betaFeedback

WorksPhysics and Astronomy

Droplet Bag Formation in Turbulent Airflows

Abstract

We present novel numerical simulations investigating the evolution of liquid droplets into bag-like structures in turbulent airflows. The droplet bag breakup problem is of significance for many multiphase processes in scientific and engineering applications. Turbulent fluctuations are introduced synthetically into a mean flow, and the droplet is inserted when the air-phase turbulence reaches a statistically stationary state. The morphological evolution of the droplet under different turbulence configurations is retrieved and analysed in comparison with laminar aerobreakup results. While the detailed evolution history of individual droplets varies widely between different realisations of the turbulent flow, common dynamic and morphological evolution patterns are observed. The presence of turbulence is found to enhance the drag coefficient of the droplet as it flattens. At late times, the droplet becomes tilted and increasingly corrugated under strong turbulence intensity. We quantify these phenomena and discuss their possible governing mechanisms associated with turbulence intermittency. Lastly, the influences of liquid-gas viscosity ratio are examined and the implications of air-phase turbulence on the later bag film breakup process are discussed.

Introduction

Atomisation involves the morphological evolution of the liquid bulk driven by external forces, followed by the formation of corrugated ligaments subject to capillary pinch-off, and terminates in a number of fragments featuring a broad size distribution [1]. The aerobreakup of spherical liquid droplets in uniform airflows is a canonical problem of atomisation and two-phase fluid dynamics [2, 3], an indispensable fragmentation mechanism found in numerous natural and industrial processes, including precipitation [4], sea spume production [5], disease transmission [6, 7], liquid fuel combustion [8] and inkjet printing [9]. In a uniform flow, droplet aerobreakup can be parameterised by the following four non-dimensional controlling parameters [10]:

We≡ρgU02d0σ,Oh≡μlρld0σ,ρr≡ρlρg,μr≡μlμg.(1)We \equiv\frac{\rho_g U_0^2 d_0}{\sigma}, \qquad Oh \equiv\frac{\mu_l}{\sqrt{\rho_l d_0 \sigma}}, \qquad\rho_r \equiv\frac{\rho_l}{\rho_g}, \qquad\mu_r \equiv\frac{\mu_l}{\mu_g}. \tag*{(1)}

Among these, We and Oh are respectively the Weber and Ohnesorge numbers quantifying the ratio of inertial to capillary and viscous to capillary forces, and ρr\rho_r and μr\mu_r are respectively the density and viscosity ratios of the liquid and gas phases. Here, ρl\rho_l, μl\mu_l and ρg\rho_g, μg\mu_g are respectively the density and viscosity for the liquid and gas phase, d0d_0 is the initial diameter of the droplet, U0U_0 is the velocity of the ambient airflow, and σ\sigma is the surface tension at the liquid-gas interface [11].

The effects of We and Oh on the morphological evolution of droplets undergoing aerobreakup have been extensively investigated in many experimental and numerical studies [12]. Various aerobreakup regimes, including bag, multi-mode (bag-stamen), sheet-thinning and catastrophic breakup, have been observed and the transition thresholds between them have been delineated using these two non-dimensional parameters [13, 14, 15]. Among these regimes, the investigation of the bag breakup regime can potentially shed light on the mechanism of ocean spray production [16], as the bag-shaped geometry of the late-time deformed drops bears a strong resemblance to the small-scale sea-surface perturbations, whose breakup dominates sea spume generation under extreme wind conditions [5, 17]. Most of the early works investigating droplet aerobreakup focus on the analysis of bag formation using hydrodynamic instability models [18], while discussions on the late-time bag breakup mechanisms and fragment statistics are still scarce [19, 20]. The recent experimental work by Jackiw and Ashgriz [2] discussed in detail the physical mechanisms governing the early-time droplet deformation patterns, which are not captured by the simplified hydrodynamic instability models. They also observed that the major pathways leading to film fragment generation are the destabilisation and collision of receding hole rims under centripetal acceleration, and an analytical model predicting the overall volume probability distribution of aerobreakup fragments is developed and verified at various We values [21]. Numerical investigations into fragment statistics, especially those exploiting the flexible Volume-of-Fluid (VoF) interface reconstruction method, have historically been handicapped by the lack of grid convergence, which arises from uncontrolled numerical perforation of bag films when they reach the minimum grid size. However, recently Chirco et al. [22] developed a Manifold Death (MD) algorithm that prevents such spurious breakup of thin films by artificially perforating them once their thickness reduces to a prescribed critical value. This algorithm improves the grid convergence behaviour of the statistics of large bag film fragments, allowing further investigation into the generation mechanism and behaviour of individual aerobreakup fragments [3, 23].

Nevertheless, despite the progress made against these limitations, a complete accounting of the realistic physical environment around droplet breakup has not yet been achieved. One important effect is that of turbulence in the freestream gas flow. In most of the natural phenomena as well as previous experimental studies, droplet breakup occurs in turbulent gas flows. For example, combustion chamber flows may feature a variety of integral length scales [24], and spume generation at the air-sea interface occurs within the turbulent atmospheric boundary layer [17], which can be further modified over time due to coupled wind-wave interactions [25, 26]. In these scenarios, apart from the aerodynamic force exerted by the mean flow, accelerating, shearing and fluctuating effects of the turbulent flow can also affect the deformation and breakup of the droplet [27]. In particular, a detailed and quantitative analysis approach is still lacking for the interaction between the droplet surface and the turbulent vortices present in the ambient gas flow. It is also noted that, although droplet-turbulence interactions have seen some investigation in recent years with important findings, e.g., droplets larger and smaller than the Hinze scale demonstrate different behaviour: the former tend to break up through a memoryless process [28] and absorb energy from the ambient flow, while the latter oscillate rapidly or coalesce to release energy to the environment [29]; these studies lack a base flow and are mostly carried out at low density ratios close to 1, which differ from the air-sea interaction scenario featuring finite wind speeds and large density contrasts.

Despite the difficulty in accurately measuring turbulent flow properties [24] and quantifying the response of droplets to external turbulence forcing, there have been a few experimental studies of turbulent aerobreakup in recent years [27, 30, 31]. Among these, Zhao et al. [30] utilised perforated plates to generate turbulent counterflows, which in turn drives the deformation of droplets falling from above. It is found that increasing turbulence intensity results in larger lengths and widths of the bag, which then breaks up to form smaller fragments compared with uniform flow conditions. Xu et al. [27, 31] placed droplets within the shear layer of a turbulent air jet, where two new breakup regimes with complicated morphological patterns are identified: butterfly and swing breakup.

Numerical studies of the turbulent aerobreakup problem are even more scarce, most likely due to the difficulty in imposing turbulent fluctuations on a uniform mean flow, and the high computational cost of resolving all turbulence length scales apart from the droplet dynamics. The linear-forcing method proposed by Rosales and Meneveau [32] cannot be directly applied to the aerobreakup problem, despite its recent successful application to turbulent bubble breakup studies [33, 34]. This is because the linear-forcing method is usually implemented with triply-periodic boundary conditions, while in the aerobreakup problem the droplet produces a separated wake region that breaks the fore-aft symmetry of the ambient flow field; in other words, a droplet undergoing aerobreakup can be affected by the wake of its preceding periodic image [35]. On the other hand, synthetic turbulence generation methods produce pseudo-turbulence fluctuation by summation of random Fourier modes (Synthetic Random Fourier Method), or superposition of spatial and temporal coherence on initially uncorrelated random number series (Synthetic Digital Filtering Method). Jiao et al. [24], to our knowledge the only investigation of the turbulent aerobreakup problem utilising direct numerical simulations, applied the LEMOS inflow generator [36] at the inlet with outflow boundary conditions. This generator represents turbulence as a series of random spots whose inner velocity distributions are determined by prescribed autocorrelation functions, which enables generation of turbulence structures with prescribed integral length scales and fluctuating velocities on the mean flow [36]. Their results suggest that superimposed air-phase turbulence leads to rapid redistribution of flow pressure around the droplet, which can cause asymmetric deformation patterns coupled with various forms of rotational motion; while the bag film breakup process and associated fragment statistics are not covered due to limited grid resolution level. It is also noted that this work is carried out at a relatively low liquid-gas density ratio ρr=20\rho_r = 20, which means that their results might not be directly applicable to air-water systems where the density ratio is much larger.

Overall, it can be seen that the currently available results of the turbulent aerobreakup studies are very limited and obtained under considerably different flow configurations. A well-defined methodological framework is still lacking, based on which systematic investigations of this problem can be carried out. Droplet bag breakup within a uniform incoming flow with a superimposed, well-characterised turbulent fluctuation features a simple configuration and therefore serves as a first step towards understanding aerobreakup in more complicated turbulent flow environments, e.g. boundary-layer flows where shear effects dominate. We investigate the turbulent aerobreakup problem using two-phase and three-dimensional direct numerical simulations. Our focus is on elucidating the droplet deformation mechanisms up to the point of bag formation, which serves as a basis for future analysis of the breakup behaviour and fragment statistics. Our study is structured as follows. We present in §II A our problem configuration and parameter space, and then introduce the numerical method in §II B, with a focus on the synthetic turbulence generation method [37] we adopt for producing turbulent incoming airflows. We proceed to present in §III a brief overview of the droplet-turbulence interactions. We then discuss in §IV the evolution of the droplet centre-of-mass dynamic properties, and the air-phase pressure distributions giving rise to such evolution trends. Afterwards, we analyse in §V various droplet deformation patterns, including the aspect ratio (§V A), the tilting behaviour (§V B) and the surface curvature and velocity increment distributions (§V C). Finally, we discuss the possible influences of turbulent airflows on the bag film fragmentation process, and conclude the study in §VII with some remarks on future work.

Formulation and methodology

Problem description

The configuration for our fully three-dimensional direct numerical simulations is shown in Figure 1. A liquid droplet with initial diameter d0d_0, density ρl\rho_l and viscosity μl\mu_l is inserted close to the left boundary, within a turbulent gas phase featuring density ρg\rho_g and viscosity μg\mu_g. The domain width DD is set as 10d010d_0, at which the influence of finite domain size on the aerobreakup process can be considered negligible. Outlet flow conditions are applied at the right boundary and a turbulent inflow with mean velocity U0U_0, fluctuating velocity u∗u^* and injection length scale L0L_0 is introduced at the left boundary, while zero-gradient conditions are applied at the other domain boundaries.

Sketch showing the configuration of the turbulent droplet aerobreakup problem

Figure 1. Sketch showing the configuration of the turbulent droplet aerobreakup problem.

Among the dimensional parameters introduced above, the fluctuation velocity u∗u^* and the injection length scale L0L_0 characterises the swirling motion of the largest eddies within the synthesised air-phase turbulence. These in turn lead to two additional non-dimensional groups [38] apart from those already introduced in Eq. (1),

We~≡2ρgε2/3d05/3σ=2ρgu∗2d05/3σL02/3(2)\widetilde{We} \equiv\frac{2\rho_g\varepsilon^{2/3}d_0^{5/3}}{\sigma} = \frac{2\rho_g u^{*2}d_0^{5/3}}{\sigma L_0^{2/3}} \tag*{(2)}
dhd0∝(σρg)3/5ε−2/5=We−3/5(u∗U0)−6/5(L0d0)2/5(3)\frac{d_h}{d_0} \propto\left(\frac{\sigma}{\rho_g}\right)^{3/5}\varepsilon^{-2/5} = We^{-3/5}\left(\frac{u^*}{U_0}\right)^{-6/5}\left(\frac{L_0}{d_0}\right)^{2/5} \tag*{(3)}

where We~\widetilde{We} and dh/d0d_h/d_0 are respectively the turbulent Weber number and the Hinze scale non-dimensionalised by the initial droplet diameter; and ε\varepsilon is the turbulence dissipation rate per unit volume, which we scale using the nominal value of u∗3/L0u^{*3}/L_0 for the turbulent fluctuations we produce. The resulting turbulence reaches statistical stationarity after an initial startup transient, is unforced in the interior of the numerical domain, and shows spatially a gradual decay in the streamwise direction (parallel to the mean flow), but is homogeneous in the spanwise plane. The turbulence is further characterised in §II B below.

u∗/U0u^{*}/U_{0}L0/R0L_{0}/R_{0}μr\mu_{r}Re\mathrm{Re}We~\widetilde{\mathrm{We}}dh/d0d_{h}/d_{0}λ/d0\lambda/d_{0}Reλ\mathrm{Re}_{\lambda}η/d0\eta/d_{0}Real. No.
00551476-----1
0.2515514762.9761.0080.14252.80.01005
0.251.55514762.2711.1860.17464.60.01105
0.2525514761.8751.3310.20274.60.01195
0.2545514761.1811.7560.2841060.01415
0.2585514760.7442.3170.4021490.01675
0.5155147611.900.4390.10174.60.00595
0.65155147620.120.3200.08885.00.00495
0.8155147630.480.2490.08094.40.00425
0.251205372.9761.0080.23631.70.02133
0.512053711.900.4390.16744.90.01273
0.6512053720.120.3200.14651.20.01043
0.812053730.480.2490.13256.80.00893
0.251102682.9761.0080.33411.20.03593

Table 1. List showing the parameter space for three-dimensional numerical simulations of turbulent bag formation carried out in this work.

