Assisted atomization of a gas-liquid jet: effect of the volumic gas fraction in the inner jet on the spray characteristics

Transcription

1 ILASS Americas 27th Annual Conference on Liquid Atomization and Spray Systems, Raleigh, NC, May 2015 Assisted atomization of a gas-liquid jet: effect of the volumic gas fraction in the inner jet on the spray characteristics J.-C. Guillard 1,2,3, A. Cartellier 1,2 and J.-P. Matas 1,2 1 Université Grenoble Alpes, LEGI, F Grenoble, France 2 CNRS, LEGI, F Grenoble, France 3 CNES, Direction des lanceurs, Paris, France Abstract We study the assisted atomization of a two-phase jet. This situation is notably encountered during the start-up phase of cryotechnic engines where the gas flow rate fraction β of the inner jet decreases from unity (pure gas) down to zero (pure liquid). The key questions concern how the spray characteristics vary with β and/or the flow configuration in the inner jet. We have carried out air-water experiments for β ranging from 0 to 1 and for external gas velocities Ug ext from 20 to 100 m/s. The physical mechanisms leading to atomization happen to be significantly altered when we vary β. First, the break-up length decreases with β. Second, the drop size at short distance from injection, which is mainly controlled by the external gas velocity at β=0, becomes less and less sensitive to Ug ext as β increases. This trend indicates that new mechanisms control the drop production. An alternative process may be the following: as β increases, the inner two-phase jet configuration evolves as well as the scales of the liquid bridges change and may control the size of the drops produced by atomization. This scenario is supported by experiments showing that drop sizes are rather well correlated to the dimension of liquid bridges. In addition, the flapping phenomenon disappears with increasing β, in agreement with the initial break-up of the jet. Corresponding Author: jeanchristophe.guillard@legi.grenoble-inp.fr

2 Introduction The assisted atomization of a liquid jet by a high-speed co-flowing gas stream has been extensively studied [1 5].However, the assisted atomization of a two-phase gas-liquid jet has received much less attention [6 8]. Atomization of a single phase liquid jet by a high speed co-current gaseous annulus occurs in many industrial applications. One of them is the atomization of a liquid oxygen jet by a gaseous hydrogen annulus in cryotechnic rocket engines. During the start-up phase, the central jet progressively shifts from pure gas to pure liquid. In between, the gas flow rate fraction defined as : β = Q gas Q gas + Q liq (1) with Q gas and Q liq respectively the gas and liquid flow rates, continuously evolves from 1 to 0. The open questions are: 1. what is the influence of β on the breakup mechanisms and on drop size? 2. is there any influence of the flow configuration on the atomization process? Let us briefly recall what is known for a single liquid phase jet assisted atomization. Droplets can be produced by different mechanisms in different locations as shown by figure 1. Figure 1. Spray for Ul sup = 0.86 m/s, Ug ext = 100 m/s and β = 0 Close to the injector nozzle the dominant mechanism is related with interfacial stripping due to the shear imposed by the external gas flow. It consists of a succession of instabilities of the gas-liquid interface. First, a Kelvin-Helmholtz instability occurs and give birth to axial waves [1, 2, 5, 9]. Then, a Rayleigh-Taylor type transverse instability destabilizes the KH wave crests and creates corrugations. These corrugations are elongated into ligaments under the gas current until they break into droplets. Numerous studies have focused on these instabilities in order to understand them and a few have been proposed to predict drop size [1, 3, 5, 9]. Only a fraction of the incoming liquid flux is stripped off the jet. The remaining liquid flux forms a distorted jet that experiences strong lateral displacements (fig. 1), similar to what is observed on liquid sheets [10,11]. That flapping motion is probably caused by non linear mechanisms such as gas vortices detached behind significant interface deformations, and it is often triggered by Kelvin-Helmholtz instabilities [12]. Such a mechanism controls the break-up length and, further downstream the injection, it produces large drops, which are typically 4 to 10 times larger than those due to stripping. Most observations indicate that drop size decreases with gas velocity, with typical trends close to the gas velocity to the power 1.5. Yet, model predictions of drop characteristics are