Вы находитесь на странице: 1из 36

Chapter 5 STEADY LAMINAR FLOW OF LIQUID-LIQUID JETS AT HIGH REYNOLDS NUMBERS

Let us take in the rst instance the problem of the eux of a liquid from a small orice in the walls of a vessel which is kept lled up to a constant level, so that motion may be regarded as steady . . . Experiment shews however that the converging motion above spoken of ceases at a short distance beyond the orice, and that (in the case of a circular orice) the jet then becomes approximately cylindrical . . . The calculation of the form of the issuing jet presents diculties which have only been overcome in a few ideal cases of motion in two dimensions. H. Lamb, Hydrodynamics (1932) Among the systems we are interested in are ones that represent key elements of one or more types of actual contactors. In this chapter we examine a liquidliquid jet that is representative of the situation above an orice on a sieve tray. The object is to model the velocity and pressure elds as well as the interface shape and location in this system that has complex free surfaces. Two regions are considered for the liquid-liquid jet: the steady region near the nozzle is discussed in this chapter, and the entire region from the nozzle to the breakup of the jet into drops in chapters 6 and 7. 5.1 Background Laminar liquid jets injected from a circular nozzle into another liquid have been studied for many years (Addison and Elliott, 1950; Scheele and Meister, 1968; Meister and Scheele, 1969b; Meister and Scheele, 1969a; Yu and Scheele, 1975; Richards, 1978; Gospodinov et al., 1979; Richards and Scheele, 1985; Anwar 71

72 et al., 1982). They are important as a means of heat and mass transfer due to the creation of large new surface area. For example, since jetting occurs in sieve plate columns, apparatuses have been specically designed for mass transfer and surface tension experiments with a single jet (Skelland and Huang, 1977; Skelland and Huang, 1979; Skelland and Walker, 1989). Qualitative features of the jet behavior are well known: for low ow rates, drops form, grow, and break o from the nozzle at regular intervals. Above a certain critical velocity a stable laminar jet is formed at the nozzle. The jet rises to a certain length, and then breaks up into drops. At still higher nozzle velocities, the jet becomes turbulent and eventually disrupts into small drops. Brief overviews of the jet literature before 1987 can be found in Vrentas and Vrentas (1982) and Gonz alez-Mendizabal et al. (1987). Previous experimental and theoretical studies can be classied into those examining the steady-state jet, such as the evaluation of the jet radius and the velocity proles in the jet (Addison and Elliott, 1950; Yu and Scheele, 1975; Richards, 1978; Gospodinov et al., 1979; Richards and Scheele, 1985; Anwar et al., 1982), and those investigating jet dynamics, such as drop volume before jetting and jetting velocity (Scheele and Meister, 1968), drop sizes produced by the jet (Meister and Scheele, 1969b), and jet length and disruption velocity (Meister and Scheele, 1969a). In the present work we focus on the investigation of the steady-state laminar axisymmetric jet ow. The major theoretical diculty in calculating jet ows, in which numerical methods are resorted to, is the unknown location of the free surface, which must be found together with the pressure and velocity elds. The jet radius and the velocity proles of a cylindrical laminar liquid jet injected into air were calculated numerically by Duda and Vrentas (1967). They used the stream-

function instead of the radial coordinate as an independent variable in their equations, thus making the jet interface a constant coordinate surface. They then simplied the steady equations of motion for the jet dispersed phase and the jet surface using boundary-layer theory, and solved the resulting problem using

73 a parabolic marching technique. Yu and Scheele (1975) extended this approach to include the continuous phase equations to describe the liquid-liquid jet, but assumed approximate velocity prole forms in their momentum integral treatment of the two phases. Gospodinov et al. (1979) relaxed this assumption and solved the boundary-layer equations for the velocity prole established within each phase, also using a parabolic marching technique. Experimentally, Richards and Scheele (1978, 1985) developed a ash photolysis dye technique to measure the velocity proles in liquid-liquid jets and compared their experimental proles with the models of Yu and Scheele (1975) and Gospodinov et al. (1979). Although they found reasonable agreement with the jet radius and the velocity proles with both models at the highest Reynolds number examined, the agreement deteriorated at lower Reynolds numbers as the interface contraction increased. These issues are further explored in this chapter. These previous numerical approaches relying on boundary-layer theory have, by necessity, only limited regions of validity. They assume that the jet is at steady-state, and the boundary-layer assumption is then used to simplify the equations of motion. This assumption necessarily limits their validity to high Reynolds numbers, where this approximation is considered valid. Vrentas and Vrentas (1982) have shown by comparing boundary-layer results of Duda and Vrentas (1967) and full equation simulations by Omodei (1980) for a free jet that a Reynolds number greater than 1000 must be reached before the boundary-layer limit is valid. This conclusion was conrmed by a more recent experimental and numerical study by Gonz alez-Mendizabal et al. (1987). So far, similar comparisons for liquid-liquid jets are not available since solutions to the full equations of motion have not yet been performed for a liquid-liquid jet. It is the objective of the work described in this chapter to ll this gap. The method discussed in chapter 3, after validation on several test problems, is used to calculate the axisymmetric steady ow of a liquid-liquid jet. The results are compared with experimental data of Richards and Scheele (1985) and