As for the four non-dimensional parameters defined in (1), namely the mean flow Weber number We\mathrm{We}, the Ohnesorge number Oh\mathrm{Oh}, the density ratio ρr\rho_{r} and the viscosity ratio μr\mu_{r}, we fix We\mathrm{We}, Oh\mathrm{Oh} and ρr\rho_{r} as 15, 0.005 and 833, as is typical for air-water droplet bag breakup. We test three values of μr\mu_{r}, namely 55, 20 and 10, which allows us to analyse the influence of the liquid-gas viscosity ratio on turbulent aerobreakup in §VI. It should be noted that the single configuration with μr=10\mu_{r}=10 only serves to examine the effects of μr\mu_{r} and can be difficult to obtain for liquid-gas systems under experimental conditions. For reference to the readers, we note that these three viscosity ratios correspond to three different gas-phase Reynolds numbers Re≡ρgU0d0/μg=1476\mathrm{Re}\equiv\rho_{g}U_{0}d_{0}/\mu_{g}=1476, 537 and 268. At this magnitude of Re\mathrm{Re}, flows past a solid sphere would begin to transit to turbulence [3, 39]. A full list showing the parameter space of this study is provided as Table I, featuring a wide range of We\mathrm{We} and dh/d0d_{h}/d_{0} values including both super- and sub-Hinze-Scale droplets. The presence of significant variations within the turbulent ambient flow necessitates ensemble averaging across multiple realisations, thus enabling us to determine the statistically converged evolution patterns or distributions for physical properties under a certain turbulence configuration. We have therefore provided the number of ensemble realisations in the last column of Table I.

Among the physical properties introduced in Table I, the Taylor microscale λ\lambda, the Taylor Reynolds number Reλ\mathrm{Re}_{\lambda}, and the Kolmogorov microscale η\eta are defined as follows [38],

λ≡15νgεu∗,Reλ≡u∗λνg,η≡(νg3ε)1/4,(4)\lambda\equiv\sqrt{\frac{15\nu_{g}}{\varepsilon}}u^{*},\quad\mathrm{Re}_{\lambda}\equiv\frac{u^{*}\lambda}{\nu_{g}},\quad\eta\equiv\left(\frac{\nu_{g}^{3}}{\varepsilon}\right)^{1/4}, \tag*{(4)}

where νg≡μg/ρg\nu_{g}\equiv\mu_{g}/\rho_{g} is the kinematic viscosity of the gas phase. According to Table I, our turbulence configurations feature Taylor Reynolds number Reλ\mathrm{Re}_{\lambda} ranging from 31.7 to 149, which becomes stronger than the turbulence-bubble interaction studies carried out by Riviere et al. [38] for simulations with large u∗u^{*} or L0L_{0}, but remains typical of two-phase turbulence simulations [40, 41].

Numerical method

We use the open-source Basilisk numerical library [42] to solve the two-phase incompressible Navier-Stokes equations, which utilises a finite-volume numerical scheme coupled with the geometric volume-of-fluid (VOF) method for fluid interface reconstruction [43]. Usage of an octree-based adaptive mesh refinement (AMR) technique significantly reduces the computational cost for such two-phase simulations. Readers are referred to our previous works [3, 44] for a detailed description of the aforementioned numerical methods. Here we focus on the synthetic digital filtering method proposed by Xie and Castro [37], which we adopt for generating turbulent fluctuations at the inlet.

As is mentioned in §I, the linear forcing method of Rosales and Meneveau [32] has seen successful appli cations in turbulence-bubble interaction studies [33, 38]. It adds a volumetric forcing term to the Navier-Stokes equation in the physical space, which leads to the transition from an ABC-flow at initialisation to well-characterised HIT fluctuations at steady-state. However, this method cannot be applied to the present turbulent aerobreakup problem, otherwise the deformation and breakup of the drop will be affected by the wake of its preceding mirror image. To show this, a scaling analysis can be made based on existing works on wake velocity defects (the difference between velocities at the same location with and without the obstacle) of spherical obstacles immersed in a turbulent carrier phase [35, 45]. Namely, as the distance to a solid spherical obstacle zz increases, the magnitude of the velocity defect Δu\Delta u decays following a power law of z−1z^{-1} within the range of z≤13d0z \le13d_0, which asymptotes to z−2z^{-2} at larger zz values. In our case where the liquid-gas density ratio ρr\rho_r is large, the droplet can be approximated as a solid obstacle, and the velocity defect caused by the droplet’s preceding mirror image reads

Δuiu∗≈U0u∗(Dd0)−1.(5)\frac{\Delta u_i}{u^*} \approx\frac{U_0}{u^*}\left(\frac{D}{d_0}\right)^{-1}. \tag*{(5)}

For a typical simulation configuration considered in the current work, where U0/u∗≈10U_0/u^* \approx10 and D/d0=10D/d_0 = 10, Eq. (5) suggests that the magnitude of velocity defect caused by the preceding periodic image of the droplet is the same as the turbulent fluctuating velocity of the ambient airflow. In other words, significant mutual interactions between neighbouring drops would arise across periodic boundaries. While such interactions may be eliminated by further increasing the domain size DD, this would incur higher computational costs for resolving turbulence in an enlarged simulation domain, which suggests that the linear forcing method cannot be applied directly with our current computational capabilities. In addition, the turbulent flows generated by this method feature a fixed injection length scale L0=0.19DL_0 = 0.19D, which does not meet our need to change it as a variable.

Based on the discussions above, it would be ideal to generate and maintain homogeneous and isotropic airphase turbulence in a simulation domain with inflow and outflow conditions. Wu [46] provides an overview of currently available inlet turbulence generation methods and their applications, where at each timestep the inlet velocity condition can be set using flow-field slices imported from an external auxiliary simulation (Strong Recycling Method), slices extracted from a downstream location within the current simulation domain (Weak Recycling Method), or synthesised using random number series or Fourier modes filtered by an approximate correlation function (Synthetic Turbulence Generation). Here we select the synthetic turbulence generation method proposed by Xie and Castro [37] due to its simplicity, high efficiency and well-validated turbulence statistics including inertial subrange and two-point velocity correlations [47].

The framework of this turbulence generation method is briefly reviewed as follows. Firstly, at each timestep, a two-dimensional random matrix RijR_{ij} with zero mean and unit variance is generated, and then filtered and normalised to produce two-dimensional space-correlated velocity fluctuation matrices ψij\psi_{ij} for each velocity component,

ψm,l=∑j=−NN∑k=−NNbjbkRm+j,l+k,0≤m,l≤Ngrid,(6)\psi_{m,l} = \sum_{j=-N}^{N}\sum_{k=-N}^{N} b_j b_k R_{m+j,l+k}, \quad0 \le m,l \le N_{\mathrm{grid}}, \tag*{(6)}

where Ngrid=2LN_{\mathrm{grid}} = 2^L is determined by the maximum grid level LL for resolving turbulence, and N=3n=3L0/ΔN = 3n = 3L_0/\Delta is proportional to the number of grid cells spanned by the turbulence integral length L0L_0. The filter coefficients bjb_j are prescribed as follows to approximate exponential velocity correlation functions,

bj=e−π∣j∣2n∑i=−NNe−π∣i∣n.(7)b_j = \frac{e^{-\frac{\pi|j|}{2n}}}{\sqrt{\sum_{i=-N}^{N}e^{-\frac{\pi|i|}{n}}}}. \tag*{(7)}

Time correlation is imposed between turbulent inlet conditions at successive timesteps by computing the new normalised fluctuating velocity components Ψ(t)m,l\Psi(t)_{m,l} based on that of the previous timestep Ψ(t−Δt)m,l\Psi(t-\Delta t)_{m,l} and newly synthesised fluctuation matrices ψm,l\psi_{m,l},

Ψ(t)m,l=Ψ(t−Δt)m,le−πΔt4TLa+ψm,l1−e−πΔt2TLa,0≤m,l≤Ngrid,(8)\Psi(t)_{m,l} = \Psi(t-\Delta t)_{m,l}e^{-\frac{\pi\Delta t}{4T_{La}}} + \psi_{m,l}\sqrt{1-e^{-\frac{\pi\Delta t}{2T_{La}}}}, \quad0 \le m,l \le N_{\mathrm{grid}}, \tag*{(8)}

where TLa=L0/U0T_{La} = L_0/U_0 is the injection time scale. Afterwards, the full inlet velocity profile is generated as follows:

Uin∣m,l=U0+urms2Ψ(t)m,l,0≤m,l≤Ngrid,(9)U_{\mathrm{in}|m,l} = U_0 + u_{\mathrm{rms}}^2\Psi(t)_{m,l}, \quad0 \le m,l \le N_{\mathrm{grid}}, \tag*{(9)}

where the constant coefficient urms2u_{\mathrm{rms}}^2 is a simplification of the Cholesky decomposition of the Reynolds stress tensor [47, 48] for homogeneous and isotropic turbulence (HIT) fluctuations. UinU_{\mathrm{in}} is then corrected following Kim et al. [48] to ensure constant mass influx, and superimposed at the left boundary of the simulation domain as a time-variant Dirichlet boundary condition.

Following Rivi\ere et al. [38], we first run precursor simulations without droplets to create statistically stationary turbulent air-phase flows within the entire simulation domain, where the maximum grid resolution level for the AMR scheme is set as $L_{\mathrm{turb}}=8$ for simulations with $\mu_r=55$, and 9 for those with $\mu_r=20$. In the latter case, we reach a minimum grid size of $\Delta_{\mathrm{turb}}=D/2^{L_{\mathrm{turb}}}=0.0195d_0$, allowing us to resolve the air-phase turbulence down to the Kolmogorov microscale $\eta$ as shown in {xref}Table I`. Afterwards, we insert the droplet which is allowed to interact freely with the ambient air-phase turbulence, and in the meantime increase the maximum grid resolution level to Ldrop=12L_{\mathrm{drop}}=12 for resolving the droplet surface, thus obtaining high-fidelity droplet deformation patterns up to the initiation of bag formation.

Finally, the droplet diameter d0d_0, incoming flow velocity U0U_0, dynamic flow pressure p0≡ρgU02/2p_0 \equiv\rho_g U_0^2/2, and the characteristic droplet deformation time τ≡ρl/ρgd0/U0\tau\equiv\sqrt{\rho_l/\rho_g}d_0/U_0 provide the natural reference scales for the length, mass and time quantities associated with the turbulent aerobreakup process. We will henceforth use these to non-dimensionalise the numerical results in the remainder of this study unless otherwise specified.

Verification of air-phase turbulence statistics

Upon the start of the simulation, a turbulent airflow is produced at the left boundary and transported downstream through the initially quiescent simulation domain, eventually reaching a statistical stationary state before the droplet is inserted. Here we inspect a few characteristics of the stationary-state turbulent airflow to evaluate its homogeneity and isotropy. In particular, Table I shows that the Kolmogorov microscale η/d0\eta/d_0 can be as small as 0.004 for simulations with viscosity ratios μr=55\mu_r=55, while the minimum bulk grid size is only Δturb/d0=D/d0/2Lturb=0.078\Delta_{\mathrm{turb}}/d_0=D/d_0/2^{L_{\mathrm{turb}}}=0.078. This suggests that for simulations with the most severe turbulence fluctuations, we are not yet able to resolve the smallest turbulence length scales where viscous dissipation dominates. Previous studies conclude that only strong airphase eddies at the large turbulence integral scale will impact droplet deformation [24, 30, 31], which indicates that under-resolved small eddies at the Kolmogorov length scale will not be likely to affect the fidelity of the droplet deformation data. Nonetheless, we evaluate the influence of reducing μr\mu_r on the turbulence statistics here, and further investigate its effects on droplet surface statistics in §VI.

Figure 2a shows the evolution of the averaged air-phase turbulent kinetic energy KK within the simulation domain at various values of LturbL_{\mathrm{turb}} for u∗/U0=0.8u^*/U_0=0.8 and L0/R0=1L_0/R_0=1, where time is non-dimensionalised by the turnover time τe≡d02/3ε−1/3\tau_e \equiv d_0^{2/3}\varepsilon^{-1/3} of eddies comparable to the size of the droplet. The turbulent kinetic energy first increases and then saturates as the turbulent airflow fully develops, with the steady-state value reaching grid convergence at a bulk resolution level of Lturb=8L_{\mathrm{turb}}=8. This suggests that at Lturb=8L_{\mathrm{turb}}=8, the mid-to-large-scale turbulent velocity fluctuations are already well-resolved, although the dissipation behaviour associated with smaller length scales requires further examination. Figure 2b shows the turbulence power spectrum E(k)E(k) sampled from the entire simulation domain; where we observe the −5/3-5/3 power law characteristic of the inertial subrange, and the wavelength corresponding to the droplet diameter 2π/d02\pi/d_0 is found to either fall within or remain close to the inertial regime. The slope of E(k)E(k) steepens at higher frequencies, and diminishes again towards the upper bound of the wavenumber kk for spectra with μr=55\mu_r=55; the latter behaviour is also present in fig. 13 of Xie and Castro’s original paper [37]. Indeed, this is most likely because the Kolmogorov microscale and the associated dissipative action at that scale are not fully resolved for μr=55\mu_r=55. This behaviour in the spectrum does not appear for μr=20\mu_r=20, where the Kolmogorov microscale and dissipation behaviour are well resolved.

