betaFeedback

WorksPhysics and Astronomy

An investigation of anisotropy in the bubbly turbulent flow via direct numerical simulations

Abstract

This study explores the dynamics of dispersed bubbly turbulent flow in a channel using interface-resolved direct numerical simulation (DNS) with an efficient Coupled Level-Set Volume-of-Fluid (CLSVOF) solver. The influence of number of bubbles (96 and 192), flow direction, and Eotvos number was examined across eight distinct cases. The results indicate that in upward flows, bubbles tend to accumulate near the wall, with smaller Eotvos numbers bringing them closer to the wall and enhancing energy dissipation through increased turbulence and vorticity. This proximity causes the liquid phase velocity to attenuate, and the bubbles, being more spherical, induce more isotropic turbulence. Conversely, in downward flows, bubbles cluster in the middle of the channel and induce additional pseudo-turbulence in the channel center, which induce additional turbulent kinetic energy in the channel center. The study further examines budget of Turbulent Kinetic Energy (TKE) and the exact balance equation for the Reynolds stresses, revealing that near-wall bubble motion generates substantial velocity gradients, particularly in the wall-normal direction, significantly impacting the turbulence structure.

Introduction

Dispersed turbulent bubbly flow plays an important role in industrial applications, such as boiling water reactors in the nuclear industry, bubble columns in the chemical metallurgical industry, and heat exchangers in thermal power plants [1, 2]. The performance of these systems depends partly on the turbulence in the liquid phase, as it significantly influences the local distribution of the dispersed phase. Furthermore, bubble fragmentation and coalescence also dictate the distribution of bubble sizes. On the other hand, it relies on the distribution of the dispersed phase, which affects both the large-scale velocity fields and the small-scale turbulence [3], thereby facilitating interactions. Additionally, parameters such as bubble velocity, bubble diameter, and the capability for bubble deformation directly impact the transfer of momentum, heat, and mass [4]. In experiments, individual measurements of bubbles and liquid velocity fluctuations present a challenge. Optical measurement techniques such as Particle Image Velocimetry (PIV), Laser Doppler Anemometry (LDA), and Particle Tracking Velocimetry (PTV) are limited to bubbly flows with a gas fraction lower than 3% due to the potential occlusion between bubbles. Overall, these methods can only provide effective measurements under limited conditions [5, 6].

Numerical studies of bubbly flows can be conducted through various approaches, primarily classified into direct numerical simulations (DNS), Euler-Lagrange (EL), and Euler-Euler (EE) methods. Among these, DNS employs an interface resolving method to accurately treat the phase boundary and fully resolve the bubble geometry along with the interfacial effects. While DNS for multiphase flows requires substantial computational resources than those of single-phase [7, 8, 9, 10, 11, 12, 13], many studies are therefore limited to the examination of individual or pairs of droplets/bubbles due to these computational demands [14, 15, 16, 17, 18]. However, with the advancement of numerical techniques, the development of DNS for multi-phase flows has advanced rapidly [19, 20, 21, 22], and studies now encompass thousands of bubbles [23, 24, 25, 26], droplets [16, 27, 28, 29], or fluid jets [30, 31]. Nevertheless, these techniques require significant computational resources and are not suitable for modeling large industrial systems. For large-scale simulations, the Euler-Euler (EE) approach coupled with Reynolds-averaged Navier-Stokes (RANS) modeling is the only viable framework [32]. The accuracy of the models largely depends on the closure models employed. Developing a suitable model for bubble-induced turbulence (BIT) is currently a central topic of investigation, providing crucial support for simulating multiphase flow engineering applications that involve bubbles.

Many studies are based on the rigorous derivation by Kataoka and Serizawa [33] of the basic equations of turbulence in gas-liquid two-phase flow. Through ensemble averaging, the local instantaneous conservation equation for averaged turbulent energy is obtained. This equation includes terms for diffusion, turbulent dissipation, turbulent production, and the interfacial transport term of turbulent energy, which incorporates the interfacial area concentration. These equations have also been utilized within the RANS modeling framework for bubbly flows, and various forms of the BIT model have been proposed, e.g., by Troshko and Hassan [34], Politano et al. [35], Lopez de Bertodano et al. [36], Colombo and Fairweather [37], and Ma et al. [38]. Based on DNS data, it is possible to accurately assess the budget of Turbulent Kinetic Energy (TKE), which provides a pathway for the modeling of BIT source terms and the evaluation of available models. However, the direct utilization of DNS data to assist BIT modeling is still rare [38]. Current BIT models frequently fail to deliver consistent accuracy across different conditions, making the use of DNS data to enhance BIT modeling a crucial area for further research.

However, EE kk-ϵ\epsilon type models for bubbly flows exhibit some issues. The numerical values required for modeling may fluctuate with changes in the void fraction, leading to deficient predictions. Simultaneously, the eddy viscosity models fail to account for the anisotropic velocity fluctuations caused by the buoyancy-generated rise of bubbles in the liquid. Measuring the anisotropy induced by bubbles based on DNS data is a key focus of this study.

An alternative to eddy-viscosity models are differential second-moment closures (SMC). Unlike eddy-viscosity models that use the kk-equation, SMC employs balance equations for the Reynolds stresses of the liquid. Kataoka et al. [39] also rigorously derived the exact Reynolds-stress equations for two-phase flows, where additional terms can similarly serve as data for DNS analysis. So far, only Ma et al. [3] has combined DNS data with the SMC framework to develop BIT models, focusing on the pressure–strain term in the application of bubbly flows.

The organization of the paper is as follows: Section II provides an overview of the DNS-related simulation details, including the governing equations, numerical methods, and the computational domain used in the study. Section III discusses key results and significant findings from the simulations. Finally, § 0.4 concludes the paper with a summary of findings.

Simulation details

Governing equations and computational method

In the analysis of incompressible flows of two immiscible fluids, the fundamental dynamics are described by the nondimensionalized Navier-Stokes equations, specified as follows:

∇⋅u=0,(1)\nabla\cdot\mathbf{u} = 0, \tag*{(1)}
∂u∂t+∇⋅(uu)=−1ρ∇p+1ρReτ∇⋅τ+1Frg+1ρWeFσ,(2)\frac{\partial\mathbf{u}}{\partial t} + \nabla\cdot(\mathbf{u}\mathbf{u}) = -\frac{1}{\rho}\nabla p + \frac{1}{\rho Re_{\tau}}\nabla\cdot\boldsymbol{\tau} + \frac{1}{Fr}\mathbf{g} + \frac{1}{\rho We}\mathbf{F}_{\sigma}, \tag*{(2)}

Here τ=μ(∇u+∇uT)\boldsymbol{\tau} = \mu(\nabla\mathbf{u} + \nabla\mathbf{u}^{T}) symbolizes the dimensionless viscous stress tensor, g\mathbf{g} represents body force and Fσ=κδ(f)n\mathbf{F}_{\sigma} = \kappa\delta(f)\mathbf{n} is surface tension force. Here, ff is identified as the Level-Set (LS) function, characterized as a signed distance from the interface. Regions with ff greater than 0 are designated as liquid, whereas regions with ff less than 0 are classified as gas. The collection of points where ff equals 0 implicitly outlines the free surface. n=∇f∣∇f∣\mathbf{n} = \frac{\nabla f}{\lvert\nabla f\rvert} denotes the unit normal vector pointing from the liquid phase to the gas phase. κ=−∇⋅n\kappa= -\nabla\cdot\mathbf{n} signifies the curvature of the interface. The purpose of the Dirac function δ(f)\delta(f) is to localize the surface tension forces Fσ\mathbf{F}_{\sigma}, ensuring that they act exclusively at the interface location. The symbols tt, ρ\rho, μ\mu, u\mathbf{u}, and pp represent the dimensionless time, density, dynamic viscosity, velocity vector and hydrodynamic pressure, respectively. The variables are normalized by the characteristic length h∗h^{*} and the friction velocity uτ∗u_{\tau}^{*}, which is defined as τw/ρ\sqrt{\tau_{w}/\rho}, τw\tau_{w} the average wall shear,in the following manner:

t=t∗h∗/uτ∗,xi=xi∗h∗,ρ=ρ∗ρc∗,μ=μ∗μc∗,ui=ui∗uτ∗,p=p∗ρc∗uτ∗2.(3)t = \frac{t^{*}}{h^{*}/u_{\tau}^{*}},\quad x_i = \frac{x_i^{*}}{h^{*}},\quad\rho= \frac{\rho^{*}}{\rho_c^{*}},\quad\mu= \frac{\mu^{*}}{\mu_c^{*}},\quad u_i = \frac{u_i^{*}}{u_{\tau}^{*}},\quad p = \frac{p^{*}}{\rho_c^{*}u_{\tau}^{*2}}. \tag*{(3)}

Here the superscript ∗* indicates dimensional quantities, while the subscripts cc and dd distinguish between the continuous and dispersed phases, respectively. Three non-dimensional numbers, Reynolds number ReτRe_{\tau}, Froude number FrFr and Weber number WeWe, are specified as follows:

Reτ=ρc∗uτ∗h∗μc∗,Fr=uτ∗2g∗h∗,We=ρc∗uτ∗2h∗σ∗,(4)Re_{\tau} = \frac{\rho_c^{*}u_{\tau}^{*}h^{*}}{\mu_c^{*}},\quad Fr = \frac{u_{\tau}^{*2}}{g^{*}h^{*}},\quad We = \frac{\rho_c^{*}u_{\tau}^{*2}h^{*}}{\sigma^{*}}, \tag*{(4)}

where σ∗\sigma^{*} is the surface tension coefficient. Using the Heaviside function

H(f)={1f>00otherwise.(5)H(f) = \begin{cases} 1 & f > 0 \\ 0 & \text{otherwise.} \end{cases} \tag*{(5)}

the fluid density and the viscosity in Eqn. (3) can be written as:

ρ=ρcH(f)+ρd(1−H(f)),(6)\rho= \rho_c H(f) + \rho_d(1-H(f)), \tag*{(6)}
μ=μcH(f)+μd(1−H(f)),(7)\mu= \mu_c H(f) + \mu_d(1-H(f)), \tag*{(7)}

Following Chang et al. [40], to regularize the viscosity, the Heaviside function HH is substituted by the subsequent smoothed version as:

H(f)={0f<−χ12(1+fχ+1πsin⁡(πfχ))∣f∣≤χ1f>χ(8)H(f) = \begin{cases} 0 & f < -\chi\\ \frac{1}{2}\left(1+\frac{f}{\chi}+\frac{1}{\pi}\sin\left(\frac{\pi f}{\chi}\right)\right) & \lvert f\rvert\le\chi\\ 1 & f > \chi \end{cases} \tag*{(8)}

where χ\chi is a function of grid size (typically χ=1.5Δ\chi= 1.5\Delta). To accurately model the moving interface of the dispersed and continuum phases, both the Lagrangian and Eulerian methods are feasible options. In the Lagrangian context, as detailed by Tryggvason et al. [41], the interface between the phases is tracked by the marker points that are defined at the interface. This method involves initially advancing the front, followed by the formulation of a grid density field tailored to align with the front’s new location, effectively capturing the dynamic evolution of interfaces within the computational domain. Front-capturing methods are represented by the level-set (LS) method and the volume-of-fluid (VOF) method [42, 43]. Sussman and Puckett [44] combined these two methods into the Coupled Level-Set Volume-of-Fluid (CLSVOF) method, which exploits the strengths of both methods. A level-set function is employed to accurately compute the position of the interface, and a VOF function is responsible for volume and mass conservation. The equations governing the interface motion are

∂f∂t+u⋅∇f=0.(9)\frac{\partial f}{\partial t} + \mathbf{u} \cdot\nabla f = 0. \tag*{(9)}

However, According to Rider and Kothe [45], the level-set method exhibits a significant mass loss problem. Therefore, the introduction of the Volume-of-Fluid (VOF) function, ψ\psi, is motivated by the necessity to conserve mass while still preserving a sharp delineation of the interface. This term denotes the discrete volume fraction within a computational cell, characterized by the integral of the Level-Set (LS) function over the cell.

ψ(t)=1Ω∫ΩH(f(x,t)) dΩ,(10)\psi(t) = \frac{1}{\Omega}\int_{\Omega} H(f(\mathbf{x},t))\,d\Omega, \tag*{(10)}

where Ω\Omega denotes the volume of a computational cell. ψ=0\psi= 0 corresponds to the continuous phase, whereas ψ=1\psi= 1 corresponds to the dispersed phase. If the cell is transected by the interface, then 0<ψ<10 < \psi< 1. The advection of the volume fraction is performed by the following mass conservation

∂ψ∂t+∇⋅(uψ)=0.(11)\frac{\partial\psi}{\partial t} + \nabla\cdot(\mathbf{u}\psi) = 0. \tag*{(11)}

The level set function is updated with the exact signed normal distance to the reconstructed interface, thereby facilitating the coupling with the volume fraction described by the VOF function. The specific iterative details and processes are detailed in Sussman and Puckett [44].

The governing equations are solved using a second-order central difference scheme for spatial discretization on a staggered and equidistant Cartesian grid. Typically used fast Poisson solvers, which employ a combination of fast Fourier transforms (FFT) and Gauss elimination, are expressed as follows:

∇⋅(1ρn+1/2∇pn+1/2)=1Δt∇⋅u^(12)\nabla\cdot\left(\frac{1}{\rho^{n+1/2}}\nabla p^{n+1/2}\right) = \frac{1}{\Delta t}\nabla\cdot\hat{\mathbf{u}} \tag*{(12)}

Due to the significant variations in the density ρ\rho at the interface between two phases in two-phase flows, the coefficient in the Poisson equation is no longer constant, causing the coefficient 1ρ\frac{1}{\rho} to vary in space and time. Consequently, iterative methods are employed to solve it, and a fast pressure-correction method developed by Dodd and Ferrante [46] is applied, as shown in the following equation:

∇2pn+1/2=∇⋅((1−ρ(0)ρn+1/2)∇p^)+ρ(0)Δt∇⋅u^,(13)\nabla^2 p^{n+1/2} = \nabla\cdot\left(\left(1-\frac{\rho^{(0)}}{\rho^{n+1/2}}\right)\nabla\hat{p}\right) + \frac{\rho^{(0)}}{\Delta t}\nabla\cdot\hat{\mathbf{u}}, \tag*{(13)}

where ρ(0)=min⁡(ρc,ρd)\rho^{(0)} = \min(\rho_c,\rho_d) for numerical stability [47], p^=2pn−1/2−pn−3/2\hat{p} = 2p^{n-1/2} - p^{n-3/2} is a linear approximation of the pressure from the previous time levels. The predicted velocity field u^\hat{\mathbf{u}} can be computed using un\mathbf{u}^n and un−1\mathbf{u}^{n-1},

u^=un+Δt(−32A(un)+12A(un−1))+Δtρn+1/2(32D(un)−12D(un−1))+ΔtFrg+ΔtρWeFσn+1/2,(14)\hat{\mathbf{u}} = \mathbf{u}^n + \Delta t\left(-\frac{3}{2}\mathcal{A}(\mathbf{u}^n) + \frac{1}{2}\mathcal{A}(\mathbf{u}^{n-1})\right) + \frac{\Delta t}{\rho^{n+1/2}}\left(\frac{3}{2}\mathcal{D}(\mathbf{u}^n) - \frac{1}{2}\mathcal{D}(\mathbf{u}^{n-1})\right) + \frac{\Delta t}{Fr}\mathbf{g} + \frac{\Delta t}{\rho We}\mathbf{F}_{\sigma}^{n+1/2}, \tag*{(14)}

where A\mathcal{A} and D\mathcal{D} are discretized forms of ∇⋅(uu)\nabla\cdot(\mathbf{u}\mathbf{u}) and 1ρReτ∇⋅τ\frac{1}{\rho Re_\tau}\nabla\cdot\boldsymbol{\tau}.

Under periodic boundary conditions, the right-hand side of equation (13) is transformed into the frequency domain via FFT. In this Fourier space, the Poisson equation becomes an algebraic equation. This equation can be efficiently solved using the Gauss elimination method, enabling the computation of pn+12p^{n+\frac{1}{2}} in Fourier space. The numerical solver’s validity and previous applications are documented in the works of Talebanfard et al. [48], Chu et al. [49], Nemati et al. [50], and Liu et al. [51].

Computational domain and boundary conditions

The DNS simulations were performed for both upward and downward flows in a rectangular channel bounded by two flat vertical walls, featuring periodic boundary conditions in the streamwise (xx) and spanwise (zz) directions, and enforcing a no-slip condition for the liquid phase on both walls. The size of the domain is Lx×Ly×Lz=6h∗×2h∗×3h∗L_x \times L_y \times L_z = 6h^* \times2h^* \times3h^* with the domain discretized for all DNS cases by a cubic mesh of 576×192×288576 \times192 \times288 points in the streamwise, wall-normal, and spanwise directions with the same step size Δ=2192\Delta= \frac{2}{192}. And h∗h^* is the half channel width. The domain length in the streamwise and spanwise directions is approximately twice that of Lu and Tryggvason [21], specifically πh∗×2h∗×πh∗2\pi h^* \times2h^* \times\frac{\pi h^*}{2}, and considerably larger than that reported by Ma et al. [38] which was 4.41h∗×h∗×2.21h∗4.41h^* \times h^* \times2.21h^*. Additionally, it is identical to the domain used in our previous publications [51]. Hence, it is assumed that the domain size is sufficiently large to enable a fully-developed turbulent field. It includes two vertical no-slip walls perpendicular to the yy-direction and utilizes periodic boundary conditions in the other directions. Figure 1(b) illustrates the domain and presents an instantaneous snapshot of the bubbly flow for one of the DNS upward cases, whereas Figure 1(a) does the same for one of the DNS downward cases. The gravitational force is oriented in the negative zz-direction. The chosen dimensionless parameters for the flow equations are as follows:

Illustration of the computational domain

Figure 1. An illustration of the computational domain: (a) shows the Instantaneous DNS data for the case 192Eö0.5U, (b) displays the case 192Eö0.5D. The vertical plane with the contour plot represents the instantaneous streamwise liquid velocity.

Reτ=ρc∗uτ∗h∗μc∗=180,Fr=uτ∗2g∗h∗=2.6422×10−2,We=ρc∗uτ∗2h∗σ∗(15)Re_\tau= \frac{\rho_c^* u_\tau^* h^*}{\mu_c^*} = 180,\qquad Fr = \frac{u_\tau^{*2}}{g^*h^*} = 2.6422 \times10^{-2},\qquad We = \frac{\rho_c^*u_\tau^{*2}h^*}{\sigma^*} \tag*{(15)}

In the simulations, a density ratio of ρd/ρc=0.03\rho_d/\rho_c = 0.03 and a viscosity ratio of μd/μc=0.018\mu_d/\mu_c = 0.018 are applied. The density of the continuous phase is specified as 1000 kg/m3^3. To describe the deformability of bubbles, two dimensionless parameters, the Eötvös number and the Morton number, are defined as follows:

Eo=(ρc∗−ρd∗)g∗D0∗2σ∗,Mo=(ρc∗−ρd∗)g∗μc∗4ρc∗σ∗3.(16)Eo = \frac{(\rho_c^*-\rho_d^*)g^*D_0^{*2}}{\sigma^*}, \qquad Mo = \frac{(\rho_c^*-\rho_d^*)g^*\mu_c^{*4}}{\rho_c^*\sigma^{*3}}. \tag*{(16)}

The specifics of the DNS cases are presented in Table I. The model is equipped with separate marker functions for each bubble within the domain, which prevents numerical coalescence in the solution; consequently, the number of bubbles remains constant over time. It is important to highlight that this work does not account for the effects of bubble breakup. A total of eight cases are primarily classified by the number of bubbles, Eötvös number, and flow direction. The initial bubble radius is kept constant, equivalent to ten grid lengths (10Δ10\Delta), and the non-dimensional initial bubble radius is defined as R0=R0∗/h∗=0.1042R_0 = R_0^*/h^* = 0.1042, consistent across all eight cases. The number of bubbles is set to either 96, corresponding to a void fraction of αG=1.263%\alpha_G = 1.263\%, or 192, corresponding to αG=2.525%\alpha_G = 2.525\%, uniformly initialized within the channel. Only two Eötvös numbers are considered: 0.5 and 2. The flow direction is categorized into upward and downward flows. For instance, ’96Eo0.5D’ indicates a case with 96 bubbles and an Eötvös number of 0.5, with ’D’ signifying downward flow, while ’U’ denotes upward flow. 20 cell points are placed on each bubble diameter D0∗D_0^*. As a reference, Lu and Tryggvason [21, 52] allocated 16 grid points per bubble diameter; Cifani et al. [23] placed 20 grid points along the bubble diameter. Additionally, Liu et al. [51] placed 31 grid points along the bubble diameter.

The driving force in the turbulent channel is represented by a constant nondimensional pressure gradient:

dpmdx=dpdx+1Frρavg,(17)\frac{d p_m}{d x} = \frac{d p}{d x} + \frac{1}{Fr}\rho_{av}g, \tag*{(17)}

where ρav\rho_{av} is the volume averaged density, and the applied driving force is dpmdx=1\frac{d p_m}{d x}=1 for the downward flow. The simulations is initially conducted for single-phase flows and has been run for a sufficient duration to ensure the full development of turbulence. Subsequently, bubbles corresponding to different cases are uniformly dispersed into the flow field. The fluid flow is then sustained, and simulations continue for over fifteen flow-through times to achieve statistical convergence.

Casebubble volume fraction αG\alpha_Gnumber of bubbles NNInitial bubble radius (R0R_0)Weber number WeWeEötvös number EoEoMorton number MoMo
96Eö0.5D1.263%960.10420.307460.51.6209×10−81.6209 \times10^{-8}
96Eö0.5U1.263%960.10420.307460.51.6209×10−81.6209 \times10^{-8}
96Eö2D1.263%960.10421.229821.0374×10−61.0374 \times10^{-6}
96Eö2U1.263%960.10421.229821.0374×10−61.0374 \times10^{-6}
192Eö0.5D2.525%1920.10420.307460.51.6209×10−81.6209 \times10^{-8}
192Eö0.5U2.525%1920.10420.307460.51.6209×10−81.6209 \times10^{-8}
192Eö2D2.525%1920.10421.229821.0374×10−61.0374 \times10^{-6}
192Eö2U2.525%1920.10421.229821.0374×10−61.0374 \times10^{-6}

Table 1. DNS case parameters for the current study are summarized as follows: The numerical labels ’96’ and ’192’ denote cases with 96 and 192 bubbles, respectively. ’Eö0.5’ and ’Eö2’ refer to Eötvös numbers of 0.5 and 2, respectively. ’U’ signifies upward flow, while ’D’ indicates downward flow.

Turbulence statistics

Figure 2 demonstrates the relationship between the bubble volume fraction αG\alpha_G and the distance to the wall yy, with subfigures (a) and (b) depicting the upward and downward cases, respectively. The volume fraction αG\alpha_G is determined by averaging over both time and spatial dimensions, including streamwise and spanwise directions. In a vertical shear flow, the lift force acting on a clean spherical bubble is directed towards the faster-moving fluid side relative to the bubble, in a frame of reference moving with the bubble. This dynamic affects the average mixture density, which depends on the number density of bubbles. Thus, lateral motion increases the density of the mixture where bubbles are lost and decreases it where bubbles accumulate. In upflow, the imposed pressure gradient that drives the flow exceeds the force of gravity, and the excess pressure gradient is balanced by the shear forces in the mixtures. As bubbles migrate from the center of the channel towards the walls, the mixture density increases until the weight of the mixture is balanced by the pressure gradient, leading to zero shear and stopping the lateral migration of the bubbles.

Figure 2. Bubble volume fraction αG\alpha_G as a function of wall distance

In the case of downward flow, the bubbles are observed to accumulate in the middle of the channel. Considering the eight cases exhibit low volume fractions, the phenomena observed align with the results presented in the studies conducted by [53] and [52]. Closer to the wall (in the region 0<y<0.20 < y < 0.2), the gas volume fraction ff is negligible for all cases. As yy continues to increase, particularly in the range of 0.2<y<0.80.2 < y < 0.8, the volume fraction profiles for all cases show a more significant rise, eventually stabilizing as they approach the central region of the channel. Since in downflow, gravity and the downward pressure gradient are in the same direction, a large pressure gradient may not be required to maintain flow. As bubbles migrate towards the center and the density decreases, the shear forces in the central area diminishes. This phenomenon is illustrated by Figure 4(b) and Figure 5(b), where both the liquid and gas velocities tend to flatten at the center of the channel, indicating that the shear force is related to the velocity gradient, i.e., the difference in speed between the bubbles and the surrounding fluid.

Figure 4. Liquid phase-averaged mean velocity profile

Figure 5. Gas phase-averaged mean velocity profile

In addition, cases with an Eötvös number of 0.5 have their peak void fraction closer to the wall relative to those with an Eötvös number of 2. The concentration of bubbles becomes more centralized within the channel as the Eötvös number rises. This phenomenon is consistent with the relationship between the Eötvös number and the void fraction described by Dabiri et al. [54]. Figure 3 displays the differences between the liquid phase-averaged mean velocity and the gas phase-averaged mean velocity in upward cases. The figure shows that, in the case of 192Eo0.5U, the velocity of the gas is higher than that of the liquid near the wall, but at y=0.1y = 0.1, the velocities of gas and liquid are the same, with this point of minimum difference coinciding with the peak of the bubbles. Similarly, in the case of 192Eo2U, the velocity of the gas is consistently higher than that of the liquid throughout, but the location where the difference between gas and liquid velocities is minimal also coincides with the peak of the bubbles.

Image not extracted

Figure 3. Comparison of velocity differences between liquid and gas phases in upward flow

The comparison of phase-averaged streamwise velocity profiles, u‾x\overline{u}_x, for the liquid phase in upward (a) and downward (b) flows, as shown in Figure 4, suggests a trend towards uniform liquid velocity in the channel’s midpoint [52]. The double bar denotes the phase-weighted averaging, which is defined by

A‾m=φmAmφ‾m,m=1,2),(18)\overline{A}_m = \frac{\varphi_m A_m}{\overline{\varphi}_m}, \quad m = 1, 2), \tag*{(18)}

The function φk\varphi_k represents the characteristic function of phase mm, which is defined as follows:

φL(x,y,z,t)=H(f(x,y,z,t)),(19)\varphi_L(x,y,z,t) = H(f(x,y,z,t)), \tag*{(19)}
φG(x,y,z,t)=1−H(f(x,y,z,t))(20)\varphi_G(x,y,z,t) = 1 - H(f(x,y,z,t)) \tag*{(20)}

In the upward cases, as shown in Figure 4(a), the mean velocity incrementally increases in the range of 0.2<y<0.80.2 < y < 0.8 and exhibits a plateau within the region of 0.8<y<10.8 < y < 1. Remarkably, there is a significant reduction in mean velocity for the cases with an Eötvös number of 0.5 when compared to those with an Eötvös number of 2. As highlighted in Lu and Tryggvason [21], upward flows with low Eötvös numbers not only exhibit a pronounced peak in the void fraction adjacent to the wall but also a substantial reduction in velocity, which reflects the trends observed. In the downward cases, as illustrated in Figure 4(b), the relatively high density of bubbles generates significant buoyancy in the channel’s bulk. This results in a smoothed-out velocity profile. The liquid phase’s mean velocity approaches a uniform value across the region where 0.4<y<10.4 < y < 1.

Figure 5 illustrates the mean velocity profiles along the streamwise direction, u‾x\overline{u}_x, for the gas phase in both upward (a) and downward (b) flows. In the downward case, the scenario with 192 bubbles, results in a flattened velocity profile in the bulk. This phenomenon occurs because the accumulation of bubbles in the bulk of the channel significantly increases buoyancy. According to the discussion in Chu et al. [49], a high volume fraction results in a noticeably more uniform mean velocity profile. The velocity profiles are noticeably reduced compared to those from the liquid phase, predominantly in scenarios with 192 bubbles. This reduction is attributed to the buoyancy effect, which acts in opposition to the direction of the main flow. The average streamwise velocity in the gas phase remains uniform with the exception of the region where 0.1<y<0.20.1 < y < 0.2. Here, the velocity gradient in the liquid phase significantly accelerates the bubbles. In the upward cases, as shown in Figure 5(a) the gas velocity with an Eötvös number of 0.5 is significantly lower than that with an Eötvös number of 2.

The turbulent intensity components ux′ux′‾\overline{u'_x u'_x}, uy′uy′‾\overline{u'_y u'_y}, uz′uz′‾\overline{u'_z u'_z} and Reynolds shear stress −ux′uy′‾-\overline{u'_x u'_y} of the liquid phase, calculated using phase averaging, are depicted for both upward and downward cases in Figure 6, compared with the single-phase DNS channel flow. The inner peaks of streamwise turbulence intensity, denoted by ux′ux′‾\overline{u'_x u'_x} in Figure 6(a), within the buffer layer at y=0.1y = 0.1, are diminished in bubbly flows, particularly exhibiting a more substantial decrease in the upward cases. In this case, the scenario with 192 bubbles is closer to the wall compared to the case with 96 bubbles, due to the higher bubble density. More bubbles serve

Figure 6. Velocity fluctuations ux′ux′‾\overline{u'_x u'_x} (a), uy′uy′‾\overline{u'_y u'_y} (c), uz′uz′‾\overline{u'_z u'_z} (e), and −ux′uy′‾-\overline{u'_x u'_y} (g) of the liquid phase for the upward case using phase averaging compared with the single-phase flow. Similarly, ux′ux′‾\overline{u'_x u'_x} (b), uy′uy′‾\overline{u'_y u'_y} (d), uz′uz′‾\overline{u'_z u'_z} (f), and −ux′uy′‾-\overline{u'_x u'_y} (h) correspond to the downward case.

as a medium for energy dissipation, helping to absorb and disperse the energy generated during fluid motion. This observation is consistent with the dissipation distribution shown in Figure 10(a). Additionally, the case with an Eötvös number of 0.5 exhibits a greater attenuation in peak, consistent with the observations by Liu et al. [51]. This is due to the inverse relationship between surface tension and the Eötvös number. A smaller Eötvös number results in a shape closer to a sphere, leading to more isotropic characteristics. In the range 0.2<y<10.2 < y < 1, the case with an Eötvös number of 0.5 more closely resembles single phase flow, due to the absence of bubbles. Meanwhile, in the range 0.2<y<0.80.2 < y < 0.8, the presence of bubbles in the case with an Eötvös number of 2 acts as a buffer, leading to a decline in ux′ux′‾\overline{u'_x u'_x}. In the downward examples, as depicted in Figure 6(b), ux′ux′‾\overline{u'_x u'_x} shows an increase at the channel center when compared to the single-phase flow. The enhanced energy in the channel center indicates that bubble clustering in that region induces extra turbulence, also referred to as pseudo-turbulence [49, 55].

Figure 10. (a) Dissipation in the upward case; (b) Dissipation in the downward case.

In the upward cases, the rise of bubbles near the wall induces a stirring motion of the liquid, specifically leading to an increase in the fluctuations of uy′uy′‾\overline{u'_y u'_y} and uz′uz′‾\overline{u'_z u'_z} within the near-wall region. In the cases studied, the void fraction for 192 bubbles closely aligns with the observations reported by Lu and Tryggvason [21], showing similar phenomena: the longitudinal vorticity is essentially zero in the middle of the channel but peaks near the walls. For 96 bubbles, within 0.2<y<10.2 < y < 1, the ux′ux′‾\overline{u'_x u'_x} is larger than in the 192-bubble cases, but the peak near the wall is lower, which is likely related to differences in the void fraction. In downward cases, an identical increase in kinetic energy at the channel centre is also evident in the wall-normal fluctuation uy′uy′‾\overline{u'_y u'_y} and spanwise fluctuation uz′uz′‾\overline{u'_z u'_z}, exhibiting a similar trend to the behaviour observed in ux′ux′‾\overline{u'_x u'_x}.

Regarding the Reynolds stress −ux′uy′‾-\overline{u'_x u'_y}, the intensity is diminished throughout the entire region, which contrasts with the central part of the channel where an increasing trend is observed in the three turbulence intensities. In single-phase turbulent flow, the Reynolds shear stress, −ux′uy′‾-\overline{u'_x u'_y} demonstrates a linear profile across a significant portion of the channel 0.2<y<10.2 < y < 1, reflecting the equilibrium between total shear stress and the pressure gradient, while the viscous contribution to the total shear stress is negligible in this region [51]. For downward bubbly flows, the linear decrease of −ux′uy′‾-\overline{u'_x u'_y} is more pronounced near the wall and diminishes more rapidly. It is notable that cases with 192 bubbles exhibit this behavior closer to the wall compared to those with 96 bubbles. Moreover, −ux′uy′‾-\overline{u'_x u'_y} approaches almost zero in the channel’s central region (y>0.8y > 0.8), where the bubble volume fraction peaks and remains relatively constant. Therefore, −ux′uy′‾-\overline{u'_x u'_y} is closely associated with the local bubble volume fraction. In upward cases, the case with 96 bubbles is more similar to a single-phase flow, whereas the case with 192 bubbles exhibits a slight linear variation. As the Eötvös number decreases, the Reynolds stress approaches zero except in a layer near the walls. Conversely, with a higher Eötvös number, the Reynolds stress maintains a linear profile throughout the channel, similar to that observed in single-phase turbulent flow. These phenomena are consistent with the observations made by Dabiri et al. [54].

As mentioned in Klein et al. [56], an increase in void fraction can increase the anisotropy, and specifically, the streamwise direction may exhibit a pronounced dominance. Therefore, it is crucial to investigate the extent of the impact of anisotropy in different cases, which will also assist in appropriately parameterizing models for RANS.

The normalized Reynolds stress anisotropy tensor, bijb_{ij}, is defined by Eq. (21), which can be utilized to describe the turbulence characteristics near the wall.

bij=ui′uj′‾uk′uk′‾−δij3(21)b_{ij} = \frac{\overline{u'_i u'_j}}{\overline{u'_k u'_k}} - \frac{\delta_{ij}}{3} \tag*{(21)}

Using the invariants of a symmetric second-order tensor, the Reynolds stress anisotropy tensor can be characterized by three principal invariants, denoted I, II, and III.

I=0,II=−bijbji2,III=bijbjkbki3.(22)I = 0, \quad\mathrm{II} = -\frac{b_{ij}b_{ji}}{2}, \quad\mathrm{III} = \frac{b_{ij}b_{jk}b_{ki}}{3}. \tag*{(22)}

Since the anisotropy tensor possesses a zero trace (bii=0b_{ii} = 0), the characterization of the Reynolds stress anisotropy tensor is fully described by its second and third invariants. The condition of anisotropy can be described using only two variables, represented by ξ\xi and η\eta, which are defined by Eq. (23) and Eq. (24), respectively. The anisotropic state of the Reynolds stresses at any point along the wall-normal direction can be represented by plotting on the ξ\xi-η\eta plane [57].

η2=−II3=bii26=bijbji6,(23)\eta^2 = -\frac{\mathrm{II}}{3} = \frac{b_{ii}^2}{6} = \frac{b_{ij}b_{ji}}{6}, \tag*{(23)}
ξ3=III2=bii36=bijbjkbki6.(24)\xi^3 = \frac{\mathrm{III}}{2} = \frac{b_{ii}^3}{6} = \frac{b_{ij}b_{jk}b_{ki}}{6}. \tag*{(24)}

Figure 7 represents the ξ\xi-η\eta plane, also known as the single-phase Lumley triangle, presents the values of (ξ,η)(\xi,\eta) derived from turbulent channel flow and a turbulent mixing layer. The Lumley triangle has three borders, two of which are straight lines connecting the origin to the points (−1/6,1/6)(-1/6,1/6) and (1/3,1/3)(1/3,1/3). The other border is a curved line defined by η=(127+2ξ3)12\eta=\left(\frac{1}{27}+2\xi^{3}\right)^{\frac{1}{2}}. In this figure, ‘iso’, ‘2C’, and ‘axi’ correspond to the isotropic state, two-component state, and axisymmetric state, respectively. The colorbar on the right corresponds to y+y^{+}, where y+=uτyνy^{+}=\frac{u_{\tau}y}{\nu} and uτu_{\tau} is the friction velocity. In the single-phase turbulent channel flow near the wall, specifically where y+<5y^{+}<5, the turbulence exhibits a two-component state. The anisotropy peaks at y+≈7y^{+}\approx7. Further along the channel, the Reynolds stress maintains a nearly axisymmetric state with a positive ξ\xi. At the core of the turbulent mixing layer, the Reynolds stresses remain nearly axisymmetric with a positive ξ\xi yet exhibit slightly less anisotropy compared to the log-law region of the channel flow.

Image not extracted

Figure 7. Sketch of the Lumley triangle on the plane of the invariants ξ\xi and η\eta. 1C, one-component; 2C, two-component.

Figure 8 presents the analysis of the anisotropy of the Reynolds stresses across all cases within the Lumley triangle. The general trends observed are largely consistent with single-phase DNS data. Notably, very close to the wall, within the viscous sublayer, the turbulence is predominantly two-component. Anisotropy peaks at a dimensionless wall distance, y+≈7y^{+}\approx7, nearing the 1C state, and it becomes progressively more isotropic towards the channel center[58]. In the upward case example, compared to the instance where the Eötvös number equals 0.5, the example with the Eötvös number equal to 2 exhibits anisotropy that is closer to that of a single phase. Figure 8(a) and (e) showcase the differences in bubble count under upward flow with an Eötvös number of 0.5. Figure 8(a) and (b) display variations under

Figure 8. Lumley triangle for 8 cases

upward flow with 192 bubbles at different Eötvös numbers, while Figure 8(e) and (f) show the differences for 96 bubbles at varying Eötvös numbers. When the Eötvös number is 0.5, bubbles are notably closer to the wall. The stirring action of the bubbles, especially in regions dense with bubbles, disrupts the orderly structure of the flow, effectively breaking up large-scale vortices and redistributing their energy into smaller-scale vortices that are more isotropic. Additionally, the upward movement of the bubbles induces extra motion in directions perpendicular to the flow, aiding in the dispersion of energy across multiple directions. This is evident from Figure 6(c) and (e), where in the case of 192 bubbles in upward flow, uy′uy′‾\overline{u'_y u'_y} and uz′uz′‾\overline{u'_z u'_z} are higher near the wall compared to other cases. Furthermore, bubble-induced turbulence enhances the isotropy of the turbulence. The distribution and motion of bubbles in the fluid increase the exchange of kinetic energy between wall-normal and spanwise directions, helping to balance velocity fluctuations across different orientations. It is noteable that in downward cases, near the wall-center region, unlike single-phase flow, the liquid tends to deviate further from isotropic behavior. This observation is associated with additional pseudo-turbulence caused by bubble aggregation, as shown in Figure 6(b), (d), and (f).

To further investigate the impact of the Eötvös number, void fraction, and flow direction on the turbulent kinetic energy (TKE), the budget terms from its transport equation are evaluated [33, 38]. The transport equation for the turbulence kinetic energy of the liquid phase is expressed as

ρLDϕLk‾Dt=Pk+Dk+ϵk−pL′uL,i′‾,i+τL,ij′uL,i′‾,j⏟Sk(25)\rho_L \frac{D\overline{\phi_L k}}{Dt}=P_k+D_k+\epsilon_k-\underbrace{\overline{p'_L u'_{L,i}}_{,i}+\overline{\tau'_{L,ij}u'_{L,i}}_{,j}}_{S_k} \tag*{(25)}

with

Pk=−ρLϕLuL,i′uL,j′‾∂u‾L,i∂xj(26)P_k=-\rho_L\overline{\phi_L u'_{L,i}u'_{L,j}}\frac{\partial\overline{u}_{L,i}}{\partial x_j} \tag*{(26)}
Dk=−∂∂xi(ϕLpL′uL,i′‾)−ρL∂∂xj(ϕLuL,i′uL,i′uL,j′‾)+∂∂xj(ϕLuL,i′τL,ij′‾)(27)D_k=-\frac{\partial}{\partial x_i}\left(\overline{\phi_L p'_L u'_{L,i}}\right)-\rho_L\frac{\partial}{\partial x_j}\left(\overline{\phi_L u'_{L,i}u'_{L,i}u'_{L,j}}\right)+\frac{\partial}{\partial x_j}\left(\overline{\phi_L u'_{L,i}\tau'_{L,ij}}\right) \tag*{(27)}
ϵk=−ϕLτL,ij′‾∂u‾L,i∂xj(28)\epsilon_k=-\overline{\phi_L\tau'_{L,ij}}\frac{\partial\overline{u}_{L,i}}{\partial x_j} \tag*{(28)}
k=12uL,i′uL,i′‾(29)k=\frac{1}{2}\overline{u'_{L,i}u'_{L,i}} \tag*{(29)}
τ‾ij=ν(∂ui∂uj+∂uj∂ui),(30)\overline{\tau}_{ij}=\nu\left(\frac{\partial u_i}{\partial u_j}+\frac{\partial u_j}{\partial u_i}\right), \tag*{(30)}

where φ‾L\overline{\varphi}_L is the indicator of the liquid phase, as shown in Eq. (19). The simple statistical averaging is indicated by a single overbar, while the double overbar denotes the phase-weighted averaging, as demonstrated in Eq. (18). Both averaging procedures involve spatial averaging in the wall-normal direction and time averaging. The variables PkP_k, DkD_k, ϵk\epsilon_k, and SkS_k represent shear production, turbulent diffusion, dissipation, and the interfacial transfer of turbulent energy between bubbles and the liquid, respectively. Additionally, ρ\rho, uu, kk and pp are the density, velocity, TKE of liquid phase and pressure of the liquid phase, respectively. The terms pL′p'_L, uL,i′u'_{L,i}, and τL,ij′\tau'_{L,ij} denote the fluctuations of pressure, the iith velocity component, and the viscous stress tensor in the liquid phase. Furthermore, nin_i represents the normal vector at the phase boundary directed towards the gas phase, and II is the interfacial area concentration.

Ini=−∂φ∂xi(31)I n_i=-\frac{\partial\varphi}{\partial x_i} \tag*{(31)}

Production is demonstrated in FIG. 9 for eight cases. Figure 9(b), which corresponds to the downward cases, shows a decay across the entire domain compared to single-phase flow. This trend is consistent with that observed in Figure 6(h). Figure 9(a) corresponds to the upward cases, where the case with an Eötvös number equal to 0.5 shows Production closer to the wall. This is associated with greater disturbances generated in the spanwise and wall-normal directions, consistent with the trends observed in Figure 6(c) and (Figure 6e). Dissipation, as illustrated by Eq. (28), is demonstrated in FIG. 10 for eight cases. Figure 10(a) represents the upward cases. For the case with an Eötvös number of 2, the curve initially rises near the wall, then decreases, reaching a minimum near y=0.1y=0.1, and subsequently monotonically increases, approaching zero after y=0.5y=0.5. For the case with an Eötvös number of 0.5, there is an overall trend of monotonic increase, and compared to the case with an Eötvös number of 2, it is closer to the wall. Additionally, the amplitude of Dissipation is greater than that of single-phase flow for y<0.2y<0.2. Figure 10(b) corresponds to the downward cases. For y>0.4y>0.4, the amplitude of Dissipation exceeds that of the single-phase flow, which can be attributed to bubble clustering.

Figure 9. (a) Production in the upward case; (b) Production in the downward case.

In Eq. (25), the single-phase term can be closed using the corresponding terms from the shear stress transport (SST) model[59]. Additionally, a source term SkRANSS_{k}^{\mathrm{RANS}} is introduced to represent the production of BIT

D(αLρLk)Dt=PkRANS+DkRANS−αLCμρLωk+CIFD(uG‾−uL‾),(32)\frac{D(\alpha_L \rho_L k)}{Dt} = P_k^{\mathrm{RANS}} + D_k^{\mathrm{RANS}} - \alpha_L C_\mu\rho_L \omega k + C_I F_D(\overline{u_G} - \overline{u_L}), \tag*{(32)}

Here, αL=φ‾\alpha_L = \overline{\varphi} is the void fraction of liquid, and ω\omega is the turbulent eddy frequency. CIC_I is a coefficient that can be determined through modeling, usually CI≤1C_I \le1. Specifically, it is assumed that the drag force, acting as the sole contributor to turbulence generation, results in all energy lost by the bubble being converted into turbulent kinetic energy in the bubble’s wake. The expression for SkRANSS_k^{\mathrm{RANS}} is given as:

SkRANS=CIFD(u‾G−u‾L),(33)S_k^{\mathrm{RANS}} = C_I F_D(\overline{u}_G-\overline{u}_L), \tag*{(33)}

where FDF_D is:

FD=34dpCDρLαG∣u‾G−u‾L∣(u‾G−u‾L).(34)F_D = \frac{3}{4d_p}C_D\rho_L\alpha_G\left|\overline{u}_G-\overline{u}_L\right|(\overline{u}_G-\overline{u}_L). \tag*{(34)}

with dpd_p, the average bubble diameter. The drag coefficient CDC_D is expressed as a function of the bubble Reynolds number Rep=∣u‾G−u‾L∣dp/νRe_p=\left|\overline{u}_G-\overline{u}_L\right|d_p/\nu and the Eötvös number EoEo, where ν\nu is the kinematic viscosity[60, 61].

CD=max⁡(CD,sphere,min⁡(CD,ellipse,CD,cap))(35)C_D = \max\left(C_{D,\mathit{sphere}},\min\left(C_{D,\mathit{ellipse}},C_{D,\mathit{cap}}\right)\right) \tag*{(35)}

where

{CD,sphere=24Re(1+0.1Re0.75)CD,ellipse=23EoCD,cap=83(36)\begin{cases} C_{D,\mathit{sphere}} = \frac{24}{Re}\left(1+0.1Re^{0.75}\right) \\ C_{D,\mathit{ellipse}} = \frac{2}{3}\sqrt{Eo} \\ C_{D,\mathit{cap}} = \frac{8}{3} \tag*{(36)} \end{cases}

Figure 11 includes four subfigures, each illustrating the distribution of the interfacial term SkS_k under different conditions and its comparison with SkRANSS_k^{\mathrm{RANS}}, based on a priori tests. From Figure 11(a) and (c), it can be observed that compared to SkRANSS_k^{\mathrm{RANS}}, the values of SkS_k are closer to the wall. This is because SkS_k represents the interfacial energy transfer between bubbles, and thus the wall-normal averaging of SkS_k reflects the position of the bubble interface. On the other hand, SkRANSS_k^{\mathrm{RANS}}, as seen from Eq. (33), is obtained through the difference in average velocities between the liquid and the gas, therefore it is closer to the center of the bubbles. Consequently, SkRANSS_k^{\mathrm{RANS}} does not align well with the SkS_k in the upward case. From Eq. (25), it can be seen that SkS_k is composed of two terms, one of which is related to the fluctuation of pressure and velocity, while the other is associated with the fluctuation of shear stress and velocity. Considering that shear stress originates from velocity gradients, SkS_k exhibits higher values near the wall-side interface. Furthermore, by comparing Figure 11(a) and (c), it is evident that in the case where the Eötvös number equals 0.5, which is closer to the wall, there is an increase in the fluctuation of both velocity and shear stress due to the proximity of the bubbles to the wall. SkS_k also neutralizes a portion of the dissipation.

Figure 11. Comparative analysis of bubble dynamics in different flow orientations and Eötvös numbers, showcasing the comparison between the interfacial term SkS_k and SkRANSS_k^{\mathrm{RANS}} based on a a priori tests: (a) upward flow with Eötvös number of 0.5, (b) downward flow with Eötvös number of 0.5, (c) upward flow with Eötvös number of 2, and (d) downward flow with Eötvös number of 2.

Thus, discussing the changes in the stress components becomes more crucial. Consequently, we introduce the exact balance equation for the Reynolds stresses, which is given as follows [39]:

DDt(ϕui′uj′‾)=Pij+Dij+ϕij+εij+SR,ij;(37)\frac{\mathrm{D}}{\mathrm{D}t}\left(\overline{\phi u_i' u_j'}\right)=\mathcal{P}_{ij}+\mathcal{D}_{ij}+\phi_{ij}+\varepsilon_{ij}+S_{R,ij}; \tag*{(37)}

To distinguish from Equation (25), an upright font is used in Equation (37). with the r.h.s. terms written as follows:

production:

Pij=−ϕui′uk′‾∂uj‾∂xk−ϕuj′uk′‾∂ui‾∂xk,(38)\mathcal{P}_{ij}=-\overline{\phi u_i'u_k'}\frac{\partial\overline{u_j}}{\partial x_k}-\overline{\phi u_j'u_k'}\frac{\partial\overline{u_i}}{\partial x_k}, \tag*{(38)}

diffusion:

Dij=−∂∂xk(1ρLϕp′(δjkui′+δikuj′)‾+ϕui′uj′uk′‾−1ρLϕ(ui′τjk′+uj′τik′)‾)(39)\mathcal{D}_{ij}=-\frac{\partial}{\partial x_k}\left(\frac{1}{\rho_L}\overline{\phi p'(\delta_{jk}u_i'+\delta_{ik}u_j')}+\overline{\phi u_i'u_j'u_k'}-\frac{1}{\rho_L}\overline{\phi(u_i'\tau'_{jk}+u_j'\tau'_{ik})}\right) \tag*{(39)}

pressure-strain:

ϕij=1ρLϕp′(∂ui′∂xj+∂uj′∂xi)‾,(40)\phi_{ij}=\frac{1}{\rho_L}\overline{\phi p'\left(\frac{\partial u_i'}{\partial x_j}+\frac{\partial u_j'}{\partial x_i}\right)}, \tag*{(40)}

dissipation:

εij=−1ρLϕτik′(∂uj′∂xk)‾−1ρLϕτjk′(∂ui′∂xk)‾.(41)\varepsilon_{ij}=-\frac{1}{\rho_L}\overline{\phi\tau'_{ik}\left(\frac{\partial u_j'}{\partial x_k}\right)}-\frac{1}{\rho_L}\overline{\phi\tau'_{jk}\left(\frac{\partial u_i'}{\partial x_k}\right)}. \tag*{(41)}

the interfacial energy transfer term:

SR,ij=−1ρL(pL′uL,j′niI‾+pL′uL,i′njI‾)+1ρL(τL,ik′uL,j′nkI‾+τL,jk′uL,i′nkI‾)(42)S_{R,ij}=-\frac{1}{\rho_L}\left(\overline{p'_L u'_{L,j}n_iI}+\overline{p'_L u'_{L,i}n_jI}\right)+\frac{1}{\rho_L}\left(\overline{\tau'_{L,ik}u'_{L,j}n_kI}+\overline{\tau'_{L,jk}u'_{L,i}n_kI}\right) \tag*{(42)}

Figure 12 presents a set of eight subfigures illustrating interfacial energy transfer. In the four downward cases, SR,ijS_{R,ij} is concentrated around the center of the channel. Among these, SR,11S_{R,11} is the dominant component, while SR,22S_{R,22} and SR,33S_{R,33} have similar values and the other three quantities are close to zero. In cases where the Eötvös number is 0.5, SR,ijS_{R,ij} is greater than in cases with an Eötvös number of 2. This is associated with the pseudo-turbulence generated by bubble clustering in the channel center, as evident from Figure 6(b), (Figure 6d), and (Figure 6f). In upward cases, the term related to shear stress makes a main contribution to SR,ijS_{R,ij}. The motion of bubbles near the wall generates substantial velocity gradients, particularly in the wall-normal direction. This increases the shear stress in that direction. Thus, in upward cases, SR,22S_{R,22} exceeds SR,33S_{R,33} in magnitude and is also closer to the wall. When the Eötvös number is 0.5, due to the bubbles being closer to the wall, SR,22S_{R,22} even surpasses SR,11S_{R,11}, becoming the dominant component in SR,ijS_{R,ij}.

Figure 12. The interfacial energy transfer SR,ijS_{R,ij} for eight cases

Apart from the interfacial term, the pressure-strain term ϕij\phi_{ij}, which includes the gradients of velocity fluctuations, stands as the sole correlation containing directional information and

plays a crucial role in capturing the anisotropy of Reynolds stress. Figure 13 displays the pressure strain for eight cases, corresponding to an a priori evaluation of the pressure-strain terms from Eq. (40). In the figures, ϕ11\phi_{11}, ϕ22\phi_{22}, and ϕ33\phi_{33} are depicted using solid lines, while the other three quantities are shown with dashed lines. Here, ϕ11\phi_{11} is negative, whereas ϕ22\phi_{22} and ϕ33\phi_{33} are positive. In downward scenarios, ϕ22\phi_{22} and ϕ33\phi_{33} display comparable values, with ϕ11\phi_{11} having a magnitude nearly twice that of ϕ22\phi_{22}. The terms ϕ13\phi_{13} and ϕ23\phi_{23} are close to zero. The cases with an Eötvös number of 0.5 exhibit a larger amplitude compared to those with an Eötvös number of 2, which is attributed to the additional pseudo-turbulence caused by bubble clustering.

Figure 13. Pressure strain for 8 cases

In the upward cases for Eötvös number 2, ϕ11\phi_{11}, ϕ22\phi_{22}, ϕ33\phi_{33}, and ϕ12\phi_{12} peak around y=0.15y = 0.15, with ϕ12\phi_{12} being more dominant than ϕ22\phi_{22} and ϕ33\phi_{33}. The term ϕ12\phi_{12} describes the momentum exchange between the streamwise and wall-normal directions, indicating significant changes in the velocity gradient in the wall-normal direction. This can be seen in Figure 6(g). Additionally, both ϕ22\phi_{22} and ϕ22\phi_{22} exhibit high values near the wall, with substantial gradients, as illustrated in Figure 6(c) and (e).

Conclusions

Dispersed bubbly turbulent flow in a channel is investigated through interface-resolved direct numerical simulation. An efficient CLSVOF solver is utilized to simulate bubble channel flow, where the initial diameter of each bubble remains constant and each bubble is represented by 20 grids per diameter. Depending on the number of bubbles (96 and 192), variations in flow direction and Eötvös number, eight cases are investigated via DNS to study the impact of these variables. The primary research findings are as follows.

In upward flow, bubbles accumulate near the wall. The smaller the Eötvös number, the closer the bubbles are to the wall, and the greater the attenuation of the liquid phase velocity. More bubbles near the wall serve as a medium for energy dissipation, helping to absorb and disperse the energy produced during fluid motion. Moreover, a smaller Eötvös number results in bubbles that are closer to spherical, producing turbulence that is more isotropic. When the Eötvös number is 0.5, uy′uy′‾\overline{u'_y u'_y} and uz′uz′‾\overline{u'_z u'_z} increase near the wall, as the wall itself is a source of disturbance, and the proximity of bubbles to the wall increases nearby turbulence and vorticity. With an Eötvös number of 0.5, bubbles are closer to the wall. The stirring action of the bubbles, especially in areas dense with bubbles, disrupts the orderly structure of the flow, effectively breaking up large-scale vortices and redistributing their energy into smaller-scale vortices that are more isotropic. Additionally, the upward motion of the bubbles induces extra motion in directions perpendicular to the flow, aiding in the dispersion of energy across multiple directions.

Analyzing the interfacial energy transfer SkS_k from the kk-equation, SkS_k consists of two components, one related to fluctuations of pressure and velocity, and the other associated with shear stress and velocity fluctuations. Given that shear stress originates from velocity gradients, SkS_k exhibits higher values near the wall-side interface. Consequently, the current model source term SkRANSS_k^{\mathrm{RANS}} in upward cases does not align well with SkS_k.

Analyzing the exact balance equation for the Reynolds stresses and its interfacial energy transfer term SR,ijS_{R,ij} and pressure-strain ϕij\phi_{ij}, the motion of bubbles near the wall generates significant velocity gradients, especially in the wall-normal direction. This results in SR,22S_{R,22} becoming a major contributor to SR,ijS_{R,ij}. ϕij\phi_{ij} points to the same conclusion.

In downward flow scenarios, bubbles are observed to cluster in the middle of the channel. The streamwise velocity of the bubbles is significantly lower than that of the liquid in the center of the channel, a consequence of the buoyancy effect due to the density difference. Bubbles in the center of the channel induce additional turbulence kinetic energy and also attenuate the energy in the buffer layer. This phenomenon is reflected in velocity fluctuations, Reynolds shear stress, the Lumley triangle, SR,ijS_{R,ij}, and ϕij\phi_{ij}.

Acknowledgments

The study is funded by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy-EXC2075-390740016. We also thank the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) for supporting this work by funding DFG-SFB 1313, Project No.327154368. X. Zhang and Y. Liu acknowledges the support from the Chinese Scholarship Council (CSC). G. Yang kindly acknowledges the support from Natural Science Foundation of China (NSFC: 52276013). The authors gratefully appreciate the access to the high performance computing facility Hawk at HLRS, Stuttgart of Germany.

References

References

  1. [1]D. Lohse, “Bubble puzzles: From fundamentals to applications,” Physical Review Fluids 3, 110504 (2018).DOI
  2. [2]S. Elghobashi, “Direct numerical simulation of turbulent flows laden with droplets or bubbles,” Annual Review of Fluid Mechanics 51, 217–244 (2019).DOI
  3. [3]Tian Ma, Dirk Lucas, Suad Jakirlić, and Jochen Fröhlich, “Progress in the second-moment closure for bubbly flow based on direct numerical simulation data,” Journal of Fluid Mechanics 883, A9 (2020).DOI
  4. [4]Yixiang Liao, Roland Rzehak, Dirk Lucas, and Eckhard Kröper, “Baseline closure model for dispersed bubbly flow: Bubble coalescence and breakup,” Chemical Engineering Science 122, 336–349 (2015).DOI
  5. [5]M. Kim, J.H. Lee, and H. Park, “Study of bubble-induced turbulence in upward laminar bubbly pipe flows measured with a two-phase particle image velocimetry,” Experimental Fluids 57, 55 (2016).DOI
  6. [6]J. Martinez Mercado, D. Chehata Gómez, D. van Gils, C. Sun, and D. Lohse, “On bubble clustering and energy spectra in pseudo-turbulence,” Journal of Fluid Mechanics 650, 287–306 (2010).
  7. [7]A. Lozano-Durán, O. Flores, and J. Jiménez, “The three-dimensional structure of momentum transfer in turbulent channels,” Journal of Fluid Mechanics 694, 100–130 (2012).DOI
  8. [8]A. Lozano-Durán and J. Jiménez, “Effect of the computational domain on direct simulations of turbulent channels up to re τ = 4200,” Physics of Fluids 26, 011702 (2014).DOI
  9. [9]X. Chu, B. Weigand, and V. Vaikuntanathan, “Flow turbulence topology in regular porous media: From macroscopic to microscopic scale with direct numerical simulation,” Physics of Fluids 30, 065102 (2018).DOI
  10. [10]X. Chu, G. Yang, S. Pandey, and B. Weigand, “Direct numerical simulation of convective heat transfer in porous media,” International Journal of Heat and Mass Transfer 133, 11–20 (2019).DOI
  11. [11]X. Chu, Y. Wu, U. Rist, and B. Weigand, “Instability and transition in an elementary porous medium,” Physical Review Fluids 5, 044304 (2020).DOI
  12. [12]S. Pandey, X. Chu, E. Laurien, and B. Weigand, “Buoyancy induced turbulence modulation in pipe flow at supercritical pressure under cooling conditions,” Physics of Fluids 30, 065105 (2018).DOI
  13. [13]W. Wang, G. Yang, C. Evrim, A. Terzis, R. Helmig, and X. Chu, “An assessment of turbulence transportation near regular and random permeable interfaces,” Physics of Fluids 33, 115103 (2021).DOI
  14. [14]J. Hasslberger, M. Klein, and N. Chakraborty, “Flow topologies in bubble-induced turbulence: A direct numerical simulation analysis,” Journal of Fluid Mechanics 857, 270–290 (2018).DOI
  15. [15]C. Pesci, A. Weiner, H. Marschall, and D. Bothe, “Computational analysis of single rising bubbles influenced by soluble surfactant,” Journal of Fluid Mechanics 856, 709–763 (2018).DOI
  16. [16]G. Yang, X. Chu, V. Vaikuntanathan, S. Wang, J. Wu, B. Weigand, and A. Terzis, “Droplet mobilization at the walls of a microfluidic channel,” Physics of Fluids 32, 012004 (2020).DOI
  17. [17]T. Yi, X. Chu, B. Wang, J. Wu, and G. Yang, “Numerical simulation of single bubble evolution in low gravity with fluctuation,” International Communications in Heat and Mass Transfer 130, 105828 (2022).DOI
  18. [18]V. Vaikuntanathan, M. Ibach, A. Arad, X. Chu, D. Katoshevski, J. B. Greenberg, and B. Weigand, “An analytical study on the mechanism of grouping of droplets,” Fluids 7, 172 (2022).DOI
  19. [19]E. A. Ervin and G. Tryggvason, “The Rise of Bubbles in a Vertical Shear Flow,” Journal of Fluids Engineering 119, 443–449 (1997).
  20. [20]I. Roghair, Y.M. Lau, N.G. Deen, H.M. Slagter, M.W. Baltussen, M. Van Sint Annaland, and J.A.M. Kuipers, “On the drag force of bubbles in bubble swarms at intermediate and high reynolds numbers,” Chemical Engineering Science 66, 3204–3211 (2011), 10th International Conference on Gas–Liquid and Gas–Liquid–Solid Reactor Engineering.DOI
  21. [21]J. Lu and G. Tryggvason, “Dynamics of nearly spherical bubbles in a turbulent channel upflow,” Journal of Fluid Mechanics 732, 166–189 (2013).DOI
  22. [22]Igor A. Bolotnov, Kenneth E. Jansen, Donald A. Drew, Assad A. Oberai, Richard T. Lahey, and Michael Z. Podowski, “Detached direct numerical simulations of turbulent two-phase bubbly channel flow,” International Journal of Multiphase Flow 37, 647–659 (2011).DOI
  23. [23]P. Cifani, J. GM. Kuerten, and B. J. Geurts, “Highly scalable DNS solver for turbulent bubble-laden channel flow,” Computers & Fluids 172, 67–83 (2018).DOI
  24. [24]X. Li, D. Gao, B. Hou, and X. Wang, “A mass-conserving lattice boltzmann method for bubble behavior estimation,” Chemical Engineering Science 193, 76–88 (2019).DOI
  25. [25]E. Trautner, M. Klein, F. Bräuer, and J. Hasslberger, “Conditional and unconditional second-order structure functions in bubbly channel flows of power-law fluids,” Physics of Fluids 33, 055121 (2021).DOI
  26. [26]L. Zhang, K. Peng, X. Shao, and J. Deng, “Direct numerical simulation of deformable rising bubbles at low reynolds numbers,” Physics of Fluids 33, 113309 (2021).DOI
  27. [27]M. S. Dodd and A. Ferrante, “On the interaction of Taylor length scale size droplets and isotropic turbulence,” Journal of Fluid Mechanics 806, 356–412 (2016).DOI
  28. [28]I. Cannon, D. Izbassarov, O. Tammisola, L. Brandt, and M. E. Rosti, “The effect of droplet coalescence on drag in turbulent channel flows,” Physics of Fluids 33, 085112 (2021).DOI
  29. [29]Yanchao Liu, Xu Chu, Guang Yang, and Bernhard Weigand, “Simulation and analytical modeling of high-speed droplet impact onto a surface,” Physics of Fluids 36 (2024).DOI
  30. [30]Y. Liu, A. Geppert, X. Chu, B. Heine, and B. Weigand, “Simulation of an annular liquid jet with a coaxial supersonic gas jet in a medical inhaler,” Atomization and Sprays 31 (2021).DOI
  31. [31]Yanchao Liu, Xu Chu, Wenkang Wang, and Bernhard Weigand, “Large-eddy simulation, convective instability, and modal causality of coaxial supersonic air–water jets considering a swirl effect,” Physics of Fluids 35 (2023).DOI
  32. [32]Mamoru Ishii and Takashi Hibiki, Thermo-fluid Dynamics of Two-Phase Flow, 1st ed. (Springer, New York, NY, 2006).
  33. [33]I. Kataoka and A. Serizawa, “Basic equations of turbulence in gas-liquid two-phase flow,” International Journal of Multiphase Flow 15, 843–855 (1989).DOI
  34. [34]A. A. Troshko and Y. A. Hassan, “A two-equation turbulence model of turbulent bubbly flows,” International Journal of Multiphase Flow 27, 1965–2000 (2001).DOI
  35. [35]M. S. Politano, P. M. Carrica, and J. Converti, “A model for turbulent polydisperse two-phase flow in vertical channels,” International Journal of Multiphase Flow 29, 1153–1182 (2003).DOI
  36. [36]M. Lopez de Bertodano, Jr. Lahey, R. T., and O. C. Jones, “Development of a k-epsilon Model for Bubbly Two-Phase Flow,” Journal of Fluids Engineering 116, 128–134 (1994).DOI
  37. [37]Marco Colombo and Michael Fairweather, “Multiphase turbulence in bubbly flows: Rans simulations,” International Journal of Multiphase Flow 77, 222–243 (2015).DOI
  38. [38]T. Ma, C. Santarelli, T. Ziegenhein, D. Lucas, and J. Fröhlich, “Direct numerical simulation–based reynolds-averaged closure for bubble-induced turbulence,” Physical Review Fluids 2, 034301 (2017).DOI
  39. [39]Isao Kataoka, D. C. Besnard, and Akimi Serizawa, “Basic equation of turbulence and modeling of interfacial transfer terms in gas-liquid two-phase flow,” Chemical Engineering Communications 118, 221–236 (1992).DOI
  40. [40]Y.C. Chang, T.Y. Hou, B. Merriman, and S. Osher, “A level set formulation of eulerian interface capturing methods for incompressible fluid flows,” Journal of Computational Physics 124, 449–464 (1996).DOI
  41. [41]G. Tryggvason, B. Bunner, A. Esmaeeli, D. Juric, N. Al-Rawahi, W. Tauber, J. Han, S. Nas, and Y. J. Jan, “A front-tracking method for the computations of multiphase flow,” Journal of Computational Physics 169, 708–759 (2001).
  42. [42]B. Lafaurie, C. Nardone, R. Scardovelli, S. Zaleski, and G. Zanetti, “Modelling merging and fragmentation in multiphase flows with SURFER,” Journal of Computational Physics 113, 134–147 (1994).DOI
  43. [43]Jan Schlottke and Bernhard Weigand, “Direct numerical simulation of evaporating droplets,” Journal of Computational Physics 227, 5215–5237 (2008).DOI
  44. [44]M. Sussman and E. G. Puckett, “A coupled level set and volume-of-fluid method for computing 3D and axisymmetric incompressible two-phase flows,” Journal of Computational Physics 162, 301–337 (2000).DOI
  45. [45]William Rider and Douglas Kothe, “Stretching and tearing interface tracking methods,” (1995), https://doi.org/10.2514/6.1995-1717.
  46. [46]M. S. Dodd and A. Ferrante, “A fast pressure-correction method for incompressible two-fluid flows,” Journal of Computational Physics 273, 416–434 (2014).DOI
  47. [47]J. Shen S. Dong, “A time-stepping scheme involving constant coefficient matrices for phase-field simulations of two-phase incompressible flows with large density ratios,” Journal of Computational Physics 231, 5788–5804 (2012).DOI
  48. [48]N. Talebanfard, H. Nemati, and B. J. Boersma, “Heat transfer in deforming droplets with a direct solver for a coupled level-set and volume of fluid method,” International Communications in Heat and Mass Transfer 108, 104272 (2019).DOI
  49. [49]X. Chu, Y. Liu, W. Wang, G. Yang, B. Weigand, and H. Nemati, “Turbulence, pseudo-turbulence, and local flow topology in dispersed bubbly flow,” Physics of Fluids 32, 083310 (2020).DOI
  50. [50]H. Nemati, W.-P. Breugem, M. Kwakkel, and B. J. Boersma, “Direct numerical simulation of turbulent bubbly down flow using an efficient clsvof method,” International Journal of Multi-phase Flow 135, 103500 (2021).DOI
  51. [51]Yanchao Liu, Wenkang Wang, Guang Yang, Hassan Nemati, and Xu Chu, “The interfacial modes and modal causality in a dispersed bubbly turbulent flow,” Physics of Fluids 35, 083309 (2023).DOI
  52. [52]J. Lu and G. Tryggvason, “Numerical study of turbulent bubbly downflows in a vertical channel,” Physics of Fluids 18, 103302 (2006).DOI
  53. [53]X. Chu, W. Wang, J. Müller, H. Von Schöning, Y. Liu, and B. Weigand, “Turbulence modulation and energy transfer in turbulent channel flow coupled with one-side porous media,” in High Performance Computing in Science and Engineering’20: Transactions of the High Performance Computing Center, Stuttgart (HLRS) 2020 (Springer, 2021) pp. 373–386.DOI
  54. [54]Sadegh Dabiri, Jiacai Lu, and Gretar Tryggvason, “Transition between regimes of a vertical channel bubbly upflow due to bubble deformability,” Physics of Fluids 25, 102110 (2013).DOI
  55. [55]V. Pandey, R. Ramadugu, and P. Perlekar, “Liquid velocity fluctuations and energy spectra in three-dimensional buoyancy-driven bubbly flows,” Journal of Fluid Mechanics 884 (2020).DOI
  56. [56]M. Klein, T. Trummler, and J. Radtke, “Multiscale analysis of the Reynolds stress, dissipation, and subgrid-scale tensor in turbulent bubbly channel flows: Characterization of anisotropy and modeling implications,” Physics of Fluids 34, 085122 (2022).DOI
  57. [57]Kwing-So Choi and John L. Lumley, “The return to isotropy of homogeneous turbulence,” Journal of Fluid Mechanics 436, 59–84 (2001).
  58. [58]Stephen B. Pope, Turbulent Flows (Cambridge University Press, Cambridge, 2000).
  59. [59]F. R. Menter, “Two-equation eddy-viscosity turbulence models for engineering applications,” AIAA Journal 32, 1598–1605 (1994).DOI
  60. [60]Marco Colombo, Roland Rzehak, Michael Fairweather, Yixiang Liao, and Dirk Lucas, “Benchmarking of computational fluid dynamic models for bubbly flows,” Nuclear Engineering and Design 375, 111075 (2021).DOI
  61. [61]Mamoru Ishii and Novak Zuber, “Drag coefficient and relative velocity in bubbly, droplet or particulate flows,” AIChE Journal 25, 843–855 (1979).DOI

Paper details

vlnc:W_cfcr87e8
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