lacking. In particular, there is not yet any definitive agreement on the scaling of the mean flapping frequency with flow parameters and injector geometry. Farther downstream, secondary atomization may play a role, either through collisions, or breakup due to aerodynamic forces or turbulence [13, 14]. We therefore undertook experiments using air and water in thermodynamic equilibrium. Note that the bubbles size was never smaller than 1 mm. By covering the range β = 0 1, we swept all two-phase flow configurations from bubbly to annular flow. 1 Experimental set-up and method 1.1 Experimental facility In order to atomize a two-phase flow jet, we choose to use the same injector that was extensively used to study the assisted atomization of a pure liquid jet [1, 2]. Cross-section views of this injector are presented figure 2. The two-phase inner jet is a 7.6 mm diameter axisymmetric round jet. It is atomized by a 1.7 mm thick annulus of gas. A 0.2 mm thickness lip separates both flows: the lip is very thin to avoid flow disturbances and ensure quasi parallel streams at the injector exit. The injector axis is a vertical with the jet flowing downward. The two-phase flow in the inner jet is obtained with a liquid-gas mixer plugged above the injector (figure 3). A 2.5 mm internal diameter tube is passed into another larger tube (8 mm internal diameter). The 2.5 mm tube is used to bring the inner gas into the 2

3 1.2 Experiment plan Control parameters are the superficial liquid velocity, Ul sup, the external gas velocity, Ug ext and the internal superficial gas velocity Ugint sup. Note Ugint sup Ugint sup+ul sup. that β defined by 1 also writes β = For the present experiments, the dynamic pressure ratio M has been kept to a fixed value, M = 16. M = ρ g.ugext 2 ρ l.ulsup 2, (2) Figure 2. Injector details liquid flowing in the 8 mm diameter tube. The inner gas passes in the water flow through 0.4 mm holes which are distributed all along the 2.5 mm tube. The total mixer length is 792 mm. with ρ g and ρ l are respectively the density of gas and of liquid, namely air and water. Hence, in our experiments, a change in Ug ext induces a change in Ul sup. Moreover the experiments were achieved at given β values. For a given β, once Ul sup is set, Ugint sup is uniquely determined. We have used five sets of parameters {Ul sup ; Ug ext } for which we made β vary from zero to the maximal value that the experimental set-up allowed. Ul sup varied from 0.17 to 0.86 m/s and Ug ext from 20 to 100 m/s. According to the range of Ul sup and β values, we were able to generate flow configurations ranging from bubbly to annular flows. A map of the configurations of the inner jet was established and is shown on figure 4. Several observations will be use- Figure 4. Inner jet flow configurations map - M=16 Figure 3. Mixer details To supply the experimentation in water we used an overflowing water tank to ensure a constant liquid pressure level. External and internal gases come from the laboratory seven bars compressed air circuit. For the inner gas a pressure regulator is used to avoid flow rate fluctuations. That pressure has to be finely regulated because minor fluctuations could lead to relatively strong variations in the inner gas flow rate when the latter is very small. ful for the discussion: the evolution of the inner jet structure as a function of β depends on the liquid superficial velocity. For the two lower Ul sup, 0.17 and 0.26 m/s, we do not observe neither bubbly or churn configurations. Instead, long slugs arise that develop into an annular flow as the inner gas is increased, since the size of liquid bridges between slug decreases. For Ul sup = 0.36 m/s, the flow structure starts as a slug flow for weak β and becomes then churn and annular as β increases. 3

4 For the two highest liquid velocities namely, 0.61 and 0.86 m/s, we observe bubbly flows at low β, which turn into churn flows for larger β. Above about β = 0.9, the inner jet structure remains annular independently of {U lsup ; U gext }. Thus, all flow configurations can be generated except mist flow. 2 Jet structure and break-up length Images have been taken with a high speed camera Miro M310 at a sample rate of 5100 images per second. Figure 5 shows the evolution of the spray for a fixed set {U lsup = 0.26 m/s; U gext = 30 m/s} while β increases. We observe a strong evolution of the spray as β increases: At β = 0, we observe small amplitude shear instability waves as well as the apparition of digitations probably derived from a RayleighTaylor instability (see a on figure 5). At β = 0.6, the