Figure 2. (a): Grid convergence test for the air-phase turbulent kinetic energy KK for the turbulence configuration of u∗/U0=0.8u^{*}/U_{0}=0.8 and L0/R0=1L_{0}/R_{0}=1. (b): Power spectrum density of fluctuating velocities uu, vv and ww calculated from time series recorded near the inlet. (c): Compensated second-order turbulence structure functions DLLD_{LL} and DNND_{NN} as a function of the dimensionless distance r/ηr/\eta, where η\eta is the Kolmogorov length scale. (d): Comparison between DNN∣isoD_{NN}|_{iso} and DNND_{NN} at different values of u∗u^{*} as a test for flow isotropy.

We now proceed to assess the isotropy of the turbulence we generate. For this purpose, we introduce the second-order structure functions measured in both the longitudinal (DLLD_{LL}) and transverse (DRRD_{RR}) directions [33, 38, 49] to characterise turbulence fluctuations,

DLL(r)=13∑i⟨[ui(x,t)−ui(x+rx^i,t)]2⟩,(10)D_{LL}(r)=\frac{1}{3}\sum_i\left\langle\left[u_i(\boldsymbol{x},t)-u_i(\boldsymbol{x}+r\hat{x}_i,t)\right]^2\right\rangle, \tag*{(10)}
DNN(r)=16∑i≠j⟨[ui(x,t)−ui(x+rx^j,t)]2⟩,(11)D_{NN}(r)=\frac{1}{6}\sum_{i\ne j}\left\langle\left[u_i(\boldsymbol{x},t)-u_i(\boldsymbol{x}+r\hat{x}_j,t)\right]^2\right\rangle, \tag*{(11)}

Figure 2c shows the structure functions DLLD_{LL} and DNND_{NN} compensated by the theoretical scaling (εr)2/3(\varepsilon r)^{2/3} for HIT fluctuations. Similar to Rivière et al. [38], we recover the DLL=3/4DNND_{LL}=3/4D_{NN} relationship and the DLL∝r2D_{LL}\propto r^{2} scaling (corresponding to the r4/3r^{4/3} scaling observed for the compensated structure functions in fig. 2c) at small length scales regardless of the value of μr\mu_{r}. The compensated structure functions reach their maxima within the range of 20≤r/η≤4020\leq r/\eta\leq40, and for μr=20\mu_{r}=20 their peak values are close to 2, agreeing with theoretical predictions [38, 49]. For μr=55\mu_{r}=55, the peak values become greater than 2, which suggests that the turbulence dissipation ε\varepsilon is underpredicted at the set grid resolution. This is because we are not yet resolving the air-phase turbulence down to the Kolmogorov microscale η\eta in these simulations. At larger values of rr both DLLD_{LL} and DNND_{NN} start to decrease, where flow isotropy is no longer expected, especially along the streamwise direction due to the decay of turbulence intensity.

As an additional test for flow isotropy, we calculate the transverse structure function DNN∣isoD_{NN}|_{iso} corresponding to HIT fluctuations from its longitudinal counterpart,

DNN∣iso=DLL+r2∂∂rDLL(r).(12)D_{NN}|_{iso}=D_{LL}+\frac{r}{2}\frac{\partial}{\partial r}D_{LL}(r). \tag*{(12)}

Figure 2d compares the computed values of DNN∣isoD_{NN}|_{iso} according to (12) and those of DNND_{NN} directly determined from our turbulence data according to (11), showing an excellent agreement between the two regardless of the specific value of μr\mu_r.

Overall, the results presented in Figure 2 suggest that the air-phase turbulence generated using the Synthetic Turbulence Generation method introduced in §II B features satisfactory homogeneity and isotropy. The turbulent energy budget might not be fully recovered due to under-resolved viscous dissipation at small length scales, especially with large values of u∗u^* and μr\mu_r. Similar to grid-generated turbulence [50], the intensity of synthetically generated turbulence also decays further away from the inlet due to the lack of turbulence forcing in the simulation domain, causing large-scale inhomogeneity in the spanwise direction. However, we anticipate that these limitations should not affect the integrity of our results at μr=55\mu_r = 55, since eddies at the Kolmogorov microscales do not have any discernible influence on droplet deformation, and we do not consider energy exchange between the droplet and ambient turbulent flow in this work.

Phenomenology of droplet-turbulence interactions

After the air-phase turbulence reaches the statistically stationary state, we insert the droplet and immediately release it to freely interact with the surrounding airflow, while recording its motion and deformation history. In this section, we discuss the influence of air-phase turbulence on droplet morphology qualitatively by inspecting snapshots produced from various individual simulations, with the aim of summarising general patterns of such influences and providing the basis for further quantitative analyses in the following sections.

In Figure 3 we first compare the morphological evolution of droplets within a laminar airflow (Figure 3a-Figure 3d) and a typical turbulent airflow (Figure 3e-Figure 3h) with u∗/U0=0.25u^*/U_0 = 0.25 and L0/R0=1L_0/R_0 = 1. It can be observed that while the droplet shape remains axisymmetric until the blow-out of the bag in the laminar aerobreakup case, this symmetry about the streamwise minor axis of the drop can be lost at very early times when turbulence is introduced. As the droplet flattens in the turbulent airflow, its minor axis no longer aligns with the streamwise direction in Figure 3h, which will be covered in §V B. Furthermore, significant corrugations arise across the entire droplet surface and disrupt the well-defined bag shape observed in laminar aerobreakup scenarios, which we will discuss in detail in §V C.

Snapshots showing the bag formation process in laminar and turbulent airflow cases

Figure 3. Snapshots showing the bag formation process in the laminar airflow case (a-d) in comparison with a typical turbulent airflow case with u∗/U0=0.5u^*/U_0 = 0.5, L0/R0=1L_0/R_0 = 1 (e-h). The snapshots are taken at t/τ=0.28t/\tau= 0.28 (a,e), 0.69 (b,f), 1.11 (c,g) and 1.52 (d,h).

Due to the chaotic nature of the ambient airflow, droplets forced by turbulence with identical configurations may exhibit different deformation histories. Figure 4 shows the bag shapes from two different turbulence configurations: (u∗/U0,L0/R0)=(0.25,1)(u^{*}/U_{0}, L_{0}/R_{0}) = (0.25, 1) (a-c), and (0.65,1)(0.65, 1) (d-f), where the snapshots in the same row correspond to bag shapes taken from different individual ensemble realisations at the same simulation time. While the bag shapes and orientations vary significantly among different realisations, some common features are still observed for those with the same turbulence configurations. Namely, at u∗/U0=0.25u^{*}/U_{0} = 0.25 we find the bags to feature well-defined peripheral rims and inflating central films similar to laminar bag breakup phenomena observed in previous studies [3, 51], despite mild stretching across the entire bag. On the other hand, we find the flattened drops severely distorted at u∗/U0=0.65u^{*}/U_{0} = 0.65, and the distinction between the peripheral rim and the central bag becomes much less prominent. Ensemble-averaging across different turbulence realisations is therefore necessary for obtaining statistically meaningful data [3, 52].

Snapshots showing the bags formed at late times

Figure 4. Snapshots showing the bags formed at late times, retrieved from different ensemble realisations with the same turbulence configurations. For (a) to (d): (u∗/U0,L0/R0)=(0.25,1)(u^{*}/U_{0}, L_{0}/R_{0}) = (0.25, 1). For (e) to (h): (u∗/U0,L0/R0)=(0.65,1)(u^{*}/U_{0}, L_{0}/R_{0}) = (0.65, 1).

Apart from the fluctuating velocity u∗u^{*}, variations in the injection length scale L0L_{0} also lead to qualitative changes in droplet deformation. Figure 5 shows the typical late-time bag shapes across a wide range of turbulence configurations (u∗,L0)(u^{*}, L_{0}). It is observed that as the turbulence intensity u∗u^{*} increases (figs. 5a-5d) or the injection length scale L0L_{0} decreases (figs. 5e-5h) such that the non-dimensionalised Hinze scale dh/d0d_{h}/d_{0} becomes smaller than unity, the shape of the deformed droplet deviates further away from the well-defined laminar aerobreakup morphology, as is already observed in Figure 4. Instead, large-scale distortion arises across the bag, and the peripheral rim becomes increasingly corrugated, forming sharp protrusions (Figure 5c) which evolve into liquid nodes (Figure 5d) [2] bearing some resemblance to the ‘handle’ structures observed by Cannon et al. [53] for ρr=μr=1\rho_{r} = \mu_{r} = 1. This also matches qualitatively with previous multiphase turbulence studies [34, 38], where the morphological evolution and breakup behaviour of the bubble or droplet relies strongly on dh/d0d_{h}/d_{0}. Similar rim protrusions are not observed in previous numerical studies of laminar aerobreakup [3, 51] but reported in experimental bag breakup studies, as they eventually pinch off to produce node drops. It is found that either the Rayleigh-Plateau (RP) or the Rayleigh-Taylor (RT) instability can explain their formation [2, 54].

Figure 5. Snapshots showing the bags formed at late times under different turbulence configurations. For (a) to (d): L0/R0=1L_0/R_0 = 1, while u∗/U0=0.25u^\ast/U_0 = 0.25 (a), 0.5 (b), 0.65 (c), and 0.8 (d). For (e) to (h): u∗/U0=0.25u^\ast/U_0 = 0.25, while L0/R0=1.5L_0/R_0 = 1.5 (e), 2 (f), 4 (g), and 8 (h). The Hinze scales dh/d0d_h/d_0 for (a)-(h) are respectively: 1.008 (a), 0.439 (b), 0.320 (c), 0.249 (d), 1.186 (e), 1.331 (f), 1.756 (g), 2.317 (h).

It is also noted that as the injection length scale L0L_{0} increases, the late-time bags recover the laminar axisymmetric shape with only an overall tilting motion, resembling the bag morphologies reported by Jiao et al. [24]. This is expected as the droplet now interacts with turbulent eddies of larger size in individual realisations, and therefore experiences less severe ambient flow velocity variation across its diameter with a fixed turbulent fluctuation intensity. Indeed, this agrees with Wang et al. [55] and Xu et al. [56] that eddies interact preferentially with particles featuring diameters closest to their sizes. Overall, the bag morphologies observed in Figure 5 suggest that ‘mild’ or ‘severe’ modifications of the bag shape by ambient turbulence can be predicted based on the non-dimensionalised Hinze scale value dh/d0d_h/d_0: the former occurs when dh/d0>1d_h/d_0 > 1, whereas the latter occurs when dh/d0≤1d_h/d_0 \le1. The validation of this heuristic criterion requires ensemble realisations at more ambient turbulence configurations, and is therefore outside the scope of the current work.

Droplet dynamics

We now seek to inspect the evolution of the centre-of-mass dynamic properties of the droplet, including both velocity and acceleration, which is closely tied to the droplet deformation history and will shed light on the overall effect of the ambient turbulence on the droplet.

In Figure 6, we present the streamwise and radial components of the droplet centre-of-mass velocity UxU_x and r˙≡Uy2+Uz2\dot{r} \equiv\sqrt{U_y^2 + U_z^2} at both viscosity ratios μr=55\mu_r = 55 and 20. Here, r˙\dot{r} is the component of the droplet velocity orthogonal to the mean flow. For example, in a laminar freestream flow which remains strictly axisymmetric, e.g. at early times, this component would be zero. The results have been ensemble-averaged to reduce variations across different individual realisations. It is observed that the introduction of air-phase turbulence leads to an increase of both UxU_x and r˙\dot{r}; in other words, ambient turbulence intermittency contributes to droplet acceleration in both the axial and radial directions. It is particularly interesting to note that according to Figure 6a and Figure 6c, the evolution patterns of the streamwise velocity UxU_x do not depend strongly on the air-phase turbulence configurations within our parameter space, including u∗u^\ast, L0L_0 and μr\mu_r; and compared with the laminar aerobreakup case, a consistent enhancement in the increasing rate of UxU_x is reached at around t=0.6τt = 0.6\tau for all turbulent aerobreakup cases, when the droplet is about to become fully flattened. On the other hand, Figure 6c suggests that increasing u∗u^* causes the radial component of the droplet velocity r˙\dot{r} to increase; although the magnitude of r˙\dot{r} still remains much smaller than its streamwise counterpart UxU_x, meaning that the droplet is transported mostly in the mean flow direction. This is expected since the large liquid-gas density ratio ρr\rho_r ensures flow separation and hence a large pressure difference between the frontal and leeward sides of the droplet, which dominates droplet acceleration even when large turbulence fluctuations in the cross-stream directions are present. It is also likely that the increase in r˙\dot{r} incorporates the contribution of irregularity in the droplet internal flows, which causes the droplet morphology to deviate further away from axisymmetry at large turbulence fluctuating speeds u∗/U0u^*/U_0.

