betaFeedback

WorksPhysics and Astronomy

Interface reconstruction and advection schemes for volume of fluid method in axisymmetric coordinates

Abstract

Volume of fluid(VOF) method is a sharp interface method employed for simulations of two phase flows. Interface in VOF is usually represented using piecewise linear line segments in each computational grid based on the volume fraction field. While VOF for cartesian coordinates conserve mass exactly, existing algorithms do not show machine-precision mass conservation for axisymmetric coordinate systems. In this work, we propose analytic formulae for interface reconstruction in axisymmetric coordinates, similar to those proposed by Scardovelli and Zaleski (J. Comput. Phys. 2000) for cartesian coordinates. We also propose modifications to the existing advection schemes in VOF for axisymmetric coordinates to obtain higher accuracy in mass conservation

Introduction

Multiphase flows are ubiquitious in several industrial applications. In the last three decades, there has been a surge in the numerical methods and algorithms for simulations of complex multiphase flows. There have been several different types of interface capturing strategies that have been proposed for two-phase flows. The most popular of these are the Level set method, Volume of Fluid (VOF) method, and Front tracking scheme [1, 2]. VOF methods with geometric advection strictly conserve the volume of the two phases.

Several improvements have been made since the inception of the method (see Hirt and Nichols[3, 4, 5, 6, 7, 8]). The most simplest and earliest representation for interface reconstruction is simple line interface calculation(SLIC) in which the interface is approximated by horizontal or vertical lines. Subsequently, piecewise line interface construction (PLIC) was introduced where the interface is approximated as a linear line at an angle in the cell [9]. Higher order interface construction have been proposed (such as Parabolic reconstruction by [4]), but considering the associated computational cost and complexity for geometric advection, PLIC is usually preferred.

Scradovilli and Zaleski[10] proposed analytical formulae for the piecewise linear reconstruction of the interface in cartesian coordinates that led to a significant speedup over the earlier iterative schemes. However, for curvilinear coordinates (such as axisymmetric coordinate system), the proposed analytical formulae cannot be employed directly. In the present work, we derive similar analytic formulae for axisymmetric coordinates, which result in a speedup of ∼ 28 over the iterative counterparts (Brent’s root finding method). Further, we demonstrate that the existing interface advection schemes in VOF for axisymmetric coordinates are not strictly mass conserving. In this study, we propose modifications to the current operator split algorithms that result in machine-precision mass conservation in axisymmetric coordinates. We show the efficacy of the proposed algorithms using several test cases.

The paper is organized as follows. We first present analytical formulae for the interface reconstruction schemes in axisymmetric coordinates in section 2. In section 3, we propose modifications in the existing interface advection algorithm for axisymmetric VOF and present test cases to show the efficacy of the scheme. Finally, in section 4, we discuss the important conclusions.

Interface Reconstruction Scheme

Interface reconstruction in the volume of fluid (VOF) method requires the volume fraction field. Using the volume fraction field, a piece-wise linear or a higher order interface is constructed in a given grid-cell. Interface reconstruction is an integral part of the geometric advection schemes to ensure mass conservation property of the VOF method [2]. Initial condition for a multiphase flow simulation requires the initial distribution of the volume fraction field, usually provided as an implicit function of the spatial coordinates. VOFI library [11] is an open source library to initialise the liquid volume fraction field in cartesian coordinate systems accurately. In VOFI, for cells cut by the interface (see Figure 1), PLIC reconstruction method [9, 6] is employed to approximate the interface as a linear line segment,

Figure 1. Coordinates of the vertices of a simple polygon cut by the interface ordered counter clockwise. Here x3,y3x_3, y_3 and x4,y4x_4, y_4 represents the end points of the PLIC line segment with interface normal m\mathbf{m} and line constant aa.

m⋅x=a,(1)\mathbf{m} \cdot\mathbf{x} = a, \tag*{(1)}

where m\mathbf{m} is the local normal at the interface, x\mathbf{x} is a point on the plane, and aa is the normal distance of the origin from the plane. Analytical relation, given by Scardovelli and Zaleski [10], between the volume fraction and the line constant is employed to determine the line constant aa. Thus, for two-dimensional and three-dimensional Cartesian coordinate systems, VOFI library can be directly employed for accurate assignment of the initial volume fraction field on a given discretized domain using an implicit equation of the interface. However, in curvilinear coordinate systems, for a given implicit function, the piece-wise linear interface constructed from VOFI would require computation of the volume fraction field using a formula specific to the curvilinear coordinates. For instance, for axisymmetric coordinate system, the modified Gauss area (shoelace) formula for computation of the area of a convex polygon is given by,

V=π3∣∑i=1n(xi+xi+1)(xiyi+1−xi+1yi)∣(2)V = \frac{\pi}{3} \left| \sum_{i=1}^{n} (x_i + x_{i+1})(x_i y_{i+1} - x_{i+1} y_i) \right| \tag*{(2)}

where (xi,yi)(x_i, y_i) for i=1,…,ni = 1, \ldots, n (with xn+1=x1x_{n+1} = x_1 and yn+1=y1y_{n+1} = y_1) are the coordinates of the vertices of a convex polygon ordered counter clockwise as shown in the Figure 1.

