO x formation under gas-turbine combustion conditions, Combust. Sci. and Tech., Vol. 87, (1992). [3] Chen G., Gomez A., Counterflow diffusion flames of quasi-monodisperse electrostatic sprays, 24 th Int. Symp. on Comb., (1992). [4] Li S.C., Libby P.A., Williams F.A., Spray structure in counterflowing streams with and without a flame, Combustion and Flame, Vol. 94, (1993). [5] Li S.C., Spray stagnation flames, Prog. Energy Combust. Sci., Vol. 23, (1997). [6] Mercier X., Orain M., Grisch F., Investigation of droplet combustion in strained counterflow diffusion flames using planar laser-induced fluorescence, Applied Physics B, Vol. 88, (2007). [7] Hayashi J., Watanabe H., Kurose R., Akamatsu F., Effects of fuel droplet size on soot formation in spray flames formed in a laminar counterflow, Combustion and Flame, Vol. 158, (2011). [8] Greenberg J.B., Albagli D., Tambour Y., An opposed jet quasi-monodisperse spray diffusion flame, Combust. Sci. and Tech., Vol. 50, (1986). [9] Continillo G., Sirignano W.A., Counterflow spray combustion modelling, Combustion and Flame, Vol. 81, (1990). [10] Gutheil E., Sirignano W.A., Counterflow spray combustion modelling with detailed transport and detailed chemistry, Combustion and Flame, Vol. 113, (1998). [11] Nakamura M., Akamatsu F., Kurose R., Katsuki M., Combustion mechanism of liquid fuel spray in a gaseous flame, Phys. Fluids, Vol. 17, (2005). [12] Watanabe H., Kurose R., Hwang S.M., Akamatsu F., Characteristics of flamelets in spray flames formed in a laminar counterflow, Combustion and Flame, Vol. 148, (2007). [13] Lacas F., Darabiha N., Versavel P., Rolon J.C., Candel S., Influence of droplet number density on the structure of strained laminar spray flames, 24 th Int. Symp. on Comb., (1992). [14] Darabiha N., Lacas F., Rolon J.C., Candel S., Laminar counterflow spray diffusion flames: a comparison between experimental results and complex chemistry calculations, Combustion and Flame, Vol. 95, (1993). 23

25 [15] Li S.C., Williams F.A., Experimental and numerical studies of two-stage methanol flames, 26 th Int. Symp. on Comb., (1996). [16] Li S.C., Libby P.A., Williams F.A., Experimental and theoretical studies of counterflow spray diffusion flames, 24 th Int. Symp. on Comb., (1992). [17] Mikami M., Miyamoto S., Kojima N., Counterflow diffusion flame with polydisperse sprays, 29 th Int. Symp. on Comb., (2002). [18] Mikami M., Mizuta Y., Tsuchida Y., Kojima N., Flame structure and stabilization of lean-premixed sprays in a counterflow with low-volatility fuel, 32 nd Int. Symp. on Comb., (2009). [19] Greenberg J.B., Sarig N., An analysis of multiple flames in counterflow spray combustion, Combustion and Flame, Vol. 104, (1996). [20] Puri I.K., Libby P.A., Droplet behaviour in counterflowing streams, Combust. Sci. and Tech., Vol. 66, (1989). [21] Chen N.H., Rogg B., Bray K.N.C., Modelling laminar two-phase counterflow flames with detailed chemistry transport, 24 th Int. Symp. on Comb., (1992). [22] Hollmann C., Gutheil E., Flamelet-modeling of turbulent spray diffusion flames based on a laminar spray flame library, Combust. Sci. and Tech., Vol. 135, (1998). [23] Gutheil E., Structure and extinction of laminar ethanol-air spray flames, Combustion Theory and Modelling, Vol. 5, (2001). [24] Gutheil E., Multiple solutions for structures of laminar counterflow spray flames, Progress in Computational Fluid Dynamics, an International Journal, Vol. 5, (2005). [25] Olguin H., Gutheil E., Influence of evaporation on spray flamelet structures, Combustion and Flame, Vol. 161, (2014). [26] Orain M., Experiments with gas and liquid-fuelled flames, PhD Thesis, University of London, UK (2002). [27] Mastorakos E., Taylor A.M.K.P., Whitelaw J.H., Extinction and temperature characteristics of turbulent counterflow diffusion flames with partial premixing, Combustion and Flame, Vol. 91, (1992). [28] Mastorakos E., Turbulent combustion in opposed jet flows, PhD thesis, University of London, UK (1994). 24