Figure 6. Evolution of the ensemble-averaged droplet centre-of-mass velocity UxU_x (a,c) and radial velocity r˙\dot{r} (b,d), measured at different fluctuating velocities u∗u^* (a,b) and injection length scales L0L_0 (c,d).

To allow for detailed analysis of the droplet streamwise acceleration patterns, especially around t/τ=0.5t/\tau= 0.5 where the significant acceleration by ambient turbulence is observed, we differentiate UxU_x numerically in time to obtain streamwise acceleration axa_x, and non-dimensionalise the results as a drag coefficient CD,xC_{D,x} [51],

CD,x≡2mdρgU02πR02ax=83ρlρgR0U02ax.(13)C_{D,x} \equiv\frac{2m_d}{\rho_g U_0^2 \pi R_0^2}a_x = \frac{8}{3}\frac{\rho_l}{\rho_g}\frac{R_0}{U_0^2}a_x. \tag*{(13)}

Due to the presence of strong fluctuations in the instantaneous streamwise droplet acceleration axa_x, we select two time-windows for comparison: 0≤t/τ≤0.1730 \le t/\tau\le0.173 and 0.485≤t/τ≤0.6930.485 \le t/\tau\le0.693, which correspond to the early and intermediate deformation stage respectively, and present in Figure 7 the distributions of instantaneous streamwise drag coefficients CD,xC_{D,x} sampled across all ensemble realisations, rather than directly showing the original time sequence of streamwise acceleration. Figure 7a show that at early times, the distributions of CD,xC_{D,x} are largely symmetric. The mean values of CD,xC_{D,x} increase over time, as indicated by the shift of the maxima of CD,xC_{D,x} towards larger values. Figure 7b suggests that while the distribution of CD,xC_{D,x} is still largely symmetric for urms/U0≤0.5u_{\mathrm{rms}}/U_0 \le0.5, a skew towards larger CD,xC_{D,x} values is observed at larger turbulence fluctuating velocities. Interestingly, while increasing turbulence intensity u∗u^* causes the CD,xC_{D,x} distribution to broaden, it does not significantly influence the mean value of CD,xC_{D,x} as shown in the inset of Figure 7b, agreeing with our earlier observations of Figure 6a. On the other hand, Figure 7c suggests that increasing L0L_0 does not cause any consistent change in the shape of the CD,xC_{D,x} distribution at early times. Interestingly, for turbulent aerobreakup cases with L0/R0≥2L_0/R_0 \ge2, Figure 7d suggests that the CD,xC_{D,x} distributions become double-peaked, which is especially prominent for L0/R0=2L_0/R_0 = 2 and 88. The distribution for L0/R0=4L_0/R_0 = 4 also features a skew towards large CD,xC_{D,x}, while the mean values for CD,xC_{D,x} remain largely unaffected by L0L_0. We consider this change in the distribution shape of CD,xC_{D,x} to be most likely associated with the increased occurrence of large-scale velocity fluctuations within the ambient turbulence. Indeed, as the injection length scale of the turbulence becomes larger than the droplet, it is expected that the interior structure of the turbulent eddies becomes more important in controlling the instantaneous droplet acceleration. A detailed investigation of this phenomenon is left for future work.

Figure 7. Distributions of droplet streamwise drag coefficients CD,xC_{D,x} measured at different turbulence configurations. (a,c) and (b,d) are sampled within time windows 0≤t/τ≤0.1730 \le t/\tau\le0.173 and 0.485≤t/τ≤0.6930.485 \le t/\tau\le0.693, respectively. For (a,b), L0/R0=2L_0/R_0 = 2 and for (c,d) u∗/U0=0.25u^*/U_0 = 0.25. The insets show averaged drag coefficients of turbulent aerobreakup cases (scattered data) compared with their laminar counterparts (dashed line).

Agreeing with our observations in Figure 6, the insets of Figure 7 show that turbulent aerobreakup cases (shown in scattered data points) indeed feature larger CˉD,x\bar{C}_{D,x} values compared with their laminar aerobreakup counterparts, while our current results do not indicate any clear dependence of CˉD,x\bar{C}_{D,x} on the turbulence parameters u∗u^* and L0L_0. We also find that while the magnitudes of instantaneous CD,yC_{D,y} and CD,zC_{D,z} values are comparable with CD,xC_{D,x} [57], their mean values remain close to zero without any obvious evolution over time (not shown here). Overall, these results confirm our observation in Figure 6 that introducing air-phase turbulence causes a nontrivial streamwise mean acceleration effect, which most likely arises from a qualitative change in the ambient flow field that does not depend significantly on specific turbulence configurations. The recorded extreme values in CD,xC_{D,x}, on the other hand, are found to appear more frequently at large values of u∗u^* and L0L_0, which are probably associated with more extreme fluctuations in the ambient turbulence.

To further elucidate the physical mechanism leading to the droplet acceleration patterns we observed, we now inspect the distribution of air-phase pressure around the droplet. In Figure 8 we show the evolution of pressure profiles within the period of 0.485≤t/τ≤0.6930.485 \le t/\tau\le0.693 for individual realisations, when the flattened droplet is further accelerated by air-phase turbulence. These pressure profiles are sampled close to the droplet surface within the reference plane z=0z = 0, and plotted as a function of polar angle θ\theta traversing the droplet surface in the clockwise direction as shown in the inset of fig. 8a. Here, θ=0\theta= 0 corresponds to the leftmost point on the windward surface, where the pressure values have been referenced to p=0p = 0 to facilitate comparison across different turbulence configurations. Fig. 8a shows the pressure profiles obtained from laminar aerobreakup, where the minima at θ=π/2\theta= \pi/2 and 3π/23\pi/2 increase compared with the potential flow solution, a known effect of wake flow separation [3]. These laminar profiles are relatively stable without significant changes over time. Unique to the laminar aerobreakup case is the pressure maximum at θ=π\theta= \pi, which is present at all times therein except t/τ=0.624t/\tau= 0.624, when it briefly disappears. This most likely corresponds to the attached wake vortices, which oppose the movement of the leeward side of the droplet at early times [3, 58].

Figure 8. The air-phase pressure profiles close to the droplet surface, sampled within the plane z=0z = 0 at different turbulent fluctuating velocities u∗u^*. From (a) to (d): laminar airflow (a), (u∗/U0,L0/R0)=(0.25,1)(u^*/U_0, L_0/R_0) = (0.25, 1) (b), (0.5,1)(0.5, 1) (c) and (0.25,4)(0.25, 4) (d). The inset of fig. 8a shows how the angle θ\theta is calculated along the drop surface, and the potential flow solution for laminar airflow around a sphere is included in all subfigures for comparison.

As airphase turbulence is introduced, the pressure maxima at θ=π\theta= \pi are no longer consistently observed, indicating that the attached vortices are now disrupted by external turbulent forcing. Moreover, figs. 8b and 8c show that the minima in the wake region decrease and become significantly more negative than the potential-flow prediction as the turbulent fluctuating velocity u∗u^* increases, corresponding to an increase in the instantaneous pressure-induced force felt by the droplet. While our problem is transient in nature and differs from the configuration of Peng and Wang [59] where the spherical particles are fixed in turbulent ambient flows, our observations match their report of pressure-drag amplification by external turbulence. An earlier study by Rind and Castro [60] discussed this pressure decrease in the wake of a disc-shaped obstacle in detail, where it is ascribed to increased entrainment of external turbulent fluid into the separated region and the wake vortices forming closer to the drop surface. The pressure profiles are also found to change significantly within and between the sampling periods when ambient turbulence is present. While our limited pressure-field data do not permit high-frequency time averaging to fully confirm this, the time-averaged turbulent flow field in the droplet wake might feature similar pressure profiles that do not depend strongly on either u∗u^* or L0L_0, thus preventing accumulation of streamwise acceleration at larger u∗u^* or L0L_0 values and leading to the universal acceleration rate we observed in Figure 7. Indeed, while increased turbulent fluctuations lead to temporary peaks in the pressure profiles (e.g. as seen in the t/τ=0.693t/\tau= 0.693 curve of Figure 8c), the rate of pressure redistribution around the droplet increases as the ambient turbulence becomes more severe [24], so that these momentary peaks are approximately smoothed out over time, thus ensuring a net acceleration effect largely insensitive to the external turbulent forcing conditions as we observed here.

Another interesting observation from Figure 8 is that the pressure maxima do not always occur at θ=0\theta= 0. In other words, the stagnation point may drift away from the centre of the droplet frontal face under the disruption of ambient turbulence. Moreover, many instantaneous pressure profiles show extreme values appearing at θ=π/4\theta= \pi/4 and 3π/43\pi/4 in an anti-symmetric pattern (e.g., t/τ=0.693t/\tau= 0.693 in Figure 8b), suggesting a net torque acting on the flattened droplet. To further explore this, we compute the evolution of the hydrodynamic torque MM acting upon the droplet,

M=∬Sr×(−pδmn+2μgSmn) dS,(14)M = \iint_{S} \boldsymbol{r} \times\left(-p\delta_{mn} + 2\mu_g S_{mn}\right)\,dS, \tag*{(14)}

where r\boldsymbol{r} is the displacement vector from the droplet centre-of-mass to a point on its surface, δmn\delta_{mn} is the Kronecker symbol, and SijS_{ij} is the strain-rate tensor of the ambient airflow [61]. It is found that the streamwise torque component of MxM_x remains close to zero at all times, regardless of the turbulence configuration. The evolution patterns of the other two cross-stream torque components MyM_y and MzM_z are similar to each other while differing considerably from that of MxM_x, and thus we present the evolution of My2+Mz2\sqrt{M_y^2 + M_z^2} in Figure 9. Initially, My2+Mz2\sqrt{M_y^2 + M_z^2} remains small and comparable in magnitude to MxM_x (not shown in the figure) for all simulation cases up to t/τ≈0.5t/\tau\approx0.5, as is the case for the fixed particle study of Wang et al. [55] (see e.g. their fig. 3). For t/τ≥0.5t/\tau\ge0.5, the fluctuation in My2+Mz2\sqrt{M_y^2 + M_z^2} becomes significantly amplified. Since according to Figure 3, the droplet is being compressed in the streamwise direction over this period, the results shown in Figure 9a suggest that the droplet experiences much greater hydrodynamic torques as it undergoes flattening. This rapid growth in My2+Mz2\sqrt{M_y^2 + M_z^2} can potentially lead to droplet rotation and is thus the direct cause of its tilting behaviour, which we will analyse in more detail later in §V B, alongside the possible origins of such growth patterns.

Figure 9. The evolution of the hydrodynamic torque My2+Mz2\sqrt{M_y^2 + M_z^2} acting on the droplet at different turbulence fluctuating velocities u∗u^* (a) and L0L_0 (b), calculated using the pressure field in the vicinity of the droplet surface.

Droplet deformation patterns

In this section, we provide a quantitative analysis of the deformation patterns of droplets interacting with turbulent ambient airflows up to the point of bag formation. We start with quantifying the global extent and characteristics of droplet deformation using spherical harmonic decomposition in §V A, and move on to examine the droplet tilting behaviour in §V B. We finish this section with a discussion on the droplet surface corrugation patterns in §V C.

Global features

We first analyse the deformation characteristics of the droplet using spherical harmonic decomposition following Perrard et al. [33], which allows us to investigate how surface deformation patterns of different length scales evolve as the droplet undergoes strongly nonlinear surface deformation. To assess the overall droplet deformation, we introduce the root-mean-squared deformation ζΩ\zeta_{\Omega} measured from an integration over all solid angles dΩ≡dA/(r−R0)2\mathrm{d}\Omega\equiv\mathrm{d}A/(r-R_0)^2,

ζΩ2=14π∬S(r−R0)2 dΩ,(15)\zeta_{\Omega}^{2} = \frac{1}{4\pi}\iint_{S}(r-R_0)^2\,\mathrm{d}\Omega, \tag*{(15)}

Where R0R_0 is the radius of the undeformed droplet, and r−R0r-R_0 reflects the local deformation extent on the droplet interface, which is assumed to be single-valued. Figure 10a and Figure 10b show the ensemble-averaged early-time evolution of ζΩ\zeta_{\Omega} for different airphase turbulence configurations. ζΩ\zeta_{\Omega} is found to undergo an accelerated increase over time for all cases recorded, without saturating or oscillating trends observed at smaller We values in the absence of ambient mean flows [33, 62]. Moreover, turbulence is found to cause a slight increase in the total deformation for most of the turbulent aerobreakup cases in comparison with their laminar aerobreakup counterpart; except (u∗/U0,L0/R0)=(0.25,8)(u^*/U_0, L_0/R_0)=(0.25, 8) where ζΩ\zeta_{\Omega} is found to decrease instead, which is probably an effect of the ensemble size. These changes in ζΩ\zeta_{\Omega} are not dramatic, which is most likely because the turbulence-induced deformation corresponds to small-amplitude high-mode perturbations, and does not significantly modify the dominant mode-2 perturbations corresponding to droplet flattening, as will be discussed below.

