betaFeedback

WorksPhysics and Astronomy

The 'Brazil-nut effect' in bidisperse particle laden flow on an incline

Abstract

We study bidisperse suspensions – suspensions where there are two particle species of the same density but different sizes – of a viscous fluid on an incline. We use a lubrication theory/thin film model to form a hyperbolic system of three conservation laws for the height and particle volume fractions. The model predicts, over a range of parameters, that the larger particles rise to the top of the layer, consistent with the well-known `Brazil-nut effect' for granular media. The model predicts well-separated fronts of the two species of particles, behind a clear fluid front, at lower inclination angles and volume fractions. This corresponds to a triple shock structure in the system of conservations. At higher inclination angles and volume fractions the particles congregate at a high concentration at the leading front corresponding to a singular shock in the model. We find excellent agreement between theory and experiments in terms of the overall dynamic structures as the parameters vary.

Introduction

Particle laden flows, mixtures of particles and fluid under the influence of forces, arise in many different manufacturing, environmental, and industrial applications. Granular flows on an incline are well studied for their impact on debris flow in large-scale phenomena such as avalanches and landslides [1] and are also modeled at a fundamental level, with theoretical interest in their rheology [2] and particle segregation [3]. The study of particle laden flows are also of interest to the food industry with the Bostwick consistometer [4, 5, 6], a device used to determine the viscosity of food products. Spiral separators are used in the mining industry to separate wet mixtures of particles based on density and/or particle size [7, 8, 9].

In this work, we consider particles in a viscous layer of fluid on an incline. In particular, we examine a slurry with two negatively buoyant particle species of the same density but different sizes. The particles are sufficiently large to avoid being influenced primarily by Brownian motion, yet small enough to stay suspended and maintain the continuum nature of the fluid. We are interested in understanding how the particles naturally segregate by size over time as they move down the inclined geometry. It is well-known for dry, granular flows that many-body interactions between different-sized particles, under gravity, often result in the larger particles rising to the top as demonstrated in Figure 1. This effect is colloquially known as the ‘Brazil-nut effect’. Investigation into the Brazil-nut effect began with the seminal work of Rosato et. al in [10] where the authors attribute the Brazil-nut effect to the ability of smaller particles to fit into the voids surrounding larger particles—spaces that larger particles themselves cannot occupy. The authors of [11] further this argument by considering a three dimensional geometry and arguing smaller particles roll over larger ones. Many other mechanisms have been proposed since then. The authors in [12] present particles shaken in a cylinder follow a convective cycle up the middle of the container and then travel down the walls. The larger particles are trapped when they reach the top of the convection zone, while the smaller ones are able to convect freely. A different study [13] indicates particle density as a factor affecting sedimentation. Another proposed mechanism is granular temperature, the velocity variance, as the segregating mechanism that behaves analogously to thermal temperature as discussed in [14]. Numerous experimental work such as [15] and [16] support these investigations into the Brazil-nut effect.

Four photographs showing bidisperse mixtures before and after shaking and settling

Figure 1. Demonstration of different effects achievable in bidisperse mixtures of particles. In Figure 1a, equal volumes of red and blue glass particles are added to a cup, with the blue particles layered above the red ones. The red glass particles are larger and have 0.6 mm diameter. The blue glass particles are smaller and have 0.3mm diameter. When vigorously shaken in all directions for 3 minutes, the larger particles rise to the top as shown in Figure 1b—an example of the Brazil-nut effect. On the other hand, when the same particles are suspended in a viscous fluid, as in Figure 1c, mixed well, and then allowed to settle, the larger particles settle at a faster rate than the smaller ones. After 15 minutes of settling, there’s a distinct layer of blue smaller particles on top in Figure 1d [17].

The Brazil-nut effect has also been studied in granular flows on an incline. The authors of [18] attribute the Brazil-nut effect to the same void filling mechanism as in [10] and contact forces squeezing particles out of their layer. In a high speed setting, the authors of [19] use gradients of granular temperature to explain particle migration. Understanding of the Brazil-nut effect is of interest to the mining and pharmaceutical industry as the tendency for large particles to aggregate at the top prevents well mixing [20].

We contrast this phenomena with Stokes’ law, where larger spheres in a viscous fluid moving at a very low particle Reynolds number (Re⁡=ρℓd2γ˙4μℓ\operatorname{Re} = \frac{\rho_{\ell}d^{2}\dot{\gamma}}{4\mu_{\ell}}) [21] move faster than smaller spheres as seen in Figure 1c and Figure 1d. We find in bidisperse particle laden flow the larger particles rise to the top like in granular flows and unlike in Stokes flow, as seen in Figure 2a. Here we develop a continuum model for the dynamic motion of particles (according to volume fraction) that explains why we observe the Brazil-nut effect in our bidisperse slurry flow.

Side profile and aerial profile of the flow

Figure 2. Side profile of an experiment in the settled regime and the bird’s eye perspective of the flow. The red particles are the larger particles.

Previous work on thin-film slurries has built on a classical model for clear fluid flow on an incline, studied by Huppert in [22]. He used a conservation law for the flow, arising from lubrication theory, and derived analytic, self-similar, formulas for the front speed and fluid height for the finite volume problem. The authors of [23] then investigated the same geometry with a fluid containing one species of non-neutrally buoyant particles of identical size and density. In that work they included hindered settling of particles in the model. Cook [24] further advanced the model by including shear-induced migration, which quantitatively explained a natural bifurcation in the experiment in which the particles preferentially settle to the substrate or congregate at the leading front depending on the volume fraction and inclination angle. This model was quantitatively validated by detailed experiments in [25]. The work [21] develops a quantitative dynamic lubrication model that shows excellent quantitative agreement with dynamic experiments in the settled regime. Using a similar framework, [26] extended the model in [21] to the case of two particle species of the same size but different densities. That work followed the equilibrium theory derivation in [27] to derive the dynamic conservation laws. As in the earlier works, the dynamic model assumes rapid equilibration of the stratification of particles in the normal direction to the substrate, compared with the speed of the front.

Other authors have studied bidisperse particle laden flow of two different sizes in other geometries. The authors of [28] model shear-induced migration of polydisperse suspensions and introduce an dynamic expression for the maximum packing fraction in the bidisperse case depending on the relative diameters of the particles and their volume fractions. In [29], the authors examine the case of bidisperse suspensions in channel flow and expand upon the modeling of the shear-induced migration flux term for the bidisperse case. In both of these works the authors find the particles segregate within the flow by size. The authors of [30] approach the modeling of neutrally buoyant bidisperse particle laden flow in a channel using the suspension balance model. Since we consider negatively buoyant bidisperse particle laden flow down a slope, we include gravitational settling in our analysis which leads to more varied qualitative outcomes. In contrast, the authors of [31] use the modeling techniques established in the granular convection literature study bidisperse particle laden flow with a generic interstitial fluid. However, they make the simplifying assumption that the particle volume fraction is constant. In the more complex spiral geometry, particle species separate by density through a similar thin-film model, with long-time behavior analyzed once the cross-sectional flow has equilibrated [32].

This paper is organized as follows. In § 2, we derive the conservation law model that governs how the slurry develops as it flows down the incline. We also present the equilibrium model that dictates how the particle species and fluids equilibrate in the direction normal to the incline. In Sec. III, we analyze the bidisperse equilibrium model which yields the counterintuitive result of coarser particles accumulating near the surface and finer particles settling at the bottom. In § 5, we go over the experimental methods and materials. In § 6, we qualitatively compare numerical simulations of our dynamic model to laboratory experiments, highlighting the similar features we observe in both. In Sec. VI, we finish with concluding remarks and potential future directions.

Model

In this section, we develop a system of conservation laws in the continuum limit that describe the evolution of the film height and particle concentration in the slurry as it slides down the incline. Figure 3 shows the geometry of the problem. The slurry is comprised of a fluid with viscosity μℓ\mu_{\ell} and density ρℓ\rho_{\ell}, and two different species of negatively buoyant particles with the same density ρs\rho_{\mathrm{s}} and different diameters d1>d2d_1 > d_2. This slurry is confined to a track of fixed width and slides down an incline of angle α\alpha. The viscosity of the slurry is a function of the particle concentration, ϕ\phi, and is given via the Krieger-Dougherty relation as μ(ϕ)=μℓ(1−ϕ/ϕm)−2\mu(\phi)=\mu_{\ell}(1-\phi/\phi_{\mathrm{m}})^{-2} [33] where ϕm\phi_{\mathrm{m}} is the maximum packing fraction of spheres in the fluid. Unlike the case with particles of the same size and different densities, ϕm\phi_{\mathrm{m}} depends on the particle concentration of particles of different sizes. This is due to the fact smaller particles can fit into the gaps between larger particles that larger particles cannot. We denote ϕ1\phi_1 as the particle concentration of larger particles and ϕ2\phi_2 as the particle concentration of smaller particles within the slurry, with the relation ϕ=ϕ1+ϕ2\phi=\phi_1+\phi_2. The dependence of ϕm\phi_m on χ=ϕ1ϕ\chi= \frac{\phi_1}{\phi}, the proportion of the larger particles to all of the particles in the slurry, is given as

Schematic of bidisperse particle laden flow developing on an incline

Figure 3. Schematic of bidisperse particle laden flow developing on an incline

ϕm=ϕm,0(1+32b23χ23(1−χ))(1)\phi_m = \phi_{m,0}\left(1+\frac{3}{2}b^{\frac{2}{3}}\chi^{\frac{2}{3}}(1-\chi)\right) \tag*{(1)}

where b=d1−d2d1+d2b = \frac{d_1-d_2}{d_1+d_2} and ϕm,0\phi_{m,0} is the maximum packing fraction of spherical particles of the same size in the fluid [28]. We choose ϕm,0=0.61\phi_{m,0}=0.61 [34].

We now state the conservation of momentum and conservation of mass equation for the slurry respectively:

ρ(∂tu⃗+u⃗⋅∇u⃗)=∇⋅(−pI+μ(∇u⃗+∇u⃗T))+ρg⃗(2)\rho\left(\partial_t\vec{u}+\vec{u}\cdot\nabla\vec{u}\right)=\nabla\cdot\left(-pI+\mu\left(\nabla\vec{u}+\nabla\vec{u}^{T}\right)\right)+\rho\vec{g} \tag*{(2)}

and

∂tρ+∇⋅(ρu⃗)=0(3)\partial_t\rho+\nabla\cdot(\rho\vec{u})=0 \tag*{(3)}

where u⃗=(u,w)\vec{u}=(u,w) represents the xx and zz components of velocity respectively and g⃗=(0,−g)\vec{g}=(0,-g) where gg is gravitational acceleration. Equation (2) describes how momentum of the fluid is conserved and equation (3) describes how mass of the fluid is conserved. However, we assume the fluid has low Reynolds number, so the momentum terms—the left hand side of Eq. (2)—can be neglected. The velocity satisfies the no-slip boundary condition (u=0u=0 when z=0z=0) and no stress (n⃗⋅(−pI+μ(∇u⃗+∇u⃗T)=0\vec{n}\cdot(-pI+\mu(\nabla\vec{u}+\nabla\vec{u}^{T})=0) occurs at the free surface. We also account for the conservation of each species of particles with

∂tϕi+u⃗⋅∇ϕi+∇⋅J⃗=0i=1,2(4)\partial_t\phi_i+\vec{u}\cdot\nabla\phi_i+\nabla\cdot\vec{J}=0 \qquad i=1,2 \tag*{(4)}

where J⃗\vec{J} represents all of the particle flux terms.

We now nondimensionalize the system using a thin film approximation as done in [27] under the following scaling:

(x^,z^)=1H(δx,z),u^=1U0(u,wδ),J^=H2δ2U0(Jxδ,Jz),(\hat{x},\hat{z})=\frac{1}{H}(\delta x,z),\qquad\hat{u}=\frac{1}{U_0}\left(u,\frac{w}{\delta}\right),\qquad\hat{J}=\frac{H^2}{\delta^2U_0}\left(\frac{J_x}{\delta},J_z\right),
p^=HμℓU0p,μ^=μμℓ,ρ^s=ρs−ρℓρℓ,s=zH,\hat{p}=\frac{H}{\mu_\ell U_0}p,\qquad\hat{\mu}=\frac{\mu}{\mu_\ell},\qquad\hat{\rho}_s=\frac{\rho_s-\rho_\ell}{\rho_\ell},\qquad s=\frac{z}{H},

where HH and LL are the characteristic film thickness and the axial length scale respectively, U0=H2ρℓgsin⁡αμℓU_0=\frac{H^2\rho_\ell g\sin\alpha}{\mu_\ell}, δ=HL≪1\delta=\frac{H}{L}\ll1, and dˉ\bar{d} is the average diameter of the particles. Going forward the hats will be dropped for brevity.

Next, we make the assumption, as done in previous work [21, 25, 26, 32], the particles equilibrate in the normal direction faster than fluid flows down the incline. We can conclude this by studying the leading order behavior of Eqs. (2) and (3) alongside the assumption that δ≪(dˉH)2≪1\delta\ll\left(\frac{\bar{d}}{H}\right)^2\ll1. The upper bound ensures the particles are not too large and is consistent with the continuum model. The lower bound ensures the particles are not too small and their motion is not dominated by Brownian motion [32].

First, we study the xx component of the momentum equation (2), which reduces to

∂∂z(μ∂u∂z)=−ρ(5)\frac{\partial}{\partial z}\left(\mu\frac{\partial u}{\partial z}\right)=-\rho \tag*{(5)}

in leading order of δ\delta. The no-slip boundary condition on the surface of the incline gives us ∂zu∣z=0=0\partial_z u|_{z=0}=0 and no stress at the free surface of the slurry gives us μ∂zu∣z=h=0\mu\partial_z u|_{z=h}=0. This leads to the following ODE in terms of the stress σ\sigma in the zz direction:

∂σ∂z=−1−ρsϕ,σ(0)=1+ρs∫0zϕ(s) ds,σ(h)=0.(6)\frac{\partial\sigma}{\partial z}=-1-\rho_s\phi,\qquad\sigma(0)=1+\rho_s\int_0^z\phi(s)\,\mathrm{d}s,\qquad\sigma(h)=0. \tag*{(6)}

Similar analysis for the conservation of particles yields ∂zJz=0\partial_z J_z=0 for both species. As particles cannot exit the slurry through the track or the free surface, this implies

Jz,i=0,J_{z,i}=0,

meaning each species of particles is in equilibrium in the zz direction.

To describe how the system evolves in the xx direction, we need to know the velocity in the xx direction, uu. From Eqs. (5) and (6), we obtain the ODE

σ=μ(ϕ)∂u∂z.(7)\sigma=\mu(\phi)\frac{\partial u}{\partial z}. \tag*{(7)}

In order to compute uu, we need to first compute the viscosity and the shear stress, which requires knowledge of the particle concentration of both species in the zz direction.

We now examine the flux balance of particles in the zz direction by describing each of the individual components of the particle flux. We claim the total flux for each particle species ii is comprised of three components: shear-induced migration, settling also known as sedimentation, and self-mixing between particle species yielding

Ji=Jshear,i+Jsettling,i+Jtracer,i.(8)J_i=J_{\mathrm{shear},i}+J_{\mathrm{settling},i}+J_{\mathrm{tracer},i}. \tag*{(8)}

We have dropped the zz subscript from the flux terms JJ for brevity. We now define each flux term in detail.

Shear-induced migration represents the migration of particles from high to low concentrations and from high to low shear stress [35, 36]. The shear-induced migration term largely remains the same from its description in [25] and [26], but the different diameters of the two particle species leads to different displacements when collision occurs. The shear-induced migration term is thus

Jshear,i=−ϕi∑j=12Aij(Kc∇(ϕjγ˙)+Kvϕjγ˙μ(ϕ)∇(μ))(9)J_{\mathrm{shear},i}=-\phi_i\sum_{j=1}^{2}A_{ij}\left(K_c\nabla(\phi_j\dot{\gamma})+K_v\frac{\phi_j\dot{\gamma}}{\mu(\phi)}\nabla(\mu)\right) \tag*{(9)}

where Kc≈0.41K_c\approx0.41 and Kv≈0.62K_v\approx0.62 are empirically determined constants [36], the coupling matrix AA is

Aij=14(di+dj)22d+1(1+didj)d1+(didj)d,(10)A_{ij}=\frac{1}{4}\frac{(d_i+d_j)^2}{2^{d+1}}\frac{\left(1+\frac{d_i}{d_j}\right)^d}{1+\left(\frac{d_i}{d_j}\right)^d}, \tag*{(10)}

γ˙=σ/μ\dot{\gamma}=\sigma/\mu is the shear rate for a Newtonian fluid, and dd is the dimension of the spheres (in this case, d=3d=3). We note that the diagonal terms of this matrix reduces to d12/4d_1^2/4 and d22/4d_2^2/4. Also, AA is symmetric as the off diagonal terms are identical:

Aij=14(di+dj)22d+1(1+didj)d1+(didj)d⋅(djdi)d(djdi)d=14(di+dj)22d+1(djdi+1)d(djdi)d+1=Aji.A_{ij}=\frac{1}{4}\frac{(d_i+d_j)^2}{2^{d+1}}\frac{\left(1+\frac{d_i}{d_j}\right)^d}{1+\left(\frac{d_i}{d_j}\right)^d}\cdot\frac{\left(\frac{d_j}{d_i}\right)^d}{\left(\frac{d_j}{d_i}\right)^d} =\frac{1}{4}\frac{(d_i+d_j)^2}{2^{d+1}}\frac{\left(\frac{d_j}{d_i}+1\right)^d}{\left(\frac{d_j}{d_i}\right)^d+1}=A_{ji}.

Equations (9) and (10) are adapted from [29]. In shear-induced migration, the collisions from the particles as they migrate cause different displacements depending on the pair of particle species. Thus, the shear-induced migration of species ii has a different contribution from the two particle species, and the total flux is the sum of the contributions from each particle species.

Now we move on to the settling flux. Since the particles are of the same density, we do not need to account for special particle interactions between species unlike in [37]. We also assume particles settle the same regardless of their size. This leads us to the settling flux as originally described in [38] with hindrance function f(ϕ)=μℓ(1−ϕ)/μ(ϕ)f(\phi)=\mu_\ell(1-\phi)/\mu(\phi):

Jsettling,i=−2di2ϕicot⁡α(1−ϕ)(ρi−ρℓ)9μ(ϕ).(11)J_{\mathrm{settling},i}=-\frac{2d_i^2\phi_i\cot\alpha(1-\phi)(\rho_i-\rho_\ell)}{9\mu(\phi)}. \tag*{(11)}

Finally, we have the tracer flux which captures the random motion of particles in a sheared mixture [38]. While the net concentration of particles ϕ\phi does not change under tracer flux, the individual particle concentrations ϕi\phi_i do. The tracer flux for a particle species is given as

Jtracer,i=−γdi24Dtr(ϕ)ϕ∇(ϕiϕ)(12)J_{\mathrm{tracer},i}=-\frac{\gamma d_i^2}{4}D_{\mathrm{tr}}(\phi)\phi\nabla\left(\frac{\phi_i}{\phi}\right) \tag*{(12)}

where DtrD_{\mathrm{tr}} is the tracer diffusivity and is given by the empirical expression Dtr(ϕ)=12min⁡{ϕ2,ϕtr2}D_{\mathrm{tr}}(\phi)=\frac{1}{2}\min\{\phi^2,\phi_{\mathrm{tr}}^2\} [27]. In the dilute limit, [35, 38] proposes the empirical expression ϕ2/2\phi^2/2 for tracer diffusivity, while [39] notes the tracer diffusivity becomes a constant ϕtr=0.4\phi_{\mathrm{tr}}=0.4 for large concentrations.

Now, we express the flux balance in Eq. (8) for each particle species as an ODE in terms of the particle concentration ϕ\phi and the species ratio χ\chi. Writing this in terms of ϕ′\phi' and χ′\chi' where the derivatives here are with respect to zz, we obtain

−[ϕ100ϕ2]A[χgh1(1−χ)gh2][ϕ′χ′]−σ4ϕϕDtr(ϕ)[d12/400d22/4][010−1][ϕ′χ′]=2cot⁡α9ρs(1−ϕ)[ϕ100ϕ2][d12/400d22/4][11]+Kc[ϕ100ϕ2]A[ϕχσ′ϕ(1−χ)σ′](13)-\begin{bmatrix}\phi_1&0\\0&\phi_2\end{bmatrix}A\begin{bmatrix}\chi g&h_1\\(1-\chi)g&h_2\end{bmatrix}\begin{bmatrix}\phi'\\\chi'\end{bmatrix} -\frac{\sigma}{4}\phi\phi D_{\mathrm{tr}}(\phi)\begin{bmatrix}d_1^2/4&0\\0&d_2^2/4\end{bmatrix}\begin{bmatrix}0&1\\0&-1\end{bmatrix}\begin{bmatrix}\phi'\\\chi'\end{bmatrix} = \frac{2\cot\alpha}{9}\rho_s(1-\phi)\begin{bmatrix}\phi_1&0\\0&\phi_2\end{bmatrix}\begin{bmatrix}d_1^2/4&0\\0&d_2^2/4\end{bmatrix}\begin{bmatrix}1\\1\end{bmatrix} +K_c\begin{bmatrix}\phi_1&0\\0&\phi_2\end{bmatrix}A\begin{bmatrix}\phi\chi\sigma'\\\phi(1-\chi)\sigma'\end{bmatrix} \tag*{(13)}

where:

g=Kc(σ−ϕσ(2ϕm−ϕ))+Kvϕσ(2ϕm−ϕ),h1=Kc(ϕσ+ϕχσ(2ϕm−ϕ)ξ),g=K_c\left(\sigma-\phi\sigma\left(\frac{2}{\phi_m-\phi}\right)\right)+K_v\phi\sigma\left(\frac{2}{\phi_m-\phi}\right),\qquad h_1=K_c\left(\phi\sigma+\phi\chi\sigma\left(\frac{2}{\phi_m-\phi}\right)\xi\right),
h2=Kc(−ϕσ+ϕ(1−χ)σ(2ϕm−ϕ)ξ),ξ=ϕϕm(ϕm−ϕm,03−5X2X(1−X)).h_2=K_c\left(-\phi\sigma+\phi(1-\chi)\sigma\left(\frac{2}{\phi_m-\phi}\right)\xi\right),\qquad\xi=\frac{\phi}{\phi_m}\left(\phi_m-\phi_{m,0}\frac{3-5X}{2X(1-X)}\right).

The system in equation (13) is linear in the variables ϕ′\phi' and χ′\chi', and recalling ϕ1=ϕχ\phi_1=\phi\chi and ϕ2=ϕ(1−χ)\phi_2=\phi(1-\chi), we obtain the following ODEs for χ′\chi' and ϕ′\phi':

χ′=2cot⁡α9ρs(1−ϕ)(1−χ)χ[(2χ−1)d12d22+a(d12−(d12+d22)χ)]×[σ(Kcdet⁡A(1−χ)ϕχ+Dtr(ϕ)[d12d22(1−χ)2+a(d12+d22)(1−χ)χ+d12d22χ2])]−1(14)\chi'=\frac{2\cot\alpha}{9}\rho_s(1-\phi)(1-\chi)\chi\left[(2\chi-1)d_1^2d_2^2+a(d_1^2-(d_1^2+d_2^2)\chi)\right] \times\left[\sigma\left(K_c\det A(1-\chi)\phi\chi+D_{\mathrm{tr}}(\phi)\left[d_1^2d_2^2(1-\chi)^2+a(d_1^2+d_2^2)(1-\chi)\chi+d_1^2d_2^2\chi^2\right]\right)\right]^{-1} \tag*{(14)}

and

ϕ′=−[(χ+ad22(1−χ))h1+((1−χ)+ad12χ)h2]χ′+2cot⁡α9ρs(1−ϕ)+Kcϕσ′(χ2+aχ(1−χ)(1d12+1d22)+(1−χ)2)×[g(χ2+a(1d12+1d22)χ(1−χ)+(1−χ)2)]−1,(15)\begin{aligned} \phi'=-\left[\left(\chi+\frac{a}{d_2^2}(1-\chi)\right)h_1+\left((1-\chi)+\frac{a}{d_1^2}\chi\right)h_2\right]\chi' \\ &+\frac{2\cot\alpha}{9}\rho_s(1-\phi)+K_c\phi\sigma'\left(\chi^2+a\chi(1-\chi)\left(\frac{1}{d_1^2}+\frac{1}{d_2^2}\right)+(1-\chi)^2\right) \times\left[g\left(\chi^2+a\left(\frac{1}{d_1^2}+\frac{1}{d_2^2}\right)\chi(1-\chi)+(1-\chi)^2\right)\right]^{-1}, \tag*{(15)} \end{aligned}

where aa is the off diagonal entry of AA with 14\frac{1}{4} factored out:

a=(d1+d1)2(d1+d2)d2d+1d1d+d2dd1d+d2d.(16)a=\frac{(d_1+d_1)^2(d_1+d_2)^d}{2^{d+1}}\frac{d_1^d+d_2^d}{d_1^d+d_2^d}. \tag*{(16)}

Coupled with the ODE for stress given in Eq. (6) and the appropriate boundary and integral conditions, we obtain the following system that describes how the particles equilibrate in the zz direction:

ϕ′=−[(χ+ad22(1−χ))h1+((1−χ)+ad12χ)h2]χ′+2cot⁡α9ρs(1−ϕ)+Kcϕσ′(χ2+aχ(1−χ)(1d12+1d22)+(1−χ)2)×[g(χ2+a(1d12+1d22)χ(1−χ)+(1−χ)2)]−1,\begin{aligned} \phi'={}&-\left[\left(\chi+\frac{a}{d_2^2}(1-\chi)\right)h_1+\left((1-\chi)+\frac{a}{d_1^2}\chi\right)h_2\right]\chi'\\ &+\frac{2\cot\alpha}{9}\rho_s(1-\phi)+K_c\phi\sigma'\left(\chi^2+a\chi(1-\chi)\left(\frac{1}{d_1^2}+\frac{1}{d_2^2}\right)+(1-\chi)^2\right)\\ &\times\left[g\left(\chi^2+a\left(\frac{1}{d_1^2}+\frac{1}{d_2^2}\right)\chi(1-\chi)+(1-\chi)^2\right)\right]^{-1}, \end{aligned}
χ′=2cot⁡α9ρs(1−ϕ)(1−χ)(χ)[(2χ−1)d12d22+a(d12−(d12+d22)χ)][σ(Kcdet⁡A(1−χ)ϕχ+Dtr(ϕ))[d12d22(1−χ)2+a(d12+d22)(1−χ)(χ)+d12d22χ2]]−1(17)\chi'=\frac{2\cot\alpha}{9}\rho_s(1-\phi)(1-\chi)(\chi)\left[(2\chi-1)d_1^2d_2^2+a\left(d_1^2-(d_1^2+d_2^2)\chi\right)\right] \left[\sigma\left(K_c\det A(1-\chi)\phi\chi+D_{\mathrm{tr}}(\phi)\right)\left[d_1^2d_2^2(1-\chi)^2+a(d_1^2+d_2^2)(1-\chi)(\chi)+d_1^2d_2^2\chi^2\right]\right]^{-1} \tag*{(17)}
σ′=−1−ρsϕ,\sigma'=-1-\rho_s\phi,
∫0hϕ dz=ϕ0,1ϕ0∫0hϕχ dz=χ0,σ(0)=1+ρs∫0zϕ(s) ds,σ(h)=0.\int_0^h\phi\,dz=\phi_0,\qquad\frac{1}{\phi_0}\int_0^h\phi\chi\,dz=\chi_0,\qquad\sigma(0)=1+\rho_s\int_0^z\phi(s)\,ds,\qquad\sigma(h)=0.

By solving the ODE system (17), we obtain vertical profiles for ϕ\phi, χ\chi, and σ\sigma, which coupled with Eq. (7), determines the velocity profile in the downstream direction of the mixture. We scale the profile quantities by the film height, so that u~=h2u\tilde{u}=h^2u. By integrating the scaled velocity in the normal direction to the plane, we define the fluxes

f(ϕ0,χ0)=∫01u~ ds,gi(ϕ0,χ0)=∫01u~ϕi ds.(18)f(\phi_0,\chi_0)=\int_0^1\tilde{u}\,ds,\qquad g_i(\phi_0,\chi_0)=\int_0^1\tilde{u}\phi_i\,ds. \tag*{(18)}

With the change of variables s=z/hs=z/h, we obtain the following hyperbolic system of conservation laws that govern the height profiles of the clear fluid and particle mixtures as it flows down the track:

∂h∂t+∂∂x(h3f(ϕ0,χ0))=0∂(hϕ0,i)∂t+∂∂x(h3gi(ϕ0,χ0))=0i=1,2.(19)\frac{\partial h}{\partial t}+\frac{\partial}{\partial x}\left(h^3f(\phi_0,\chi_0)\right)=0\qquad\frac{\partial(h\phi_{0,i})}{\partial t}+\frac{\partial}{\partial x}\left(h^3g_i(\phi_0,\chi_0)\right)=0\qquad i=1,2. \tag*{(19)}

Analysis

In Sec. III A we analyze equilibrium model presented in Eq. (20). We observe larger particles tend to settle at the top as seen in example solutions presented in Figure 4 and we will show in Proposition 1 why we expect larger particles to settle at the top. In Sec. III B we discuss the general structure of solutions to systems of hyperbolic conservation laws and the types of shocks we expect to see in the solution.

Example equilibrium profiles for ridged and settled regimes

Figure 4. (a) Ridged Regime at χ0=0.3\chi_0 = 0.3; (b) Ridged Regime at χ0=0.7\chi_0 = 0.7; (c) Settled Regime at χ0=0.3\chi_0 = 0.3; (d) Settled Regime at χ0=0.7\chi_0 = 0.7. Example equilibrium profiles that solve system (17) for the given parameters. Figs. 4a and 4b show the case of a concentrated suspension and Figs. 4c and 4d show a dilute suspension. The concentration of larger particles is red, and the concentration of smaller particles is blue.

Equilibrium Model

First, we discuss how the fluid and particles equilibrate in the zz direction according to system (17). As in [25] and [27], we define a solution to (17) as ridged if particles of either species accumulate at the top of the profile (if ϕ→ϕm\phi\to\phi_m as s→1s\to1). On the other hand, we define a solution as settled if the particle concentration decreases monotonically as ss increases. This leads to a layer of clear fluid on top of the profile (ϕi=0\phi_i=0 for all s≥s∗s\ge s^* for some s∗<1s^*<1).

Figure 4 shows example equilibrium profiles that solve system (17). Figure 4a is an example of a ridged profile while Figure 4c is an example of a settled profile. In Figure 4a, as zz increases from the substrate to the free surface, the volume fraction of larger particles, ϕ1\phi_1, increases monotonically to the maximum packing fraction, ϕm\phi_m. On the other hand, the volume fraction of smaller particles, ϕ2\phi_2, decreases monotonically to 0. Physically, this corresponds to particles on top of the fluid profile, corresponding to ϕ=ϕ1+ϕ2≠0\phi=\phi_1+\phi_2\ne0 when z=1z=1. Additionally, only larger particles are present at the top of the fluid profile as ϕ2=0\phi_2=0 when z=1z=1. On the other hand, in Figure 4c, both concentrations of particles, ϕ1\phi_1 and ϕ2\phi_2, vanish as zz increases from the substrate to the free surface, eventually reaching 0 before z=1z=1. Physically, we see a mix of larger and smaller particles in the fluid profile, but the top of the fluid profile contains only clear fluid.

As in the single species and bidensity cases, different choices of integral conditions (ϕ0\phi_0 and χ0\chi_0) lead to different behaviors for the equilibrium solution. In the bidensity case, the authors of [26] are able to show the ODE for χ′\chi' takes the essential form of χ(1−χ)\chi(1-\chi), and subsequently χ\chi is a monotonically increasing function. While this result is intuitive in the bidensity case as it implies lighter particles will rise to the top, we show a similar result for the bidisperse case, instead showing larger particles rise to the top.

Proposition 1 (Brazil-nut Proposition). Given an integral condition ϕ0∈(0,ϕm)\phi_0 \in(0,\phi_m), χ0∈(0,1)\chi_0 \in(0,1) and assuming 0≤ϕ≤10 \le\phi\le1 and det⁡A≥0\det A \ge0, then the solution χ(s)\chi(s) to system (17) is monotonically increasing.

Note the assumption that 0≤ϕ≤10 \le\phi\le1 is natural as ϕ\phi represents the particle concentration. However, the assumption det⁡A≥0\det A \ge0 requires some explanation. Unfortunately, it is not always true that det⁡A≥0\det A \ge0. When d2/d1≲0.1d_2/d_1 \lesssim0.1 the determinant is actually negative. In experiments, we choose particle sizes well away from this limit. To prove Proposition 1, we will use the following lemma that orders the entries of the coupling matrix AA.

Lemma 1. In the coupling matrix AA defined in (10),

d224<A12<d124\frac{d_2^2}{4} < A_{12} < \frac{d_1^2}{4}

where A12A_{12} is the off diagonal entry.

The proof of lemma 1 can be found in the appendix. We now proceed with the proof of proposition 1.

Proof. Our goal is to determine the equilibrium points of χ′\chi' in terms of χ\chi and the sign on the interval between said equilibrium points. We note the denominator of χ′\chi' is positive (as we assume det⁡A\det A is positive). This leaves us with examining the numerator of χ′\chi', where we immediately note equilibria at χ=0\chi= 0 and χ=1\chi= 1. Since we assume ϕ<1\phi< 1, we only need to check the expression (2χ−1)d12d22+a(d12−(d12+d22)χ)(2\chi- 1)d_1^2d_2^2 + a(d_1^2 - (d_1^2+d_2^2)\chi), which is a linear expression in χ\chi and can be rewritten as (2d12d22−a(d12+d22))χ+(ad12−d12d22)(2d_1^2d_2^2 - a(d_1^2+d_2^2))\chi+ (ad_1^2 - d_1^2d_2^2). Recall that aa is the off diagonal entry of the coupling matrix AA divided by 14\frac{1}{4} as stated in equation (16). We compute this linear term at values χ=0\chi= 0 and χ=1\chi= 1. At χ=0\chi= 0, the expression evaluates to ad12−d12d22=d12(a−d22)ad_1^2 - d_1^2d_2^2 = d_1^2(a-d_2^2). As established in Lemma 1, d22<ad_2^2 < a, so the expression is positive at χ=0\chi= 0. Similarly, at χ=1\chi= 1, the expression evaluates to d12d22−ad22=d22(d12−a)d_1^2d_2^2-ad_2^2=d_2^2(d_1^2-a). But, since d12>ad_1^2>a, the expression is also positive at χ=1\chi= 1. Since the linear expression is positive at χ=0\chi= 0 and χ=1\chi= 1, it must be positive for all χ∈(0,1)\chi\in(0,1). Thus, χ′\chi' is positive for χ∈(0,1)\chi\in(0,1), and χ\chi as a solution to system (20) is monotone increasing.

Proposition 1 establishes that regardless of whether the solution is ridged or settled, the ratio of larger particles to the total number of particles increases as the height of the fluid profile increases. We can heuristically explain this observation by arguing sedimentation flux, which would otherwise promote stokes settling, is counterbalanced by the shear and tracer fluxes when on an incline. As the inclination angle decreases, the shear rate decreases and so do the shear and tracer fluxes. The settling flux does not scale with shear rate and thereby dominates the other two. In practice, the smaller particles vanish at the top of the equilibrium profiles which leaves only the larger particles at the top of the particle profile as in Figure 4.

Our model is consistent with the observations in granular convection and inclined granular flows that larger particles rise to the top, and provides a mathematical explanation in the equilibrium theory for why this should be the case. We note that our model does not include mechanisms that try to capture the ability of smaller particles being able to fit into the void space larger particles are not, unlike in other works [10, 18, 31]. In particular, our sedimentation term in Eq. (11) for a particle species ii does not explicitly depend on the concentration of the other particle species; we only include a hindered settling term that depends on the presence of both species of particles. While the maximum packing fraction ϕm\phi_m is a function of χ\chi, it does not impact differently sized particles differently. Our model also does not include wall effects which [12] and [15] propose as one of the potential explanations behind the Brazil-nut effect. While we do not claim to fully explain the Brazil-nut effect, we find it encouraging the Brazil-nut effect naturally arises in our model without including terms that try to account for it. However, our model assumes the particles are suspended in a viscous fluid and many simplifying assumptions were made with this assumption. It is unclear how dependent our Brazil-nut conclusion is on such assumptions. Furthermore, the Brazil-nut effect is obviously observed in more general settings as well. Still, it is promising the Brazil-nut effect arises in our model without specific consideration.

Conservation Law Model

With the ability to compute the equilibrium profiles that solve system (17), we are also able to compute the fluxes in Eq. (18) and thus advance the PDE system described in Eq. (19). We establish the structure of solutions to a three system of hyperbolic conservation laws as presented in [40]. We write the system in vector form so let U=(h,hϕ0,1,hϕ0,2)TU=(h,h\phi_{0,1},h\phi_{0,2})^T and F(U)=h3(f,g1,g2)TF(U)=h^3(f,g_1,g_2)^T resulting in the system

∂U∂t+∂∂x(F(U))=0.(20)\frac{\partial U}{\partial t}+\frac{\partial}{\partial x}\left(F(U)\right)=0. \tag*{(20)}

To understand the solution structure, we study the solution of Eq. (20) under Riemann initial conditions:

U(x,0)={ULx<0URx>0(21)U(x,0)= \begin{cases} U^L & x<0 \\ U^R & x>0 \end{cases} \tag*{(21)}

with the superscripts denoting the left or right end states respectively. We call the system (20) with initial conditions (21) the Riemann problem. Although our experiments and thus our numerical simulations have a finite volume initial condition, understanding of the solution structure for Riemann initial condition can be bridged to understanding for finite volume initial condition as done in the monodisperse case in [41]. Denoting JJ as the Jacobian of the system, we order the eigenvalues 0<λ1<λ2<λ30<\lambda_1<\lambda_2<\lambda_3 assuming the eigenvalues are real and positive. The eigenvalues depend on the state UU.

The weak solution to the Riemann problem transports the discontinuity from ULU^L to URU^R with a speed ss prescribed by the Rankine-Hugoniot jump condition:

F(UL)−F(UR)=s(UL−UR)(22)F(U^L)-F(U^R)=s\left(U^L-U^R\right) \tag*{(22)}

We impose the Lax entropy condition to define what are considered permissible shocks:

λk(UL)>s>λk(UR),λk+1(UR)>s>λk−1(UL)(23)\lambda_k(U^L)>s>\lambda_k(U^R),\qquad\lambda_{k+1}(U^R)>s>\lambda_{k-1}(U^L) \tag*{(23)}

for some kk. So, for a given state UU, the Hugoniot locus is the smooth curve that passes through UU and consists of all other states that solve equation (22) under the Lax entropy condition (23). The state UU can be taken as either the left or right state.

For a given left state ULU^L and right state URU^R we don’t expect a single ss to satisfy the Rankine-Hugoniot jump condition in (22) as that would require each component of the equality to be satisfied. This corresponds geometrically to there being no direct Hugoniot curve connecting the states ULU^L and URU^R. However, we can find intermediary states such that there is a sequence of intermediary states such that there exists a connection between ULU^L and URU^R by traveling along the Hugoniot curves of ULU^L, the intermediary states, and URU^R. In general, we expect there to be three intermediary states which leads to the triple shock structure we see in the settled regime [26]. However, in the ridged regime, we see a singular shock, a shock whose height is singular, instead. In the monodisperse [42] and bidensity case [26], this is due to the Hugoniot curves of the states approaching each other asymptotically in the far field instead of intersecting. We pose as an open problem that as the system transitions from the settled to the ridged regime, the Hugoniot curves of the states approach each other asymptotically in a way such that a single singular shock forms, as opposed to the multiple shock structure in the bidensity case. In the next sections, we present the laboratory experimental design and a few experiments where the Brazil-nut effect and evolution of the particle and liquid fronts can be seen.

Experiments

The goal of the experiments is to capture the separation of different-sized particles suspended in a liquid thin-film moving down an incline. Figure 5 shows the inclined track and table setup for the constant volume inclined plane experiments. The experiments are carried out on a flat one-meter-long acrylic track that can be adjusted at angles of 10−65∘10 - 65^\circ. This is the same track used in previous work [23, 25, 26]. Before running the experiment, the track is wiped with paper towels, and a gate is placed at the front of the reservoir seen at the top of the track. A well-mixed slurry is poured behind the gate and then quickly released to start the experiment. An aerial view of the experiment is filmed to capture the distance of each front (liquid, large particles, small particles) as seen in the sample frame in Figure 2b.

Setup for the constant volume inclined plane experiments

Figure 5. Setup for the constant volume inclined plane experiments.

The particles are hard, spherical beads made with soda lime glass (Ceroglass). They are 65 – 95% round with a density of 2.45−2.502.45-2.50 g/cm3^3. The beads used were ‘deco beads’ which are colored and processed by the manufacturer. The two species sizes we chose were light blue with a diameter of 0.2−0.40.2 - 0.4 mm and red with a diameter of 0.5−0.750.5 - 0.75 mm. The particles are mixed with PDMS silicone oil (Clearco Products) with a medium viscosity of 1,000cSt and a density of 0.971 g/cm3^3. In the next section, we show a few experiments and compare them with the model.

Model Case Studies Compared with Laboratory Experiments

We make qualitative comparisons between numerical simulations and physical experimental data. Figure 6c, Figure 6d, and Figure 6g are experiments and Figure 6a, Figure 6b, and Figure 6e are simulations done in the settled regime with parameters α=20∘\alpha= 20^\circ and ϕ0=0.4\phi_0 = 0.4. The parameter χ0\chi_0 is varied across the different runs. Figure 6h shows an example of the ridged regime with parameters α=55∘\alpha= 55^\circ, ϕ0=0.4\phi_0 = 0.4, and χ0=0.5\chi_0 = 0.5. For the numerical simulations the large particle diameter used is 0.625 mm and the small particle diameter used is 0.200 mm.

Figure 6. Comparison between numerical simulations and experimental results for various parameter choices. Figs. 6a and 6c: ϕ0=0.4\phi_0 = 0.4, χ0=0.25\chi_0 = 0.25. Figs. 6b and 6d: ϕ0=0.4\phi_0 = 0.4, χ0=0.5\chi_0 = 0.5. Figs. 6e and 6g: ϕ0=0.4\phi_0 = 0.4, χ0=0.75\chi_0 = 0.75. Figs. 6f and 6h: ϕ0=0.55\phi_0 = 0.55, χ0=0.5\chi_0 = 0.5. The track is 1 meter long. Note the simulation times do not match exactly to the experimental time.

For the numerical simulation, we assume a fixed volume initial condition and use an upwind scheme. The fixed volume is a rectangle of volume ϕ0\phi_0 supported from −8.4⋅10−2-8.4 \cdot10^{-2}m to 0m to model how the slurry initially sits in the reservoir. The numerical domain ranges from −8.4⋅10−2-8.4 \cdot10^{-2} to 1.2 with 514 grid points. Similar to previous numerical simulations [25] and [26], we assume the slurry remains well-mixed until a transition time t∗=150t^* = 150s and which afterwards the particles are assumed to evolve according to equation (19).

Overall, we observe excellent qualitative agreement between the numerical simulations and experimental data, with our model accurately capturing the key qualitative features of the experiments. First, the larger particles in both the settled and ridged regimes are always at the front of the particle ridge. This is explained with the understanding of the Brazil-nut effect established in Proposition 1. Since larger particles rise to the top according to the established equilibrium theory, the higher velocity field at the free surface of the fluid moves the larger particles faster downstream to the front of the slurry. When examining the side profiles of our experiments, the larger particles congregate at the top as shown in Figure 2.

When looking at the experiments in Figure 6, recall the larger (red) particles will appear on top of the smaller (blue) particles. So in aerial profiles, it can be harder to view the separation when there’s a high concentration of larger particles. Also, the simulations use a fixed diameter for each particle size. In the experiments, we have distributions of particle sizes and we use the mean to represent the diameter for comparison. Having distributions of sizes in the experiment can account for the lack of distinct separation.

Settled regime, χ0=0.25\chi_0 = 0.25

We begin with comparisons between the experimental data and numerical simulations in the settled regime. In Figure 6a and Figure 6c, we have the case with the parameters above and χ0=0.25\chi_0 = 0.25. In the numerical simulation in Figure 6a, the simulation predicts a settled regime. Behind the clear fluid front, we expect a sharp yet thin front of the larger red particles, followed by smaller light blue particles that make up the bulk of the slurry upstream. This is exactly what we see in Figure 6c, although the clear fluid front predicted in the numerical simulation appears much less than what is actually observed in experiments. However, the strong fingering instability seen in Figure 6c may contribute to this discrepancy.

Settled regime, χ0=0.5\chi_0 = 0.5

Next, in Figure 6b and Figure 6d, we have the case where χ0=0.5\chi_0 = 0.5. The numerical simulation in Figure 6b is similar to the previous case where we have a settled regime and a front of the larger red particles leading the particle front. However, this red front is thicker than the particle front we saw in the case where χ0=0.25\chi_0 = 0.25. Additionally, the boundary of the larger red particle front smoothly blends with the smaller light blue particles as we continue to go upstream. Again, this is a different from the case in figure 6c where the difference between the smaller and larger particle fronts is stark. We see in Figure 6d this is the case. Upstream from the clear fluid front is the red particle front. By comparing this particle front in Figure 6d to the particle front in Figure 6c, we do see the particle front in Figure 6d is thicker, and this is corroborated when comparing the numerical results in Figure 6b and Figure 6a. This large, red particle ridge blends with the small, light blue particles as we go upstream, which again is predicted in the numerical simulations.

Settled regime, χ0=0.75\chi_0 = 0.75

Finally, Figure 6b and Figure 6d show the case where χ0=0.75\chi_0 = 0.75. Once again, the larger red particles lead the particle front in the numerical simulation shown in Figure 6e. Similar to the case where χ0=0.5\chi_0 = 0.5, the large, red particle front blends with the small, light blue particles as we go upstream and appears well-mixed near the top of the incline. This behavior is shown in Figure 6g. The red particle ridge is distinctly thicker than the particle ridge in Figure 6c and Figure 6d, and smoothly transitions to a well-mixed blend of red and light blue particles as predicted in the numerical simulations. Overall, our numerical simulations are not only able to predict the flow being settled, but also predict detailed behavior of the particle fronts.

Ridged regime

We also have an example of an experiment in the ridged regime in Figure 6h. There are not as many distinct fronts to capture in the ridged case compared to the settled case which is why the majority of our comparisons so far have focused on the settled case. We see our numerical simulation in Figure 6f predicts the flow is in the ridged regime as the particles are present at the very front of the slurry. Again, the particle front is comprised mostly of the larger, red particles. Our numerical simulations also predict a singular shock. In the experimental results shown in Figure 6h, we see the particles at the very front of the flow establishing this experiment is in the ridged regime. Furthermore, the particle front is comprised mostly of the larger, red particles as we expect from the numerical simulations.

We do not see the explicit presence of a singular shock at the particle front, which is expected. Our model is a hyperbolic conservation law where singular shocks can occur, while in reality there are surface tension effects which dampen shock formation. We can see in both the simulations and in the experiment the particle concentration at the front is much higher than that in the settled regime, a hallmark of the ridged regime. In the numerical simulations, we are able to determine the particle volume fraction approaches the particle maximum packing fraction. Although we are not able to determine this in the experiment, visual inspection of the red particle ridge in Figure 6h suggests the particle volume fraction also approaches the particle maximum packing fraction.