26 [29] Hardalupas Y., Taylor A.M.K.P., Wilkins J.H., Experimental investigation of sub-millimetre droplet impingement onto spherical surfaces, Int. J. Heat and Fluid Flow, Vol. 20, (1999). [30] Berglund R., Liu B.B., Generation of monodispersed aerosol standards, Envir. Sci. Technol., Vol. 7, (1973). [31] Scheffer R.W., Namazian M., Kelly J., OH, CH and CH 4 concentration measurements in a lifted turbulent-jet flame, 23 rd Int. Symp. on Comb., (1990). [32] Kojima J., Ikeda Y., Nakajima T., Detailed distributions of OH *, CH * and reaction zone of laminar premixed methane/air flames, AIAA (2000). * C 2 chemiluminescence in the [33] Yip B., Samaniego J.M., Direct C 2 radical imaging in combustion instabilities, Combust. Sci. and Tech., Vol. 84, (1992). [34] Lawn C.J., Distribution of instantaneous heat release by the cross-correlation of chemiluminescent emissions, Combustion and Flame, Vol. 123, (2000). [35] Chou T., Patterson D.J., In-cylinder measurement of mixture maldistribution in L-head engine, Combustion and Flame, Vol. 101, (1995). [36] Sandrowitz A.K., Cooke J.M., Glumac N.G., Flame emission spectroscopy for equivalence ratio monitoring, Appl. Spect., Vol. 52, (1998). [37] Kojima J., Ikeda Y., Nakajima T., Spatially resolved measurement of OH *, CH * and C * 2 chemiluminescence in the reaction zone of laminar methane/air premixed flames, 28 th Int. Symp. on Comb., (2000). [38] Muruganandam T.M., Kim B.H., Morell M.R., Nori V., Patel M., Romig B.W., Seitzman J.M., Optical equivalence ratio sensors for gas turbine combustors, 30 th Int. Symp. on Comb., (2005). [39] Cheng T.S., Wu C.Y., Li Y.H., Chao Y.C., Chemiluminescence measurements of local equivalence ratio in a partially premixed flame, Combust. Sci. and Tech., Vol. 178, (2006). [40] Hardalupas Y., Orain M., Local measurements of the time-dependent heat release rate and equivalence ratio using chemiluminescent emission from a flame, Combustion and Flame, Vol. 139, (2004). [41] Orain M., Hardalupas Y., Effect of fuel type on equivalence ratio measurements using chemiluminescence in premixed flames, Comptes Rendus Mécaniques, Vol. 338, (2010). 25