Figure 10. Evolution of the ensemble-averaged droplet total spherical harmonic deformation ζΩ\zeta_{\Omega} (a,b) and mode-2 deformation a2a_2 (c,d) at different values of u∗u^{*} (a) and L0L_0.

Here we introduce the spherical harmonics Ylm(θ,ϕ)Y_l^m(\theta,\phi), and the deformation coefficient corresponding to each spherical harmonic mode al,ma_{l,m} can be calculated as follows,

al,m=14π∬S(r−R0)Ylm(θ,ϕ) dΩ,(16)a_{l,m} = \frac{1}{4\pi}\iint_{S}(r-R_0)Y_l^m(\theta,\phi)\,\mathrm{d}\Omega, \tag*{(16)}

which is realised through a Voronoi decomposition of the droplet surface during postprocessing. In the absence of a mean flow and within the limits of linear deformation, the coefficients al,ma_{l,m} are solutions to a set of damped oscillator equations [33, 63]. As harmonic modes with the same value of ll feature the same oscillation frequency, we introduce the global coefficient ala_l quantifying the total potential energy contained in all spherical harmonic modes ll,

al2=∑m=−llal,m2,(17)a_l^2 = \sum_{m=-l}^{l} a_{l,m}^2, \tag*{(17)}

which is in turn associated with the root-mean-squared global deformation ζΩ\zeta_{\Omega} (15) as follows via the orthogonality of the spherical harmonic bases,

ζΩ2=∑l=1+∞al2.(18)\zeta_{\Omega}^{2} = \sum_{l=1}^{+\infty} a_l^{2}. \tag*{(18)}

Among these global coefficients, we select the droplet centre position rc\mathbf{r}_c as the origin for the spherical coordinate system such that a1=0a_1 = 0. The mode-2 deformation coefficient a2a_2 is found to dominate all global coefficients ala_l and evolve with an increasing growth rate similar to that of ζΩ\zeta_{\Omega}. To facilitate comparison between a2a_2 and ζΩ\zeta_{\Omega}, we plot the evolution of their ratios over time in Figure 10c and Figure 10d. For most of the turbulent aerobreakup cases, a2/ζΩa_2/\zeta_{\Omega} reaches a plateau value between 0.7 and 0.8, smaller than its laminar aerobreakup counterpart of 0.9. Since mode-2 spherical harmonic deformation corresponds to prolate-oblate shape changes, this slight decrease in a2/ζΩa_2/\zeta_{\Omega} suggests that while the flattening of the droplet still largely follows a prolate-oblate pattern, the introduction of air-phase turbulence fluctuations excites higher-mode deformation patterns with smaller length scales, as we have already observed in Figure 3. Figure 10c also suggests that with increasing values of u∗u^{*}, the growth of a2/ζΩa_2/\zeta_{\Omega} for t/τ≤0.3t/\tau\le0.3 becomes slower. This might be associated with stronger surface corrugation patterns as the ambient turbulent fluctuations intensify. The influence of L0L_0 on the evolution of a2/ζΩa_2/\zeta_{\Omega} is less obvious. The evolution patterns of a2/ζΩa_2/\zeta_{\Omega} at L0/R0≥4L_0/R_0 \ge4 closely resemble their laminar aerobreakup counterpart, which is expected as the droplets experience much milder ambient flow velocity variation at dh/d0>1d_h/d_0 > 1 (see our discussions of Figure 5). For (u∗/U0,L0/R0)=(0.25,2)(u^{*}/U_0, L_0/R_0) = (0.25, 2), a2/ζΩa_2/\zeta_{\Omega} becomes much smaller at very early times as shown in Figure 10d, but have exceeded 0.5 by t/τ=0.3t/\tau= 0.3. These smaller values of a2/ζΩa_2/\zeta\Omega at early times correspond to a phase reversal in some of the individual realisations, i.e., for these cases the droplets first elongate upon insertion, and higher-mode deformation arises as the droplet relaxes from this initial elongation and begins to flatten.

Given that the early-time droplet deformation can still be largely described as a prolate-oblate shape change despite corrugation formation, we now seek to quantify the global influence of turbulence corresponding to the mode-2 spherical harmonics, while leaving local higher-order corrugation features for the curvature distribution analysis in §V C. It should be noted that, since the droplet can become distorted and/or tilted during its deformation process under strong air-phase turbulence, the definitions of bag length and width [3, 20, 51] as the streamwise and spanwise extents of the bag are not strictly applicable here. Instead, following Cannon et al. [53], we define the droplet aspect ratio I1/I3\sqrt{I_1/I_3}, which is the ratio between the smallest and largest eigenvalues of the moment-of-inertia tensor for the droplet,

TI=∫Vρ(r2I−r⊗r) dV,(19)\mathbf{T}_I = \int_V \rho(r^2\mathbf{I} - \mathbf{r} \otimes\mathbf{r})\,\mathrm{d}V, \tag*{(19)}

and show their evolution in Figure 11. We note here that I1I_1 and I3I_3 are proportional to the lengths of the minor and major axes of the spheroid, respectively.

Figure 11. Evolution of the ensemble-averaged droplet aspect ratio I1/I3\sqrt{I_1/I_3} at different values of u∗u^* (a) and L0L_0.

It is first observed that I1/I3\sqrt{I_1/I_3} initially decreases over time and stabilises around t/τ=0.75t/\tau= 0.75, for laminar and turbulent airflows satisfying u∗/U0≤0.5u^*/U_0 \le0.5. This corresponds to the flattening stage of the droplet where the bag width increases and the bag length decreases. At later stages, the laminar flow case shows a slight increase in the aspect ratio, corresponding to the late-time bag blowout phenomenon where the bag length significantly increases. The introduction of air-phase turbulence does not significantly affect the evolution of I1/I3\sqrt{I_1/I_3} at early times except for u∗=0.65u^* = 0.65; while at mid-to-late times it is found to cause a decrease in I1/I3\sqrt{I_1/I_3} compared with the laminar airflow case. Taking into account the observation of Zhao et al. [30] that liquid bags are stretched longer and wider in turbulent ambient flows, a decrease in I1/I3\sqrt{I_1/I_3} then suggests that the increase in bag width exceeds that of its length. Moreover, for turbulent ambient flows featuring u∗/U0≥0.65u^*/U_0 \ge0.65, Figure 11a shows a continued decrease of I1/I3\sqrt{I_1/I_3}, which is possibly because the air-phase turbulence is now strong enough to distort the flattened droplet so significantly that both its orientation and bag blowout dynamics are modified, as observed in Figure 5c and Figure 5d.

We also present in Figure 11 the aspect ratio at the onset of bag formation predicted by Jackiw and Ashgriz [20] for We=15\mathrm{We} = 15, which is determined as follows. According to that study, the major axis lengths for bag morphologies are approximately 2Ri/d0≈1.62R_i/d_0 \approx1.6; whereas the minor axis length is given as follows,

did0=4Werim+10.4,(20)\frac{d_i}{d_0} = \frac{4}{\mathrm{We}_{\mathrm{rim}} + 10.4}, \tag*{(20)}

where the rim Weber number is defined as Werim≡ρlR˙2d0/σWe_{\mathrm{rim}} \equiv\rho_l \dot{R}^{2}d_0/\sigma. Based on an analysis of pressure balance at the drop surface, Jackiw and Ashgriz [20] proposed the following prediction for the bag spanwise growth rate R˙\dot{R},

R˙d0/2=1τa24(1−128a2We)Tbal,(21)\frac{\dot{R}}{d_0/2}=\frac{1}{\tau}\frac{a^2}{4}\left(1-\frac{128}{a^2 We}\right)T_{\mathrm{bal}}, \tag*{(21)}

where a=6a=6 is the stretching rate at the frontal stagnation point of the droplet, and the initial flow balancing time TbalT_{\mathrm{bal}} is set as τ/8\tau/8. As such, we are able to determine the theoretical aspect ratio I1/I3\sqrt{I_1/I_3} assuming an ellipsoidal shape with axis lengths ai=bi=Ria_i=b_i=R_i and ci=di/2c_i=d_i/2,

I1I3=bi2+ci2ai2+bi2=0.712,(22)\sqrt{\frac{I_1}{I_3}}=\sqrt{\frac{b_i^2+c_i^2}{a_i^2+b_i^2}}=0.712, \tag*{(22)}

which we plotted in fig. 11 for reference. It is found that the prediction of (22) is very close to the minimum in our laminar airflow case, thus validating our aspect ratio calculations. Overall, we consider the deviation of turbulent aerobreakup aspect ratios from (22) to result from the increased droplet fore-aft pressure difference, which modifies the spanwise growth rate R˙\dot{R} (21), while the generation of higher-mode interfacial deformation might also play a role.

Droplet tilting behaviour

As the droplet flattens in the ambient air-phase turbulence, it may exhibit the tilting behaviour formerly reported by Jiao et al. [24]; namely, the principal axes of the deforming droplet no longer align with the streamwise direction. Similar phenomena have been observed in previous experimental studies of droplet aerobreakup [2, 20] and also investigations of sea spume generation [64]; which are analogous to the tumbling behaviour of anisotropic solid spheroids where the particle rotates around a non-symmetry axis [65]. In our current study, this behaviour is usually first observed at intermediate times as shown in Figure 12, when the droplet has fully flattened and not yet evolved into bag-rim structures at later times; although for one particular realisation with (u∗/U0,L0/R0)=(0.25,2)(u^*/U_0, L_0/R_0)=(0.25, 2) this occurred at very early times when the droplet is still largely spherical. Here, we have coloured the background flow field using the magnitude of airflow velocity u∗/U0u^*/U_0. These snapshots reveal that the droplet tilting phenomenon is closely associated with turbulence intermittency: while the principal axis of the droplet initially aligns largely with the streamwise direction, as shown in Figure 12a, a cluster of high-speed air parcels approaches (Figure 12b) and bypasses (Figure 12c-Figure 12d) the droplet asymmetrically from above. This high-speed air parcel modifies the pressure distribution on the droplet surface to form a net torque (see our discussion in Sec. IV), causing it to tilt over in Figure 12e. It is also noted that this high-speed air parcel does not directly impact the droplet, which might correspond to the earlier observations of Vela-Martín and Avila [66] that non-local actions by outer eddies away from the drop surface contribute significantly to the surface energy increment. For most simulation cases, the droplet will retain this oblique orientation afterwards as the interaction with these high-speed air parcels is a rare event.

Snapshots showing the tilting of a droplet in turbulent airflow

Figure 12. Snapshots showing the tilting of a droplet with air-phase turbulence characteristics u∗/U0=0.25u^{*}/U_{0}=0.25, L0/R0=2L_{0}/R_{0}=2, where the background flow field is coloured based on the magnitude of the local airflow velocity. Warmer colour represents higher magnitudes of airflow velocity in these snapshots. From (a) to (f): t/τ=1.03t/\tau=1.03, 1.09, 1.14, 1.20, 1.26, and 1.32.

To quantify the tilting phenomenon described above, we measure the droplet principal axis e1≡(θ11,θ12,θ13)\boldsymbol{e}_1\equiv(\theta_{11},\theta_{12},\theta_{13}), which can be obtained by computing the eigenvectors of the moment of inertia tensor TI\boldsymbol{T}_I (19). Here, θij\theta_{ij} is the angle between the droplet principal axis ei\boldsymbol{e}_i and the lab-frame coordinate jj; and we are particularly interested in the orientation angle θ11\theta_{11} between the droplet minor axes and the streamwise direction, since its change most clearly reflects the droplet tilting dynamics.

We then explore the possible connections between the tilting behaviour and the turbulent airflow characteristics at the scale of the droplet diameter d0d_0. Following Masuk et al. [67, 68], we introduce the slip velocity uslip\boldsymbol{u}_{\mathrm{slip}},

uslip=uˉi(x0)−uc.(23)\boldsymbol{u}_{\mathrm{slip}}=\bar{\boldsymbol{u}}_i(\boldsymbol{x}_0)-\boldsymbol{u}_c. \tag*{(23)}

Here, uˉi(x0)\bar{\boldsymbol{u}}_i(\boldsymbol{x}_0) is the coarse-grained air-phase velocity at x0\boldsymbol{x}_0, computed by averaging all velocity records within a shell enclosing the droplet given by 1.5R0≤r≤2R01.5R_0\le r\le2R_0; and we have checked that changing the upper and lower boundaries of this sampling shell does not have any discernible effect on the value of uˉi(x0)\bar{\boldsymbol{u}}_i(\boldsymbol{x}_0). As uc\boldsymbol{u}_c is the droplet centre-of-mass velocity, the slip velocity uslip\boldsymbol{u}_{\mathrm{slip}} represents the coarse-grained speed of the airflow felt by the droplet.

Figure 13 shows the evolution of the magnitude of slip velocity uslipu_{\mathrm{slip}}, where a decreasing trend is observed for all cases reported. This decrease is due to the acceleration of the droplet by the ambient airflow, and is most prominent for the early-time flattening stage where t/τ≤0.6t/\tau\leq0.6. Interestingly, we observe that at the same non-dimensional time t/τt/\tau, the slip velocities for the turbulent airflow cases mostly exceed their laminar airflow counterpart, and there does not appear to be any strong dependence of uslipu_{slip} on the turbulence configurations. This increase in uslipu_{slip} is most likely due to the high-speed turbulent fluid parcels bypassing the droplet, as we observed in Figure 12; which is also associated with the increased centre-of-mass acceleration and deformation rates we discussed earlier in §IV and §V A.