wave amplitude becomes much larger (see b) and we observe a pinching of the water jet(c). On this set of images, the jet remains connected to injection downstream the pinching. In addition, large liquid lumps are present along the jet axis(d): they experience limited radial displacements compared to the case β = 0 indicating that flapping is damped by the presence of gas in the inner jet. We notice also the apparition of a gas pocket trapped in the liquid jet just at the exit of the nozzle injector, which is not well visible on single images but that is easily detectable on videos as soon as we added inner gas. Figure 5. Sequences of spray evolution with β- Flow conditions:{u lsup = 0.26 m/s; U gext = 30 m/s} - Top images set : β = 0 - Middle images set : β = Bottom images set : β = 0.95 At β = 0.95, wave amplitudes are still very large. The water jet can be drilled by the inner gas (e) and it breaks closer to the nozzle injector at the pinching location (f). Thus, the detached liquid lump, which was initially a wave, flows downstream as a disconnected liquid mass (g). When this mechanism is observed, the jet break up is usually associated with a strong pulsation. defined as the axial extent of the liquid tongue continuously connected to the injector. We filmed the spray with a sample rate equal to 60 images per second in order to collect uncorrelated events. For each measuring point we took 2000 images. Lb has been measured on each image using a Matlab code image processing. The images are recorded with 8-bit grey scale pixels. First of all, we remove the image background in order to isolate the spray. Then spray edges are detected with the Matlab Canny filter. We add the edges twice to the image to make them more pronounced. Finally a binary version of the image is created with the Matlab function graythresh. These three steps are illustrated on figure 7. The break-up length is calculated as the vertical distance between the end of the spray and the injector nozzle exit. Figure 6 shows the spray evolution as a function of increasing {U lsup ; U gext }, for a fixed β. On the top set, for β = 0 one can observe that for each set {Ulsup ;Ugext } flapping is present. For the set {U lsup = 0.17 m/s; U gext = 20 m/s} we do not observe any stripping close to the nozzle injector, only shear-instability waves appear. On the bottom set, for β = 0.6, flapping is not visible anymore. Besides a pinching is present at all velocities. We now discuss the break-up length Lb, which is 4

5 Figure 7. MATLAB image processing sample. Left: Raw image - Middle: Image without background - Right: Binary image to one HL. However, the maximum break-up length LbM AX remains constant, typically about 9HL from β = 0 to β = 0.8. For β > 0.4, the number of events of short Lb increases, while the minimum value LbM IN itself decreases, and reaches LbM IN < HL for β = 0.8. Moreover we observe for each {U lsup ; U gext } an increase of the histogram width when β is increased from 0.2 to 0.9. It demonstrates the large variability of Lb in this range of β. Added to this variability in Lb, we have observed a spray pulsation which begins softly at β (depending on {U lsup ; U gext }) and becomes very strong from β 0.8 to β Above β = 0.95, the pulsation phenomenon subsides. As β rises to values larger than 0.9, the histogram becomes thinner and thinner and shifts toward shorter Lb values, in the range 1HL 6HL. However, events where Lb is larger than 3HL are relatively rare. At this point, the inner jet is a pure annular flow. Figure 9 shows Lb /HL versus β for Figure 6. Spray evolution with {U lsup ; U gext } From left to right: {U lsup = 0.17 m/s; U gext = 20 m/s} {U lsup = 0.36 m/s; U gext = 43 m/s} {U lsup = 0.86 m/s; U gext = 100 m/s}- Top images set : β = 0 - Bottom images set : β = Different magnifications, the diameter of the nozzle injector exit is the same HL = 7.6 mm. This value was averaged over 2000 images. This method was called automatic. For high values of β, this treatment was not very effective: due to the weaker contrast, the determination of interface position was less accurate. We used therefore a more basic method, called manual, which was also used to verify the validity of the automatic one. In the manual way, we directly measured Lb on the pictures for a set of 500 images for each flow condition. A set of 500 images were determined to be enough to obtain a good convergence of Lb. The convergence was verified for each condition. When flow conditions allowed to use both methods, results between them showed a good agreement. Examples of Lb histograms are presented on figure 8 where {U lsup ; U gext} is fixed and β evolves from zero to One can notice that as soon as the inner gas is added (β = 0.2) we observe events where the jet breaks very close to the nozzle injector, down Figure 8. Lb /HL histograms for different β - Flow conditions {(U lsup = 0.26 m/s; U gext = 30 m/s} Number of events by histogram =

