Influence of angled dispersion gas on coaxial atomization, spray and flame formation in the context of spray-flame synthesis of nanoparticles

To quantify possible fluctuations of the free droplet surface or spray density, a predefined section of the spray between \(5\text { mm}<\text {HAB}<8\text { mm}\) (\(7<\Delta \text {z}/\text {D}_\text {l}<11.5\)) is analyzed. In this spatial section, primary atomization is completed and only dispersed droplets are visible, as exemplarily shown in Fig. 3a. Data evaluation is conducted based on 5000 consecutive binarized images of a time-series that is captured at 30k fps. Thereby, the normalized projected cross-sectional area of passing droplets as well as the normalized mean gray level (NMG-normalized with the maximum NMG value) of the entire section has been determined with identical outcome. In the following, the NMG is considered further for data evaluation due to advantages with regard to computational cost.

In Fig. 3c, the resulting temporal distribution of NMG for projected droplets is presented for a short exemplary time interval. Therein, high NMG values are an indicator for high amounts of liquid in the evaluated image section. The overall arithmetic mean signal amplitude is low at NMG\(_\text {m,d}\) = 0.17 compared to the highest peak value that is normalized to NMG\(_\text {d,max}\) = 1. Randomly distributed high peaks indicate large lumps of liquid being shed from the initial liquid bulk at nozzle exit. Determination of specific shedding frequencies by means of an FFT analysis leads to a broad distribution of frequencies without a distinct peak due to the non-harmonic behavior and noise. Instead, a mean time interval \(\Delta \text {t}_\text {m,d}\) can be determined that represents the average time interval between two consecutive peaks. To determine \(\Delta \text {t}_\text {m,d}\), a “find-peak” algorithm is applied. To attenuate noise, two thresholds are utilized as input values for the algorithm: i) minimum time interval distance of two consecutive peaks \(\Delta \text {t}_\text {min,d}=10/\text {fps}\) and ii) minimum peak amplitude threshold. The latter is set to the arithmetic mean value of the data signal (\(\text {NMG}_\text {min,d}=\text {NMG}_\text {m,d}\)), indicated by the dark-grey shaded background in Fig. 3c and d. To determine the threshold for minimum time interval distance, a sensitivity analysis is conducted, by means of evaluating the gradient of the resulting characteristic mean time interval \(\Delta \text {t}_\text {m,d}(\Delta \text {t}_\text {min,d})\) over \(\Delta \text {t}_\text {min,d}\). The final threshold \(\Delta \text {t}_\text {m,d}\) is chosen where the gradient reached a minimum. Identified peaks are highlighted with circles as shown in Fig. 3c and d. Based on the find peak analysis, it can be concluded that the liquid shedding and consequently the detected droplet surface area fluctuates with a characteristic mean time interval of \(\Delta \text {t}_\text {m,d}=0.73\) ms.

Analogous to integral fluctuations of free droplet surface, the flame activity is analyzed using the same “find-peak” algorithm. However, data acquisition is performed further downstream in the region between \(5\text { mm}<\text {HAB}\) \(<45\text { mm}\) (\(7<\Delta \text {z}/\text {D}_\text {l}<64\)), i.e.,  height above nozzle exit at which changes in flame intensity are most significant (see Fig. 3b. The corresponding thresholds for the “find-peak” algorithm are set to \(\Delta \text {t}_\text {min,f}=10/\text {fps}\) and \(\text {NMG}_\text {min,f}= 0.5 \cdot \text {NMG}_\text {m,f}\). Note that in this case, the NMG can be directly evaluated from the raw image data without the necessity of previous binarization, since high values of NMG already portray high flame activity (Bieber et al. 2019).

From qualitative comparison of the NMG data signals of each case, timescales of flame activity fluctuations appear to be significantly larger than the previously presented fluctuations of projected droplet surface, which is also reflected in determined mean event time intervals. Consequently, the mean time interval between states of high flame activity is determined to \(\Delta \text {t}_\text {m,f}=1.5\) ms with a maximum value of \(\Delta \text {t}_\text {max,f}=5.9\) ms. Thus, the previously made assumption of different quasi-stationary states during the time interval of a single droplet passing the flame region is supported since the residence time of a single droplet is estimated to \(\Delta \text {t}=0.7\) ms. As the characteristic mean time interval between states of high flame activity is twice as large, flame activity cannot be directly correlated to fluctuating liquid mass flow at nozzle exit.

Initially, this appears to be contradictory, since qualitative comparison of the video data suggests a connection between flame fluctuation and the formation of large liquid lumps from the initial jet at nozzle exit. Corresponding high-speed images show that these lumps and subsequent secondary atomization always leads to ignition of the spray phase, as soon as droplets surpass a radial position that exceeds the outer diameter of the dispersion gas. This phenomenon can be explained by considering not only frequencies but also local variations in spray pattern associated with shedding: Resulting liquid lumps or single droplets, do not always spread in the same spatial direction. Instead, they are shed with varying radial velocity share. If the radial momentum is large enough to allow the liquid to “break” through the coaxial dispersion gas stream, it ignites in the pilot flame and presumably acts as conductor between pilot flame and evaporating droplet phase. Ultimately, this leads to ignition of the entire droplet phase in case of a flammable mixture.

In Fig 3c, the occurrence of exclusively big lumps of liquid is characterized by extreme peak values of NMG. These events can be emphasized and evaluated by shifting the amplitude threshold to higher NMG values. An increased threshold to \(\text {NMG}_\text {min,d}\) = 2.5 \(\cdot \,\text {NMG}_\text {m,d}\) is indicated by the light grey background and represents shedding of large liquid lumps. As a consequence, exclusively the black peak circles are taken into account for evaluation which leads to \(\Delta \text {t}_\text {m,d}=1.5\) ms, i.e., the mean time interval of two consecutive flame ignitions which could be an explanation for the qualitatively observed interdependence of droplet mass and flame.

Summarizing, the coaxial \(\text {O}_\text {2}\) dispersion gas flow presumably acts as a protective barrier that pushes back the pilot flame, encapsulates the spray phase and thus prevents spray ignition. Hence, ignition depends on two different effects associated with the liquid shedding mechanism: high peaks of temporally fluctuating shed liquid volume flow and high radial momentum. For the present nozzle configuration, however, the second effect appears to predominate.

On the basis of this knowledge, two alternative nozzle designs are tested, as shown in Fig. 4, that aim for a reduction of large-scale flame fluctuations and prevention of flame extinction.

Approach (B) aims for a reduced distance between pilot flame and spray phase. The elevated position of the nozzle exit relative to the surrounding sinter matrix enables a guaranteed injection into the fully developed pilot flame. However, apart from an improved visual access, data evaluation of measurements with identical operation conditions has shown no impact on timescales of flame and droplet phase fluctuations compared to the data of the original planar SpraySyn configuration (see Fig. 1); therefore, this configuration will not be further considered in the following.

The second approach C, shown in Fig. 4 bottom, also features improved visual access due to an elevated position of the nozzle exit but in addition, the gas-phase is redirected by 45° at nozzle exit in order to i) reduce the axial velocity share at nozzle exit, ii) increase the average value of radial momentum. This way, the gas impingement region is shifted closer towards the nozzle exit and shearing and turbulence are increased. As a result, improved atomization and mixture formation are expected and the development of a protective \(\text {O}_\text {2}\) hollow cone barrier might be avoided. While similar configurations have already been studied in the literature (cf. e.g. Mädler et al. 2002), in Configuration C, key parameters such as characteristic length of gas and liquid flow and corresponding cross section areas remain identical to the original SpraySyn configuration. This way, comparability is ensured since identical gas and liquid mass flow rates lead to the same dimensionless quantities (e.g., Re, We) and stoichiometry at the same time, which is not the case with similar nozzle configurations in the literature. If key nozzle dimensions are changed (e.g., gap width of dispersion gas (Mädler et al. 2002), identical flow parameters (Re\(_\text {l}\),Re\(_\text {g}\)) will lead to different Mach numbers and/or stoichiometry.

In Fig. 5, a qualitative comparison of macroscopic and microscopic primary atomization shadowgraphy images at nozzle exit are presented for the original SpraySyn (left) and the modified nozzle configuration C (right) with identical operation conditions (see Table 1).

Compared to the SpraySyn nozzle, the resulting spray of nozzle configuration C appears to be characterized by more homogeneous and smaller droplet sizes. In addition, the formation of large liquid lumps is avoided. To compare the nozzle configurations quantitatively, the “find peak” algorithm (see Fig. 3) is applied for both configurations. In addition to the characteristic mean time intervals (\(\Delta \text {t}_\text {m}\)), arithmetic mean values (\(\overline{\text {NMG}}\)) and standard deviations (\(\sigma\)) normalized to the arithmetic mean value are determined and compared for both cases, i.e., droplet and flame fluctuations. In addition, the dimensionless active flame time (\(\text {t}^*_\text {act. flame}\)) is determined by counting all frames for which \(\text {NMG}_\text {f} >=0.5 \cdot \overline{\text {NMG}}_\text {f}\) applies and subsequent normalization with the total number of frames. In Table 2, the calculated data regarding the projected droplet area and flame activity of the SpraySyn nozzle configuration and the modified configuration C are compared.

The nozzle modification significantly decreases the characteristic mean time interval between high flame activity (\(\Delta \text {t}_\text {m,f}\)), which indicates the avoidance of unwanted large-scale fluctuations (cf. Fig. 3, e). In addition, the flame and spray characteristic mean time intervals follow the previously expected trend and show very similar values (\(\Delta \text {t}_\text {m,d}\approx 0.52\) ms and \(\Delta \text {t}_\text {m,f}\approx 0.57\) ms). Corresponding high-speed videos of the resulting flame support the assumption that the flame is significantly stabilized by the modified nozzle configuration C, due to visible flame activity in every frame. In addition, the share of flame active time is increased from 50.11% to 78.05%. Hence, it can be assumed for configuration C that fluctuations in flame activity are influenced by the following two remaining effects: i) gas-phase turbulence or ii) fluctuations of the available free reactive surface. The fact that the modified nozzle leads to more stable flame operation can also be shown by analysing the medians and standard deviations. For configuration C, the median (\(\tilde{\text {NMG}}\)) of the NMG for both, flame and droplets, is very close to corresponding arithmetic mean values (\(\overline{\text {NMG}}\)), but deviates considerably from the arithmetic mean for the SpraySyn configuration. Accordingly, the standard deviations of the SpraySyn configuration are approximately equal to the arithmetic mean, while for configuration C, the standard deviation is only approximately half of the arithmetic mean.

In Fig. 6, quantitative spray and particle measurement data of the SpraySyn nozzle configuration (circle-shape) and the improved configuration C (rhombus-shape) are presented in dependency on HAB. Left, the axial velocity (red) and Sauter mean droplet diameter (blue) are shown for 0 mm < HAB < 50 mm and right, the corresponding Sauter mean mobility particle diameter (green) is presented for 50 mm < HAB \(\le\) 50 mm. Starting at the nozzle exit, droplet sizes (blue lines) are decreasing due to a combination of evaporation and secondary atomization processes until they are increasing again from HAB > 15 mm, since small droplets are already completely evaporated while bigger droplets are still present (Bieber et al. 2019).

This characteristic shape matches well with the corresponding velocity distribution. Towards HAB = 15 mm, the mean axial velocity increases, due to droplet acceleration, until it decreases again, due to air entrainment and an increasing cross sectional flow area.

The Sauter mean droplet diameter distributions along the center axis represent a quantification of the qualitative observation of refined spray with nozzle configuration C (cf. Fig. 4). While both distributions have a similar characteristic shape, mean droplets are smaller for configuration C with a reduced diameter range as shown in corresponding exemplary droplet size histograms for HAB = 40 mm. Overall, relative velocity and thus shearing are increased at nozzle exit, which leads to an increased grade of atomization.

Since the nozzle variation leads to differences in atomization, mixture formation, and flame activity, an influence of the nozzle geometry on particle characteristics was expected. Thus, syntheses are performed with both nozzles (SpraySyn and configuration C) and final powder properties are determined by BET and TGA-MS measurements (Table 3). Additionally, samples are extracted by utilization of a HIAT probe at various HAB in situ (Fig. 6). Our investigations have shown that particle structure properties as well as the elemental composition of the final particle output can be influenced by solely varying the nozzle geometry.

The production of particles with certain tailored structure properties is one major challenge in the field of FSP synthesis. Our study indicated that the agglomerate size and the primary particle sizes are sensitive to the nozzle geometry. Physisorption measurements by BET method revealed the powders specific surface area (SSA). The average primary particle size d\(_{\mathrm{P}}\) was estimated subsequently by assuming the presence of ideal spherical, monodisperse primary particles. In gas phase synthesis, the development of primary particle sizes mainly takes place during the particle residence in high-temperature regions within the flame (Gröhn et al. 2014). Thus, d\(_{\mathrm{P}}\) can reveal information regarding differences in the initial particle growth by collision and sintering in the hot flame regions. The powder comparison revealed strong differences for both nozzles, as the SSA for nozzle configuration C (290 \(\text {m}^2/\text {g}\)) significantly exceeds the value for the SpraySyn case (195 \(\text {m}^2/\text {g}\)). By assuming a bulk density of 5.24 g/cm\(^{3}\) (\(\gamma -\hbox {Fe}_{2}\hbox {O}_{3}\) as product output), the mean primary particle size can be calculated to 5.9 nm for the SpraySyn nozzle and 3.9 nm for configuration C. Therefore, configuration C leads to a reduced sintering dominance which can be argued qualitatively by time averaged average spray/flame characteristics. By comparing the flame shape and mean droplet velocities for both nozzles, one can roughly estimate differences in the particles residence times in the hot flames. Our study has shown that SpraySyn produces a larger flame while droplets move with a lower axial velocity (ref. Table 4). Calculated maximum residence time of small droplets (\(\le 4\,\upmu \hbox {m}\)) is significantly larger for the SpraySyn nozzle compared to configuration C (\(\Delta \text {t}_\text {res,SpraySyn}=670\,\upmu \text {s}\), \(\Delta \text {t}_\text {res,C}=479\,\upmu \text {s}\)). Assuming that the particle residence time correlates qualitatively to droplet residence times, configuration C potentially reduces particle sintering. However, further investigation of flame temperature fields is necessary for a more detailed investigation of the nozzle design influence on the particle sintering.

Although the BET values match qualitatively to the comparison of the particle residence time in hot flame regions, we want to emphasize that the BET values should be reviewed critically for two reasons. First, BET reveals a surface equivalent, volume-related average particle size. However, due to flame pulsations, different synthesis conditions are temporally apparent and a certain (unknown) polydispersity in the final product has to be expected in general. Secondly, during FSP, particle precipitation within the hot droplet phase can be present, leading to the production of larger, dense and nearly spherical particles, which is referred to as droplet to particle-pathway. This side pathway can lead to a second particle mode, ranging from the two digit to three digit nm regime (Strobel and Pratsinis 2011; Jossen et al. 2005). Both cases lead to an unknown degree of polydispersity within the product, and in particular, the latter one may affect the physisorption measurements sensitively. At this point, we want to emphasize that there is still a lack of data in literature describing the exact (mass) ratio of droplet originated particles in respect to gas borne particles. Third, TGA-MS measurements indicate the presence of an unneglectable amount of carbon in the final powder outcome for both configurations. Product impurities may affect the physisorption measurement as well. However, the potential origin of carbon residuals as detected from TGA-MS measurement is described in the next section.

Beside the effect on the primary particle size evolution, it was shown that the change in the nozzle design also influences the evolution of resulting particle agglomerate/aggregate sizes. By means of a HIAT sampling probe, particle ensembles were diluted in situ and particle size distributions were determined by SMPS measurements (Tischendorf et al. 2020). Along the center line between 5 to 11 cm HAB, a linear raise in the mobility diameter can be observed. In this area configuration C leads to larger particle diameter which is in accordance to the BET results. Thus, configuration C may lead to reduced sintering and smaller primary particle sizes, but their agglomeration leads consequently to the evolution of larger and likely more fractal aggregates and agglomerates. The linear raise of the particle diameter is an indication that particle growth is already solely dominated by agglomeration/aggregation Tsantilis and Pratsinis (2004). It is likely that the primary particle size evolution by collision and sintering might be finished already at lower HAB. Interestingly, from 12 cm HAB and above the SpraySyn nozzle leads to the formation of larger particles and the corresponding size distributions are broadened due to the presence of particle sizes up to 50 nm (Fig. 6). The origin of those larger particles is not yet fully understood. However by transmission electron microscopy of in situ extracted particle samples, it was revealed earlier that visibly larger and partially sintered aggregates evolve in regions above the flame tip, presumably as a result of particle re-circulation (Tischendorf et al. 2020). At higher HAB, the standard deviations for both nozzle geometries are converging asymptotically, likely caused by the aerosols self-preservation at later synthesis stages Vemury and Pratsinis (1994), while the SpraySyn case still indicates larger agglomerates.

During TGA-MS measurement, powders of both nozzles have been heated in air until reaching their mass steady state. The MS revealed an initial release of adsorption water, followed by an emission of CO\(_{2}\) until 300 \(^\circ \text {C}\), whereby the mass loss was higher for SpraySyn (\(17,3\%\)) in comparison with the powder obtained from synthesis with nozzle configuration C (\(13,3\%\)). The emission of CO\(_{2}\) during the powders thermal treatment may have several origins. Organic material might be present, leading to an emission of CO\(_{2}\) during its oxidation under oxygen-rich air atmosphere. At one side, carbon containing material might be present as soot. Soot production might be caused by flame pulsations, since those may lead to flame states with variable flame stoichiometry, despite lean combustion operating conditions in our test setup (Tischendorf et al. 2020). On the other side, organic species of the precursor formulation might pass through the synthesis chain without fully reacting. Especially larger particles from the droplet to particle pathways might carry significant amounts of organic material, in case they do not pass high temperature regions for sufficient residence time. In this case, droplet size distributions and droplet trajectories would predominantly affect the product composition. At this point, we want to emphasize that the CO\(_{2}\) emission can also be present, if free Carbondioxide is integrated as bidentate at the particles surfaces (Grimm et al. 1997). However, surface bidentates are unlikely the dominant origin for the CO\(_{2}\) emission as the relative mass loss during the TGA does not scale with the free surface area of the powders. Although the origin of organic residuals is not clear so far, the atomization and flame characteristics do affect their quantitative presence. Therefore, flame activity and corresponding droplet characteristics seem to play important roles in the particle synthesis chain during FSP. However, to the authors knowledge they have not yet been focused and described in literature and might explain possible literature discrepancies of particle characteristics produced under similar operating conditions but with different nozzle types. For instance, ambivalent results can be found by solely comparing the values in this work to product purities derived in literature under comparable synthesis and atomization conditions. In particular, Grimm et al. 1997 produced \(\gamma -\hbox {Fe}_{2}\hbox {O}_{3}\) under lean combustion conditions as well. However, the authors concluded organic impurities corresponding to values in the low single-digit regime mass percentage (Grimm et al. 1997). During the synthesis of Zirconium Oxides and Silanol also even highest product purities (with practically zero carbon content) have been derived in literature (Spyrogianni et al. 2017; Mueller et al. 2004). However, this point one has to mention that the evolution of product impurities might be also highly sensitive to the applied precursor composition.