Thus, for initialization of the volume fraction field, CC, once we obtain the linear interface in each grid cell using the VOFI library, we use the above formula to compute the volume, VV, and assign volume fraction in each grid cell as,

C=V2πrcΔxΔy.(3)C = \frac{V}{2\pi r_c \Delta x \Delta y}. \tag*{(3)}

Here rcr_c is the distance of the center of the cell from the axis of symmetry, and Δx\Delta x and Δy\Delta y are the grid-cell sizes in the radial (rr) and axial (yy) directions, respectively. We note that the above procedure is followed essentially to minimize the error in the volume-fraction during initialization. To illustrate this, we initialise a torus of minor radius rt=0.25r_t = 0.25 and major radius r=0.50r = 0.50 in the center of a computational domain of size 1×11 \times1 as shown in the Figure 7. The volume of the torus can be analytically computed as Vt=2π2rrt2V_t = 2\pi^2 r r_t^2, where the major radius, rr, is the distance to the center of the torus from the axis of symmetry. We compare the results for various grid sizes with the results obtained using the popular VOF based open source flow solver, Gerris[12], given in Table 1.

GridCurrent SolverGerris Solver
16 × 165.4×10−165.4 \times10^{-16}9.4×10−39.4 \times10^{-3}
32 × 327.1×10−167.1 \times10^{-16}2.6×10−32.6 \times10^{-3}
64 × 641.8×10−161.8 \times10^{-16}5.5×10−45.5 \times10^{-4}
128 × 1285.4×10−165.4 \times10^{-16}1.6×10−41.6 \times10^{-4}

Table 1. Relative error in volume: E=∣V−Vt∣VtE = \frac{\lvert V - V_t\rvert}{V_t}

Figure 7. Interface shape after 1000 time steps forward and backward after advecting the torus of radius .25 with a grid Courant(CFL) number of 1.

Figure 2. Reconstructed interface for 16 mesh points of a torus of radius rt=0.25r_t = 0.25 and major axis r=0.50r = 0.50 initialised at the center of 1×11 \times1 domain where the dots represent the mid–points of the reconstructed PLIC line segments.

Thus, we have shown that for curvilinear coordinates volume fraction field can be initialized up to machine accuracy. Now, we derive an analytic relation for PLIC reconstruction for axisymmetric coordinates on the lines of Scardovelli and Zaleski [10]. We use Youngs method[9] to get the interface normal (mm in equation.1) from fluid-1 (C=1C = 1) to fluid-2 C=0C = 0 and is given by m=−∇C/∣∇C∣m = -\nabla C/\lvert\nabla C\rvert. To complete the PLIC interface reconstruction for a given CC, in addition to the normal mm, we also need to obtain the line constant, aa, which is the normal distance of the interface from one of the vertices of the computational cell. In what follows, we present a methodology to get the line constant(aa) analytically for a given interface normal vector and the volume fraction of a mixed cell.

As discussed in [10], using an analytical relation between the volume fraction(CC), interface normal (mm) and the line constant (aa), we can implement an if − else − if − end − if construct to determine the line constant, aa. This approach is computationally much more efficient compared to the alternative iterative approach to get the line constant. Given the equation of the interface, m1x+m2y=am_1x + m_2y = a, all combinations of m1m_1, m2m_2 (such that m12+m22=1m_1^2 + m_2^2 = 1) can be reduced to one of the cases shown in Figure 3, either by changing the origin or by changing the reference fluid from fluid-1 to fluid-2, such that both m1m_1 and m2m_2 are positive and the left bottom corner of the mixed cell under consideration is contained in fluid-1. Figure 3 shows all the possible configurations for interface arrangement with m1≥0m_1 \ge0 and m2≥0m_2 \ge0.

Figure 3. Various cases which can arise in the standard configuration of the interface with m1,m2≥0m_1,m_2 \ge0 and fluid 1 is occupying the bottom left corner of the cell.