27 [42] Akamatsu F., Wakabayashi T., Tsushima S., Katsuki M., Mizutani Y., Ikeda Y., Kawahara N., Nakajima T., The development of a light-collecting probe with high spatial resolution applicable to randomly fluctuating combustion fields, Meas. Sci. Technol., Vol. 10, (1999). [43] Hardalupas Y., Orain M., Panoutsos C.S., Taylor A.M.K.P., Olofsson J., Seyfried H., Richter M., Hult J., Aldén M., Hermann F., Klingmann J., Chemiluminescence sensor for local equivalence ratio of reacting mixtures of fuel and air (FLAMESEEK), Applied Thermal Engineering, Vol. 24, (2004). [44] Hardalupas Y., Panoutsos C.S., Taylor A.M.K.P., Spatial resolution of a chemiluminescence sensor in a model gas turbine combustor, AIAA (2005). [45] Hardalupas Y., Panoutsos C., Taylor A.M.K.P., Spatial resolution of a chemiluminescence sensor for local heat release rate and air-fuel ratio measurements in a model gas turbine combustor, Exp. in Fluids, Vol. 49, (2010). [46] Orain M., Hardalupas Y., Measurements of local mixture fraction of reacting mixture in swirl-stabilised natural gas fuelled burners, Applied Physics B, Vol. 105, (2011). [47] Panoutsos C.S., Hardalupas Y., Taylor A.M.K.P., Numerical evaluation of equivalence ratio measurement using OH* and CH* chemiluminescence in premixed and non-premixed methane air flames, Combustion and Flame, Vol. 156, (2009). [48] A.E. Lutz, R.J. Kee, A FORTRAN program for computing opposed-flow diffusion flames, Sandia Report No. SAND , Sandia National Laboratories (1996). [49] R.J. Kee, F.M. Rupley, J.A. Miller, CHEMKIN-II: A Fortran chemical kinetics package for the analysis of gas phase chemical kinetics, Sandia Report No. SAND , Sandia National Laboratories (1989). [50] Marinov N., A detailed chemical kinetic model for high temperature ethanol oxidation, Int. J. Chem. Kin., Vol. 31, (1999). [51] Benvenutti L.H., Marques C.S.T., Bertran C.A., Chemiluminescent emission data for kinetic modeling of ethanol combustion, Combust. Sci. and Tech., Vol. 177, 1-26 (2005). [52] Hardalupas Y., Laker J.R., Description of the Thermofluids Section Model 3 Phase Doppler Counter, Thermofluids Section, Mechanical Engineering Department, Imperial College, London, UK: TF/93/15 (1993). [53] Hardalupas Y., Taylor A.M.K.P., Phase validation criteria of size measurements for the phase Doppler technique, Exp. in Fluids, Vol. 17, (1994). 26

28 [54] Hardalupas Y., Taylor A.M.K.P., On the measurement of particle concentration near a stagnation point, Exp. in Fluids, Vol. 8, (1989). [55] Hardalupas Y., Taylor A.M.K.P., Whitelaw J.H., Mass flux, mass fraction and concentration of liquid fuel in a swirl-stabilised flame, Int. J. Multiphase Flow, Vol. 20, (1994). [56] Lewis B., Von Elbe G., Combustion, Flames and Explosions of Gases, (3 rd ed), Academic Press, London, (1987). [57] Suzuki T., Chiu H.H., Multi-droplet combustion of liquid propellants, Proceeding of the 9 th international symposium on space technology and science, (1971). [58] Chiu H.H., Liu T.M., Group combustion of liquid droplets, Combust. Sci. and Tech., Vol. 17, (1977). [59] Candel S., Lacas F., Darabiha N., Rolon J.C., Group combustion in spray flames, Invited lecture, International Conference on Multiphase Flow (ICMF 1998), Lyon, France (1998). 27

29 TABLES Droplets moving downwards Droplets moving upwards z/h D 10 (µm) V d (m/s) Table 1. Evolution of droplet arithmetic mean diameter D 10 and droplet velocity V d at different axial locations on the axis (r/r=0) of the counterflow burner, for equivalence ratio of natural gas premixed flame of Φ 0 =0.7 and injection of 125 µm droplets. Droplets moving downwards z/h D 10 (µm) V d (m/s) Table 2. Evolution of droplet arithmetic mean diameter D 10 and droplet velocity V d at different axial locations on the axis (r/r=0) of the counterflow burner, for equivalence ratio of natural gas premixed flame of Φ 0 =0.7 and injection of 200 µm droplets. D d (µm) N (s -1 ) L f (mm) T t (ms) N t R b (mm) N b d/r d G Table 3. Characteristics of droplet Group Combustion regime with the injection of 125 and 200 µm droplets. 28

