betaFeedback

WorksPhysics and Astronomy

When is settling important for particle concentrations in wall-bounded turbulent flows?

Abstract

We explore the role of gravitational settling on inertial particle concentrations in a wall-bounded turbulent flow. While it may be thought that settling can be ignored when the settling parameter Sv≡vs/uτSv\equiv v_s/u_τ is small (vsv_s - Stokes settling velocity, uτu_τ - fluid friction velocity), we show that even in this regime the settling may make a leading order contribution to the concentration profiles. This is because the importance of settling is determined, not by the size of vsv_s compared with uτu_τ or any other fluid velocity scale, but by the size of vsv_s relative to the other mechanisms that control the vertical particle velocity and concentration profile. We explain this in the context of the particle mean-momentum equation, and show that in general, there always exists a region in the boundary layer where settling cannot be neglected, no matter how small SvSv is (provided it is finite). Direct numerical simulations confirm the arguments, and show that the near-wall concentration is highly dependent on SvSv even when Sv≪1Sv\ll 1, and can reduce by an order of magnitude when SvSv is increased from O(10−4)O(10^{-4}) and O(10−2)O(10^{-2}). The results also show that the preferential sampling of ejection events in the boundary layer by inertial particles when Sv=0Sv=0 is profoundly altered as SvSv is increased, and is replaced by a preferential sampling of sweep events due to the onset of the preferential sweeping mechanism.

Introduction

In many particle-laden, wall-bounded turbulent flows, it is important to characterize the near-wall distribution of a dispersed phase. In the environment, surface emission of particulate matter (e.g. dust, aerosols) is often estimated by assuming a relationship between the mean concentration and surface flux. This flux-profile relationship is also the basis for wall models of heavy scalar transport [1].

While many past studies have focused on the phenomena of turbophoresis [2, 3] and the interactions of particles with near-wall coherent structures [4, 5, 6], much less attention has been given to the settling of particles through wall-bounded turbulence and the impact this has on the particle concentrations. Indeed, although it is well-known that particles can experience an inertial enhancement to their settling velocity in isotropic, homogeneous turbulence [7, 8], the interplay between settling and inertial effects, especially near the wall, is much less understood. In a recent study we explored this question using theory and direct numerical simulations (DNS) to consider how the mechanisms governing the particle settling and concentration profiles vary with distance from the wall [9]. It was shown that sufficiently far from the wall, the main mechanism that modifies the settling compared to the Stokes settling velocity is the enhancement due to the preferential sweeping mechanism of Maxey [7]. Closer to the wall, however, this mechanism becomes subleading and instead the turbophoretic drift velocity [2] dominates the drift of the particles towards the wall. These changes in the mechanisms controlling the particle settling velocity directly impact the particle concentration profiles, since the two quantities are connected via the particle continuity equation [9]. These results highlight the particular mechanisms that models which account for settling, but ignore particle inertia (e.g. Rouse [10], Li et al. [11]) must incorporate if they are to be extended to the regime of finite particle inertia.

In many studies that are aimed at understanding the behavior of inertial particles in wall-bounded turbulence, the effect of gravity is often neglected, usually under the assumption that the settling velocity is small. The relevant nondimensional parameter which describes this, however, is often assumed to be the ratio of the Stokes settling velocity vs≡τpgv_s \equiv\tau_p g (where τp\tau_p is the particle response time and gg is the gravitational acceleration) to the flow friction velocity uτu_\tau. What we aim to demonstrate both theoretically and numerically, however, is that near the wall, even small Stokes settling velocities can be comparable or larger than the other contributions to the particle settling velocity (e.g. the turbophoretic velocity) that generate concentration buildup near the wall. This in turn disrupts commonly assumed balances which ultimately determine the wall-normal distribution of the particles.

Theoretical analysis

We consider the vertical motion of small, heavy particles subject to Stokes drag and gravitational forces [12]

ddtwp(t)=1τp(up(t)−wp(t))−g,(1)\frac{d}{dt}w^p(t) = \frac{1}{\tau_p}\left(u^p(t)-w^p(t)\right)-g, \tag*{(1)}

where wp(t)w^p(t) is the vertical particle velocity and up(t)u^p(t) is the vertical fluid velocity at the particle position. The system will also be assumed to have sufficiently low volume fraction so that one-way coupling can be assumed together with the absence of particle-particle collisions.

In assessing the importance of gravitational settling on particle motion in turbulent flows, it is typical to define a settling parameter such as τpg/uη\tau_p g/u_\eta [13], where uηu_\eta is the Kolmogorov velocity scale] or in the context of boundary layers τpg/uτ\tau_p g/u_\tau [3], and then to conclude that the effect of settling can be neglected when these non-dimensional parameters are small. Nevertheless, in some works, the effect of gravitational settling has been shown to be important even when such parameters are small. For example, DNS results in Bragg et al. [9], Richter and Chamecki [14] showed that even when τpg/uτ=O(10−2)\tau_{pg}/u_{\tau}=O(10^{-2}), the effect of gravitational settling on the particle motion in the boundary layer was strong, and in some parameter regimes, made a leading order contribution to the particle motion. As we will now show, this is because quantities such as τpg/uτ\tau_{pg}/u_{\tau} are inappropriate measures of the importance of settling on the particle motion in the boundary layer.

In the following, all quantities are normalized using the fluid friction timescale τ\tau and velocity scale uτu_{\tau} to express them in wall units, usually denoted by the superscript ++. However, in what follows, we drop the superscript for notational simplicity.

Using phase-space Probability Density Function (PDF) equations, we can construct transport equations for the average concentration ρ=⟨δ(zp(t)−z)⟩\rho=\langle\delta(z^{p}(t)-z)\rangle, where zp(t)z^{p}(t) is the vertical particle position, zz is the time-independent vertical position coordinate (with z=0z=0 corresponding to the wall), and ⟨⋅⟩z\langle\cdot\rangle_{z} denotes an ensemble average conditioned on zp(t)=zz^{p}(t)=z. The continuity equation governing ρ(z,t)\rho(z,t) is [9]

∂tρ+∇z[ρ⟨wp(t)⟩z]=0.(2)\partial_{t}\rho+\nabla_{z}\left[\rho\langle w^{p}(t)\rangle_{z}\right]=0. \tag*{(2)}

Equation (2) can be solved by specifying ⟨wp(t)⟩z\langle w^{p}(t)\rangle_{z}, for which an expression can be obtained from the particle mean-momentum equation [9]

⟨wp(t)⟩z=⟨up(t)⟩z−Sv−St(Dt⟨wp(t)⟩z+∇zW+Wρ−1∇zρ),(3)\langle w^{p}(t)\rangle_{z}=\langle u^{p}(t)\rangle_{z}-Sv-St\left(D_{t}\langle w^{p}(t)\rangle_{z}+\nabla_{z}\mathcal{W}+\mathcal{W}\rho^{-1}\nabla_{z}\rho\right), \tag*{(3)}

where Dt≡∂t+⟨wp(t)⟩z∇zD_{t}\equiv\partial_{t}+\langle w^{p}(t)\rangle_{z}\nabla_{z}, W≡⟨(wp(t)−⟨wp(t)⟩z)2⟩z\mathcal{W}\equiv\langle(w^{p}(t)-\langle w^{p}(t)\rangle_{z})^{2}\rangle_{z} is the variance of the vertical particle velocity, St≡τp/τSt\equiv\tau_{p}/\tau is the Stokes number, Sv≡τpg/uτSv\equiv\tau_{p}g/u_{\tau} is the settling number. Hereafter, the mean vertical velocity ⟨wp(t)⟩z\langle w^{p}(t)\rangle_{z} will sometimes be referred to as the settling velocity, which is to be distinguished from the Stokes settling velocity SvSv.

The term ⟨up(t)⟩z\langle u^{p}(t)\rangle_{z} in (3) is a mean velocity that arises from the particles preferentially sampling the underlying turbulent flow [3], that vanishes for fully-mixed fluid particles [15]. The term Dt⟨wp(t)⟩zD_{t}\langle w^{p}(t)\rangle_{z} is the mean particle acceleration. The term −St∇zW-St\nabla_{z}\mathcal{W} is the turbophoretic velocity which arises due to the combination of turbulence inhomogeneity and particle inertia [2]. Finally, the term −StWρ−1∇zρ-St\mathcal{W}\rho^{-1}\nabla_{z}\rho is a diffusive velocity that arises from decoupling between the particle velocity and the local fluid velocity, and is only finite when the concentration field is non-uniform. Detailed explanations of each of the terms in (3) may be found in Bragg et al. [9].

We will focus on the steady-state regime for which the solution to (2) is that the flux ρ⟨wp(t)⟩z\rho\langle w^{p}(t)\rangle_{z} is a constant (that depends upon the boundary conditions and system parameters). For Sv=0Sv=0, a possible steady-state is the zero-flux configuration, for which ⟨wp(t)⟩z=0\langle w^{p}(t)\rangle_{z}=0. For Sv>0Sv>0, unless resuspension mechanisms at the wall are sufficiently strong to overcome the weight of the particles, a zero-flux configuration will not be established, and instead a constant, negative flux will be established with ⟨wp(t)⟩z<0\langle w^p(t)\rangle_z < 0. In this paper we consider this constant negative flux regime, although the analysis and its implications could easily be extended to the zero-flux configuration.

The regime St≪1St \ll1 but finite SvSv corresponds to the regime of negligible particle inertia, but non-negligible settling that was analyzed by Rouse [10]. Our interest, by contrast, is the regime Sv≪1Sv \ll1 and finite StSt, and to understand whether in this regime the contribution from SvSv to the total vertical velocity ⟨wp(t)⟩z\langle w^p(t)\rangle_z in (3) may be ignored. When the contribution of SvSv to ⟨wp(t)⟩z\langle w^p(t)\rangle_z can be ignored, then it follows from (2) that the concentration profile will be independent of SvSv. Equation (3) is regularly perturbed with respect to SvSv, i.e. the limit of (3) for Sv→0Sv \to0 is equal to (3) when setting Sv=0Sv = 0. To explore if SvSv may be neglected if it is small but non-zero we introduce

Λ(z)≡⟨up(t)⟩z−St(Dt⟨wp(t)⟩z+∇zW+Wρ−1∇zρ),(4)\Lambda(z) \equiv\langle u^p(t)\rangle_z - St(D_t\langle w^p(t)\rangle_z + \nabla_z W + W\rho^{-1}\nabla_z\rho), \tag*{(4)}

so that ⟨wp(t)⟩z=Λ−Sv\langle w^p(t)\rangle_z = \Lambda- Sv, and expand Λ\Lambda in SvSv to obtain

Λ(z)=∑n=0∞SvnΛn(z),(5)\Lambda(z) = \sum_{n=0}^{\infty} Sv^n\Lambda_n(z), \tag*{(5)}
⟨wp(t)⟩z=Λ0(z)+Sv(Λ1(z)−1)+O(Sv2),(6)\langle w^p(t)\rangle_z = \Lambda_0(z) + Sv(\Lambda_1(z) - 1) + O(Sv^2), \tag*{(6)}

where Λ0(z)=Λ(z)∣Sv=0\Lambda_0(z) = \Lambda(z)|_{Sv=0}. For settling to be ignored we require that Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv). To consider when this condition is satisfied, we will first focus on the regime z≪1z \ll1 where analytical results are possible, and which is also the regime of most interest since this is where the particle concentration is highest. It is also worth emphasizing that although the following analysis is strictly for the limited regime z≪1z \ll1, the asymptotic results to be used in the analysis are known to hold up to z=O(10)z = O(10) [3, 16], and therefore our results should also apply up to z=O(10)z = O(10).

To show whether the condition Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv) is satisfied, we must describe each of the terms contributing to Λ0(z)\Lambda_0(z) in the near wall region. It was shown in Sikovsky [16] using asymptotic analysis that for z≪1z \ll1 and Sv=0Sv = 0, ρ∼z−γ\rho\sim z^{-\gamma} (in what follows we ignore the coefficients in the asymptotic relationships since it is the dependence on zz that will be of interest) where γ=max⁡[α0,3]\gamma= \max[\alpha_0,3], and α0(St)∈[0,4]\alpha_0(St) \in[0,4] so that γ(St)∈[3,4]\gamma(St) \in[3,4]. The asymptotic analysis also shows that for sufficiently large StSt, an additional contribution in the expansion for ρ\rho becomes important which generates ρ∼constant\rho\sim\text{constant} in the limit St→∞St \to\infty, corresponding to particles moving ballistically through the flow. For simplicity, we will restrict our focus in this analysis to low to moderate inertia particles for which ρ∼z−γ\rho\sim z^{-\gamma} describes the correct behavior. Since ⟨wp(t)⟩z\langle w^p(t)\rangle_z is constant, the result ρ∼z−γ\rho\sim z^{-\gamma} then implies ⟨wp(t)⟩z∼zγ\langle w^p(t)\rangle_z \sim z^\gamma and Dt⟨wp(t)⟩z∼z2γ−1D_t\langle w^p(t)\rangle_z \sim z^{2\gamma-1}. Using these results, we will now consider whether the condition Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv) is satisfied for weak to moderate inertia particles.

In the weak inertia regime St≪1St \ll1 with Sv=0Sv = 0, W∼z4\mathcal{W} \sim z^4 [3]. Furthermore, the model in Sikovsky [16] for ⟨up(t)⟩z\langle u^p(t)\rangle_z yields the behavior ⟨up(t)⟩z∼z3\langle u^p(t)\rangle_z \sim z^3 for arbitrary StSt in the regime where ρ∼z−γ\rho\sim z^{-\gamma}. Therefore, substituting these asymptotic results into the definition of Λ0\Lambda_0 we obtain

Λ0(z)∼z3−St(z2γ−1+z3+z3).(7)\Lambda_0(z) \sim z^3 - St(z^{2\gamma-1} + z^3 + z^3). \tag*{(7)}

For St≪1St \ll1, γ=3\gamma= 3 [16] and so (7) has the limiting form Λ0∼z3\Lambda_0 \sim z^3. This shows that the condition Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv), which must be satisfied if settling is to be ignored, will be violated for z→0z \to0 if Sv>0Sv > 0. Hence for St≪1St \ll1 there will always exist a region near the wall where gravitational settling cannot be ignored, even if Sv≪1Sv \ll1 (but finite).

For particles with moderate inertia and for z≪1z \ll1, Sikovsky [16] showed that the particle velocity moments obey the asymptotic result ⟨(wp(t)−⟨wp(t)⟩z)n⟩z∼zγ\langle(w^p(t) - \langle w^p(t)\rangle_z)^n\rangle_z \sim z^\gamma for n≥2n \ge2, and therefore W∼zγ\mathcal{W} \sim z^\gamma. Using this we obtain

Λ0(z)∼z3−St(z2γ−1+zγ−1+zγ−1).(8)\Lambda_0(z) \sim z^3 - St(z^{2\gamma-1} + z^{\gamma-1} + z^{\gamma-1}). \tag*{(8)}

Since γ∈[3,4]\gamma\in[3,4], the leading behavior is Λ0∼zγ−1\Lambda_0 \sim z^{\gamma-1}. Therefore, just as for the St≪1St \ll1 case, this also shows that the condition Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv) will be violated for moderately inertial particles for z→0z \to0 if Sv>0Sv > 0.

We have therefore shown that for both weakly and moderately inertial particles, even if Sv≪1Sv \ll1, there will always exist a region close to the wall where settling cannot be ignored, and indeed where settling makes a leading order contribution to ⟨wp(t)⟩z\langle w^p(t)\rangle_z and therefore ρ\rho. Further way from the wall it is also possible that the condition Λ0(z)≫O(Sv)\Lambda_0(z) \gg O(Sv) could be violated since Λ0(z)\Lambda_0(z) depends on gradients in the flow statistics, and these become weak for sufficiently large zz (at least if the flow Reynolds number is large enough for a quasi-homogeneous region to emerge sufficiently far from the wall). To explore the role of settling on the particle dynamics throughout the boundary layer, we will now consider DNS results for inertial particle motion in an open channel flow. These results will enable us to go beyond the analysis in this section by allowing for a quantitative assessment of the impact of settling on the particle concentration profiles in the regime Sv≪1Sv \ll1.

Direct Numerical Simulations

DNS is used to solve the incompressible Navier-Stokes equations, which are then used in a one-way coupled scenario to solve for the motion of heavy, inertial, point-particles that are subject to drag and gravitational forces via the equation of motion

Schematic of particle-laden flow with particle inlet, outlet, wall, free surface, and downward flux

Figure 1. Particles are initialized at the top of the domain and settle down. When they reach the wall they are removed, and replaced by another particle that is introduced at the top of the domain.

d2dt2xp(t)≡ddtvp(t)=1τp(up(t)−vp(t))−g,(9)\frac{d^2}{dt^2}x^p(t) \equiv\frac{d}{dt}v^p(t) = \frac{1}{\tau_p}\left(u^p(t)-v^p(t)\right)-\boldsymbol{g}, \tag*{(9)}

where xp(t),vp(t)x^p(t),v^p(t) are the particle position and velocity, up(t)u^p(t) is the fluid velocity at the particle position, and g\boldsymbol{g} is the gravitational acceleration in the vertical direction. The flow has a friction Reynolds number of Reτ=315Re_\tau=315, and is generated by applying a constant streamwise pressure gradient to force the flow. The streamwise xx and spanwise yy directions are periodic, and the wall at z=0z=0 imposes a no-slip condition on the fluid velocity field. At the upper wall, z=1z=1, a free-slip (i.e., zero-stress) condition is imposed on the fluid velocity. The domain size is Lx×Ly×Lz=2π×2π×1L_x \times L_y \times L_z = 2\pi\times2\pi\times1 with a corresponding grid of Nx×Ny×Nz=128×256×128N_x \times N_y \times N_z = 128 \times256 \times128. The grid is stretched in the wall-normal direction and thus the simulations have a resolution of Δx+×Δy+×Δz+=15.4×7.73×0.25(wall),4.49(center)\Delta_x^+ \times\Delta_y^+ \times\Delta_z^+ = 15.4 \times7.73 \times0.25(wall), 4.49(center). The unladen flow field from DNS has been tested and validated by comparison with published data for Reτ=40−950Re_\tau=40-950 [17, 18]. Inertial particles are introduced into the flow at z=1z=1 with an initial velocity equal to the Stokes settling velocity plus the local fluid velocity. When they eventually settle to the wall at z=0z=0 they are removed and replaced by another particle injected at z=1z=1. This system is then simulated until the flow reaches steady-state. The point-particle DNS code has been validated for inertial particles in the range St=30−2000St=30-2000 [19]. This setup provides a canonical case of steady-state, particle-laden wall-bounded turbulence which provides a uniform downwards flux of particle concentration; figure 1 provides a schematic of the configuration.

In order to test the idea presented in the previous section, it would be desirable to compare results for Sv=0Sv=0 with those from 0<Sv≪10<Sv\ll1 in order to see the impact of settling when SvSv is small. However, at finite StSt and with the absorbing wall boundary condition described above, particles accumulate in the vicinity of the wall and free-surface for simulations using Sv=0Sv = 0, since there exists no gravitational settling with which to exit the laminar viscous layer, and resuspension events are rare. This results in a situation where the simulations cannot reach a steady state in a reasonable amount of time. Therefore, we instead consider results for finite SvSv in the range 0<Sv≪10 < Sv \ll1. If the results are insensitive to SvSv then this would show that settling can be ignored in this regime. If, however, there are regions of the flow where the results are sensitive to SvSv even when Sv≪1Sv \ll1, then this would confirm the prediction of the theoretical analysis in the previous section. In view of this, we consider simulations with Sv=3×10−4,3×10−3,3×10−2,3×10−1Sv = 3 \times10^{-4}, 3 \times10^{-3}, 3 \times10^{-2}, 3 \times10^{-1}, and for each of these we consider four different Stokes numbers St=0.93,2.80,9.32,46.67St = 0.93, 2.80, 9.32, 46.67. Roughly speaking, this value of SvSv would correspond to a sand/dust grain with a diameter of O(1−100 μm)O(1-100\,\mu\mathrm{m}) suspended over a windy surface (e.g. uτ≈O(0.1 m/s)u_{\tau} \approx O(0.1\,\mathrm{m/s})). Note that Sv=3×10−1Sv = 3 \times10^{-1} could be considered to be too large to satisfy the requirement Sv≪1Sv \ll1; we nevertheless include it to better identify the limits of the above theory.

Results & Discussion

We begin by considering in Figure 2 the results for concentration ϱ(z)\varrho(z) for different SvSv and StSt. For each StSt, the results show a very strong effect of SvSv on ϱ\varrho, with strong reductions in the near-wall concentration as SvSv is increased. Indeed, for the cases with St=2.8St = 2.8, 9.39.3, 46.546.5, the near wall value of ϱ\varrho reduces by an order of magnitude as SvSv is increased from Sv=3×10−4Sv = 3 \times10^{-4} to Sv=3×10−2Sv = 3 \times10^{-2}. This remarkable sensitivity of ϱ\varrho to SvSv even when Sv≪1Sv \ll1 confirms the prediction of the analysis in §II. Conceptually, that gravity leads to a reduction in the value of the concentration near the wall is because with an absorbing wall, settling reduces the residence time of the particles in the near-wall region as their vertical velocity through the viscous sublayer is larger than it would be in the absence of settling. For the case of particles which elastically collide with the wall, the steady-state would be characterized by zero flux (⟨wp(t)⟩z=0\langle w^{p}(t)\rangle_z = 0), and settling would actually lead to an increase of the near wall concentration in general since their weight would keep them trapped in the near wall region until they experience large enough fluctuations of the vertical fluid velocity to resuspend them into the flow. The analysis of §II can also be extended to this zero-flux case and would again show that even if Sv≪1Sv \ll1, settling can make a non-negligible contribution to the near wall concentration of the particles (this has been confirmed in results not shown here).

Figure 2. Plot of ϱ\varrho as a function of zz for different SvSv, and (a) St=0.93St = 0.93, (b) St=2.80St = 2.80, (c) St=9.30St = 9.30, (d) St=46.5St = 46.5.

Not only do the results in Figure 2 show a strong effect of settling on the magnitude of the particle concentrations even when Sv≪1Sv \ll1, but they also indicate that settling affects the scaling behavior of ρ(z)\rho(z). In particular, as discussed in §II, the asymptotic analysis of Sikovsky [16] shows that for Sv=0Sv = 0, as the wall is approached the concentration behaves as a power law ρ(z)∼z−γ\rho(z) \sim z^{-\gamma}, which was also confirmed using DNS data. Unlike those DNS results (or those in Johnson et al. [3]), the results in Figure 2 show that as zz decreases, there is a noticeable kink in the curves for ρ\rho around z=O(1)z = O(1), below which ρ\rho appears to flatten out. This change in behavior is due to the role of SvSv in (3).

We now turn to consider how settling affects each of the terms in Λ(z)\Lambda(z) as defined in (4). In Figure 3, the results for ⟨up(t)⟩z\langle u^p(t)\rangle_z are shown for each of the St,SvSt, Sv combinations. In the absence of settling, this term is known to be positive in the near wall region [3], which is associated with the preference of the particles to accumulate in the near-wall ejection events of the turbulent boundary layer [4, 5, 20]. Our results also show ⟨up(t)⟩z>0\langle u^p(t)\rangle_z > 0 near the wall for the smallest SvSv case Sv=3×10−4Sv = 3 \times10^{-4}. However, as SvSv is increased, the size of ⟨up(t)⟩z\langle u^p(t)\rangle_z reduces, and is negative near the wall for Sv=3×10−1Sv = 3 \times10^{-1}. Moreover, the inset plots in Figure 3 indicate that even for Sv=3×10−2Sv = 3 \times10^{-2}, ⟨up(t)⟩z\langle u^p(t)\rangle_z is negative for z≤O(1)z \le O(1) (at which point the curve disappears on the logarithmic plot). Hence the preferential sampling of ejection events in the boundary layer by the inertial particles is very sensitive to SvSv even when Sv≪1Sv \ll1, and transitions to a preferential sampling of sweep events as SvSv increases (over a range of zz that depends on StSt and SvSv). This striking change in behavior can be understood as being due to the onset of the preferential sweeping mechanism [7, 21] as SvSv increases, and has been seen in other DNS of particles settling through wall-bounded turbulence [22]. This change in the sign of ⟨up(t)⟩z\langle u^p(t)\rangle_z as SvSv increases also means that the role of this term in governing the particle concentration changes. When ⟨up(t)⟩z>0\langle u^p(t)\rangle_z > 0 this term hinders the settling of the particles towards the wall and acts to reduce their near wall accumulation [3]. However, when ⟨up(t)⟩z<0\langle u^p(t)\rangle_z < 0, this term contributes to the settling of the particles towards the wall and so can contribute to their near wall accumulation.

Figure 3. Plots of the average fluid velocity sampled by the particles ⟨up(t)⟩z\langle u^p(t)\rangle_z as a function of zz for different SvSv, and (a) St=0.93St = 0.93, (b) St=2.80St = 2.80, (c) St=9.30St = 9.30, (d) St=46.5St = 46.5. The insets correspond to the same plots but in a log-log scale to highlight the behavior for small zz (and in these plots the symbols disappear for small zz once the quantity becomes negative). Legend is the same as that in figure 2.

In Figure 4, the results for −StDt⟨wp(t)⟩z-StD_t\langle w^p(t)\rangle_z are shown for each of the St,SvSt,Sv combinations. Since the flow has constant negative flux ρ⟨wp(t)⟩z<0\rho\langle w^p(t)\rangle_z<0, then since ρ\rho increases as the wall is approached, it follows that −StDt⟨wp(t)⟩z-StD_t\langle w^p(t)\rangle_z is negative, as shown in the results. In Bragg et al. [9] it was shown that for Sv=3×10−2Sv = 3 \times10^{-2} this acceleration term makes a negligible contribution to ⟨wp(t)⟩z\langle w^p(t)\rangle_z. Our results for different SvSv reveal that although this term generally makes a small contribution to ⟨wp(t)⟩z\langle w^p(t)\rangle_z, it can become important for large StSt as SvSv increases. However, for these larger St,SvSt,Sv cases, the concentration is almost uniform (see Figure 2), and so while −StDt⟨wp(t)⟩z-StD_t\langle w^p(t)\rangle_z can be important for predicting the settling velocity ⟨wp(t)⟩z\langle w^p(t)\rangle_z, it is not important in the regime of St,SvSt,Sv for which there is a significant build up of particle concentration near the wall.

Figure 4. Plots of the average vertical particle velocity ⟨wp(t)⟩z\langle w^p(t)\rangle_z as a function of zz for different SvSv, and (a) St=0.93St = 0.93, (b) St=2.80St = 2.80, (c) St=9.30St = 9.30, (d) St=46.5St = 46.5. Legend is the same as that in figure 2. Note that inset plots showing the results in a log-log scale are not shown for this quantity as the data is very noisy for small zz due to the spatial differentiation of the data for ⟨wp(t)⟩z\langle w^p(t)\rangle_z involved in computing Dt⟨wp(t)⟩zD_t\langle w^p(t)\rangle_z.

In Figure 5, the results for the turbophoretic drift velocity −St∇zW-St\nabla_z\mathcal{W} are shown for different StSt and SvSv. Compared to the results for ⟨up(t)⟩z\langle u^p(t)\rangle_z and −StDt⟨up(t)⟩z-StD_t\langle u^p(t)\rangle_z, the results show that −St∇zW-St\nabla_z\mathcal{W} is relatively insensitive to SvSv. The insets to the plots reveal, however, that there is a strong dependency on SvSv in the near wall region z≤O(1)z \le O(1), with −St∇zW-St\nabla_z\mathcal{W} increasing as SvSv increases for Sv<3×10−1Sv < 3 \times10^{-1}. The corresponding results for W\mathcal{W} (not shown) also show that W\mathcal{W} increases as SvSv increases. This behavior may be partially understood by considering the expression for W\mathcal{W} obtained using the formal solution to (9), namely

Figure 5. Plots of the turbophoretic velocity −St∇zW-St\nabla_z\mathcal{W} as a function of zz for different SvSv, and (a) St=0.93St = 0.93, (b) St=2.80St = 2.80, (c) St=9.30St = 9.30, (d) St=46.5St = 46.5. The insets correspond to the same plots (except that the vertical axis is now St∇zWSt\nabla_z\mathcal{W}) but in a log-log scale to highlight the behavior for small zz. Legend is the same as that in Figure 2.

W(z)=∫−∞0∫−∞0e(s+s′)/St(⟨up(s)up(s′)⟩z−⟨up(s)⟩z⟨up(s′)⟩z) ds′ ds.(10)\mathcal{W}(z) = \int_{-\infty}^{0}\int_{-\infty}^{0} e^{(s+s')/St}\left(\langle u^p(s)u^p(s')\rangle_z-\langle u^p(s)\rangle_z\langle u^p(s')\rangle_z\right)\,ds'\,ds. \tag*{(10)}

Settling reduces the correlation timescale of the flow seen by the particle [23], meaning that increasing SvSv will reduce the time span ∣s−s′∣\lvert s-s'\rvert over which the covariance ⟨up(s)up(s′)⟩z\langle u^p(s)u^p(s')\rangle_z is finite. This in turn would have the effect of reducing W\mathcal{W} with increasing SvSv (see [24] for the discussion of an analogous effect in the context of how settling impacts particle-pair relative velocities). The results in Figure 3 show, however, that increasing SvSv leads to a reduction in the preferential sampling of the flow for Sv≤3×10−2Sv \le3 \times10^{-2}. If this reduction in the preferential sampling leads to a greater reduction in ⟨up(s)⟩z⟨up(s′)⟩z\langle u^p(s)\rangle_z\langle u^p(s')\rangle_z than the reduction in the correlation timescale associated with ⟨up(s)up(s′)⟩z\langle u^p(s)u^p(s')\rangle_z, then according to (10) the overall effect of increasing SvSv would be to increase W\mathcal{W}, as observed. Such an argument also shows the way in which the preferential sampling of the flow impacts ⟨wp(t)⟩z\langle w^p(t)\rangle_z not only explicitly through the term ⟨up(t)⟩z\langle u^p(t)\rangle_z in (3), but also implicitly through its effect on −St∇zW-St\nabla_z\mathcal{W} which itself depends on how the particles interact with and sample the flow.

Finally, in Figure 6, the results for the diffusive velocity −StWρ−1∇zρ-St\mathcal{W}\rho^{-1}\nabla_z\rho are shown for different StSt and SvSv. Near the wall this velocity is positive since ∇zρ<0\nabla_z\rho< 0 there, so that this diffusive velocity hinders the settling velocity of the particles. This term is also very sensitive to SvSv even for Sv≪1Sv \ll1, which is mainly due to the sensitivity of ρ\rho to SvSv that was shown in Figure 2. Throughout much of the boundary layer this term makes a sub-leading contribution in (3). However, comparing the logarithmic scale inset plots to those of the other quantities reveals that very close to the wall where the gradients of ρ\rho are strongest, this term can be significant compared to the other contributions in (3).

Figure 6. Plots of the diffusive velocity −StWϱ−1∇zϱ-St\mathcal{W}\varrho^{-1}\nabla_z\varrho as a function of zz for different SvSv, and (a) St=0.93St=0.93, (b) St=2.80St=2.80, (c) St=9.30St=9.30, (d) St=46.5St=46.5. The insets correspond to the same plots but in a log-log scale to highlight the behavior for small zz. Legend is the same as that in figure 2.

Conclusions

We have considered the role of gravitational settling on the concentration profiles of small inertial particles in wall-bounded turbulent flows. We provided theoretical arguments and DNS results that show that settling can play a leading order role in determining the concentrations, even when the Stokes settling velocity is very small compared with the fluid friction velocity. The reason is that the dynamical relevance of settling is determined by the size of the Stokes settling velocity compared with the other mechanisms contributing to the particle vertical velocity, not compared with the fluid friction velocity (or any other fluid velocity scale). In the theoretical analysis, this corresponds to saying that settling can only be ignored if Λ0=Λ∣Sv=0\Lambda_0 = \Lambda|_{Sv=0} (see (4) for the definition of Λ(z)\Lambda(z)) is much larger than O(Sv)O(Sv). However, as we have shown, this condition will always be violated for Sv>0Sv>0 in some region close to the wall since in the viscous sub-layer, lim⁡z→0Λ0→0\lim_{z\to0}\Lambda_0\to0.

Quantitatively, the DNS results showed that the near-wall concentration is highly dependent on SvSv even when Sv≪1Sv \ll1, and indeed the concentration can be reduced by an order of magnitude when SvSv is increased from O(10−4)O(10^{-4}) and O(10−2)O(10^{-2}). The results also show that the preferential sampling of ejection events in the boundary layer by inertial particles that occurs for Sv=0Sv = 0 is profoundly altered as SvSv is increased, and is replaced by a preferential sampling of sweeping events due to the onset of the preferential sweeping mechanism. The results are very consequential for understanding and predicting the concentration profiles of inertial particles in wall-bounded turbulent flows, since many previous studies neglected the effect of settling under the assumption that its effect is negligible for Sv≪1Sv \ll1.

One practical issue is that it is not always possible to reliably predict Λ0(z)\Lambda_0(z) a priori, and therefore when performing numerical simulations, it may not be clear whether the condition Λ0≫O(Sv)\Lambda_0 \gg O(Sv) will be satisfied so as to justify neglecting particle settling. Our results suggest that it is probably best to always retain the particle settling, and then one can check a posteriori whether the settling plays any important role.

Acknowledgements

The authors acknowledge grant G00003613-ArmyW911NF-17-0366 from the US Army Research Office. Computational resources were provided by the High Performance Computing Modernization Program (HPCMP), and by the Center for Research Computing (CRC) at the University of Notre Dame.