74 the approximate numerical results of Yu and Scheele (1975) and Gospodinov et al. (1979). Also, interface shapes are compared with a new macroscopic momentum balance and previous macroscopic energy balances (Addison and Elliott, 1950; Anwar et al., 1982). The steady-state jet is a special case of the more general problem of the analysis of the dynamic jet and the same numerical technique presented is used in chapters 6 and 7 to examine this more general problem, which involves not only the calculation of the steady jet radius and velocity, but also the transient regions including drop volume before jetting, jetting velocity, drop sizes produced by the jet, jet length, and disruption velocity. 5.2 Problem Denition and Formulation The ow conguration is shown in Figure 5.1. It corresponds to the

experimental setup of Richards and Scheele (1985), details of which can be found in that reference. Briey, a jet of xylene is injected vertically from a circular nozzle upwards into a tank of mutually saturated stationary immiscible water. Figure 5.1 shows the stationary tank of water (uid 1) with density 1 and viscosity 1 with the jet of xylene (uid 2) owing upward with density 2 and viscosity 2 , from a nozzle of inner radius R and outer radius Ro with average velocity v . The interfacial surface tension is , and the gravitational acceleration, g , is directed downward. The distance from the nozzle tip to the top of the tank is L1 and to the bottom is L2 . The distance from the centerline of the nozzle to the outer tank wall is L3 and the nozzle is of total length L4 . 5.2.1 Continuum Formulation of the Equations using the CSF Method It is assumed that the ow in each phase is axisymmetric, viscous, and incompressible. The continuity equation is given in cylindrical, axisymmetric coordinates (r, z ) by equation (3.1) where (u, v ) are the radial, axial components of the velocity eld respectively.2 The dynamic momentum equations are given
2

Note that (v, u) is used by Yu and Scheele (1975) for the velocity eld.

75

Continuous Fluid 1 F=0 1, 1

Outflow Boundary

Tank Wall r=a(z) Jet Interface

Dispersed Fluid 2 F=1 2, 2 Axis of Symmetry

L3 L1 g z n r L2

L4 Inflow Boundary v R Ro Nozzle

Figure 5.1: Liquid-liquid jet ow conguration (not to scale).

76 by equations (3.2) and (3.3) with p the pressure, gr [= 0], gz [= g ] the radial and axial components of the gravitational acceleration, and rr , zr , rz , zz are the components of the Newtonian stress tensor given by equation (3.4). The curvature of the liquid-liquid interface is given by equations (3.5) (3.7) and (3.62) (Kothe et al., 1991; Brackbill et al., 1992). The basis of the VOF method is the fractional volume of uid scheme for tracking free boundaries. In this technique, a scalar function, F (r, z, t), is dened by equation (2.44). The evolution equation (3.8) for the uid function marker eld, F , shows that the interface moves with the uid. The density and viscosity elds are obtained from equations (3.9) and (3.10) from which it is seen that F has the physical signicance of being the relative volume fraction of uid 1. Note that it is possible to relate the steady-state VOF function F (r, z ) to an equation of the interface used by other workers (Omodei, 1980; Reddy and Tanner, 1978; Georgiou et al., 1988; Malamataris and Papanastasiou, 1991; Adachi et al., 1990) of the form r = a(z ), where a(z ) is the radial distance to the interface at axial position z (see Figure 5.1): F (r, z ) = 1 H(r a(z )) and H(x) is the Heaviside step function dened by: H(x) 1, x 0 0, x < 0 (5.2) (5.1)

Equation (5.1) may be used in equations (3.5) to (3.7) to calculate curvature for this case (Omodei, 1980; Reddy and Tanner, 1978): = 1 a 1+ da dz
2

da dz 1+ 2 da dz (5.3)

d dz

77 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 relations (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 (3.1)) in equation (3.12). Then the streamfunction can be calculated from Poisson equation (3.13), given the velocity eld. Equations (3.1) to (3.13) are solved with appropriate boundary conditions in axisymmetric coordinates (r, z ), with the boundary conditions as follows. For the solid walls no-slip conditions are used: u=v=0 For the axis of symmetry at r = 0: u = 0, v =0 r (5.5) (5.4)

For inow into the nozzle, fully developed ow is assumed: u = 0, v = 2 v 1 r2 (5.6)

where v is the average velocity in the nozzle and r r/R, the dimensionless axial distance. The experimental ratio L4 /(2R) was assumed to be suciently large to guarantee this condition (Richards and Scheele, 1985). However, it is seen in section 5.3 that this may not have always been the case. The choice of the outow boundary condition can pose a problem, particularly for low ow rate calculations, because it can inuence the entire ow eld adversely. In the VOF approach, uid is allowed to ow through the mesh with a minimum of upstream inuence. For the outow boundary at the top of the mesh it is assumed that there is no change in the axial direction (Nichols et al., 1980;