6 each {Ul sup ; Ug ext } set. For every cases, L b /H L monotonously decreases with increasing β. L b /H L is strongly reduced as soon as inner gas is added, except for the weaker {Ul sup ; Ug ext } couple. Then L b /H L softly decreases with increasing β and for β > 0.8, it strongly decreases until the maximum β value. As already pointed out in the discussion of Figure 10. Optical probe tip refractive index of the medium surrounding its tip. A characteristic signal obtained when a drop hits the probe is shown on figure 11. When the probe tip is Figure 9. L b /H L versus β for different {Ul sup ; Ug ext } values. figures 5 and 6, the inner gas flow introduces new break up mechanisms, which become dominant as β increases. In particular, the jet flapping, which is strongly present at β = 0, progressively disappears when β increases. The impact of this change in mechanisms can be seen in the histograms of figure 8. At low β the tearing of the envelope of the trapped gas pocket induces a very short class of L b on the histogram that does not exist for a single phase liquid jet. Then, for intermediate β, breakup occurs at the pinching location, a spray pulsation appears: the fact that different mechanisms can contribute to break-up results in an increase of the width of the histograms. Finally, at very high values of β, a liquid annular jet remains opened and it is atomized by both inner and outer gas flows. At this point, L b remains very short and the pulsatile character subsides, the histogram becomes thinner and concentrated around low values of L b. To conclude, we see a clear and strong effect of β on the break-up length. Changes in the break-up mechanisms occur. L b is not much sensitive to Ug ext (Figure 9). This is the first evidence of a clear impact of the two-phase character of the inner jet on the assisted atomization process. 3 Drop size In order to measure droplet size we used a conical monofiber optical probe from A2 Photonic Sensors (Fig.10). The probe response is an image of the Figure 11. Optical probe typical response signal in the air, the signal has a high voltage, when the medium is liquid the voltage is lower. Two durations are estimated from the signal processing [15], the time spent by the probe in the drop, i.e. the residence time, T R, and the dewetting time, i.e. the rise time, T M. The droplet velocity is then derived from: V Drop = L s T b M, (3) where b is set to -1 and L s, the probe sensitive length, is determined during its calibration. L s equals to 20 µm in our case. The residence time is then used to derive the droplet chord as: C Drop = V Drop T R, (4) By post-processing, one can also obtain the size distribution (assuming spherical droplets), various moments such as mean diameter, d 10, Sauter mean diameter, d 32, and others spray characteristics such as the drop volumetric flux and the number flux [2, 15, 16]. In the present work, we will focus on the mean chord size and we will use C 10, knowing that, for spheres, the Sauter mean diameter can be directly deduced from the mean chord using d C 10 [16]. Drop sizes have been investigated at two locations : Close to the nozzle injector to analyze droplets formed by stripping. In the following this point 6

7 will be called the stripping point. It is located 11 m downsteam the injector and its radial displacement is 4 mm from the axis (under the separator liquid jet/outer gas lip). At a distance z = L b + L bst D from the nozzle injector on the axis, a position where drops arising from jet flapping are expected [17]. 3.1 Inner gas effect on drop size measured at the stripping point Normalized chord pdfs We investigate first the evolution of the pdfs of chords normalized by C 10 for varying β (Figure 12), and for two sets of conditions {Ul sup ; Ug ext }. Note that the lower drop size that the probe can measure with a very good accuracy is L s /2. Drop sizes of the order of L s /3 can still be detected but with some bias on the statistics. This limitation affects the first point to the left of all pdfs presented. Hence, we will not take into account these points in the discussion. For all phasic velocities, we observe a pretty good collapse between chord pdfs, demonstrating thus that the size