We first discuss Case A shown in the figure 3 and similar procedure can be followed to obtain relations for other cases. For the axisymmetric coordinate system, the shaded area shown in the figure 3 for Case A is given by,

V=π3(x0+x2)3m1m2−π3(x0+x1)3m1m2−πx02y1.(4)V = \frac{\pi}{3}(x_0 + x_2)^3 \frac{m_1}{m_2} - \frac{\pi}{3}(x_0 + x_1)^3 \frac{m_1}{m_2} - \pi x_0^2 y_1. \tag*{(4)}

Using the equation of line m1x+m2y=am_1x + m_2y = a we have x1=a/m1−(m2Δy)/m1x_1 = a/m_1 - (m_2\Delta y)/m_1 and x2=a/m1x_2 = a/m_1. In the present study, we assume Δx=Δy\Delta x = \Delta y, but the same analysis can be easily extended for Δx≠Δy\Delta x \ne\Delta y. Substituting x1x_1 and x2x_2 in the equation (4) and collecting the terms in powers of aa, we obtain,

(πΔym12)a2+(−πm2Δy2m12+2πΔyx0m1)a+(πm22Δy33m12−πΔy2m2x0m1)=V.(5)\left(\frac{\pi\Delta y}{m_1^2}\right)a^2+\left(-\frac{\pi m_2\Delta y^2}{m_1^2}+\frac{2\pi\Delta y x_0}{m_1}\right)a+\left(\frac{\pi m_2^2\Delta y^3}{3m_1^2}-\frac{\pi\Delta y^2m_2x_0}{m_1}\right)=V. \tag*{(5)}

Thus, we have an analytical relation between the volume and the line constant aa. We note that the above relation holds true only when the interface cuts through the top and the bottom edges of the cell shown for Case A in Figure 3: (x2−x0)≤Δx(x_2-x_0)\le\Delta x, y1=Δyy_1=\Delta y and y2=0y_2=0. These conditions yield the bounds on the values of aa: m2Δy≤a≤m1Δxm_2\Delta y\le a\le m_1\Delta x. Substituting the above bounds for aa in the equation (5), yield the bounds on the limiting volumes:

V1=πΔy(3Δx2m12+3Δxm1(2m1x0−Δym2)+Δym2(Δym2−3m1x0))3m12.(6)V_1=\frac{\pi\Delta y\left(3\Delta x^2m_1^2+3\Delta xm_1\left(2m_1x_0-\Delta ym_2\right)+\Delta ym_2\left(\Delta ym_2-3m_1x_0\right)\right)}{3m_1^2}. \tag*{(6)}

and

V2=πΔy2m2(Δym2+3m1x0)3m12.(7)V_2=\frac{\pi\Delta y^2m_2\left(\Delta ym_2+3m_1x_0\right)}{3m_1^2}. \tag*{(7)}

For a given volume fraction CC and interface normal (m1,m2)(m_1,m_2), the volume occupied by fluid-1 in the configurations shown in the Figure 3 is given by: V=2πrΔxΔyCV=2\pi r\Delta x\Delta yC where r=x0+Δx/2r=x_0+\Delta x/2 is the distance from the axis of symmetry to the cell center. If V1≤V≤V2V_1\le V\le V_2 then the analytical relation given by equation (5) can be used to determine the line constant aa. For the quadratic equation in aa given by equation. (5), we note that only one of the roots will satisfy the required bounds on aa for case A. For cases B and C, we obtain cubic equations that can be solved for aa using Cardano’s formula or using Brent’s method to find the appropriate root with necessary bounds for the line constant. Following the same approach we can get bounds on volume for case D as:

V3=πΔy2(−2Δym1+3Δym2−3m1x0+6m2x0)3m2.(8)V_3=\frac{\pi\Delta y^2\left(-2\Delta ym_1+3\Delta ym_2-3m_1x_0+6m_2x_0\right)}{3m_2}. \tag*{(8)}

and

V4=πΔy2m1(Δy+3x0)3m2.(9)V_4=\frac{\pi\Delta y^2m_1\left(\Delta y+3x_0\right)}{3m_2}. \tag*{(9)}

Figure 4 shows all the possible configurations of the interface and the volume bounds which separate each case as the interface normal in radial direction varies from minimum to a maximum. Figure 4 clearly shows that the various bounds for the cases shown in Figure 3 do not overlap and provide a unique criterion for computing the line constant aa. Thus, we can use the following algorithm to classify each case.

Figure 4. Bounds for different cases shown in Figure 3 as a function of the x-component of the interface normal, m1m_1. The interfacial cell is placed at a distance of 1 from the axis of symmetry with Δx=Δy=1\Delta x = \Delta y = 1. The total volume of the cell is Vcell=2πV_{\mathrm{cell}} = 2\pi given by the top boundary in the plot.