78 Hirt and Nichols, 1981): v u = =0 z z (5.7)

It is expected that the higher the speed of ow, the less inuence this boundary condition will have on the upstream ow, as pointed out by Patankar (1980). Alternatively, a free boundary condition could be used (Malamataris and Papanastasiou, 1991; Papanastasiou et al., 1992). The jet interface is assumed to be pinned at the nozzle lip, a fact that is observed experimentally, so no contact angle need be specied in this problem. However, 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 (5.8)

tw are the unit normal and tangent to the wall at the contact line where n w and respectively. 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. 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 at the right face and top face of each
2 2

cell respectively, whereas the pressure, pi,j , and marker function, Fi,j , are located at the center. Further details on the nite-dierence expressions used can be found chapter 3.

79 5.2.2 Macroscopic Balance Approximations Simple relations for the radial interface position as a function of axial position can be obtained by integrating the continuum equations of the previous section over the volume of the liquid-liquid jet, resulting in the jet macroscopic momentum and energy balances (Bird, 1957). All dimensionless variables in this chapter use R as the characteristic length and v as the characteristic velocity. Equations (3.1) to (3.11) can be simplied if it is assumed that the jet is at steady-state, boundary-layers exist at the interface in the continuous and dispersed phases, the interface corresponds approximately to a cylinder with dimensionless local radius a (z ) (see Figure 5.1) and dimensionless axial position z z/R, a parabolic velocity prole exists at the nozzle exit, a fully relaxed, at, velocity prole exists downstream, and viscous terms are neglected. Using these assumptions, and integrating equation (3.3) for the dispersed phase over the volume of the jet, we have the macroscopic momentum balance (derived in appendix D): 0= Nj 4 2 4 a + a3 a2 + a2 1 2 We 3 (5.9)

where z /Re2 is the axial position downstream scaled with the dispersed /2 , F r v 2 /2Rg is the Froude number, phase Reynolds number, Re2 2R 2 v Re2 1 W e 2Rv 2 2 / is the Weber number, and Nj 1 is the buoyancy F r 2 number appropriate for a liquid-liquid jet pointed upwards (+) or downwards () Re2 respectively, which reduces to Nj = for a free jet (1 = 0). This simple Fr form of the momentum balance for a liquid-liquid jet has not to our knowledge appeared before in the literature. Equation (5.9) can be obtained as a limiting case of equation (14) in Anwar et al. (1982) by neglecting the viscous terms and integrating over from the nozzle tip to a at prole downstream. Equation (5.9) reduces to the Slattery and Schowalter (Slattery and Schowalter, 1964; Duda and Vrentas, 1967) and Gavis (1964) result for Nj = 0: 0= 2 4 a3 a2 + a2 1 We 3 (5.10)

80 Finally, equation (5.10) in turn reduces to the Harmon (1955) result that a = 3/2 as W e . Equation (5.9) has an advantage over other macroscopic balance equations in that its limiting form, equation (5.10), was found to be in excellent agreement with the calculations performed by Duda and Vrentas (1967). A similar approach has been used to develop the macroscopic mechanical energy balance for free jets by Addison and Elliott (1950) by assuming an initially at velocity prole at the nozzle, and by neglecting viscous dissipation terms: 0 = Nj a4 + 4 a4 a3 + a4 1 We (5.11)

Scriven and Pigford (1959) obtained the limiting form of this equation for W e . Anwar et al. (1982) assumed a parabolic prole at the nozzle and obtained: 0 = Nj a4 + 4 a4 a3 + 2a4 1 We (5.12)

This equation, with Nj = 0, was also derived earlier by Gavis (1964), but he has an error in the exponents of a in the surface tension term. Note that in equation (5.12) as W e , a = 1/ 4 2, a slightly dierent asymptotic result to that given by equation (5.9). A more recent attempt to predict interface position at low Re2 is discussed by Adachi et al. (1990). Interface positions predicted by equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) are compared in section 5.3 with experimental data and numerical results from the present simulations (R), as well as from previous approximate results of Duda and Vrentas (1967) (DV), Yu and Scheele (1975) (YS) and Gospodinov et al. (1979) (GRP). 5.3 Results and Discussion If equations (3.1)-(3.11) are made dimensionless only ve independent dimensionless groups result (see appendix E). Those used by Yu and Scheele (1975) are the dispersed phase Reynolds number Re2 , Froude number F r, Weber number W e, buoyancy number Nj , and the continuous phase Reynolds number,

81 Re1 2R1v /1 . The Re2 , Re1 numbers represent the ratio of inertial to viscous forces, F r represents the ratio of inertial to gravitational forces, W e represents the ratio of inertial to surface forces, and Nj represents the ratio of buoyancy to viscous forces. In the single phase jet, only three groups, Re2 , W e, and Nj are needed. Other possibilities for dimensionless groups, such as the viscosity ratio 2 /1 , or the capillary number, Ca2 W e/Re2 representing the ratio of viscous to surface forces may also be used. A theory of how these groups aect the radius and velocities in the liquidliquid jet was developed by Yu and Scheele (1975). Briey, they concluded that an increase in jet momentum resulting from a force acting in the same direction as the jet motion will increase jet contraction, while any force that opposes jet motion will reduce jet contraction. Their results show that jet contraction should increase with increasing values of Re2 , Nj , W e, 2 /1 , and decreasing values of Re1 . Table 5.1 illustrates the experimental cases studied by Richards and Scheele (Richards, 1978; Richards and Scheele, 1985) and Duda and Vrentas (1967). For comparison purposes, the previous approximate numerical solutions were either obtained by running the actual programs (Yu and Scheele, 1975; Gospodinov et al., 1979) or have been published (Duda and Vrentas, 1967) and are indicated in Table 5.1. All jets in the present study contract; however, Goren and Wronski (1966) showed experimentally that horizontal (Nj = 0) free jets expand below Re2 of about 14 to 20. Although not shown here, the present simulation (R) does reproduce this result. Non-Newtonian jets can also expand and are discussed by Metzner et al. (1961). In the numerical simulation of the ow in this work, the ow rate at the nozzle was instantaneously increased from zero to its nal value from the initial condition with xylene lling the nozzle and water lling the tank, in much the same way as the physical system was started (Richards, 1978; Richards and Scheele, 1985) and the simulation followed in time until no further appreciable changes

82 Table 5.1: Experimental cases studied. Experimental System 1 2 3 4 8a,b 13c Duda & Vrentas Case 4 Nozzle R, cm Ro , cm v , cm/s L4 /(2R) 2L4 /(R Re2 ) Qerr , % Re2 Re1 We Fr Nj 2 /1 2 /1 Ca2 Temp., C Numerical: Yu & Scheele Gospodinov Duda & Vrentas
a b c

Richards & Scheele (1985)

2 0.47 0.635 12.0 151 0.38 -0.3 1585 1257 3.396 0.1562 1564.2 0.687 0.866

2 0.47 0.635 14.81 151 0.31 -7.6 1957 1551 5.172 0.2379 1267.4 0.687 0.866

1 0.319 0.48 16.33 215 0.57 -6.9 1502 1221 4.263 0.4262 549.8 0.704 0.865

1 0.319 0.48 23.37 215 0.40 3.9 2149 1748 8.731 0.8729 384.2 0.704 0.865

1 0.319 0.48 23.37 215 0.41 0.1 2079 1628 9.003 0.8729 369.8 0.678 0.866

3 0.58 0.64 7.17 123 0.42 -5.7 1160 908 1.541 0.0452 3984.1 0.678 0.866 1038 64 5.400 8.064 128.6 47.9 775 41.9 115 0.44 0.111

0.00214 0.00264 0.00284 0.00406 0.00433 0.00133 0.00520 25 25 27 27 25 25

System 8 has identical conditions to systems 5 to 12. System 11 (identical to 8) used a 5.25 cm radius cylinder to enclose the jet. System 13 has identical conditions to system 14.

83 were noted in the solution. Thus, an advantage of the dynamic simulation is that generally no special initial guess (therefore, no parameter continuation procedure such as from an asymptotic solution) is necessary to arrive at the nal steady-state solution. A typical (coarse) mesh is shown in Figure 5.2, with the increased mesh renement conned, as illustrated, in the dispersed phase. A ner local mesh is used inside the jet region, whereas a coarser mesh is employed in the much more slowly moving outer region. The actual mesh size used for all runs involved 34 cells in the radial direction (with 25 cells being conned to the interval 0 r 1) and 68 cells in the axial direction. This cell distribution has been optimized through mesh sensitivity studies. In addition, the eects of the euent boundary distance, L1 , the bottom distance, L2 , and the outer wall distance, L3 , were investigated by sensitivity studies. The actual tank dimensions were L1 = 23 cm, L2 = 9 cm, and L3 = 10.2 cm. Yu and Scheele (1975) found experimentally that placing an annular disk ush with the nozzle tip (L2 = 0) had little eect on the measured jet radii. In system 11, L3 = 5.25 cm due to placing a glass cylinder around the jet, which had no observed eect on the experimental jet radii or velocity proles (Richards and Scheele, 1985) as compared to systems 5 to 12. These dimensions result in dimensionless lengths of L1 /R > 40, L2 /R > 16, and L3 /R > 9 for the three nozzles. It was found that, for these conditions, the use of L1 /R = 10, L2 /R = 1, and L3 /R = 3 were large enough to establish insensitivity of the results to the actual values of L1 /R, L2 /R, and L3 /R. For example, comparing L3 /R = 3 to 6, no signicant dierence was found in the dispersed phase solution, which is the sole region of the experimental measurements. Likewise, because of the high speed of the jet, comparing L1 /R = 5 to 10, no signicant dierence was found in the dispersed phase solution so that L1 /R = 10 was found to be sucient not to inuence the upstream ow eld, even though the jet velocity prole is not completely relaxed. We can conjecture that the reason that boundary condition

84

10

Symmetry Axis

z*

Wall

-1

1 r*

Nozzle

Figure 5.2: Coarse mesh layout.

85 (5.7) works well at nite L1 /R distances is that upwinding is used in the inertial terms so that very little information is used at the outow in the calculation (Patankar, 1980). Note that not all of the ow domain is shown in Figures 5.5 to 5.10, since the region of experimental results is generally L1 /R < 6. The present simulations were calculated to a length of L1 /R = 12.5 in Figure 5.3. It may be that by using the free boundary condition of Malamataris et al. (Malamataris and Papanastasiou, 1991; Papanastasiou et al., 1992) instead of equation (5.7) as the outow boundary condition, this distance L1 /R = 10 for Figures 5.5 to 5.10 could be further decreased. This shortening of the computational domain could become important if operating at Reynolds numbers Re2 lower than 1000. Table 5.2: Key to data gures 5.3 5.16. System Re2 Interface Position a (z ) Duda & Vrentas (1967) 8 2 13 1038 2079 1957 1160 5.3 5.5 5.6 5.7 5.8 5.9 5.10 Dye Trace z (r , t) Axial Radial

Velocity Velocity v (r ) 5.4 5.11 5.12 5.13 5.14 5.15 5.16 u (r )

A key to the Figures 5.3 5.16 showing the obtained simulation results, is given in Table 5.2. The comparison of the results of the present simulations (R), of case 4 from Duda and Vrentas (1967) (DV), and the interface positions predicted by equations (5.9) (MOM), (5.11) (AE) (Addison and Elliott (1950), and (5.12) (ABDW) (Anwar et al., 1982) with the experimental interfacial data of Duda and Vrentas (open squares) is shown in Figures 5.3 and 5.4. The

comparison of the results of the present simulations (R) to experiments of Richards and Scheele (Richards, 1978; Richards and Scheele, 1985) (RS data points),

86 and simulations using the programs of Yu and Scheele (1975) (YS model) and Gospodinov et al. (1979) (GRP) and the interface positions predicted by equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) are shown in Figures 5.5 to 5.16. The experiments did not study the ve dimensionless groups systematically, but represent a sampling of typical jetting conditions for three nozzle diameters and several ow rates. System 8 was replicated by systems 5 to 12 and system 13 was replicated by system 14, while system 2 has no replicate. As Table 5.2 shows, numerical results of Yu and Scheele (1975) are available for systems 1 to 14, and numerical results of Gospodinov et al. (1979) are available only for systems 2 and 8. However, system 13 has the lowest Reynolds number and greatest contraction. For these reasons we focus our analysis on systems 2, 8, and 13. These three systems also represent typical experimental results of all 14 systems. The main comparison with experimental data involves the interface position and the evaluation of the location of marker particles released above the nozzle that are then tracked in time (equations (3.11)). These marker particle traces compare directly to the laser dye trace experiments from which the velocities have been evaluated (Richards and Scheele, 1985). The jet radius and dye trace positions inside the jet, obtained from high speed motion pictures, were measured. The dye trace positions were optically corrected for the dierence in index of refraction between xylene and water. Axial and radial velocity proles were then calculated by a mass balance technique that is described in subsection 5.3.4. 5.3.1 Duda and Vrentas Case 4 Case 4 from Duda and Vrentas (1967) was selected as a strict validation test of the numerical method developed in the present work. This case corresponds to a water jet injected downward into air, with the conditions used shown in Table 5.1. This case corresponds to a gas-phase outer liquid and is characterized by a moderate Re2 , low Re1 , moderate W e, high 2 /1 , and low Nj . However, it also has high surface tension, 73 dyne/cm, with large viscosity and density dierences.

87 The case is simulated as a two uid ow with the continuous phase having the physical properties of air, L1 /R = 12, L2 /R = 0, and L3 /R = 3. Thus, comparison of the results with those presented by Duda and Vrentas could be used to roughly estimate the errors introduced by the boundary-layer assumptions and neglecting the air phase. Figure 5.3 compares the interface position a (z ) for the DV model and the present simulation (R), and those predicted by equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) as well as the interface position corresponding to the available experimental data. Excellent agreement is shown for the DV and R models and the experimental data. The DV model results exhibit a slightly greater contraction than the R model, which is consistent with the Re2 = 1000 numerical simulations of Omodei (1980) as reported in Vrentas and Vrentas (1982). The slight deviations between the current method predictions and those of DV can be attributed to the fact that 1 = 0 and 1 = 0 for the present simulation, and that the DV model is based on boundary-layer approximations obtained by eliminating certain small (as Re ) terms in the momentum equations. The MOM, AE, and ABDW equations are qualitatively accurate, which is somewhat unexpected considering their simplicity. The MOM and ABDW models are close to the DV model at large axial position, where the assumptions made in deriving them are more appropriate, while the AE model is more accurate near the nozzle since it is guaranteed to match exactly at the nozzle lip. Gonz alez-Mendizabal et al. (1987) state that it is remarkable that [their equation (9), which is the AE model] is generally more accurate than the more elaborate Duda and Vrentas model based on their experimental data. This is clearly not true in the present case by examining Figure 5.3, and it is unreasonable to expect these macroscopic balances to be quantitatively accurate considering the severity of the assumptions listed in section 5.2. Figure 5.4 shows the axial velocity v (r ) as a function of radial coordinate for axial positions z = 5.19 and 10.38, which corresponds to axial positions scaled

88

50 DV Case 4 MOM 40

30 z* 20 AE 10 DV R ABDW

0 0.5 0.6 0.7 r*


Figure 5.3: Jet interface experimental data (squares) and numerical results (DV) of Duda and Vrentas case 4 compared to the present model (R), equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW).

0.8

0.9

1.0

by the Reynolds number z /Re2 = 0.005 and 0.01, respectively. Here again good agreement is obtained between the DV and R models, with the DV model slightly faster due to the increased jet contraction predicted by the DV model, as mentioned above.

89

3 DV Case 4

z* = 10.38 z* = 5.19 2 DV v* z* = 0 1 R

0 0.0 0.2 0.4 r*


Figure 5.4: Axial velocity proles for VOF (solid line) compared to Duda and Vrentas case 4 (dashed line) for axial positions z = 0, 5.19, and 10.38. Interface positions are indicated by circles. 5.3.2 Interface Positions The remainder of the results presented here are based on the experiments of Richards and Scheele (1985) and the corresponding simulations. Figures 5.5, 5.6, and 5.7 compare the experimental interface positions a (z ) computed with the R, YS, GRP programs and the MOM, AE, and ABDW equations for systems 8,

0.6

0.8

1.0

90 2, and 13 respectively. Note that the experimental uncertainty in radial position at a given axial position was about 1% for experiments using nozzles 1 and 3 (Richards and Scheele, 1985). Thus, for systems 8 and 2, the R, YS, and GRP models are all within experimental data uncertainty and not signicantly dierent from each other, although for system 2 the R and the GRP model are closer to, and lie on the same side of, the data. For system 13, whereas the R results are within experimental error and cannot be distinguished from the experimentally determined interfacial position, there is a greater disagreement with the YS model at large axial position. As with Figure 5.3, Figures 5.5, 5.6, and 5.7 show qualitative agreement of the macroscopic balance equations MOM, AE, and ABDW with the data, but no pattern of which one of the three, if any, is best seems to emerge. All that can be said in reviewing these data is that the MOM, AE, and ABDW equations give a qualitative estimate of where the interface is with very little computational eort, but they should not be used to replace the more rigorous estimates of the present simulation, DV, YS, and GRP as is perhaps suggested by Gonz alez-Mendizabal et al. (1987). Anwar et al. (1982) simplied Yu and Scheeles analysis, reducing the problem to solving two nonlinear, coupled ODEs, but we have found by direct numerical comparison that this approach gives similar results to the YS model. We do not report the Anwar et al. numerical simulation results since their approach does not appear to represent an improvement in accuracy over that of the YS model. Although it is not apparent from Figures 5.6 or 5.7, the interfacial position is not as smooth for the present numerical simulation results as for the other two approximate models for systems 2 and 13. This roughness has been smoothed out by a smoothing spline before plotting and appears as slight wiggles on the order of the mesh spacing. It is more pronounced with higher jet contraction with the same mesh spacing, and can be reduced by further mesh renement. Figure 5.5 has not been smoothed with a spline and illustrates the magnitude of the roughness.

91

5 System 8

R AE

4 YS 3 z* 2 ABDW

MOM GRP

0 0.5 0.6 0.7 r*


Figure 5.5: Jet interface experimental data of Richards and Scheele (squares) compared to the present model (R), Yu and Scheele (YS), Gospodinov et al. (GRP), equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) for system 8.

0.8

0.9

1.0

The accuracy obtained is consistent with the fact that the present simulations are accurate to between rst and second order in mesh spacing h and to rst order in time increments t. To minimize this roughness, and in practice to minimize the number of iterations taken for each cycle, some smoothing of the F marker

92

5 System 2

4 YS 3 z* 2 MOM 1 GRP 0 0.5 0.6 0.7 r*


Figure 5.6: Jet interface experimental data of Richards and Scheele (squares) compared to the present model (R), Yu and Scheele (YS), Gospodinov et al. (GRP), equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) for system 2.

ABDW

AE R

0.8

0.9

1.0

function is done internally in the program only to calculate the curvature from equation (3.5), as discussed in the CSF references (Kothe et al., 1991; Brackbill et al., 1992). However, it has been found that too much smoothing can lead to nonphysical curvatures.

93

5 System 13

R MOM

3 z* 2 AE 1 ABDW 0 0.5 0.6 0.7 r*


Figure 5.7: Jet interface experimental data of Richards and Scheele (squares) compared to the present model (R), Yu and Scheele (YS), Gospodinov et al. (GRP), equations (5.9) (MOM), (5.11) (AE), and (5.12) (ABDW) for system 13.

YS

0.8

0.9

1.0

5.3.3 Dye Trace Positions Another approach to comparing experimental data with the numerical procedure involves a direct comparison of the dye trace positions. Dye traces were used in the Richards and Scheele (1985) experiments as a means to identify

94 the location of particles on the streamlines at given increments of time. These data were later analyzed to obtain estimates of the velocities as discussed in the next section. Since this last step involves certain simplifying assumptions, we consider as a more faithful test of the comparison between the numerical and the experimental results the direct comparison of the shape of a dye trace. Figures 5.8, 5.9, and 5.10 show comparisons of dye trace and interface position data with the present simulated dye traces using massless marker particles (equations (3.11)) for systems 8, 2, and 13 respectively. As can be seen from Table 5.1, system 8 has the highest Re2 , Re1 , W e, and lowest Nj , indicating that overall, inertial, surface tension and buoyancy forces are much larger than viscous forces. System 2 has about the same Re2 as system 8, but a lower W e and higher Nj , with a resulting larger contraction in the interface. Apparently the threefold increase in Nj (which increases contraction) aects jet contraction more than the approximately twofold decrease in W e (which decreases contraction) for these two systems. System 13 is notable for having the lowest Re2 , Re1 , W e, and highest Nj , with the result that it has the largest contraction in the interface. Since the ow is mainly in the axial direction, parabola-like shapes are observed in the dye traces for each time step. Experimental dye trace data do not exist near the interface due to interference by optical refraction. As can be observed from the shapes of the traces in Figure 5.8, a trend appears of increasing error with time, about 2%. This error can be explained by an error in the experimentally measured ow rate (by rotameter), which had a reproducibility of about 1.8% (Richards, 1978). Also, as a check on the experimental data reduction done by Richards and Scheele (1985), a ow rate percentage error dened as Qerr 100(Qcalc Qexp )/Qerr was calculated from the velocity prole data vs. the experimentally measured ow rate. As can be seen in Table 5.1, this error ranged from about 0.01% to 8.9% in absolute value for the 14 systems investigated experimentally (Richards and Scheele, 1985), with system 8 having a low calculated error of +0.1%. So, as can be seen from Figure 5.8,

95

5 System 8 Dye Trace 4 Interface 3 z* 2

0 0.0 0.2 0.4 r*


Figure 5.8: Dye trace (crosses) and interface (squares) experimental data of Richards and Scheele compared to the present model (R) for system 8 with t = 0.003735 s. for system 8, the actual ow rate may be higher than the rotameter reading, and from Figure 5.10 for system 13 lower than the rotameter reading. The experimental traces appear to have the same shape as observed in the present simulation for systems 8 and 13. However, in system 2, the experimental traces also appear to be atter

0.6

0.8

1.0

96

6 System 2 Dye Trace Interface

z*

0 0.0 0.2 0.4 r*


Figure 5.9: Dye trace (crosses) and interface (squares) experimental data of Richards and Scheele compared to the present model (R) for system 2 with t = 0.006849 s.

0.6

0.8

1.0

than the present simulation prediction would suggest, Figure 5.9. This atness appears to be associated only with nozzle 2 for system 2, and to some extent for system 1. A possible explanation is that this nozzle was not long enough to develop the velocity prole fully. Langhaar (1942) has solved the entrance ow in

97

5 Dye Trace 4 System 13

3 Interface z* 2

0 0.0 0.2 0.4 r*


Figure 5.10: Dye trace (crosses) and interface (squares) experimental data of Richards and Scheele compared to the present model (R) for system 13 with t = 0.01583 s. a tube approximately assuming a at prole at the inlet and found that: u = 0, v = I0 ( ) I0 (r ) I2 ( ) (5.13)

0.6

0.8

1.0

where I0 , I2 are modied Bessel functions of the rst kind of orders 0 and 2 respectively, and is an decreasing function of 2L4 /(R Re2 ), a table of which