distributions are nearly insensitive to the inner gas flow rate fraction. The only exception is the case β = 0.99, that corresponds to a pure annular flow configuration with an internal gas velocity about 36 m/s that competes with the outer gas velocity. In this case, atomization arises by stripping on both sides of the liquid sheet. This configuration is very specific and will not be analyzed further in this paper. Let us now consider the influence of phasic velocities at fixed β. Figure 13 indicates that except for chords below C 10 /30, all curves are relatively well collapsed though, for larger velocities, pdfs are slightly shifted toward lower sizes. For chords less than C 10 /3 - C 10 /4 (depending on β), the pdf increases when velocities increase, except for {Ul sup = 0.17 m/s; Ug ext = 20 m/s}. For chords above C 10 /3 C 10 /4, this trend is reversed. Therefore, phasic velocities have a weak impact on pdf shape. Overall, normalized size pdfs at the so-called stripping point are marginally sensitive to both flow conditions and β. We therefore examine now the evolution of a single moment, namely the mean chord C 10, with flow conditions. Effect of β on C 10 Figure 14 shows the evolution of C 10 with Figure 12. Normalized Chord Pdf as a function of β for fixed {Ul sup ; Ug ext }, measured at the stripping point - Top : {Ul sup = 0.36 m/s; Ug ext = 43 m/s} - Bottom : {Ul sup = 0.86 m/s; Ug ext = 100 m/s} β for our different sets {Ul sup ; Ug ext }. The same trend is observable for all of them: until β = 0.9, C 10 softly decreases with β, down to 25 30% of its value at β = 0, except for {Ul sup = 0.17 m/s; Ug ext = 20 m/s}. Above β = 0.9, which corresponds to the transition towards an annular flow (Fig.4), the impact of β is stronger and C 10 is drastically reduced. Effect of Ug ext on C 10 In the monophasic case (β = 0), some studies have proposed models for the drop size dependency to Ug ext. In [13], Lasheras, Villermaux and Hopfinger give d 32 Ug n with 0.8<n<1.3. In [1] and [4], Marmottant et al., found d 10 /δ W e 1/3 δ where the Weber number is defined as: W e δ = ρ gas.δ.(ug ext U c ) 2, (5) σ with U c the waves convective velocity, U c = ρliq.u liqsup + ρ gaz.u gext ρ liq +ρ gaz, and δ the vorticity thickness in the gas at the exit of the injector. In this experiment, δ was varying as Ug 1/2 ext. Hence, they observed d Ugext. 1 Hong et al. in [5] found on the same injector geometry than ours d 32 /δ W e 1/2 δ. 7

8 Figure 14. C 10 as a function of β for fixed {Ul sup ; Ug ext }, measurements at the stripping point. considers only the largest gas flow rates. To sum up, the stripping point is a location Figure 13. Normalized chord pdf as a function of {Ul sup ; Ug ext } for β fixed measured at the stripping point - Top : β = 0 - Middle : β = Bottom : β = 0.95 Since δ Ug 1/2 ext, the dependency of mean drop size to Ug ext becomes in this case a power law with an exponent equal to 5/4. The key point in all these studies is that the mean drop size is governed by Ug ext and that it is sensitive to the injector internal design via δ. Are these conclusions still valid for a two-phase inner jet? Let us now analyze how β impacts the above dependencies. Figure 15, shows C 10 as a function of Ug ext. The curve corresponding to the single phase jet i.e. β = 0 case has a slope with approximately the same Ug ext power law than that found by Hong et al. [5]. As β increases, the slope decreases, indicating an attenuation of the C 10 dependency to Ug ext. The β = 0.97 case exhibits the weakest sensitivity of C 10 to the outer gas velocity, with a Ug ext power law exponent smaller than 3/4, and even less if one Figure 15. C 10 as a function of Ug ext for fixed β, measurements at the stripping point. where strong changes occur as β increases; C 10 can decrease down to 25 30% of its monophasic case value. The mean chord decreases steeply when the inner jet becomes annular. We believe that the reduction of C 10 dependency to Ug ext could be caused by the lesser role of the stripping mechanism when β increases. With increasing β, the liquid interface can be deformed by the passing of gas inclusions (slugs, bubbles, etc..). In addition we observe strong pulsations for larger β (see section 2). For β > 0.9 the outer gas meets a beating liquid sheet and not a round jet anymore, as at β = 0. Due to these perturbations of the interface between the jet and the outer gas, the conditions analyzed here depart from a co-flowing parallel flow situation that characterizes the atomization of a pure liquid jet. We discuss in the following section if at larger distances from the nozzle injector β has as much impact as it has at the stripping point. 8

9 3.2 Inner gas effect on drop size measured at L b + L bst D We now carry out measurements on the axis, but at a downstream position located at z = L b + L bst D where L bst D corresponds to the standard deviation of L b. For this location, we measure and plot the same quantities as previously at the stripping point, namely size pdfs and C 10 for varying values of β and of phasic velocities. Normalized chord pdfs Figure 16 shows two plots of the normalized chord pdfs at fixed {Ul sup ; Ug ext } for all β. We still Figure 17. Normalized chord pdf as a function of {Ul sup ; Ug ext } for fixed β measured at L b + L bst D - Top : β= 0 - Middle : β= Bottom : β= Figure 16. Normalized chord pdf as a function of β for fixed {Ul sup ; Ug ext } measured at L b + L bst D - Top : {Ul sup = 0.36 m/s; Ug ext = 43 m/s} - Bottom : {Ul sup = 0.86 m/s; Ug ext = 100 m/s}. observe a good collapse between curves except for the smallest chords i.e. less than C 10 /80. Note that the location of the measuring point L b + L bst D moves as a function of flow conditions: L b + L bst D varies between and m. Thus, the measuring point can be either in far field or in near field of the spray. Despite these changes, the collapse of pdfs is remarkable. On figure 17, normalized chord pdfs are plotted at a fixed β for all the {Ul sup ; Ug ext } sets. For chords smaller than C 10 /30, the curves do not collapse but the differences are not drastic and they tend to diminish as β increases. Above C 10 /30, curves are relatively well collapsed. As {Ul sup ; Ug ext } increases, more droplets with chords smaller than C 10 /30 and less droplets with chord above C 10 /30 are produced. Therefore, at L b + L bst D the size distribution around the mean drop size does not drastically change when β and flow conditions vary. This is exactly the same conclusion as the one made at the stripping point. We can now analyze the sensitivity of the mean chord values to them. Effect of β on C 10 Figure 18 shows the evolution of C 10 measured at L b + L bst D versus β for our different {Ul sup ; Ug ext } sets. 9

10 Two different trends are observed. For the two lower {Ul sup ; Ug ext } sets, C 10 increases until intermediate values of β ( ) and then decreases. For these two sets, we have noticed that no stripping occurs, only waves and digitizations are visible (see figure 5 and 6). At β = 0, flapping is the dominant phenomenon, displacing laterally the jet before it breaks. As soon as inner gas is added, flapping is strongly reduced. Figure 5 and 6 show how in the β = 0 case the spray impacts the probe transversely (due to flapping) whereas in the β = cases, the spray flows longitudinally on the probe (absence of flapping). Therefore, the probe detects larger liquid lumps at β = than at β = 0. This explains the increase of C 10 reported figure 18. Above β 0.6, the jet begins to pulse and it breaks at the pinching point: there is not anymore this kind of continuous liquid flow on the probe and drop sizes decrease. For the three higher {Ul sup ; Ug ext } sets, C 10 softly decreases until β = , then stabilizes and finally decreases strongly for β above 0.9 to For all {Ul sup ; Ug ext } sets, C 10 experiments a drastic decrease above β = , down to 85% compared to β = 0. To recap, the main trend is a mean chord decrease with increasing β except for lowest liquid velocities. This trend is similar to the one observed at the stripping point. Figure 18. C 10 as a function of β for a fixed {Ul sup ; Ug ext } measured at L b + L bst D Effect of Ug ext on C 10 Figure 19 shows that the dependency of C 10 to Ug ext at L b + L bst D decreases as Ug ext increases. Until Ug ext = 71 m/s, curves follow nearly the same trend (similar slopes) for β in the range , the impact of Ug ext is strong with C 10 Ugext. 1 The annular cases, β = , present a slightly steeper dependency, C 10 Ugext 1.8 at least until Ug ext 40 m/s. Note that 1 and 1.8 power laws are close to the ones observed for a pure liquid jet [12]. Above Ug ext 70 m/s, C 10 becomes quasi independent of Ug ext. To conclude, we have Figure 19. C 10 measured at L b + L bst D as a function of Ug ext for fixed β seen two distinct effects of β on C 10 depending on {Ul sup ; Ug ext }. In addition, above β = 0.9, we observed a strong decrease of C 10 for all phasic velocities sets, see figure 18. Whatever β the effect of Ug ext on C 10 follows the same trend, it is strong until Ug ext = 72 m/s and then reduces drastically. 3.3 Comparison between data at stripping point and data at L b + L bst D Normalized chord pdfs Figure 20 compares the evolution with β of the normalized chord pdfs at both measurement locations. At β = 0, the two pdfs are different. This is not surprising because for these flow conditions, the stripping point and the point at L b +L bst D are vertically relatively distant ( m). Indeed, in the single phase inner jet case, β = 0, drops arising from stripping are 2 to 4 times smaller than drops detected at the flapping location. As β increases, the pdfs become identical. At β = 0.97 the pdfs almost collapse on each other. This indicates that as β increases drop sizes become spatially homogeneous. Figure 21 shows the ratio of C 10 at L b + L bst D to C 10 at the stripping point when β varies. It shows that drops detected at L b + L bst D are 2.5 at 3.5 times larger than the ones detected at the stripping point (once again apart from the case {Ul sup = 0.17 m/s; Ug ext = 20 m/s} already discussed in section 3.2). As β increases, the ratio tends to 1, it confirms the C 10 spatial homogenization. The key question is which mechanisms lead to 10

11 Figure 21. C 10Lb +L bst D /C 10stripping of β as a function These measurements have been done without external gas. Figure 22 shows the evolution of the typical size of the liquid bridges present in the incoming jet, S frag, as a function of β. The reference size at β = 0 is arbitrarily fixed to the inner jet diameter, i.e. H L = m. As soon as gas is added to the inner jet, the size of liquid bridges becomes much smaller than H L (except for Ul sup = 0.86 m/s for which, at β = 0.2, the flow structure is a bubbly flow with small or dispersed bubbles). At β= 0.4, the size has approximately lost one order of magnitude compared with the jet diameter, and it is in the range µm. Above β = 0.9, the size of liquid bridges strongly decreases down to about 100µm. Let us now compare these length scales Figure 20. Normalized Chord pdf at stripping point and L b + L bst D - Fixed flow conditions: {Ul sup = 0.61 m/s;ug ext = 72 m/s} - Top : β= 0 - Middle : β= Bottom : β= 0.97 this phenomenon. 4 Relationship between drop sizes and inner flow length scales Concerning mechanisms, both the similarity of pdfs and the spatial homogenization of the mean drop size we have just discussed may indicate that the atomization process is strongly connected to the flow configuration in the inner jet. A plausible assumption could be that drop sizes are controlled by or strongly correlated with the liquid length scales already present in the incoming jet. In order to test this assumption we put the optical probe inside the inner jet, just upstream the injector exit and on the axis, to quantify the spatial extent of the liquid bridges between gas inclusions. Figure 22. Inner jet sizes of the liquid bridges, S frag, as a function of β. with the mean drop size. Figure 23 shows, on the top, the ratio C 10Stripping to S frag as a function of Ug ext. The figure below shows the ratio C 10Lb +L bst D to S frag again as a function of Ug ext. We notice that, on both graphs, the data are mostly concentrated in narrow bands, delimited by dotted lines in figure 23. For the stripping point, the ratio is mostly in 11

12 Figure 23. top: C 10Stripping /S frag as a function of Ug ext - Bottom: C 10Lb +L bst D /S frag as a function of Ug ext the range : the droplets detected at the stripping point are therefore not less than five times smaller than S frag and at most of the same size. For the L b + L bst D point, the ratio is in the range : droplets detected at L b +L bst D are not less than three times smaller than S frag and at most 1.5 times larger. In both cases, all data within the bands shown figure 23 are gathered around a ratio close to unity, C 10 S frag : this confirms the idea that drop size are directly connected to and possibly controlled by the size of liquid bridges in the incoming jet. In addition, the ratio is weakly dependent to Ug ext, showing that, contrary to the pure liquid jet case, the external gas has a marginal influence on the drop size. Two special cases do not follow this trend. The β = 0.97 and 0.99 cases show a ratio higher than 1. Here we can assume that the inner gas atomizes the annular jet liquid film: the thickness of the liquid film, which has not been measured, is probably a more relevant scale than S frag. The β = 0.2 and 0.4 cases at the largest liquid velocities Ul sup correspond to bubbly flow configurations, and even to a dispersed bubbly flow configuration for the lowest β. In these conditions, it is not surprising that we measured the largest