Figure 13. Evolution of the slip velocity uslipu_{slip} over time at different air-phase turbulence configurations.

We now present the evolution patterns of the droplet orientation angle θ11\theta_{11} at different turbulence configurations in Figure 14a and Figure 14b. While θ11\theta_{11} remains zero for the laminar airflow case, when turbulence is introduced, we first observe large initial values of θ11\theta_{11}, which are most likely caused by small-scale droplet surface perturbations imprinted by the turbulent airflow just after the insertion of the drop. We have already noted the effects of these small perturbations when discussing the decrease in a2/ζΩa_2/\zeta_\Omega observed in Figure 10c and Figure 10d, which decay rapidly as the droplet flattens. For most of the cases shown in Figure 14a and Figure 14b, we observe an increase in θ11\theta_{11} for t/τ≥1t/\tau\ge1; since the droplet has already finished flattening by this stage, we consider this increase in θ11\theta_{11} to represent the late-time tilting behaviour observed in Figure 12. It is also within this late stage that θ11\theta_{11} appears to mildly increase with increasing u∗u^* and L0L_0, although exceptions exist when (u∗/U0,L0/R0)=(0.5,1)(u^*/U_0, L_0/R_0) = (0.5, 1) and (0.25,2)(0.25, 2).

Figure 14. Evolution of ensemble-averaged droplet orientation angle θ11\theta_{11} (a,b), and the alignment angle ⟨e1,uslip⟩\langle\mathbf{e}_1, \mathbf{u}_{slip}\rangle between the droplet principal axis e1\mathbf{e}_1 and the slip velocity uslip\mathbf{u}_{slip} (c,d).

The previous study by Masuk et al. [68] reveals a strong preferential alignment between bubbles deforming in turbulence and the slip velocity. This is measured by the alignment angle ⟨e1,uslip⟩\langle\mathbf{e}_1, \mathbf{u}_{slip}\rangle between the minor axis of the flattening drop and the slip velocity, which is defined as,

⟨e1,uslip⟩≡arccos⁡(e1⋅uslip∥e1∥∥uslip∥)(24)\langle\mathbf{e}_1, \mathbf{u}_{slip}\rangle\equiv\arccos\left(\frac{\mathbf{e}_1 \cdot\mathbf{u}_{slip}}{\lVert\mathbf{e}_1\rVert\lVert\mathbf{u}_{slip}\rVert}\right) \tag*{(24)}

whose evolution we present in Figure 14c and Figure 14d. While the values of ⟨e1,uslip⟩\langle\mathbf{e}_1, \mathbf{u}_{slip}\rangle calculated are generally close to those of θ11\theta_{11}, larger fluctuations are observed in its evolution, which are expected due to the presence of variations in the ambient turbulent airflow and vortex shedding in the wake of the drop. Interestingly, despite the presence of scatter and the apparent increase of the alignment angle with increasing L0L_0 shown in Figure 14d, we still observe a largely decreasing trend in ⟨e1,uslip⟩\langle\mathbf{e}_1,\mathbf{u}_{slip}\rangle at early time, similar to that of θ11\theta_{11}. This suggests a close connection between the two quantities, as the coarse-grained airflow velocity felt by the droplet becomes increasingly aligned with its minor primary axis e1\mathbf{e}_1 as it flattens. This agrees with Cui et al. [65], as they observed that under the influences of the fluid inertial torque, the symmetry axes of oblate spheroids tend to align with the slip velocity.

Referring back to Figure 9a, one observes that the amplitude of the hydrodynamic torque acting on the droplet begins to rapidly increase at around the same time when θ11\theta_{11} and ⟨e1,uslip⟩\langle\mathbf{e}_1,\mathbf{u}_{slip}\rangle minimise. We believe this is due to the increased susceptibility of the flattened droplet to ambient flow field perturbations originating from the bypassing of high-speed ambient flow parcels as shown in Figure 12, even though the magnitude of the slip velocity felt by the droplet continues to decrease, as shown in Figure 13. To better demonstrate this argument, we note that previous studies prescribe the following functional form for hydrodynamic torques experienced by immersed oblate particles [61, 65],

∣MH∣=12ρg∣uslip∣2a3F(λ)sin⁡2⟨e1,uslip⟩,(25)\lvert M_H\rvert= \frac{1}{2}\rho_g\lvert\mathbf{u}_{slip}\rvert^2 a^3 F(\lambda)\sin2\langle\mathbf{e}_1,\mathbf{u}_{slip}\rangle, \tag*{(25)}

where aa is the length of the droplet major axis, and F(λ)F(\lambda) is a shape function dependent on the droplet aspect ratio λ\lambda. This prescription is strictly valid only for Stokes flow scenarios where the particle Reynolds number Rep≡ρg∣uslip∣a/μg≪1Re_p \equiv\rho_g\lvert\mathbf{u}_{slip}\rvert a/\mu_g \ll1, but available results support its usage up to Rep≈102Re_p \approx10^2 [61, 69]. We do not expect our simulations, which have RepRe_p around 10310^3, to match this result; but we note a similarity in how it connects the turbulent properties of the surrounding flow to the torque experienced by the droplet. Since it is found in Figure 14 that when the droplet fully flattens, the alignment angle ⟨e1,uslip⟩\langle\mathbf{e}_1,\mathbf{u}_{slip}\rangle remains small and does not change significantly across different turbulence configurations, whereas the slip velocity amplitude ∣uslip∣\lvert\mathbf{u}_{slip}\rvert keeps decreasing, the rapid growth of torque exerted on the droplet may be attributable to a significant growth in a3F(λ)a^3F(\lambda), which is determined by the droplet geometry. Indeed, the shape function F(λ)F(\lambda) increases rapidly with decreasing aspect ratio within the range of 0.1≤λ≤10.1 \le\lambda\le1, as shown in Figure 3 of Ref. [65], which supports our reasoning that the rapid torque growth and the droplet tilting behaviour are associated with decreasing droplet aspect ratio due to flattening. At the current stage, a more rigorous examination of the droplet tilting dynamics remains difficult compared with the tumbling behaviour of solid particles [65, 70], since the droplet is deforming and tilting at the same time. However, we consider that the discussion of θ11\theta_{11}, and the comparison with both ⟨e1,uslip⟩\langle\mathbf{e}_1,\mathbf{u}_{slip}\rangle and the hydrodynamic torque shed light on the influences of nearby extreme turbulence events on the droplet dynamics.

Formation of surface corrugations

In this section, we direct our attention towards more local morphological features present on the drop surface; more specifically the distributions of the surface velocity increment δus\delta u_s [33] and the local droplet interface curvature κ\kappa [71, 72]. Here, the surface velocity increment δus\delta u_s is defined on a reference envelope Γ\Gamma encompassing the droplet, as the difference between the radial components of the local air-phase velocity u\mathbf{u} and the mean air-phase velocity of all points located on this envelope; namely

δus(R,θ,ϕ)=u~(R)⋅r^,(26)\delta u_s(R,\theta,\phi)=\tilde{\mathbf{u}}(R)\cdot\hat{\mathbf{r}}, \tag*{(26)}
u~(R)=u−14π∬Γu(R,θ,ϕ) dΩ.(27)\tilde{\mathbf{u}}(R)=\mathbf{u}-\frac{1}{4\pi}\iint_{\Gamma}\mathbf{u}(R,\theta,\phi)\,d\Omega. \tag*{(27)}

The distance between any point on Γ\Gamma and the droplet surface is fixed as Δl=0.05R0\Delta l=0.05R_0. It should be noted that, in contrast to Perrard et al. [33] where Γ\Gamma is a fixed spherical surface, in our case Γ\Gamma follows the droplet and deforms along with it due to the presence of the mean flow. Positive δs\delta_s indicates the ambient airflow moving at velocities higher than the average value, while negative δs\delta_s values suggest lower-than-average ambient velocities. For the early stages of droplet aerobreakeup characterised by spanwise flattening, it is expected that the peripheral region of the drop mainly features positive δs\delta_s, while the frontal and back faces of the drop feature negative δs\delta_s instead.

In Figure 15a we present the probability distribution functions of δus\delta u_s normalised by its standard deviation ⟨δus⟩\langle\delta u_s\rangle at different air-phase turbulence configurations before the insertion of the droplet, where the turbulent airflow has reached a statistical steady-state. The velocity increments are computed on the spherical surface r=1.05R0r=1.05R_0 and are found to abide by a normal distribution. In Figure 15b we inspect the correlation between δus\delta u_s and the droplet local curvature κ\kappa at a fixed time t/τ=1.109t/\tau=1.109 after droplet insertion, when the droplet has fully flattened. It is found that δus\delta u_s mostly increases with κ\kappa, except for (u∗/U0,L0/R0)=(0.25,4)(u^*/U_0, L_0/R_0) = (0.25, 4) and κd0/2≥15\kappa d_0/2 \ge15. Moreover, this positive correlation does not appear to depend on the specific turbulence configurations within the range of −2d0/hi≤κd0/2≤2d0/hi-2d_0/h_i \le\kappa d_0/2 \le2d_0/h_i, where 2d0/hi=2d0/di2d_0/h_i = 2d_0/d_i is the curvature of the peripheral rim according to (20). Figs. 15c-15e present droplet surfaces at the point of bag initiation, which have been coloured to reveal the spatial distribution of the local curvature κ\kappa. Large values of κ\kappa are observed primarily at the outer edge of the peripheral rim, where large δus\delta u_s is also expected. Conversely, negative κ\kappa is found either on the ‘neck’ connecting the bag film and peripheral rim or the dimples on the bag film, and elsewhere on the bag film κ\kappa remains close to zero. This suggests that on the droplet surface, the development of highly-curved structures is closely associated with extreme events in ambient turbulence; whereas structures with medium to low local curvature are more likely associated with mean flow effects.

Figure 15. (a): Probability distribution functions of the surface velocity increment δus\delta u_s immediately before droplet insertion, measured at different u∗u^* values. (b): Correlation between δus\delta u_s and local interface curvature κ\kappa, sampled at t/τ=1.109t/\tau= 1.109 and at different u∗u^* values. The secondary axis on the top of fig. 15b shows reference curvature values corresponding to the turbulence grid size d0/Δ8d_0/\Delta_8 and droplet peripheral rim 2d0/hi2d_0/h_i. (c)-(f): Colour-mapped droplet surfaces showing the spatial distribution of local interface curvature κ\kappa for (u∗/U0,L0/R0)=(0.65,1)(u^*/U_0, L_0/R_0) = (0.65, 1) and t/τ=1.11t/\tau= 1.11, where warmer colour corresponds to larger κ\kappa.

We further inspect the evolution of δus\delta u_s over time in fig. 16a. Initially, the distribution of δus\delta u_s spans over a quite wide range of −0.75≤δus/U0≤0.5U0-0.75 \le\delta u_s/U_0 \le0.5U_0 with a flat plateau in the middle bordered by steep tails on either side, reflecting a large range of velocities present in the separated airflow around a largely spherical droplet. As time elapses, the distribution of δus\delta u_s narrows down with an increasingly steep left tail. Interestingly, while there are already two distinctive peaks present in the distribution of δus\delta u_s at a relatively early time t/τ=0.069t/\tau= 0.069, which gradually approach one another as time elapses, the monotonically decreasing right tail has developed a shoulder around δus/U0=0.25\delta u_s/U_0 = 0.25 by t/τ=0.901t/\tau= 0.901, which persists up to t/τ=1.109t/\tau= 1.109. Taking into account that the two early-time peaks are almost symmetric about δus=0\delta u_s = 0, we consider that they most likely correspond to the stagnation and separation points at the drop surface, since it is at these locations that the ambient airflow velocity minimises and maximises, respectively. The formation of the shoulder around δus/U0=0.25\delta u_s/U_0 = 0.25 suggests a breakdown of symmetry in the probability distribution function so that large positive δus\delta u_s values are more likely to be observed. Since positive δus\delta u_s is associated with the droplet periphery, the formation of this shoulder can be understood as an indicator of the torque growth and droplet tilting patterns we discussed earlier in §IV and §V B.

In contrast to the narrowing trend observed in the distributions of δus\delta u_s, Figure 16b suggests that the distributions of the local interface curvature κ\kappa broadens over time. Initially, the distribution of κ\kappa is symmetric and peaks around κd0/2=2\kappa d_0/2 = 2, which suggests the presence of small-scale perturbations on the largely spherical droplet surface due to interactions with the turbulent airflow. As the droplet deforms over time, the distribution of κ\kappa broadens and develops a skew towards large positive curvature values, representing bumps or ripple-like irregularities [71]. The peak of κ\kappa distributions has shifted to κ=0\kappa= 0 by t/τ=0.901t/\tau= 0.901, which most likely corresponds to the flat surface on either side of the disk-shaped droplet before the bag blows out.