The authors report no conflict of interest.

References

References

  1. [1]M. Chamecki, C. Meneveau, and M. B. Parlange, Large eddy simulation of pollen transport in the atmospheric boundary layer, J. of Aerosol Sci. 40, 241 (2009).DOI
  2. [2]M. W. Reeks, The transport of discrete particles in inhomogeneous turbulence, J. Aerosol Sci. 14, 729 (1983).DOI
  3. [3]P. L. Johnson, M. Bassenne, and P. Moin, Turbophoresis of small inertial particles: theoretical considerations and application to wall-modelled large-eddy simulations, J. Fluid. Mech. 883, A27 (2020).DOI
  4. [4]M. Rashidi, G. Hetsroni, and S. Banerjee, Particle-turbulence interaction in a boundary layer, International Journal of Multiphase Flow 16, 935 (1990).
  5. [5]J. K. Eaton and J. R. Fessler, Preferential concentration of particles by turbulence, Int. J. Multiphase Flow 20, 169 (1994).
  6. [6]A. Soldati and C. Marchioli, Physics and modelling of turbulent particle deposition and entrainment: Review of a systematic study, Int. J. Multiphase Flow 35, 827 (2009).DOI
  7. [7]M. R. Maxey, The gravitational settling of aerosol particles in homogeneous turbulence and random flow fields, J. Fluid Mech. 174, 441 (1987).DOI
  8. [8]L.-P. Wang and M. R. Maxey, Settling velocity and concentration distribution of heavy particles in homogeneous isotropic turbulence, J. Fluid Mech. 256, 27 (1993).DOI
  9. [9]A. D. Bragg, D. H. Richter, and G. Wang, Mechanisms governing the settling velocities and spatial distributions of inertial particles in wall-bounded turbulence, Phys. Rev. Fluids 6, 064302 (2021).DOI
  10. [10]H. Rouse, Modern conceptions of the mechanics of turbulence, Trans. Am. Soc. Civ. Eng. 102, 463 (1937).
  11. [11]S. Li, A. D. Bragg, and G. Katul, A co-spectral budget model links turbulent eddies to suspended sediment concentration in channel flows (2021), arXiv:2108.10102 [physics.flu-dyn].DOI
  12. [12]M. R. Maxey and J. J. Riley, Equation of motion for a small rigid sphere in a nonuniform flow, Phys. Fluids 26, 883 (1983).DOI
  13. [13]S. Balachandar, A scaling analysis for point particle approaches to turbulent multiphase flows, Int. J. Multiph. Flow 35, 801 (2009), special Issue: Point-Particle Model for Disperse Turbulent Flows.DOI
  14. [14]D. H. Richter and M. Chamecki, Inertial effects on the vertical transport of suspended particles in a turbulent boundary layer, Bound.-Layer Meteorol. 167, 235 (2018).DOI
  15. [15]A. Bragg, D. C. Swailes, and R. Skartlien, Drift-free kinetic equations for turbulent dispersion, Phys. Rev. E 86, 056306 (2012).pubmed.ncbi.nlm.nih.gov/23214875
  16. [16]D. P. Sikovsky, Singularity of inertial particle concentration in the viscous sublayer of wall-bounded turbulent flows, Flow Turbul Combust 92 (2014).DOI
  17. [17]G. Wang and D. H. Richter, Modulation of the turbulence regeneration cycle by inertial particles in planar couette flow, J. Fluid Mech. 861, 901 (2019).
  18. [18]G. Wang and D. H. Richter, Two mechanisms of modulation of very-large-scale motions by inertial particles in open channel flow, J. Fluid Mech. 868, 538?559 (2019).DOI
  19. [19]G. Wang, K. O. Fong, F. Coletti, J. CapeceLatro, and D. H. Richter, Inertial particle velocity and distribution in vertical turbulent channel flow: a numerical and experimental comparison, Int. J. Multiphase Flow , 103105 (2019).
  20. [20]C. Marchioli and A. Soldati, Mechanisms for particle transfer and segregation in a turbulent boundary layer, J. Fluid Mech. 468, 283 (2002).DOI
  21. [21]J. Tom and A. D. Bragg, Multiscale preferential sweeping of particles settling in turbulence, Journal of Fluid Mechanics 871, 244 (2019).
  22. [22]J. Lee and C. Lee, The effect of wall-normal gravity on particle-laden near-wall turbulence, Journal of Fluid Mechanics 873, 475 (2019).DOI
  23. [23]G. T. Csanady, Turbulent Diffusion in the Environment (Reidel, Boston, 1973).DOI
  24. [24]P. J. Ireland, A. D. Bragg, and L. R. Collins, The effect of reynolds number on inertial particle dynamics in isotropic turbulence. part 2. simulations with gravitational effects, Journal of Fluid Mechanics 796, 659 (2016).

Paper details

vlnc:W_qy6ttx57
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