liquid length scales S frag which are of the same order of magnitude as H L. Liquid length scales between bubbles S frag happen to be much larger ( one order of magnitude) than the mean drop size C 10, and thus C 10 S frag. Such a result is similar to the single phase situation. Indeed, drop sizes approach those found for a pure liquid jet (the latter are represented by open triangles in figure 23). In addition, and as in single phase jets, the drop size remains significantly sensitive to Ug ext. Clearly, when the two-phase jet is in a bubbly flow regime, the size of drop is not correlated with the extent of liquid bridges, and the atomization mechanisms are close to those identified on pure liquid jets. Another question concerns the spatial development of the spray. Measurements taken at L b + L bst D show that large liquid fragments are present at this position. Because of their size, they may evolve because of aerodynamic forcing by the external gas flow. To evaluate if secondary atomization is to be expected, we estimated the aerodynamic Weber number W e defined as: W e = ρ gs frag (Ug ext Ul sup ) 2, (6) σ where σ is the surface tension. Within the band shown figure 23 bottom ; W e varies between 1 up to 90. Thus, at least for gas velocities above 30 m/s, these liquid fragments should experience break-up due to aerodynamics forcing. It would be interesting to check this effect by carrying out measurements farther downstream. Conclusion We investigated the assisted atomization of a two-phase gas-liquid jet. We have shown that break-up mechanisms are strongly affected by the presence of a dispersed gas phase in the liquid jet: break-up length and drop sizes significantly decrease when gas flow rate fraction increases. In addition the dependency of drop size on external gas velocity is neatly weakened compared with a pure liquid jet situation. Moreover, and contrary to pure liquid jets, drop sizes produced at different locations are similar, leading to spatially homogeneous sprays. These results clearly indicate that new atomization mechanisms take place when considering a two-phase inner jet. Although the very nature of these new mechanisms has not been unveiled, we brought some experimental evidence that drop sizes are mainly controlled by the flow configuration in the jet, as they are close to the size of liquid bridges present at injection, except in the bubbly flow regime. This conclusion needs further confirmation, in particular 12

13 by considering others dynamic pressure ratio. Another issue is the role of secondary atomization: the fact that droplet production occurs earlier when β increases (see strong decrease in L b ) shows that secondary atomization may occur closer to the injection at finite β. This effect should be tested by examining the spatial variation of the drop size farther downstream injection. Acknowledgments The authors would like to acknowledge the support to this project by SAFRAN Snecma Space Engines Division and CNES. References [1] Philippe Marmottant and Emmanuel Villermaux. Journal of fluid mechanics, 498:73 111, [13] JC Lasheras, E Villermaux, and EJ Hopfinger. Journal of Fluid Mechanics, 357: , [14] JC Lasheras and EJ Hopfinger. Annual Review of Fluid Mechanics, 32(1): , [15] Moongeun Hong, Alain Cartellier, and Emil J Hopfinger. International Journal of multiphase flow, 30(6): , [16] A Cartellier. International journal of multiphase flow, 25(2): , [17] Antoine Delon, Jean-Philippe Matas, and Alain Cartellier. 8th International Conference on Multiphase Flow, ICMF 2013, Jeju, Korea, May 26-31, 2013, pp. ICMF , [2] M Hong. PhD thesis, Universit e de Grenoble, France, [3] CM Varga, Juan C Lasheras, and EJ Hopfinger. Journal of Fluid Mechanics, 497: , [4] Philippe Marmottant. PhD thesis, Grenoble, INPG, [5] M Hong, A Cartellier, and EJ Hopfinger. Proc. 4th Int. Conf. on Launcher Technology, Liege Belgique, [6] Sina Ghaemi, Payam Rahimi, and David S Nobes. Atomization and Sprays, 20(3), [7] Dancho Konstantinov, Richard Marsh, Phil J Bowen, and Andrew Crayford. Atomization and Sprays, 20(6), [8] K Ramamurthi, UK Sarkar, and BN Raghunandan. Atomization and Sprays, 19(1), [9] F Ben Rayana, Alain Cartellier, Emil Hopfinger, et al. Proceedings of the International Conference on Liquid Atomization and Spray Systems (ICLASS), Kyoto, Japan, [10] Antonio Lozano, Felix Barreras, Guillermo Hauke, and Cesar Dopazo. Journal of Fluid Mechanics, 437: , [11] A Lozano, F Barreras, C Siegler, and D Löw. Experiments in fluids, 39(1): , [12] Jean-Philippe Matas and Alain Cartellier. Comptes Rendus Mécanique, 341(1):35 43,