Figure 16. Evolution of the ensemble-averaged probability distribution functions of surface velocity increment δus\delta u_s (a) and local interface curvature κ\kappa (b) over time for u∗/U0=0.25u^*/U_0 = 0.25, L0/R0=1L_0/R_0 = 1. We also include in fig. 16b the theoretical prediction of Qi et al. [72] for comparison, where the fitting parameter C1C_1 in (28) is set as 0.4.

We also include in fig. 17b the theoretical prediction of Qi et al. [72] for turbulence-bubble interactions, where it is found to partially describe the intermediate-time curvature distribution of negative κ\kappa values. This theory assumes that the concave regions on the droplets result from their interactions with energetic, sub-droplet-scale eddies with size De<d0D_e < d_0, whose inertia ρgue2\rho_g u_e^2 is sufficient to overcome the increase in surface energy σ/De\sigma/D_e caused by the local deformation; namely

ue,d=C1σρgDe,(28)u_{e,d} = \sqrt{\frac{C_1\sigma}{\rho_g D_e}}, \tag*{(28)}

where C1C_1 is a fitting parameter. The distribution of instantaneous eddy velocity ueu_e is given as follows,

P(ue∣De)=322εe2/3De−1/3P(εe),(29)P(u_e|D_e) = \frac{3\sqrt{2}}{2}\varepsilon_e^{2/3}D_e^{-1/3}P(\varepsilon_e), \tag*{(29)}

where the distribution of the local energy dissipation rate εe\varepsilon_e is in turn approximated as a log-normal function according to a multi-fractal model. Consequently, the distribution of the sizes of the eddies strong enough to deform the drop surface is expressed as

P(De)∝De2ωc∫ue,d∞P(ue∣De) due,(30)P(D_e) \propto D_e^2\omega_c\int_{u_{e,d}}^\infty P(u_e|D_e)\,du_e, \tag*{(30)}

where the droplet surface area deformed by the eddy is assumed to scale with De2D_e^2, and the frequency of collision between the droplet and the eddy is approximated as ωc∝⟨ε⟩1/3D2De−11/3\omega_c \propto\langle\varepsilon\rangle^{1/3}D^2D_e^{-11/3} by analogy with the gas kinetic theory. As the local deformation curvature can be estimated by κ≈−2/De\kappa\approx-2/D_e, this allows one to write the distribution of the local negative curvature κ\kappa as

P(κ)∝κ−2P(De).(31)P(\kappa) \propto\kappa^{-2}P(D_e). \tag*{(31)}

Similar to Qi et al. [72], we find that when setting C1=0.4C_1 = 0.4, (31) matches partially with the left tail of the simulation results up to κd0/2=−2.5\kappa d_0/2 = -2.5, where the former shows a kink. While the agreement here suggests that collision with sub-droplet-scale eddies is an important physical mechanism responsible for the dimple formation in our study, we note that the contribution of other mechanisms, such as the balance between the deforming effects of the mean flow and surface tension, cannot be accounted for by this model and still awaits further analysis. Moreover, we note that while fig. 3 in Ref. [72] also shows a skew towards positive curvature values, the right tail of their curvature distribution functions are concave-shaped and extends to very large positive values (κd0/2≈40\kappa d_0/2 \approx40); whereas our results at t/τ=1.109t/\tau= 1.109 feature convex-shaped right tails that tend to fall off abruptly at a finite value around κd0/2≈20\kappa d_0/2 \approx20. This is likely because droplets take much longer to deform and elongate, and feature much fewer ‘sharp’ regions at the time of bag formation compared with bubbles with negligible inertia.

In Figure 17 we further show the influence of air-phase turbulence on the distribution patterns of κ\kappa and δus\delta u_s. Figure 17a and Figure 17b suggest that the right tails of the κ\kappa distributions are most strongly influenced by air-phase turbulence; as compared with their laminar airflow counterpart, all turbulent aerobreakup cases feature an increased number of large positive curvature records. This indicates that local small-scale structures forming on the bag, e.g. the protruding nodes observed in Figure 5c and Figure 5d, become increasingly prominent as u∗u^* increases. According to Jackiw and Ashgriz [2], the formation of these nodes and the subsequent formation of node drops can be attributed to either RT or RP instability developing on the periphery rim. The influence of air-phase turbulence on the left tails of the curvature distributions is not as prominent as the right tails, as the left tails only slightly broaden when u∗u^* increases. This is most likely because the formation of convex regions on the drop surface with large local curvatures (rims and nodes) benefits jointly from the suction of low ambient gas pressure and capillary pinching effects, whereas creating the concave regions (dimples) requires strong small-scale eddies to act against capillary forces, as we have discussed for Figure 16b. Interestingly, we observe that the curvature distribution within the range of 0≤κd0/2≤80 \le\kappa d_0/2 \le8 does not appear to depend on the turbulence characteristics.

Figure 17. Ensemble-averaged probability distribution functions of local interface curvature κ\kappa (a,b) and surface velocity increment δus\delta u_s (c,d) sampled at t/τ=1.11t/\tau= 1.11 for air-phase turbulence with different fluctuating velocities u∗u^* (a) and injection length scales L0L_0 (b).

Figure 17c and Figure 17d show the velocity increment distributions at different air-phase turbulence configurations. Similar to the curvature distributions, it is found that the introduction of air-phase turbulence leads to a broadening effect, which becomes the strongest at the right tail for δus/U0≥0.5\delta u_s/U_0 \ge0.5. Also, similar to the curvature distribution for 0≤κd0/2≤80 \le\kappa d_0/2 \le8, the velocity increment distribution for 0≤δus/U0≤0.50 \le\delta u_s/U_0 \le0.5 does not appear to depend strongly on the turbulence characteristics. We consider that both these turbulence-independent regimes correspond to the same droplet morphological feature present in all simulations carried out in this study, which is the flattened frontal and back faces of the droplet; since it is here that the most probable values of both κ\kappa and δus\delta u_s are found.

Influence of the liquid-gas viscosity ratio μr\mu_r

As can be seen from (4), the Kolmogorov microscale η\eta is directly linked with the air-phase kinematic viscosity νg\nu_g; therefore a decrease of the viscosity ratio μr\mu_r can cause an increase in η\eta and allow the simulation to better resolve air-phase turbulence dissipation. The effects of μr\mu_r on the air-phase turbulence characteristics and the droplet centre-of-mass properties have already been covered in §II B 1 and §IV, respectively. Here, we present distributions of droplet interface curvature κ\kappa and surface velocity increment δus\delta u_s obtained at different times with reduced μr\mu_r values of 20 and 10 (only for u∗/U0=0.25u^*/U_0 = 0.25), and compare with those obtained for μr=55\mu_r = 55 in Figure 18. This allows us to provide some insight into the effects of the liquid-gas viscosity contrast on the deformation characteristics of the droplet and its interaction with the ambient turbulence.

Figure 18. Comparison of ensemble-averaged probability distribution functions of local curvature and velocity increment at t/τ=1.11t/\tau= 1.11 and μr=10\mu_r = 10, 20 and 55. (a,b): u∗/U0=0.25u^*/U_0 = 0.25; (c,d): u∗/U0=0.5u^*/U_0 = 0.5; (e,f): u∗/U0=0.65u^*/U_0 = 0.65; while for all simulation cases L0/R0=1L_0/R_0 = 1.

It is observed that for all three turbulence configurations shown in Figure 18, the local interface curvature distribution does not show any consistent change with decreasing μr\mu_r, with the only noticeable difference occurring in the left and right tails from time to time. Alongside results presented in §II B 1 and §IV, these data suggest that the liquid-gas viscosity ratio and small eddies at the Kolmogorov microscale do not significantly affect the droplet dynamics and global deformation patterns. Furthermore, the insensitivity of the droplet behaviour to μr\mu_r justifies our use of Lturb=8L_{turb}=8 for resolving only large eddies in the turbulent bulk flow. On the other hand, the velocity increment distributions show a stronger dependence on μr\mu_r; especially at a very early time t/τ=0.069t/\tau= 0.069 immediately after droplet insertion, as we observe the distribution of δus\delta u_s broadens as μr\mu_r increases. This is expected since larger values of μr\mu_r correspond to larger gas-phase Reynolds numbers ReRe, which in turn lead to stronger turbulence and more extreme velocity values recorded around the droplet.

Concluding remarks

We have presented novel direct numerical simulations of droplet deformation in a turbulent mean flow, systematically investigating various aspects of the dynamic and morphological droplet evolution patterns. Utilising a robust synthetic turbulence generation method [37], we obtained ambient airflows with validated turbulence statistics, and reproduced various aspects of droplet behaviour observed in pioneering works by Zhao et al. [30] and Jiao et al. [24], including the increase in bag size, tilting and formation of high-curvature local interfacial geometries. Our results shed light on the detailed physical mechanisms governing the droplet acceleration and deformation processes, thus laying the foundation for future works investigating the detailed turbulent bag breakup dynamics and associated fragment statistics. Here, we further discuss the possible implications of introducing air-phase turbulence on the fragment statistics resulting from bag breakup, which will be the focus of more detailed follow-up work.

Firstly, our Figure 6 and Figure 7 suggest that introducing turbulence causes the droplet to further accelerate in the streamwise direction, with some degree of randomness in the radial direction resulting from turbulence lift effects. We therefore expect promoted droplet surface instability development and draining of bag films at later times, leading to an earlier perforation time of the latter; which is already reported in previous experimental

studies [30].

Secondly, as is observed in the snapshots in Figure 5, the flattened droplet exhibits diverse morphological features not observed in laminar aerobreakup simulations [3, 51]. Among these are the distortion of the bordering rim and formation of protruding nodes, which we expect to eventually give rise to ‘node drops’ similar to the experimental observations of Jackiw and Ashgriz [2], and result in a reduction of the final volume fraction of rim drops resulting from Plateau-Rayleigh breakup [20]. Additionally, the formation of small-scale, highly-curved features on the droplet at late times, as observed in Figure 16 and Figure 17, suggests a more corrugated bag film which will disrupt the hole expansion process governed by the Taylor-Culick mechanism [3], modifying the development of hole rim instabilities and the subsequent drop shedding processes [2, 44, 73, 74]. We therefore expect the final fragment size distribution arising from turbulent aerobreakup to become broader compared with laminar aerobreakup scenarios, since the parent features of such fragments become more varied. Nonetheless, it remains a challenge to properly resolve the bag film perforation process and obtain high-fidelity fragment statistics, even with the aid of the MD algorithm for establishing grid-converged size distributions for large fragments [3, 22, 23].

Lastly, we note that the tilting behaviour of the droplet, as observed and analysed in Figure 12 and Figure 14, implies more widespread trajectories and spatial distributions of the fragments compared with the laminar bag breakup case, as they are now ejected from a highly inclined and corrugated bag surface. A more detailed understanding of the tilting dynamics would therefore deepen our understanding of the breakup of bag-shaped sea surface perturbations which also feature oblique orientations, as discussed by Troitskaya et al. [5, 17, 64].

Acknowledgment

Insights from Professor R. J. Hearst have helped in improving the analysis of air-phase turbulence characteristics in this work. The authors would like to thank EPSRC for the computational time made available on the UK supercomputing facility ARCHER2 via the UK Turbulence Consortium (EP/R029326/1). Use of the University of Oxford Advanced Research Computing (ARC) facility is also acknowledged. K. Tang is supported by a Research Studentship from the University of Oxford.