Data: VV, m1m_1

Result: Identification of the case in which the standard interface belongs to.

if m1≥12m_1 \ge\frac{1}{\sqrt{2}} then if V≥V1V \ge V_1 then Case B else if V≤V2V \le V_2 then Case C else Case A // V1>V>V2V_1 > V > V_2 end else if V≥V3V \ge V_3 then Case B else if V≤V4V \le V_4 then Case C else Case D // V3>V>V4V_3 > V > V_4 end end

Algorithm 1: Classification of the standard case of the reconstructed interface where m1,m2≥0m_1, m_2 \ge0.

We list below the analytical relation between the line constant(aa) and the volume(VV) for each case:

Case A

(πΔym12)a2+(−πm2Δy2m12+2πΔyx0m1)a+(πm22Δy33m12−πΔy2m2x0m1)=V.(10)\left(\frac{\pi\Delta y}{m_1^2}\right)a^2+\left(-\frac{\pi m_2\Delta y^2}{m_1^2}+\frac{2\pi\Delta y x_0}{m_1}\right)a+\left(\frac{\pi m_2^2\Delta y^3}{3m_1^2}-\frac{\pi\Delta y^2m_2x_0}{m_1}\right)=V. \tag*{(10)}

Case BB

−(π3m12m2)a3+(πΔym12−πx0m1m2)a2+(π(Δy+x0)2m2−πm2Δy2m12+2πΔyx0m1−πx02m0)a+(−πΔy(Δy+x0)2m1m2+πm2Δy33m12−πΔy2m2x0m1−πm1x033m2+πm1(Δy+x0)33m2)=V(11)-\left(\frac{\pi}{3m_1^2m_2}\right)a^3+\left(\frac{\pi\Delta y}{m_1^2}-\frac{\pi x_0}{m_1m_2}\right)a^2+\left(\frac{\pi(\Delta y+x_0)^2}{m_2}-\frac{\pi m_2\Delta y^2}{m_1^2}+\frac{2\pi\Delta yx_0}{m_1}-\frac{\pi x_0^2}{m_0}\right)a+\left(-\frac{\pi\Delta y(\Delta y+x_0)^2m_1}{m_2}+\frac{\pi m_2\Delta y^3}{3m_1^2}-\frac{\pi\Delta y^2m_2x_0}{m_1}-\frac{\pi m_1x_0^3}{3m_2}+\frac{\pi m_1(\Delta y+x_0)^3}{3m_2}\right)=V \tag*{(11)}

Case CC

(π3m12m2)a3+(πx0m1m2)a2=V(12)\left(\frac{\pi}{3m_1^2m_2}\right)a^3+\left(\frac{\pi x_0}{m_1m_2}\right)a^2=V \tag*{(12)}

Case DD

a=2πΔy3m1+3πΔy2m1x0+3m2V3πΔy(Δy+2x0)(13)a=\frac{2\pi\Delta y^3m_1+3\pi\Delta y^2m_1x_0+3m_2V}{3\pi\Delta y(\Delta y+2x_0)} \tag*{(13)}

We note here that the other cases can be readily transformed into one of the cases listed in the Figure 3 by either changing the fluid (by using (1−C)(1-C) instead of CC to compute the volume and inverting the interface normal mm) or by changing the origin (keeping the location of the axis-of-symmetry same but inverting its direction).

We now compare the above described analytical method with the iterative method for finding the line constant for the case A given in Figure 3. The relative error in the line constant is given in Table 2 for the analytical and iterative methods with different tolerances. We note that the iterative method to reach an error with a tolerance of 10−810^{-8} is about 28 slower compared to the analytical method.

Comparision between the analytical and iterative reconstruction methods
MethodToleranceRelative Errortiterative/tanalyticalt_{iterative}/t_{analytical}
Analytical—01
Iterative1.0×10−41.0 \times10^{-4}1.8×10−41.8 \times10^{-4}14.2
Iterative1.0×10−61.0 \times10^{-6}1.1×10−61.1 \times10^{-6}21.2
Iterative1.0×10−81.0 \times10^{-8}6.7×10−116.7 \times10^{-11}28.3

Table 2. Relative error in line constant aa and the ratio of CPU time required by iterative method (Brent’s algorithm) to that required by analytical method for 10000 repetitions for case A shown in Figure 3 for different tolerances used in the iterative method.

Once the line constant, aa, is obtained, the position of the endpoints of the linear approximation of the interface can be computed thus completing the construction of a planar interface in a given computational cell. As discussed earlier, this more precise description of the interface within the grid cell allows geometric advection which gives the VOF method its strict mass conservation property while maintaining a sharp interface.