30 FIGURES Figure 1. Sketch of the counterflow burner fuelled with natural gas. The exit of the monodisperse droplet generator is located 120 mm from the stagnation plane. Figure 2. Optical arrangement for the local measurement of OH*, CH* and C 2 * chemiluminescent intensity (see ref [40] for details). 29

31 Figure 3. Position of the Cassegrain optics with respect to the droplet flame and mask used in front of the optics. The arrow indicates the direction the Cassegrain is traversed when performing measurements of radial profiles of chemiluminescent intensity. Figure 4. Dependence of the OH*/CH* chemiluminescent intensity ratio upon equivalence ratio Φ 0 in premixed and non-premixed prevaporised ethanol-fuelled counterflow flames for a bulk flow velocity V 0 =2 m/s at the exit of the ducts of the counterflow burner (adapted from ref [41]). 30

32 (a) (b) Figure 5. Photographs of droplet trajectories in the counterflow burner without combustion, for the injection of 125 µm (a) and 200 µm (b) droplets. (a) (b) Figure 6. Probability density function of droplet velocity at different axial locations on the burner axis (r/r=0) for the injection of 200 µm (a) and 125 µm (b) droplets in the counterflow burner, and at different radial locations on the stagnation plane (z/h=0) for the injection of 125 µm droplets (c). Measurements performed in an isothermal flow at 293 K. (c) 31

33 (a) (b) Figure 7. Droplet size distributions in the counterflow burner at different axial locations on the burner axis (r/r=0) for 200 µm (a) and 125 µm (b) droplets moving downwards, and 125 µm droplets moving upwards (c). The equivalence ratio of natural gas premixed flame is Φ 0 =0.7. (c) Figure 8. Radial profile of droplet volume mean diameter D 30 at the stagnation plane of the counterflow burner (z/h=0) for droplets moving downwards and upwards; for the injection of 125 μm droplets and an equivalence ratio of natural gas premixed flame of Φ 0 =

34 Figure 9. Radial profile of liquid volume flux at the stagnation plane of the counterflow burner (z/h=0) for droplets moving downwards and upwards; for the injection of 125 μm droplets and an equivalence ratio of natural gas premixed flame of Φ 0 =0.7. Figure 10. Schematic of droplet trajectories in the counterflow burner. 33

35 (a) (b) (c) Figure 11. Photographs of the counterflow flame (equivalence ratio of the natural gas premixed flame is Φ 0 =0.7) without droplet injection (a) and with injection of 125 µm (b) and 200 µm (c) droplets. Figure 12. Instantaneous CCD images of CH* chemiluminescence emitted from the flame with injection of 125 μm droplets and an equivalence ratio of natural gas premixed flame of Φ 0 =0.7, showing that the second flame due to droplets is not always present. 34

36 (a) (b) Figure 13. Radial profile of probability of presence of the fuel vapour flame (a) and local mean equivalence ratio of the reacting mixture (b); for the injection of 125 μm droplets. (a) (b) Figure 14. Probability density function of local equivalence ratio of the reacting mixture of the fuel vapour flame at different radial locations for two axial positions z/h=-0.10 (a) and z/h=-0.34 (b); for the injection of 125 μm droplets. Figure 15. Radial profile of degree of non-stoichiometric reaction of the reacting mixture at different axial locations in the fuel vapour flame; for the injection of 125 μm droplets. 35

37 (a) (b) Figure 16. Radial profile of probability of presence of the fuel vapour flame (a) and local mean equivalence ratio of the reacting mixture (b); for the injection of 200 μm droplets. (a) (b) Figure 17. Probability density function of local equivalence ratio of the reacting mixture of the fuel vapour flame at different radial locations for two axial positions z/h=-0.10 (a) and z/h=-0.34 (b); for the injection of 200 μm droplets. Figure 18. Radial profile of degree of non-stoichiometric reaction of the reacting mixture at different axial locations in the fuel vapour flame; for the injection of 200 μm droplets. 36

38 Figure 19. Schematic of the flame structure in the counterflow burner. Figure 20. Spray combustion diagram (adapted from ref [59]). 37