Conclusion

In this work, we have a presented a conservation law model for particle laden flows on an incline of two differently sized species of particles by extending the thin film equilibrium and conservation law model explored by [25] and [26]. For our equilibrium model, we expand upon the flux balances presented by [26] but include coupling effects caused by the different sizes of particles introduced by [29] and the dependence of the maximum packing fraction based on the presence of differently sized particles [28]. In our analysis of the equilibrium model, we find larger particles are expected to rise to the top of the profile regardless of whether the profile is in the settled or ridged regime. This observation is supported by our experimental results where both the larger particles rise to the top of side profiles during experiments and where larger particles congregate at the end of the particle front. This finding is similar to the Brazil-nut effect observed in granular flows. In our model, the Brazil-nut effect arises as a natural consequence in the equilibrium theory without having to make specific concessions for it when modeling the flux terms. Our setting is quite different than traditional granular flow as the particles are submerged in a viscous fluid and we are able to reduce the complexity of the model by leveraging thin film approximations. However, we pose as an open question if the Brazil-nut effect can arise naturally in granular flow models without the express need to include it in the mathematical model.

We next perform numerical simulations of the conservation law with a fixed volume initial condition and compare them with experimental results. Overall, our qualitative agreement is excellent. Not only is our model able to accurately predict whether the flow is in the settled or ridged regime, but our model is also able to predict finer features of the flow particularly in the settled regime such as particle ridge width and well-mixed behavior upstream. This qualitative matching is encouraging and suggests the overall modeling approach is sound.

However, we believe there are improvements to be made in the model. In our equilibrium models, although we adapt the shear-induced migration for two particle species of different sizes from [29], we use the same settling and tracer fluxes from [26] except we change the particle diameters to be different depending on the species. It is unclear whether more accurate modeling needs to be done for these two effects. For example, the original formulation of tracer flux in [26] has the property that the sum of tracer fluxes of the two species cancels each other out, which is not true in our formulation of the tracer flux because of the differing diameters of the particle species. Additionally, as stated previously, our settling flux does not include size effects. It would be interesting to include size effects in the settling flux and see whether the same qualitative features emerge in the numerical simulations.

This goal of this work is to provide a fundamental model that accurately describes particle-laden flow behavior in thin-films. Although we do not qualitatively match the front positions between numerical simulations and experiments, we are still able to present a model that is simple enough to solve numerically and make predictions that well match the behavior of the experiments. We also rigorously explain the phenomena of large particles rising to the free surface and appearing at the front of the particle region by proving that larger particles proportionally increase as the distance from the substrate increases. We are interested to see how the modeling choices or improvements suggested in the previous paragraph lead to different qualitative observations or quantitative agreements.

Acknowledgments

This material is based upon work supported by the U.S. National Science Foundation under award No. DMS-2407006. This work is also supported by Simons Math + X Investigator Award number 510776. Sarah C. Burnett was supported by the 2022 L’Oréal USA for Women in Science Postdoctoral Fellowship. Preliminary experimental work supporting this research was done during the Research Experience for Undergraduates (REU) program at University of California, Los Angeles during Summer 2022, 2023, & 2024.

Appendix A

We give the proof of lemma 1, stated again below.

Lemma 2. In the coupling matrix AA when d=3d = 3 defined in (10),

d224<a12<d124\frac{d_{2}^{2}}{4} < a_{12} < \frac{d_{1}^{2}}{4}

where a12a_{12} is the off diagonal entry.

Proof. We note we can divide d12/4d_{1}^{2}/4 from all terms in the desired inequality. For the off diagonal entry, we can factor out d12/4d_{1}^{2}/4 yielding

a12=d124(1+d2/d1)22d+1(1+d2/d1)d1+(d2/d1)d.a_{12} = \frac{d_{1}^{2}}{4}\frac{(1+d_{2}/d_{1})^{2}}{2^{d+1}}\frac{(1+d_{2}/d_{1})^{d}}{1+(d_{2}/d_{1})^{d}}.

So, we wish to show the equivalent inequality

d22d12<(1+d2/d1)22d+1(1+d2/d1)d1+(d2/d1)d<1\frac{d_{2}^{2}}{d_{1}^{2}} < \frac{(1+d_{2}/d_{1})^{2}}{2^{d+1}}\frac{(1+d_{2}/d_{1})^{d}}{1+(d_{2}/d_{1})^{d}} < 1

Now that we have reduced the inequality to one functional variable d2d1\frac{d_{2}}{d_{1}}, we set d2d1=1−ϵ\frac{d_{2}}{d_{1}} = 1-\epsilon, with 0<ϵ<10 < \epsilon< 1. This is justified as we assume 0<d2<d10 < d_{2} < d_{1}. Thus, we simplify the modified off diagonal entry as

a~12=(2−ϵ)22d+1(2−ϵ)d1+(1−ϵ)d.\tilde{a}_{12} = \frac{(2-\epsilon)^{2}}{2^{d+1}}\frac{(2-\epsilon)^{d}}{1+(1-\epsilon)^{d}}.

To see that a~12<1\tilde{a}_{12} < 1, we observe

(2−ϵ)22d+1(2−ϵ)d1+(1−ϵ)d<2d+22⋅2d+1<1.\begin{aligned} \frac{(2-\epsilon)^{2}}{2^{d+1}}\frac{(2-\epsilon)^{d}}{1+(1-\epsilon)^{d}} < \frac{2^{d+2}}{2\cdot2^{d+1}} \\ &< 1. \end{aligned}

The other inequality, to show that a~12>(1−ϵ)2\tilde{a}_{12} > (1-\epsilon)^{2} is trickier. We will show that when ϵ∈(0,1)\epsilon\in(0,1), the inequality

(2−ϵ)22d+1(2−ϵ)d1+(1−ϵ)d−(1−ϵ)2>0\frac{(2-\epsilon)^{2}}{2^{d+1}}\frac{(2-\epsilon)^{d}}{1+(1-\epsilon)^{d}} - (1-\epsilon)^{2} > 0

holds. This is equivalent to showing the polynomial

(1+x5)−24x2(1+x3)(1+x^{5}) - 2^{4}x^{2}(1+x^{3})

where x=1−ϵx = 1-\epsilon is positive on the domain (0,1)(0,1). We may factor the polynomial as

−(x−1)(x+1)(15x3−5x2+5x+1).-(x-1)(x+1)(15x^{3}-5x^{2}+5x+1).

We note −(x−1)(x+1)>0-(x-1)(x+1) > 0 when x∈(0,1)x \in(0,1), so now all we have to do is verify the sign of (15x3−5x2+5x+1)(15x^{3}-5x^{2}+5x+1) on (0,1)(0,1). To do so, we will show the polynomial is monotone by examining its derivative, 5(9x2−2x+1)5(9x^{2}-2x+1). Since the discriminant 4−4⋅9⋅1=−324-4\cdot9\cdot1=-32 is negative, the derivative has no real roots and since it is positive at x=0x=0, the derivative is positive for all xx. Thus, the original polynomial is monotone increasing. The polynomial evaluated at x=0x=0 is 1, and thus the polynomial is positive on the interval (0,1)(0,1), and we conclude (1−ϵ)2<a~122(1-\epsilon)^{2} < \tilde{a}_{12}^{2}. This establishes the desired inequality (1−ϵ)2<a~12<1(1-\epsilon)^{2} < \tilde{a}_{12} < 1.

References

  1. [1]N. Santangelo, G. Forte, M. De Falco, G. B. Chirico, and A. Santo, New insights on rainfall triggering flow-like landslides and flash floods in Campania (Southern Italy), Landslides 18, 2923 (2021).DOI
  2. [2]É. Guazzelli, Rheology of dense granular suspensions across flow regimes, Physical Review Fluids 9, 090501 (2024).DOI
  3. [3]M. Larcher and J. T. Jenkins, The influence of granular segregation on gravity-driven particle-fluid flows, Advances in Water Resources 129, 365 (2019).
  4. [4]N. J. Balmforth, R. V. Craster, P. Perona, A. C. Rust, and R. Sassi, Viscoplastic dam breaks and the Bostwick consistometer, Journal of Non-Newtonian Fluid Mechanics Viscoplastic fluids: From theory to application, 142, 63 (2007).DOI
  5. [5]C. Mouquet, V. Greffeuille, and S. Treche, Characterization of the consistency of gruels consumed by infants in developing countries: assessment of the Bostwick consistometer and comparison with viscosity measurements and sensory perception, International Journal of Food Sciences and Nutrition 57, 459 (2006).pubmed.ncbi.nlm.nih.gov/17162325
  6. [6]M. M. Tehrani and A. Ghandi, Modification of Bostwick method to determine tomato concentrate consistency, Journal of Food Engineering 79, 1483 (2007).DOI
  7. [7]D. C. Wright, A spiral separator (EU Patent 0039139). (1981).
  8. [8]Q. Dehaine and L. O. Filippov, Modelling heavy and gangue mineral size recovery curves using the spiral concentration of heavy minerals from kaolin residues, Powder Technology 292, 331 (2016).DOI
  9. [9]A. B. Holland-Batt and P. N. Holtham, Particle and fluid motion on spiral separators, Minerals Engineering 4, 457 (1991).DOI
  10. [10]A. Rosato, K. J. Strandburg, F. Prinz, and R. H. Swendsen, Why the Brazil nuts are on top: Size segregation of particulate matter by shaking, Physical Review Letters 58, 1038 (1987).pubmed.ncbi.nlm.nih.gov/10034316
  11. [11]R. Jullien and P. Meakin, A mechanism for particle size segregation in three dimensions, Nature 344, 425 (1990).DOI
  12. [12]J. B. Knight, H. M. Jaeger, and S. R. Nagel, Vibration-induced size separation in granular media: The convection connection, Physical Review Letters 70, 3728 (1993).pubmed.ncbi.nlm.nih.gov/10053947
  13. [13]M. E. Möbius, B. E. Lauder dale, S. R. Nagel, and H. M. Jaeger, Size separation of granular particles, Nature 414, 270 (2001).DOI
  14. [14]Y. Fan and K. M. Hill, Phase Transitions in Shear-Induced Segregation of Granular Materials, Physical Review Letters 106, 218301 (2011).pubmed.ncbi.nlm.nih.gov/21699345
  15. [15]J. B. Knight, E. E. Ehrichs, V. Y. Kuperman, J. K. Flint, H. M. Jaeger, and S. R. Nagel, Experimental study of granular convection, Physical Review E 54, 5726 (1996).pubmed.ncbi.nlm.nih.gov/9965760
  16. [16]P. Gajjar, C. G. Johnson, J. Carr, K. Chrysafis, J. M. N. T. Gray, and P. J. Withers, Size segregation of irregular granular materials captured by time-resolved 3D imaging, Scientific Reports 11, 8352 (2021).DOI
  17. [17]S. Burnett, Q. Luan, S. Bloom, L. Ding, and A. L. Bertozzi, Separation of bidisperse particles in viscous thin-film flow down an incline, Gallery of Fluid Motion, APS Division of Fluid Dynamics Meeting 2024 (2024).
  18. [18]S. B. Savage and C. K. K. Lun, Particle size segregation in inclined chute flow of dry cohesionless granular solids, Journal of Fluid Mechanics 189, 311 (1988).DOI
  19. [19]A. Neveu, M. Larcher, R. Delannay, J. T. Jenkins, and A. Valance, Particle segregation in inclined high-speed granular flows, Journal of Fluid Mechanics 935, A41 (2022).DOI
  20. [20]J. R. Johnson, Particle segregation and what to do about it, Chemical Engineering 8, 183 (1978).
  21. [21]N. Murisic, B. Pausader, D. Peschka, and A. L. Bertozzi, Dynamics of particle settling and resuspension in viscous liquid films, Journal of Fluid Mechanics 717, 203 (2013).
  22. [22]H. E. Huppert, Flow and instability of a viscous current down a slope, Nature 300, 427 (1982).DOI
  23. [23]J. Zhou, B. Dupuy, A. Bertozzi, and A. Hosoi, Theory for shock dynamics in particle-laden thin films, Physical Review Letters 94, 117803 (2005).pubmed.ncbi.nlm.nih.gov/15903891
  24. [24]B. P. Cook, Theory for particle settling and shear-induced migration in thin-film liquid flow, Physical Review E 78, 045303 (2008).pubmed.ncbi.nlm.nih.gov/18999483
  25. [25]N. Murisic, J. Ho, V. Hu, P. Latterman, T. Koch, K. Lin, M. Mata, and A. Bertozzi, Particle-laden viscous thin-film flows on an incline: Experiments compared with a theory based on shear-induced migration and particle settling, Physica D: Nonlinear Phenomena 240, 1661 (2011).DOI
  26. [26]J. T. Wong and A. L. Bertozzi, A conservation law model for bidensity suspensions on an incline, Physica D: Nonlinear Phenomena 330, 47 (2016).DOI
  27. [27]S. Lee, J. Wong, and A. L. Bertozzi, Equilibrium Theory of Bidensity Particle-Laden Flows on an Incline, in Mathematical Modelling and Numerical Simulation of Oil Pollution Problems, edited by M. Ehrhardt (Springer International Publishing, Cham, Switzerland, 2015) pp. 85–97.DOI
  28. [28]A. Shauly, A. Wachs, and A. Nir, Shear-induced particle migration in a polydisperse concentrated suspension, Journal of Rheology 42, 1329 (1998).DOI
  29. [29]P. Kanehl and H. Stark, Hydrodynamic segregation in a bidisperse colloidal suspension in microchannel flow: A theoretical study, J. Chem. Phys. 142, 214901 (2015).DOI
  30. [30]A. A. Howard, M. R. Maxey, and S. Gallier, Bidisperse suspension balance model, Physical Review Fluids 7, 124301 (2022).DOI
  31. [31]A. R. Thornton, J. M. N. T. Gray, and A. J. Hogg, A three-phase mixture theory for particle size segregation in shallow granular free-surface flows, Journal of Fluid Mechanics 550, 1 (2006).DOI
  32. [32]L. Ding, S. C. Burnett, and A. L. Bertozzi, Equilibrium theory of bidensity particle-laden suspensions in thin-film flow down a spiral separator, Physics of Fluids 37, 023397 (2025).arxiv.org/abs/2410.23568
  33. [33]I. M. Krieger and T. J. Dougherty, A mechanism for non-newtonian flow in suspensions of rigid spheres, Transactions of The Society of Rheology 3, 137 (1959).DOI
  34. [34]T. Ward, C. Wey, R. Glidden, A. E. Hosoi, and A. L. Bertozzi, Experimental study of gravitation effects in the flow of a particle-laden thin film on an inclined plane, Physics of Fluids 21, 083305 (2009).DOI
  35. [35]D. Leighton and A. Acrivos, The shear-induced migration of particles in concentrated suspensions, Journal of Fluid Mechanics 181, 415 (1987).DOI
  36. [36]R. J. Phillips, R. C. Armstrong, R. A. Brown, A. L. Graham, and J. R. Abbott, A constitutive equation for concentrated suspensions that accounts for shear-induced particle migration, Physics of Fluids A: Fluid Dynamics 4, 30 (1992).DOI
  37. [37]A. Tripathi and A. Acrivos, Viscous resuspension in a bidensity suspension, International Journal of Multiphase Flow 25, 1 (1999).DOI
  38. [38]D. Leighton and A. Acrivos, Measurement of shear-induced self-diffusion in concentrated suspensions of spheres, Journal of Fluid Mechanics 177, 109 (1987).DOI
  39. [39]A. Sierou and J. F. Brady, Shear-induced self-diffusion in non-colloidal suspensions, Journal of Fluid Mechanics 506, 285 (2004).DOI
  40. [40]P. D. Lax, 1. Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock Waves, in Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock Waves, CBMS-NSF Regional Conference Series in Applied Mathematics (Society for Industrial and Applied Mathematics, 1973) pp. 1–48.DOI
  41. [41]L. Wang, A. Mavromoustaki, A. L. Bertozzi, G. Urdaneta, and K. Huang, Rarefaction-singular shock dynamics for conserved volume gravity driven particle-laden thin film, Physics of Fluids 27, 033301 (2015).
  42. [42]L. Wang and A. L. Bertozzi, Shock solutions for high concentration particle-laden thin films, SIAM Journal on Applied Mathematics 74, 322 (2014).DOI

Paper details

vlnc:W_2c3889n3
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