In what follows, we discuss an operator split algorithm for the geometric advection of the interface in axisymmetric coordinates. We note that the straightforward extension of the 2D cartesian algorithm does not yield accurate results, as also indicated by the results obtained from the existing open source codes.

Advection of the Interface

We present here a scheme for accurate geometric advection of the volume fraction in the axisymmetric coordinates. We have used a uniform grid to describe the variables with volume fraction being stored at the cell centers (Ci,j)(C_{i,j}). The incompressible fluid flow is determined by the velocity field which is defined at the cell faces (ui+1/2,j,vi,j+1/2)(u_{i+1/2,j}, v_{i,j+1/2}). Here, uu denotes the radial direction velocity and vv is the axial velocity. The velocity field satisfies the discrete divergence free condition given by:

(ru)i+12,j−(ru)i−12,jriΔx+vi,j+12−vi,j−12Δy=0.(14)\frac{(ru)_{i+\frac{1}{2},j}-(ru)_{i-\frac{1}{2},j}}{r_i\Delta x}+\frac{v_{i,j+\frac{1}{2}}-v_{i,j-\frac{1}{2}}}{\Delta y}=0. \tag*{(14)}

Motion of the interface is governed by the advection equation for the volume fraction field,

∂C∂t+u⋅∇C=0(15)\frac{\partial C}{\partial t}+\mathbf{u}\cdot\nabla C=0 \tag*{(15)}

For incompressible fluids, conservation of the individual volumes of the two fluids results in the conservation of mass. Thus, in the volume of fluid method, geometric advection of the volume fraction field is expected to yield machine-precision mass conservation. Given a volume fraction field, reconstructed interface and solenoidal velocity field, we can solve the equation 15 using an operator splitting algorithm consisting of an xx−sweep and a yy−sweep following [13]. In order to employ an operator splitting algorithm, the advection of the interface (equation.15), using ∇⋅u=0\nabla\cdot\mathbf{u}=0, can be written as:

∂C∂t+∇⋅(uC)=C(∇⋅u).(16)\frac{\partial C}{\partial t}+\nabla\cdot(\mathbf{u}C)=C(\nabla\cdot\mathbf{u}). \tag*{(16)}

The above form of the advection equation is essential for performing volume conserving xx−direction and yy−direction sweeps separately (see [7]). Given a volume fraction (Ci,jn)(C_{i,j}^{n}) and velocity field (ui+1/2,jn,vi,j+1/2n)(u_{i+1/2,j}^{n},v_{i,j+1/2}^{n}) at the nnth time step, the discretised equation 16 is given by,

Ci,jn+1=Ci,jn+Δtri,jΔx(δVi−1/2,j−δVi+1/2,j)+ΔtΔy(δVi,j−1/2−δVi,j+1/2)+Ci,jn(Δtri,jΔx(ri+1/2,jui+1/2,jn−ri−1/2,jui−1/2,jn)+ΔtΔy(vi,j+1/2n−vi,j−1/2n))(17)\begin{aligned} C_{i,j}^{n+1}={}&C_{i,j}^{n}+\frac{\Delta t}{r_{i,j}\Delta x}\left(\delta V_{i-1/2,j}-\delta V_{i+1/2,j}\right)+\frac{\Delta t}{\Delta y}\left(\delta V_{i,j-1/2}-\delta V_{i,j+1/2}\right)\\ &+C_{i,j}^{n}\left(\frac{\Delta t}{r_{i,j}\Delta x}\left(r_{i+1/2,j}u_{i+1/2,j}^{n}-r_{i-1/2,j}u_{i-1/2,j}^{n}\right)+\frac{\Delta t}{\Delta y}\left(v_{i,j+1/2}^{n}-v_{i,j-1/2}^{n}\right)\right) \tag*{(17)} \end{aligned}

where δVi+1/2,j=(ruC)i+1/2,j\delta V_{i+1/2,j}=(ruC)_{i+1/2,j} is the amount of volume fraction fluxed through the right cell face. Similarly, fluxes δVi−1/2,j\delta V_{i-1/2,j}, δVi,j+1/2\delta V_{i,j+1/2} and δVi,j−1/2\delta V_{i,j-1/2} can be computed for other cell faces.