References

  1. [1]E. Villermaux, Fragmentation, Annual Review of Fluid Mechanics 39, 419 (2007).
  2. [2]I. M. Jackiw and N. Ashgriz, Prediction of the droplet size distribution in an aerodynamic droplet breakup, Journal of Fluid Mechanics 940 (2022).DOI
  3. [3]K. Tang, T. A. A. Adcock, and W. Mostert, Bag film breakup of droplets in uniform airflows, Journal of Fluid Mechanics 970, A9 (2023).arxiv.org/abs/2211.14518
  4. [4]E. Villermaux and B. Bossa, Single-drop fragmentation determines size distribution of raindrops, Nature Physics 5, 697 (2009).DOI
  5. [5]Y. Troitskaya, A. Kandaurov, O. Ermakova, D. Kozlov, D. Sergeev, and S. Zilitinkevich, Bag-breakup fragmentation as the dominant mechanism of sea-spray production in high winds, Scientific Reports 7, 1 (2017).DOI
  6. [6]L. Bourouiba, The fluid dynamics of disease transmission, Annual Review of Fluid Mechanics 53, 473 (2021).
  7. [7]P. Kant, C. Pairetti, Y. Saade, S. Popinet, S. Zaleski, and D. Lohse, Bag-mediated film atomization in a cough machine, Physical Review Fluids 8, 074802 (2023).arxiv.org/abs/2202.13949
  8. [8]B. Boyd, S. Becker, and Y. Ling, Simulation and modeling of the vaporization of a freely moving and deforming drop at low to moderate weber numbers, International Journal of Heat and Mass Transfer 218, 124735 (2024).DOI
  9. [9]D. Lohse, Fundamental fluid dynamics challenges in inkjet printing, Annual review of fluid mechanics 54, 349 (2022).DOI
  10. [10]M. Jalaal and K. Mehravaran, Transient growth of droplet instabilities in a stream, Physics of Fluids 26, 012101 (2014).DOI
  11. [11]D. R. Guildenbecher, C. López-Rivera, and P. E. Sojka, Secondary atomization, Experiments in Fluids 46, 371 (2009).DOI
  12. [12]F. Marcotte and S. Zaleski, Density contrast matters for drop fragmentation thresholds at low ohnesorge number, Physical Review Fluids 4, 103604 (2019).DOI
  13. [13]L.-P. Hsiang and G. M. Faeth, Drop deformation and breakup due to shock wave and steady disturbances, International Journal of Multiphase Flow 21, 545 (1995).
  14. [14]T. G. Theofanous, Aerobreakup of newtonian and viscoelastic liquids, Annual Review of Fluid Mechanics 43, 661 (2011).DOI
  15. [15]W. Yang, M. Jia, Z. Che, K. Sun, and T. Wang, Transitions of deformation to bag breakup and bag to bag-stamen breakup for droplets subjected to a continuous gas flow, International Journal of Heat and Mass Transfer 111, 884 (2017).DOI
  16. [16]F. Veron, Ocean spray, Annual Review of Fluid Mechanics 47, 507 (2015).
  17. [17]Y. Troitskaya, A. Kandaurov, O. Ermakova, D. Kozlov, D. Sergeev, and S. Zilitinkevich, The ”bag breakup” spume droplet generation mechanism at high winds. part i: Spray generation function, Journal of Physical Oceanography 48, 2168 (2018).DOI
  18. [18]S. Sharma, N. K. Chandra, S. Basu, and A. Kumar, Advances in droplet aerobreakeup, The European Physical Journal Special Topics , 1 (2022).
  19. [19]H. Zhao, H. Liu, J. Xu, and W. Li, Experimental study of drop size distribution in the bag breakup regime, Industrial and Engineering Chemistry Research 50, 9767 (2011).DOI
  20. [20]I. M. Jackiw and N. Ashgriz, On aerodynamic droplet breakup, Journal of Fluid Mechanics 913 (2021).DOI
  21. [21]S. S. Ade, L. D. Chandrala, and K. C. Sahu, Size distribution of a drop undergoing breakup at moderate weber numbers, Journal of Fluid Mechanics 959, A38 (2023).DOI
  22. [22]L. Chirco, J. Maarek, S. Popinet, and S. Zaleski, Manifold death: a volume of fluid implementation of controlled topological changes in thin sheets by the signature method, Journal of Computational Physics 467, 111468 (2022).DOI
  23. [23]Y. Kulkarni, C. Pairetti, R. Villiers, S. Popinet, and S. Zaleski, The atomizing pulsed jet, arXiv preprint arXiv:2405.01959 (2024).arxiv.org/abs/2405.01959
  24. [24]D. Jiao, K. Jiao, F. Zhang, and Q. Du, Direct numerical simulation of droplet deformation in turbulent flows with different velocity profiles, Fuel 247, 302 (2019).DOI
  25. [25]P. P. Sullivan and J. C. McWilliams, Dynamics of winds and currents coupled to surface waves, Annual Review of Fluid Mechanics 42, 19 (2010).DOI
  26. [26]J. Wu, S. Popinet, and L. Deike, Revisiting wind wave growth with fully coupled direct numerical simulations, Journal of Fluid Mechanics 951, A18 (2022).DOI
  27. [27]Z. Xu, T. Wang, and Z. Che, Droplet deformation and breakup in shear flow of air, Physics of Fluids 32, 10.1063/5.0006236 (2020).DOI
  28. [28]A. Vela-Martín and M. Avila, Memoryless drop breakup in turbulence, Science Advances 8, eabp9561 (2022).
  29. [29]M. Crialesi-Esposito, S. Chibbaro, and L. Brandt, The interaction of droplet dynamics and turbulence cascade, Communications Physics 6, 5 (2023).DOI
  30. [30]H. Zhao, D. Nguyen, D. J. Duke, D. Edgington-Mitchell, J. Soria, H.-F. Liu, and D. Honnery, Effect of turbulence on drop breakup in counter air flow, International Journal of Multiphase Flow 120, 103108 (2019).DOI
  31. [31]Z. Xu, T. Wang, and Z. Che, Droplet breakup in airflow with strong shear effect, Journal of Fluid Mechanics 941, A54 (2022).DOI
  32. [32]C. Rosales and C. Meneveau, Linear forcing in numerical simulations of isotropic turbulence: Physical space implementations and convergence properties, Physics of fluids 17, 095106 (2005).DOI
  33. [33]S. Perrard, A. Rivière, W. Mostert, and L. Deike, Bubble deformation by a turbulent flow, Journal of Fluid Mechanics 920, A15 (2021).DOI
  34. [34]A. Rivière, D. J. Ruth, W. Mostert, L. Deike, and S. Perrard, Capillary driven fragmentation of large gas bubbles in turbulence, Physical Review Fluids 7, 083602 (2022).arxiv.org/abs/2112.06480
  35. [35]A. Loisy and A. Naso, Interaction between a large buoyant bubble and turbulence, Physical Review Fluids 2, 014606 (2017).DOI
  36. [36]N. Kornev and E. Hassel, Method of random spots for generation of synthetic inhomogeneous turbulent fields with prescribed autocorrelation functions, Communications in numerical methods in engineering 23, 35 (2007).DOI
  37. [37]Z.-T. Xie and I. P. Castro, Efficient generation of inflow conditions for large-eddy simulation of street-scale flows, Flow Turbulence and Combustion 81, 449 (2008).DOI
  38. [38]A. Rivière, W. Mostert, S. Perrard, and L. Deike, Sub-hinze scale bubble production in turbulent bubble break-up, Journal of Fluid Mechanics 917, A40 (2021).
  39. [39]I. Rodriguez, R. Borrell, O. Lehmkuhl, C. D. P. Segarra, and A. Oliva, Direct numerical simulation of the flow over a sphere at re = 3700, Journal of Fluid Mechanics 679, 263 (2011).
  40. [40]M. S. Dodd and A. Ferrante, On the interaction of taylor length scale size droplets and isotropic turbulence, Journal of Fluid Mechanics 806, 356 (2016).DOI
  41. [41]S. Elghobashi, Direct numerical simulation of turbulent flows laden with droplets or bubbles, Annu. Rev. Fluid Mech 51, 217 (2019).DOI
  42. [42]S. Popinet, Basilisk flow solver and pde library (2019), available at: http://basilisk.fr.
  43. [43]S. Popinet, Numerical models of surface tension, Annual Review of Fluid Mechanics 50, 49 (2018).DOI
  44. [44]K. Tang, T. A. A. Adcock, and W. Mostert, Fragmentation of colliding liquid rims, Journal of Fluid Mechanics 987, A18 (2024).arxiv.org/abs/2311.07300
  45. [45]D. Legendre, A. Merle, and J. Magnaudet, Wake of a spherical bubble or a solid sphere set fixed in a turbulent environment, Physics of Fluids 18, 048102 (2006).DOI
  46. [46]X. Wu, Inflow turbulence generation methods, Annual Review of Fluid Mechanics 49, 23 (2017).DOI
  47. [47]Y. Chen, K. Dijdjeli, and Z.-T. Xie, Freestream turbulence effects on the aerodynamics of an oscillating square cylinder at the resonant frequency, Fluids 7, 329 (2022).DOI
  48. [48]Y. Kim, I. P. Castro, and Z.-T. Xie, Divergence-free turbulence inflow conditions for large-eddy simulations with incompressible flow solvers, Computers & Fluids 84, 56 (2013).DOI
  49. [49]S. B. Pope, Turbulent flows (Cambridge university press, 2000).
  50. [50]Y. Zhou, K. Nagata, Y. Sakai, H. Suzuki, Y. Ito, O. Terashima, and T. Hayase, Relevance of turbulence behind the single square grid to turbulence generated by regular-and multiscale-grids, Physics of Fluids 26 (2014).DOI
  51. [51]Y. Ling and T. Mahmood, Detailed numerical investigation of the drop aerobreakup in the bag breakup regime, Journal of Fluid Mechanics 972, A28 (2023).arxiv.org/abs/2305.00124
  52. [52]W. Mostert, S. Popinet, and L. Deike, High-resolution direct simulation of deep water breaking waves: transition to turbulence, bubbles and droplets production, Journal of Fluid Mechanics 942 (2022).DOI
  53. [53]I. Cannon, G. Soligo, and M. E. Rosti, Morphology of clean and surfactant-laden droplets in homogeneous isotropic turbulence, Journal of Fluid Mechanics 987, A31 (2024).DOI
  54. [54]H. Zhao, H.-F. Liu, W.-F. Li, and J.-L. Xu, Morphological classification of low viscosity drop bag breakup in a continuous air jet stream, Physics of Fluids 22 (2010).DOI
  55. [55]Y. Wang, A. Sierakowski, and A. Prosperetti, Rotational dynamics of a particle in a turbulent stream, Physical review fluids 4, 064304 (2019).DOI
  56. [56]X. Xu, Y. Qi, S. Zhong, S. Tan, Q. Wu, and R. Ni, Intermittency of bubble deformation in turbulence, arXiv preprint arXiv:2410.20826 (2024).DOI
  57. [57]L. Botto and A. Prosperetti, A fully resolved numerical simulation of turbulent flow past one or several spherical particles, Physics of fluids 24 (2012).DOI
  58. [58]M. Jain, R. S. Prakash, G. Tomar, and R. V. Ravikrishna, Secondary breakup of a drop at moderate weber numbers, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 471, 20140930 (2015).DOI
  59. [59]C. Peng and L.-P. Wang, Mechanisms and models of particle drag enhancements in turbulent environments, Journal of Fluid Mechanics 959, A30 (2023).DOI
  60. [60]E. Rind and J. P. Castro, On the effects of free-stream turbulence on axisymmetric disc wakes, Experiments in fluids 53, 301 (2012).DOI
  61. [61]F. Jiang, L. Zhao, H. I. Andersson, K. Gustavsson, A. Pumir, and B. Mehlig, Inertial torque on a small spheroid in a stationary uniform flow, Physical Review Fluids 6, 024302 (2021).DOI
  62. [62]I. Roa, M.-C. Renoult, C. Dumouchel, and J. C. Brändle de Motta, Droplet oscillations in a turbulent flow, Frontiers in Physics 11, 1173521 (2023).DOI
  63. [63]A. Prosperetti, Free oscillations of drops and bubbles: the initial-value problem, Journal of Fluid Mechanics 100, 333 (1980).DOI
  64. [64]Y. Troitskaya, A. Kandaурov, A. Zotova, E. V. Korsukova, and D. Sergeev, Statistical characteristics of droplets formed due to the ‘bag breakup’ fragmentation event at the interface between water and high-speed air flow, Journal of Physical Oceanography 53, 2331 (2023).DOI
  65. [65]Z. Cui, J. Qiu, X. Jiang, and L. Zhao, Effect of fluid inertial torque on the rotational and orientational dynamics of tiny spheroidal particles in turbulent channel flow, Journal of Fluid Mechanics 977, A20 (2023).DOI
  66. [66]A. Vela-Martín and M. Avila, Deformation of drops by outer eddies in turbulence, Journal of Fluid Mechanics 929, A38 (2021).DOI
  67. [67]A. U. M. Masuk, A. K. Salibindla, and R. Ni, Simultaneous measurements of deforming hinze-scale bubbles with surrounding turbulence, Journal of Fluid Mechanics 910, A21 (2021).DOI
  68. [68]A. U. M. Masuk, A. K. Salibindla, and R. Ni, The orientational dynamics of deformable finite-sized bubbles in turbulence, Journal of Fluid Mechanics 915, A79 (2021).DOI
  69. [69]R. Ouchene, Numerical simulation and modeling of the hydrodynamic forces and torque acting on individual oblate spheroids, Physics of Fluids 32 (2020).DOI
  70. [70]G. A. Voth and A. Soldati, Anisotropic particles in turbulence, Annual Review of Fluid Mechanics 49, 249 (2017).
  71. [71]F. Mangani, G. Solbigo, A. Roccon, and A. Soldati, Influence of density and viscosity on deformation, breakage, and coalescence of bubbles in turbulence, Physical Review Fluids 7, 053601 (2022).arxiv.org/abs/2309.13995
  72. [72]Y. Qi, X. Xu, S. Tan, S. Zhong, Q. Wu, and R. Ni, Breaking bubbles across multiple time scales in turbulence, Journal of Fluid Mechanics 983, A24 (2024).DOI
  73. [73]A. Vledouts, J. Quinard, N. Vandenberghe, and E. Villermaux, Explosive fragmentation of liquid shells, Journal of Fluid Mechanics 788, 246 (2016).DOI
  74. [74]B. Néel, H. Lhuissier, and E. Villermaux, ‘fines’ from the collision of liquid rims, Journal of Fluid Mechanics 893 (2020).

Paper details

vlnc:W_b7c79fss
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents