You are on page 1of 23

Chapter 3 NUMERICAL IMPLEMENTATION OF THE VOLUME OF FLUIDCONTINUOUS SURFACE FORCE METHOD

The purpose of computing is insight, not numbers. This motto is often thought to mean that the numbers from a computing machine should be read and used, but there is much more to the motto. The choice of the particular formula, or algorithm, inuences not only the computing but also how we are to understand the results when they are obtained . . . Thus computing is, or at least should be, intimately bound up with both the source of the problem and the use that is going to be made of the answers it is not a step to be taken in isolation from reality. R. W. Hamming (1986) The solutions to complex free surface ow problems involve the solution of the equations of motion and continuity for the two uids subject to specied boundary conditions and their numerical solution is briey reviewed in section 1.3. In this dissertation we use the Eulerian Volume of Fluid (VOF) method (Hirt and Nichols, 1981), in which a marker function convected by the ow is used to track the interface. The major incentive for using the VOF method is that it allows for the description of highly complex interfaces such as those encountered in multiphase ows. Reasonable accuracy is attainable with elemental control volume balances, yet the method is relatively simple to implement. This algorithm is publicly available in a two-uid, 2-D program called SOLA-VOF (Nichols et al., 1980). A 3-D version called FLOW-3D (Hirt, 1988) is also available commercially. More recently a one-uid program, RIPPLE, which implements the Continuum 33

34 Surface Force (CSF) algorithm and incorporates various improvements in the oneuid VOF algorithm, has been introduced by Kothe et al. (Kothe et al., 1991; Brackbill et al., 1992). In this chapter we describe how we have combined the twouid capability of the SOLA-VOF algorithm with the free surface implementation of the CSF algorithm to investigate problems that involve transient 2-D free surface ows with two immiscible uids. The source code for SOLA-VOF has been extensively modied and the resulting code tested for consistency and accuracy with numerous test problems. It has been rewritten to accommodate the addition of the second uid with arbitrary viscosity, to correct the incorporation of the viscous stress terms, and to add a superior surface tension algorithm. The original SOLA-VOF 2-D program (Nichols et al., 1980) is well suited for high Reynolds number ows, including those involving free surfaces. Among the latter, however, it is better suited for gas-liquid than for liquid-liquid systems, and relaxing this limitation has been an important part of our eorts. Thus, we have implemented extensive modications in the original program. Both planar and axisymmetric 2-D ows can be simulated. This chapter outlines the numerical methods used in combining the VOF (Hirt and Nichols, 1981) and CSF (Brackbill et al., 1992) algorithms. Although the VOF references describe the algorithm for the iteration procedure, very little justication is provided there. This chapter addresses this issue.

3.1

Governing Equations in a Cylindrical Coordinate System It is assumed that the ow in each phase is axisymmetric, viscous, and

incompressible. The continuity equation (2.30) given in cylindrical, axisymmetric coordinates (r, z ) as: u v u + + =0 r z r (3.1)

35 where (u, v ) are the radial, axial components of the velocity eld respectively. The dynamic momentum equation (2.63) becomes: u u u +u +v t r z v v v +u +v t r z = = p rr zr rr F + + + + gr + r r z r r (3.2)

p rz zz rz F + + + + gz + z r z r z

(3.3)

where p is the pressure, gr , gz are the radial and axial components of the gravitational acceleration, rr , zr , rz , zz are the components of the Newtonian stress tensor from equations (2.32) and (2.35): rr = 2 u , r zr = rz = v u + , r z zz = 2 v z (3.4)

and is the curvature of the liquid-liquid interface (Kothe et al., 1991; Brackbill et al., 1992) from equation (2.56), = (n ) where the unit normal (directed into uid 2), n = is derived from a normal vector, n = F The evolution equation (2.73) for the uid function marker eld, F , is: F F F +u +v =0 t r z The density and viscosity elds are obtained from: = 1 (1 F ) + 2 F (3.9) (3.8) (3.7) n |n | (3.6) (3.5)

36 = 1 (1 F ) + 2 F (3.10)

Note that the interfacial surface forces are incorporated as body forces per unit volume in the momentum equations (3.2) and (3.3) rather than as boundary conditions. Instead of a surface tensile force boundary condition applied at a discontinuous interface of the two uids, a volume force is used which acts on uid elements lying within a transition region of nite thickness. The CSF formulation makes use of the approach taken in numerical simulations that discontinuities can be approximated, without increasing the overall error of approximation, as continuous transitions within which the uid properties vary smoothly from one uid to the other over a distance of O(h), where h is a length comparable to the resolution of the computational mesh. Surface tension, therefore, is felt

everywhere within the transition region through the volume force included in the momentum equations. A derivation is given in appendix A which shows that the CSF formulation is equivalent in the limit of innitesimal interfacial thickness h to the classical description of these forces as boundary conditions at a discontinuous interface, and supplements the previous derivation in the CSF references (Kothe et al., 1991; Brackbill et al., 1992). It can be seen that this formulation is similar to the approach taken in the nite element method in that the free surface forces are included in the momentum integral residual equation (Georgiou et al., 1988; Cuvelier and Schulkes, 1990; Malamataris and Papanastasiou, 1991). Also note that interfacial surface tension is presumed constant, and surface tension gradient induced stresses are not included in equations (3.2) and (3.3). Eects such as surface tension gradient and surface dilatational and shear viscosities can, in principle, be included, (Scriven, 1960) and may be important if surface active agents are present. Massless marker particles can be placed in the ow to follow uid elements if necessary for comparison with experimental results. Local velocities (up , vp ) of the marker particles are used to update the positions by use of the kinematic

37 relations: dr = up , dt dz = vp dt (3.11)

For 2-D axisymmetric ows the streamfunction can be dened, as usual, in order for the velocity to be divergence free (i.e., by satisfying the incompressibility condition): u= 1 , r z v= 1 r r (3.12)

Then the streamfunction can be calculated from the following Poisson equation (obtained by dierentiating equations (3.12)), given the velocity eld: 2 2 + 2 =r r2 z u v z r v (3.13)

These equations are to be solved subject to appropriate boundary conditions discussed in section 3.6 (see, for example, chapter 5). 3.2 Momentum and Continuity Finite Dierencing The momentum equations are nite-dierenced on a locally variable, staggered mesh using the control volume approach, as illustrated in Figure 3.1. As Figure 3.1 shows, the radial velocity ui+ 1 ,j and axial velocity vi,j + 1 are centered
2 2

at the right face and top face of each cell respectively, whereas the pressure, pi,j , and marker function, Fi,j , are located at the center. The details can be found in Nichols et al. (1980) and Hirt and Nichols (1981), while an introduction to control volume approaches on staggered meshes can be found in Patankar (1980). The equations (3.1) to (3.4) are solved for the variables u, v , and p at each time step by an iterative method, a form of successive overrelaxation (SOR) that will now be described. We may rewrite the equations explicitly in velocity and implicitly in pressure for the (i, j )th cell with an Euler scheme from the current time n to the new time n + 1, accurate to rst-order in time: u v u + + r z r
n+1

=0

(3.14)

38

vi,j+1 2 pi,j Fi,j

Physical Boundary

zj

ui+1,j 2

ri Fictitious Cell Boundary

Figure 3.1: Staggered mesh.

39
+1 un = un i+ 1 ,j i+ 1 ,j
2 2

p r

n+1

u u 1 +t u v + r z

rr zr rr + + r z r

F + gr + r

(3.15)

n+1 n vi,j = vi,j +1 +1


2 2

p z

n+1

v 1 v +t u v + r z
n

zr zz zr + + r z r
n

F + gz + z
n

(3.16)

n = 2 rr

u r

n zr =

v u + r z

n zz = 2

v z

(3.17)

The viscous terms in the momentum equations are calculated using central nite dierences accurate to second-order, taking into account the variable mesh. In the original SOLA-VOF, the viscosity was assumed constant throughout the entire domain and the viscous terms were calculated as in the Navier-Stokes equations. The viscous term of the form 2 v were then nite dierenced.

However, a test problem showed that the tangential stresses were not equal at the interface. In the present work the viscous terms were nite dierenced using the expressions (3.17), which allow for a variable viscosity in the interface region. If the inertial terms are approximated by second-order accurate central dierencing, the algorithm is numerically unstable (Hirt, 1968). If they are

approximated by rst-order accurate upwinding or donor-cell dierencing, the algorithm is stable but it might be overstable due to the presence of excessive articial viscosity (Hirt, 1968). In the SOLA-VOF method a linear combination of central (HN = 0) and donor-cell (HN = 1) dierencing is used to approximate the inertial terms with controlling parameter HN (Hirt and Nichols, 1981). To develop it consider using second-order accurate central dierencing for the rst

40 convective term in the radial momentum equation: u u r


n

un i+ 1 2 ,j

xi+1

un un i+ 1 ,j i 1 ,j
2 2

xi

+ xi

un un i+ 3 ,j i+ 1 ,j
2 2

(3.18)

xi+1

xi+1 + xi

However, this will lead to a numerically unstable algorithm. We may use rstorder accurate upwinding where the derivative upwind of the point in question is used. In the case of ow in the positive r direction (sgn(ui,j ) = +1) this means: u u r
n

un i+ 1 2 ,j

un un i+ 1 ,j i 1 ,j
2 2

xi

(3.19)

In the case of ow in the negative r direction (sgn(ui,j ) = 1): u u r


n

= un i+ 1 ,j
2

un un i+ 3 ,j i+ 1 ,j
2 2

xi+1

(3.20)

These two equations are stable, but accurate to only rst-order. We may combine the above three equations into one for stability and accuracy: u u r
n

= un un i+ 1 ,j i 1 ,j
2 2 + xi xi un un i+ 1 ,j i 1 ,j 2 2

xi+1

un un i+ 3 ,j i+ 1 ,j
2 2

+HN sgn(ui,j ) xi+1 un i+ 1 ,j


2

xi

xi+1 un un i+ 3 i+ 1 2 ,j 2 ,j xi xi+1

(3.21)

xi+1 + xi + HN sgn(ui,j ) (xi+1 xi )

where HN is the interpolation factor between the methods such that when HN = 0 or HN = 1 central dierencing or upwinding are recovered respectively.

41 3.3 Pressure Iteration This section, which describes the pressure iteration at each time step, follows the procedure described by Gerhart et al. (1992). Pressure does not

appear explicitly for the new time step in the continuity equation so it must be recast so that pressure appears as a variable at the new time step. If we solve the momentum equations, the velocity eld does not necessarily satisfy continuity because we would be using the old time step pressure eld pn . Therefore we must use continuity to correct the pressure eld to the new values so that the velocities un+1 , vn+1 will satisfy continuity at the new time level. The pressure is split into an old time-level component and a correction such that:
+1 n pi,j pn i,j pi,j

(3.22)

Then we can decompose the pressure derivatives: p r p z


n+1

p pn+1 pn + pn = + r r p pn+1 pn + pn = + z z

p r p z

(3.23)

n+1

(3.24)

Substituting these in the momentum equations gives:


+1 un = un i+ 1 ,j i+ 1 ,j
2 2

t p r rr zr rr + + r z r + gr + F r
n

+t

1 p u u 1 u v + r r z t p z

(3.25)
n+1 n vi,j = vi,j +1 +1
2 2

+t

1 p v v 1 u v + z r z

zr zz zr + + r z r

+ gz +

F z

(3.26)

42 Dening the following quantities: Dn 1 v (ru) + r r z


n

(3.27)

Qn r

u u 1 1 p u v + r r z v v 1 1 p u v + z r z

rr zr rr + + r z r zr zz zr + + r z r

F + gr + r F z

(3.28)
n

Qn z

+ gz +

(3.29)

we may rewrite the momentum equations as:


+1 un = un i+ 1 ,j i+ 1 ,j
2 2

t p +Qn r r t p +Qn z z

(3.30)

n v n+1 ! = vi,j +1 i,j + 2


2

(3.31)

We may start the iterative method by calculating a rst estimate of the velocities with a fully explicit guess (p(1) = 0): ui+ 1 = un + Qn r i+ 1 ,j ,j
2 2

(1)

(3.32)

n vi,j + 1 = vi,j + Qn z +1
2 2

(1)

(3.33)

For an improved guess we must include the pressure correction p(2) : 1 p r


(2)

ui+ 1 ,j = un t i+ 1 ,j
2 2

(2)

+Qn r = ui+ 1 ,j t
2

(1)

1 p r

(2)

(3.34)

(2) vi,j + 1 2

n vi,j +1 2

1 p t z

(2)

+Qn z

(1) vi,j + 1 2

1 p t z

(2)

(3.35)

43 To solve for the pressure correction, substitute these into the continuity equation which we require to be satised at the new iteration: D
(2)

1 v ru + r r z

(2)

1 1 p (1) rui+ 1 rt r r r 2 ,j 1 r r rui+ 1 ,j


2

(2)

1 p (1) vi,j + 1 t z z 2 r p(2) r t z

(2)

(1)

vi,j + 1

(1)

t r r

1 p(2) z (3.36)

=0 So that a Poisson-like equation results: r 1 p(2) r + z 1 p(2) z 1 p(2) D(1) = + r r t (3.37)

This equation must be solved for p(2) . We may use nite dierences taking into consideration the variable mesh to obtain: (1) (2) (2) (2) (2) Di,j p p p p 2 i+1,j i,j i,j i1,j = t ri (r)i+1 + (r)i (r)i + (r)i1 + 2 zj (z )j +1 + (z )j (z )j + (z )j 1 + 2 ri (r)i+1 + 2(r)i + (r)i1 pi+1,j pi1,j
(2) (2)

pi,j +1 pi,j

(2)

(2)

pi,j pi,j 1

(2)

(2)

(3.38)

where mesh-weighted averages are used to obtain the quantities: (r)i+1 + (r)i = i+1,j ri + i,j ri+1 (z )j +1 + (z )j = i,j +1 zj + i,j zj +1 (3.39) (3.40)

44 i,j = 1 (1 F i,j ) + 2 F i,j i,j = 1 (1 Fi,j ) + 2 Fi,j


(2) (2)

(3.41) (3.42)

We have to solve this (linear) system (3.38) for pi,j . We may use an SOR iterative procedure where the o-diagonal terms are neglected (e.g., pi+1,j so that the system may solved directly for the new pi,j : pi,j = Di,j
2t ri 1 (r)i+1 +(r)i (1) 1 (z )j +1 +(z )j (2) (2)

0)

1 (r)i +(r)i1

2t zj

1 (z )j +(z )j 1

(3.43) where is the relaxation factor and the old divergence is:
(1) Di,j

ui+ 1 ,j ui 1 ,j
2 2

(1)

(1)

ri
(1) (1)

vi,j + 1 v i,j 1
2 2

(1)

(1)

zi

ui+ 1 ,j +ui 1 ,j
2 2

(1)

(1)

2ri

(3.44)

We may now update ui,j , vi,j and pi,j using (3.34): pi,j = pi,j + pi,j
(2) ui+ 1 2 ,j (2) ui 1 ,j 2 (2) vi,j + 1 2 (2) vi,j 1 2 (2) (1) (2)

(1)

(3.45)
(2)

(1) ui+ 1 ,j 2 (1) ui 1 ,j 2 (1) vi,j + 1 2 (1) vi,j 1 2

+ +

2tpi,j 2tpi,j 2tpi,j 2tpi,j

(r)i+1 + (r)i
(2)

(3.46)

(r)i1 + (r)i
(2)

(3.47)

(z )j +1 + (z )j
(2)

(3.48)

(z )j 1 + (z )j

(3.49)

k < and the variables are ready for the next Iteration is continued until max Di,j

time step. The SOR parameter must meet the criterion: 1 <2 (3.50)

45 As is discussed in Ciarlet (1991), this parameter is adjusted to obtain the minimum number of iterations by sensitivity studies depending upon the particular case of interest. 3.4 Kinematic Condition The kinematic condition for the advection of the VOF function can be written in nite-dierence form as:
+1 n Fn i,j = Fi,j t u

F F +v r z

(3.51)

The convective terms may be rearranged to divergence form, minus a divergence correction term: u F 1 F +v = (F ru) + (F v ) F r z r r z 1 v (ru) + r r z (3.52)

Substituting (3.52) into (3.51) we get the divergence part: 1 n i,j = Fi,j t F (rF u) + (F v ) r r z which is then updated with the correction:
+1 Fn i,j n i,j + tFi,j =F n

(3.53)

1 v (ru) + r r z

n+1

(3.54)

This sequence insures conservative advection of F (Kothe et al., 1991). Ordinarily the correction in (3.54) would be zero due to continuity, but it has been found desirable to include it numerically (Brackbill et al., 1992) because although it is small, but is non-zero and of the order of t. In the VOF code, this equation is utilized at the end of the pressure iteration using the new un+1 , vn+1 values. A brief description of the Hirt and Nichols (1981) advection algorithm, which takes special care in preserving the discontinuous nature of F , is now given; further details can be found in Hirt and Nichols (1981). The divergence

46 equation for F can be nite dierenced for the radial advection term in terms of an upstream donor (d) cell at (id, j ) and a downstream acceptor (a) cell at (ia, j ) (if
+1 un i,j > 0, idm = i 1, id = i, ia = i + 1, otherwise, ia = i, id = i + 1, idm = i + 2):

r 1 t (F u) id,j = F n 1 i+ 2 F id,j rid xid t (F u) 1 ri+ 1 n 2 ia,j = Fia,j F + ria xia where the amount of F uxed across the cell face in t is:
+1 t + CFr , Fid,j xid t (F u) = min Fiad,j un i+ 1 ,j
2

(3.55)

(3.56)

(3.57)

with the correction factor: CFr = max Fidm,j Fiad,j


+1 un t Fidm,j Fid,j xid , 0 i+ 1 ,j
2

(3.58)

In these equations, the subscript iad = ia if the uid is more vertical than horizontal, otherwise iad = id. The min feature prevents the uxing of more of uid 2 (F = 1) from the donor cell than it has to give, while the max feature accounts for an additional ux, CFr , if the amount of uid 1 (F = 0) to be uxed exceeds the amount available. A subsequent pass is made for the axial convective term: i,jd = F n t (F v ) F i,jd zjd t (F v ) n i,ja = Fi,ja + F zja
n+1 t (F v ) = min Fi,jad vi,j t + CFz , Fi,jd zjd +1
2

(3.59) (3.60) (3.61) (3.62)

CFz = max

Fi,jdm Fi,jad

n+1 vi,j t Fi,jdm Fi,jd zjd , 0 +1


2

Finally, the divergence correction is added:


n+1 n n+1 i,j + tFi,j F i,j =F Di,j

(3.63)

47 3.5 Surface Force Terms This section is a brief summary of the CSF calculation of the surface force terms in the momentum equations (Kothe et al., 1991; Brackbill et al., 1992). The surface force in the momentum equations is: Fsv = i,j n i,j (3.64)

i,j

and is the curvature of the liquid-liquid interface which is calculated as: = (n ) = 1 |n | n |n | (n ) |n | (3.65)

Brackbill et al. (1992) have rewritten the curvature in terms of n and |n | (equation (3.7)) instead of n (equation (3.6)) to insure that principal contributions to nitedierence approximation of come from the center of the transition region rather than the edges. The normal vector at cell corners is obtained by nite dierencing, for example at the top right corner of cell (i, j ): n r i+ 1 ,j + 1 =
2 2

(Fi+1,j +1 Fi,j +1 ) zj + (Fi+1,j Fi,j ) zj +1 (zj + zj +1 ) ri+ 1


2

(3.66)

n z i+ 1 ,j + 1 =
2 2

(Fi+1,j +1 Fi+1,j ) ri + (Fi,j +1 Fi,j ) ri+1 (ri + ri+1 ) zj + 1


2

(3.67) n 2 2 r +n z ):

The calculation of the rst term in the curvature expression is (|n | = n |n | = |n | = n r |n |


2 i,j

n r |n |

i,j

|n | r n rn z |n |2

+
i,j

n z |n |

i,j

|n | z +
i,j

i,j

n r r

+
i,j

i,j

n r n z + z r

n z |n |

2 i,j

n z z

i,j

(3.68) The cell-centered partial derivatives follow from knowledge of n at the corners, for

48 example : n r r n r i+ 1 ,j + 1 + n r i+ 1 ,j 1 n r i 1 ,j + 1 + n r i 1 ,j 1
2 2 2 2 2 2 2

=
i,j

2ri

(3.69)

The cell-centered normal is the average at the corners: n r i,j = n z i,j = 1 r i 1 ,j + 1 + n r i+ 1 ,j 1 + n r i 1 ,j 1 n 1 1 +n 2 2 2 2 2 2 4 r i+ 2 ,j + 2 1 n z i 1 ,j + 1 + n z i+ 1 ,j 1 + n z i 1 ,j 1 1 1 +n 2 2 2 2 2 2 4 z i+ 2 ,j + 2 (3.70)

(3.71)

The second term in the curvature expression is calculated: (n )i,j = where: ri+ 1 n r i+ 1 ,j + 1 + n r i 1 ,j + 1 + n r i+ 1 ,j 1 ri 1 n r i 1 ,j 1 1 2 2 2 2 2 2 2 2 2 2 (r n r )i,j = ri r 2riri (3.73) The face centered values of Fsv i,j at the right and top faces are required for the momentum equations and are calculated from cell centered values: Fsv r i+1,j = ri Fsv r i+1,j + ri+1 Fsv r i,j ri + ri+1 zj Fsv z i,j +1 + zj +1 Fsv z i,j zj + zj +1 (3.74) 1 (r n r )i,j + ri r n z z (3.72)
i,j

Fsv z i,j +1 =

(3.75)

The CSF method has the ability to use the smoothed or mollied VOF i,j introduced in equation (2.45) for the calculation of the curvature function F i,j in the volume force equation (3.64), which is dierent from the unsmoothed function Fi,j used to calculate the normal vector in equation (3.64). This enables the algorithm to calculate a smoother curvature for accuracy and, we have found, for decreased number of pressure iterations. The smoothed VOF function is

49 computed by convolving F with a B-spline of degree l (de Boor, 1967; Brackbill et al., 1992),
(l) (|x

x| ; h), (with l 2 in this dissertation) where

(l)

= 0 only

for |x x| < (l + 1) h/2 = 3h/2. The smoothed VOF function is given by:
k

i,j = F
i ,j =1

Fi,j

(l)

xi ,j xi,j ; h

(l)

yi ,j yi,j ; h

(3.76)

where the sum gathers contributions from the nine values (for l = 2 in 2-D) of Fi,j within the support of
(2) .

In our case this formulation becomes simply:

i,j = 9 Fi,j + 3 (Fi+1,j + Fi1,j + Fi,j +1 + Fi,j 1 ) F 16 32 1 + (Fi+1,j +1 + Fi+1,j 1 + Fi1,j +1 + Fi1,j 1 ) 64

(3.77)

This formula may be applied iteratively by multiple passes through the mesh for increased degrees of smoothing. Our experience has shown that one to three passes are optimal, with most calculations carried out with one pass. 3.6 Boundary Conditions The equations are solved with appropriate boundary conditions in axisymmetric coordinates (r, z ), with the boundary conditions as follows. The conditions which we are concerned with are no-slip, free-slip, and continuative. For the solid walls no-slip conditions are used: u=v=0 For example, at a bottom wall: ui+ 1 ,1 =
2

(3.78)

ui+ 1 ,2 z1
2

z2
2

(3.79) (3.80)

vi, 3 = 0

where the z1 and z1 appear as interpolation factors in equation (3.79) so that the radial velocity may be exactly zero on the boundary. For the axis of symmetry

50 at r = 0 (same as a free slip wall): u = 0, v =0 r (3.81) (3.82)


2

u 3 ,j = 0
2

v1,j + 1 = v2,j + 1
2

(3.83)

For a continuative boundary at the bottom of the mesh, for example, there is no change in the axial direction: u v = =0 z z ui+ 1 ,1 = ui+ 1 ,2
2 2

(3.84) (3.85) (3.86)

vi, 3 = vi, 5
2

For problems involving the eects of wall adhesion characterized by a given equilibrium or dynamic contact angle, , such as the evaluation of the interface separating two stationary immiscible uids, this condition can be imposed by requiring that the unit normal to the interface at the wall contact line, n , satisfy (Kothe et al., 1991): n =n w cos + tw sin (3.87)

where n w and tw are the unit normal and tangent to the wall at the contact line respectively. In the jet simulations, special attention was given to force the F function to intersect, or be pinned, to the nozzle lip as shown in Figure 5.1, while using the Hirt-Nichols acceptor/donor method (Nichols et al., 1980). This was accomplished by the simple expedient of not allowing uid 2 to ux across the top of the nozzle (if it so desired), thus imitating the experimentally observed condition. Note that this procedure does not inuence the steady-state solution reported in section 5.3 since the corresponding condition that activates it is not applicable there. Thus, in practice, this condition was only necessary in order to stabilize the initial transient during startup.

51 3.7 Numerical Stability The numerical dierence equations are subject to linear numerical stability conditions that are detailed in Hirt and Nichols (1981). Material cannot move more than one cell in one time step yielding the Courant condition: t < min ri ui+ 1 ,j
2

, zi vi,j + 1
2

(3.88)

Second, momentum must not diuse more than about one cell in one time step:
2 2 1 t 1 ri zj < 2 + z 2 1 2 ri j 2 2 1 ri zj 2 t < 2 + z 2 2 2 ri j

(3.89)

(3.90)

Third, when surface tension is included there must be a limit to prevent capillary waves from traveling more than one cell width in one time step (Hirt and Nichols,1981): t2
3 , z 3 min(1 , 2) min ri j

<

1 4 (1 + )

(3.91)

where = 1 for cylindrical and = 0 for Cartesian geometry. Finally, with t chosen, a stability condition on HN is needed: 1 HN > max ui+ 1 ,j t
2

ri

vi,j + 1 t
2

zi

(3.92)

In the present code, this value is automatically adjusted to be: HN = min 1.2 max 3.8 ui+ 1 ,j t
2

ri

vi,j + 1 t
2

zi

,1

(3.93)

Solution Algorithm and Validation There are three major stages followed at each time step (see Figure 3.2):

52 (1) Explicit nite volume representations of the momentum equations (3.15) and (3.16) are used to calculate new approximate velocities (u n+1 , v n+1 ) at the new time level n +1 using initial conditions or previous time level n values (u n , v n ) and p n . The volume forces at the interface are calculated using equations (3.64) to (3.75) and equation (3.87) through the CSF algorithm (Kothe et al., 1991). Inertial terms use upwind nite dierences for stability, and all nite-dierence equations are written to take into account a variable-size mesh. (2) Pressures and velocities in each mesh cell are adjusted iteratively to satisfy the continuity equation (3.14) by using the successive overrelaxation (SOR) update equations (3.42)(3.48). Boundary conditions (3.78) to (3.86) are also satised at this stage, and all quantities are now at the new time level n + 1: (u n+1 , v n+1 , p n+1 ). (3) The kinematic condition (3.8) is used to update the F n marker function to F n+1 using the Hirt and Nichols advection algorithm (3.55)(3.62), with divergence correction (3.63) as discussed by Kothe et al. (1991). Marker particles, whenever used, are also updated at this stage using explicit forms for equations (3.11). The above cycle is repeated for any subsequent time interval. Stability criteria based on viscous, inertial, and surface tension term analyses establish the size of the basic time step (Hirt and Nichols, 1981) using equations (3.88)(3.93). Also, the allowed iterations per cycle are used to decrease the time step interval further if necessary. The streamfunction is obtained at the nal time by solving the nite dierence form of equation (3.13). Several test problems that have known analytical or numerical solutions were used to test the accuracy and validate the modied algorithm. These included entry ow in a circular tube (Langhaar, 1942) (testing pressure and viscous terms), the wall startup (Rayleigh) problem (Bird et al., 1960) (transient terms), two-phase cocurrent ow between parallel plates (Bird et al., 1960) (interface boundary conditions), Kovasznay (1948) ow (inertial terms), forced cavity ow

53
Momentum Equations (TILDE2) RIPPLE Surface Tension Algorithm? Read Input Data Generate Mesh (MESHSET) Initial Conditions (SETUP) Initialize Free Surface (INITVOF) t=0 Restart? (DATAIN) Set Boundary Conditions (BC2) Yes CSF Two Phase Forces (TENSION) No One Phase Forces (TMS102)

Start

Set Boundary Conditions (BC2) Pressure Iteration (PRESSIT3) No Converged? Yes VOF Advection (VFCONV2) Set Boundary Conditions (BC2) Move Particles (PARMOV) Surface Orientation (PETACAL) Print/Plot (PRTPLT2) (DRAW) Adjust t (DELTADJ) Update Restart File (DATAOUT) t = t + t Time to Stop? No Yes Solve for the Streamfunction (STR)

Stop

Figure 3.2: VOF-CSF program owchart.

54 (Gupta, 1991) (inertial terms), four-to-one sudden contraction ow (Kim-E et al., 1983) (inertial terms), meniscus shape in a capillary tube (Concus, 1968) (surface tension terms), static and dynamic drop problems (Lamb, 1932) (surface tension and transient terms), Rayleigh capillary jet breakup (Kothe et al., 1991; Lafrance, 1975) (highly distorted interfaces), the inviscid free liquid-liquid cylinder breakup of Christiansen and Hixson (1955, 1957), and Duda and Vrentas (1967) case 4, involving a single phase jet of water injected downwards into air. This latter case, representing a limiting case of a liquid-liquid jet corresponding to an inviscid, low density, continuous phase, is analyzed separately in section 5.3. The results from these test problems indicate reasonable accuracy with coarse meshes. For example, for the entry ow problem (Langhaar, 1942), the convergence is second-order in mesh spacing for the velocity components and the streamfunction, and rst-order in mesh spacing for the pressure. For a 20 20 uniform mesh , at the dimensionless axial the centerline dimensionless axial velocity, v v/v distance, 2z/(R Re), of 0.304, is 1.986 vs. the approximate Langhaar (1942) solution of 1.993, yielding a dierence of 0.4%, and the centerline dimensionless streamfunction, /(4R2v ), is 0.1242 vs. the exact value of 0.125, for a dierence of 0.6%. 3.9 Conclusions In this chapter we have discussed the combination of the Volume of Fluid (VOF) method with the Continuous Surface Force (CSF) algorithm for the calculation of possibly high Reynolds numbers free surface ows. The numerical method is used to solve for the velocity and pressure and the advection of the free surface is described. We have detailed the boundary conditions used, and discussed numerical stability issues. It is important to stress here the generality with which we have approached this problem and the development of the resulting code. Many types of free-surface problems can now be solved with the aid of our program, as can be seen by the

55 breadth and scope of the test problems described in the preceeding section. Not only can these simple problems be attacked but more complex ows (possibly at high Reynolds numbers) of direct interest in liquid-liquid systems can now be described in great detail.

You might also like