Using operator splitting, we can split the above equation as following:

Ci,j∗=Ci,jn+Δtri,jΔx(δVi−1/2,j−δVi+1/2,j)+Ci,j∗(Δtri,jΔx(ri+1/2,jui+1/2,jn−ri−1/2,jui−1/2,jn))(18)C_{i,j}^{*}=C_{i,j}^{n}+\frac{\Delta t}{r_{i,j}\Delta x}\left(\delta V_{i-1/2,j}-\delta V_{i+1/2,j}\right)+C_{i,j}^{*}\left(\frac{\Delta t}{r_{i,j}\Delta x}\left(r_{i+1/2,j}u_{i+1/2,j}^{n}-r_{i-1/2,j}u_{i-1/2,j}^{n}\right)\right) \tag*{(18)}
Ci,jn+1=Ci,j∗+ΔtΔy(δVi,j−1/2−δVi,j+1/2)+Ci,j∗(ΔtΔy(vi,j+1/2n−vi,j−1/2n))(19)C_{i,j}^{n+1}=C_{i,j}^{*}+\frac{\Delta t}{\Delta y}\left(\delta V_{i,j-1/2}-\delta V_{i,j+1/2}\right)+C_{i,j}^{*}\left(\frac{\Delta t}{\Delta y}\left(v_{i,j+1/2}^{n}-v_{i,j-1/2}^{n}\right)\right) \tag*{(19)}

where Ci,j∗C_{i,j}^{*} is the intermediate value of the volume fraction. An implicit scheme is used in the first direction and an explicit scheme in the second direction to maintain the conservation of volume fraction[14]. The order of sweep of direction is alternated every timestep [15]("Strang spliting") to achieve second order accuracy in time.

The volume flux through cell faces, δVcell−face\delta V_{cell-face}, is computed geometrically. Consider the schematic in figure 5, where the shaded region shows the volume of fluid-1 in the cell to be fluxed through the right face (δVi+1/2,j)(\delta V_{i+1/2,j}). Considering the face velocity (ui+1/2,j)(u_{i+1/2,j}) to be positive, the flux can be computed as,

δVi+12,j=(ru)i+1/2,jV2πrΔrΔy(20)\delta V_{i+\frac{1}{2},j}=\frac{(ru)_{i+1/2,j}V}{2\pi r\Delta r\Delta y} \tag*{(20)}

where VV is the volume of fluid 1 fluxed through the right face (shown as the shaded region in figure 5), Δr\Delta r is the distance in the radial direction which contains the volume advected in this timestep and rr is the distance to the center of this volume from the axis of symmetry. We can calculate Δr\Delta r by considering the conservation of volume fluxed through the right face and solving the resulting quadratic equation, which yields, Δr=ri+12,j−ri+12,j2−2ri+12,jueΔt\Delta r=r_{i+\frac{1}{2},j}-\sqrt{r_{i+\frac{1}{2},j}^{2}-2r_{i+\frac{1}{2},j}u_e\Delta t}.

Using the section of the piece-wise reconstructed interface lying in the volume to be fluxed through the cell-face over Δt\Delta t time-step and employing the Gauss area formula, given by equation 2, we can calculate the volume cut by this region.

The above small correction in computing Δr\Delta r along with the accurate Gauss formula for axisymmetric simulations allows us to improve upon the existing volume fraction advection schemes. Existing schemes modify the velocity in 2D2D algorithms by using ruru for velocity field and use 2D2D geometric advection scheme which results in a third order error O(Δt2h)O(\Delta t^{2}h), where hh is the grid size and Δt\Delta t is the timestep) in mass conservation. We illustrate this by considering xx−direction advection of a small volume of fluid through the right face of the cell with a velocity ueu_e, shown as the shaded region in the figure 6.

The existing schemes compute the volume as Vc=2π(re−ueΔt2)(ueΔt)ΔyV_c=2\pi\left(r_e-\frac{u_e\Delta t}{2}\right)(u_e\Delta t)\Delta y, where rer_e is the distance of the east cell face from the axis of symmetry, Δt\Delta t is the timestep, and Δy\Delta y is the height of the cell. Whereas, the proposed scheme yields the exact volume, V=2π(re−Δr2)ΔrΔyV = 2\pi\left(r_e - \frac{\Delta r}{2}\right)\Delta r\Delta y with Δr=re−re2−2reueΔt\Delta r = r_e - \sqrt{r_e^2 - 2r_eu_e\Delta t}. Thus, the error in volume calculation is given by, E=π(ueΔt)2ΔyE = \pi(u_e\Delta t)^2\Delta y.

Figure 5. The fluxed volume through the right face of the cell when ui+1/2,ju_{i+1/2,j} is positive.

Figure 6. The shaded region is the volume advected through the east cell face in a single timestep.

We validate the proposed modifications with the following test cases and compare with the results obtained using the open source multiphase flow solver Gerris[12].

Advection of a torus

In this test case, a torus of radius 0.25 is initialised at (0.35, 0.5) in a computational domain of size (2.0, 1.0). The torus is advected under the steady state velocity of u=0.1/ru = 0.1/r for r>0.05r > 0.05 and v=0v = 0, where rr is the distance from the axis of symmetry. The fluid is advected 1000 timesteps forward in time and then the velocity is reversed to compute 1000 timesteps backwards in time. The grid size is 1/128, the grid Courant number(CFL) is chosen to be 1 which corresponds to a time step of Δt=0.0078125\Delta t = 0.0078125. As seen from the Figure 7, after 1000 timesteps the torus is highly compressed during the advection as less area (due to axisymmetry) occupies the same volume as we move away from the axis of symmetry. We note that the final interface shape matches very well with the initial position of the torus, thus validating our algorithm.

The relative error in the volume between the initial and final distribution of fluid-1 for various number of forward and backward advection time steps is given in Table 2. The corresponding relative change in the volume obtained for the same test case simulated using Gerris flow solver are also given for comparison. We note that the error obtained from the present schemes are highly accurate in comparison to those obtained from Gerris.

As suggested by Kothe et al.[16], simple linear advection test cases do not reveal the efficacy of advection algorithms appropriately. Thus, we further test the efficacy of the algorithm by subjecting it to a more severe test case of advection of a torus in a Hill’s vortex. This is axisymmetric equivalent of the circle in a vortex test case for 2D cases [16]. For this velocity field, the interface undergoes strong topological changes including fragmentation and merging due to strong shear effects. Here we use a modified form of Hills’s vortex with a superimposed radial flow field. A torus of radius 0.1 is initialised at (0.2,0.8)(0.2, 0.8) in a computational domain of size (1.0,1.0)(1.0, 1.0) with L=0.5L = 0.5. The fluid is advected under highly strained steady state velocity field given by

u=0.1(rLy−LL+0.05r)(21)u = 0.1\left(\frac{r}{L}\frac{y-L}{L}+\frac{0.05}{r}\right) \tag*{(21)}
v=0.1[1−(y−LL)2−2(rL)2].(22)v = 0.1\left[1-\left(\frac{y-L}{L}\right)^2-2\left(\frac{r}{L}\right)^2\right]. \tag*{(22)}

The fluid is advected 4000 timesteps forward in time and then the velocity is reversed to advect 4000 timesteps backwards in time. The grid size is 1/1281/128 and the time step is 1.0×10−31.0 \times10^{-3}.

As seen from Figure 8 after 4000 timesteps the shape of the interface is highly distorted due to highly strained velocity field. The final interface shape matches very well with the initial position of the toroid thus validating our algorithm. The relative error in change in volume between the initial and the final distribution of fluid 1 for various number on time steps is given in Table 3. The corresponding relative change in volume for the same test case in Gerris flow solver. We note the error in the proposed scheme, even for larger number of timesteps, is an order smaller compared to the results from Gerris flow solver.

Relative error in volume
Number of timestepsCurrent SolverGerris Solver
15.2×10−155.2 \times10^{-15}1.1×10−51.1 \times10^{-5}
102.9×10−142.9 \times10^{-14}7.2×10−57.2 \times10^{-5}
1004.8×10−134.8 \times10^{-13}2.1×10−42.1 \times10^{-4}
10003.4×10−103.4 \times10^{-10}3.8×10−43.8 \times10^{-4}

Table 3. Results for relative error in the volume for a torus of radius 0.25 advected in radial direction forward and backward in time for different number of timesteps.

Figure 8. Interface shape after 4000 time steps forward and backward after advecting the torus placed in a vortex. The relative error in change in volume is 1.7×10−61.7 \times10^{-6}.

Relative error in volume
Number of timestepsCurrent SolverGerris Solver
12.7×10−102.7 \times10^{-10}2.2×10−62.2 \times10^{-6}
105.2×10−105.2 \times10^{-10}4.5×10−64.5 \times10^{-6}
1002.5×10−82.5 \times10^{-8}1.8×10−51.8 \times10^{-5}
10001.3×10−61.3 \times10^{-6}4.5×10−54.5 \times10^{-5}
40001.7×10−61.7 \times10^{-6}4.8×10−54.8 \times10^{-5}

Table 4. Results for relative error in volume for a torus of radius 0.1 placed in a complex velocity field advected forward and backward in time for different number of timesteps.

Bubble in a Stagnation Point Flow

In this test case we implement the VOF algorithm presented in this paper for a more complex flow. We solve Navier-Stokes equations in one fluid form given by:

ρ(C)(∂u∂t+u⋅∇u)=−∇p+∇⋅[μ(C)(∇u+(∇u)T)]+ρ(C)g+fvγ(23)\rho(C)\left(\frac{\partial\mathbf{u}}{\partial t}+\mathbf{u}\cdot\nabla\mathbf{u}\right)=-\nabla p+\nabla\cdot\left[\mu(C)\left(\nabla\mathbf{u}+(\nabla\mathbf{u})^{T}\right)\right]+\rho(C)\mathbf{g}+\mathbf{f}_{v}^{\gamma} \tag*{(23)}

where u\mathbf{u} and pp are the velocity vector and pressure, respectively, ρ(C)\rho(C) and μ(C)\mu(C) are the fluid density and viscosity which are a function of void fraction field, CC. We use Chorin’s projection method [17] to solve the above equation (23) where we discretise the advection term using a second order ENO scheme [18] and the diffusion terms using central differencing. Surface tension forces, fvγ\mathbf{f}_{v}^{\gamma} are acting only at the interface and have been modeled as volumetric body force using the continuum surface force model of Brackbill, Kothe, and Zemach [19]. The interface is captured using CLSVOF algorithm given by Sussman and Puckett [13]. This algorithm is mass conserving and calculates the curvature and surface normal with high accuracy which is used for surface tension force calculation. The interface is advected by solving the advection equations for level-set function, ϕ\phi, and volume fraction, CC. We initialize a toroidal bubble of radius 0.1 at (0.2, 0.5) in a computational domain of unit size, 1×11 \times1. The bottom boundary has an inlet velocity of unity in the upward axial direction and the right boundary has outflow boundary conditions. The top boundary acts as a rigid wall with no-slip and impermeable surface. The density ratio and viscosity ratio is 10 with the Laplace number of the bubble, La=ρDσμ2=0.048La=\frac{\rho D\sigma}{\mu^2}=0.048. The incoming axial velocity drags the bubble and flattens it against the top wall, stretching it in the axial direction considerably. Even though the fluid interface undergoes drastic change in its shape, the volume is conserved with a high degree of accuracy with relative error in the volume of 2.1×10−62.1 \times10^{-6}.

Figure 9. Interface shape after time t=1t = 1. The interface is flattened against the top wall and stretched due to the underlying velocity. Even when the cross-sectional area changes the volume of the toroidal bubble is maintained very accurately.

Conclusions

In the present work, we have presented several improvements for the implementation of volume of fluid method in axisymmetric coordinates. We have presented analytical relations for the reconstruction of piecewise linear interface in axisymmetric coordinates similar to those given by Scardovelli and Zaleski[10] for cartesian coordinates. The proposed scheme substantially reduces the computational cost in comparison to the iterative schemes usually employed for the reconstruction. Further, we showed that even for axisymmetric coordinate system, machine-precision advection of volume fraction field can be achieved. We illustrated the improvements by comparing the results with the popular open source multiphase flow solver Gerris. Finally, we would like to note that similar modifications in the advection scheme for volume of fluid method in other curvillinear coordinate systems(such as elliptic coordinates) can be derived using the approach presented in this work.

References

  1. [1]Andrea Prosperetti and Grétar Tryggvason. Computational methods for multiphase flow. Cambridge university press, 2009.
  2. [2]Grétar Tryggvason, Ruben Scardovelli, and Stéphane Zaleski. Direct numerical simulations of gas–liquid multiphase flows. Cambridge University Press, 2011.
  3. [3]Cyril W Hirt and Billy D Nichols. Volume of fluid (vof) method for the dynamics of free boundaries. Journal of computational physics, 39(1):201–225, 1981.DOI
  4. [4]Yuriko Renardy and Michael Renardy. Prost: a parabolic reconstruction of surface tension for the volume-of-fluid method. Journal of computational physics, 183(2):400–421, 2002.
  5. [5]David L Youngs. Time-dependent multi-material flow with large fluid distortion. Numerical methods for fluid dynamics, 1982.
  6. [6]James Edward Pilliod. An analysis of piecewise linear interface reconstruction algorithms for volume-of-fluid methods. U. of Calif., Davis, 1992.
  7. [7]James Edward Pilliod Jr and Elbridge Gerry Puckett. Second-order accurate volume-of-fluid algorithms for tracking material interfaces. Journal of Computational Physics, 199(2):465–502, 2004.DOI
  8. [8]Ruben Scardovelli and Stephane Zaleski. Interface reconstruction with least-square fit and split eulerian–lagrangian advection. International Journal for Numerical Methods in Fluids, 41(3):251–274, 2003.DOI
  9. [9]David L Youngs. An interface tracking method for a 3d eulerian hydrodynamics code. Atomic Weapons Research Establishment (AWRE) Technical Report, 44(92):35, 1984.

Paper details

vlnc:W_ygrw4cd8
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