98 is given by Langhaar (1942). The parabolic prole (equations (5.6)) is recovered when 2L4 /(R Re2 ) = ( = 0), and a at prole when 2L4 /(R Re2 ) = 0 ( = ). In order for the dimensionless centerline velocity v (0) to be at least 1.993 ( = 0.4), Langhaar found that 2L4 /(R Re2 ) had to be at least 0.304. All experimental systems are above 2L4 /(R Re2 ) = 0.4, except systems 1 and 2 (see Table 5.1). This again suggests that the atness may be due to an incompletely developed prole. To test this hypothesis, the prole indicated by equation (5.13) was used as an initial condition for the present simulation instead of the parabolic equation (5.6). It was found that it is possible to reproduce the atness in the traces in Figure 5.9 by using = 2.25, such that v (0) was 1.83, and simultaneously increasing the ow rate by 3%. 5.3.4 Axial Velocities Figures 5.11, 5.12, and 5.13 show the axial velocity v (r) as a function of radial position at axial position z = 2, 4, and 2, for systems 8, 2, and 13 respectively. Velocities were calculated using the dye trace measurements, corrected for optical refraction at the interface using a mass balance technique (Richards and Scheele, 1985). Briey, the technique assumes a parabolic velocity distribution at the nozzle exit, then constructs streamline position on each dye trace to satisfy the mass balance between that streamline and the axis. Streamline polynomials are then t from these positions, and dierentiated to give axial and radial velocities. This procedure makes the velocities more prone to error than the dye trace measurements. The maximum standard deviation in the axial velocities expressed as a percentage was 7.9% and 5.8% for experiments using nozzle 1 and nozzle 3 respectively (Richards and Scheele, 1985). However, unlike the dye trace data, the velocity prole data can be compared to all three models. As can be seen in Figure 5.11, all three models appear to be within experimental error and almost indistinguishable in system 8. Dierences are observed to occur among the predictions of the three models for systems 2 and 13, however. Here

99 the present simulation and the GRP model are much closer to the data (which are much atter) and to each other than the YS model. This can be explained by the failure of the parabolic approximation of the velocity prole used in the YS model momentum integral method to represent the actual prole adequately. Also, the present simulation and the GRP model dier from each other near the interface, with the GRP model approaching zero faster with radial position. 5.3.5 Radial Velocities Figures 5.14, 5.15, and 5.16 show the radial velocity prole u (r ) at axial position z = 2, 4, and 2, for systems 8, 2, and 13 respectively. Since radial velocities are about two orders of magnitude lower than axial velocities, the error in their measurement is considerable. The maximum standard deviation error in the radial velocities was about 100% for nozzles 1 and 3 (Richards and Scheele, 1985). Therefore they are to be used only as a measure of the order of magnitude of the radial velocities. All three models show similar trends and magnitudes, except near the interface where the GRP model shows a more pronounced kink in the prole than the present simulation or the YS model, which could be due to mesh renement. The velocity prole dierences noted in the previous sections between the YS model and the data can best be understood in the following way. Both the GRP model and YS model assumed boundary-layer theory, which will be correct at high Reynolds numbers. However, the YS model assumed forms for the velocity proles in the two phases, and solutions were obtained using a momentum integral approach. The jet phase proles were assumed to be parabolic, which can be seen clearly in Figure 5.13. The GRP model relaxed this assumption, but kept the boundary-layer assumption. The lack of t is most probably not due to the boundary-layer assumptions, which are valid at least at Reynolds number greater than 1000 (Vrentas and Vrentas, 1982), but due to the assumption of the form of the velocity proles, since the GRP model does about as well as the present

100

3 System 8 z* = 2 YS 2

v*

GRP 0 0.0 0.2 0.4 r*


Figure 5.11: Axial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), Yu and Scheele (YS), and Gospodinov et al. (GRP) at axial position z = 2 for system 8. Interface positions are indicated by circles.

0.6

0.8

1.0

simulation in predicting the data, and the present simulation assumes neither boundary-layer theory nor forms for the proles. This study also illustrates that interfacial shapes alone are not sucient to distinguish between models, since in most cases the ts between YS and the data were quite good with the exception,

101

3 System 2 YS z* = 4

GRP

v*

0 0.0 0.2 0.4 r*


Figure 5.12: Axial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), Yu and Scheele (YS), and Gospodinov et al. (GRP) at axial position z = 4 for system 2. Interface positions are indicated by circles. perhaps, of Figure 5.7.

0.6

0.8

1.0

102

4 System 13 YS 3 z* = 2

R v* 2

0 0.0 0.2 0.4 r*


Figure 5.13: Axial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), and Yu and Scheele (YS) at axial position z = 2 for system 13. Interface positions are indicated by circles.

0.6

0.8

1.0

5.4

Conclusions A robust and stable direct numerical simulation has been developed by

combining the SOLA-VOF two-uid algorithm (Hirt and Nichols, 1981) with the Continuum Surface Force (CSF) algorithm (Kothe et al., 1991). It has

103

0.0 GRP -0.01 YS

-0.02 R u* -0.03

-0.04 System 8 z* = 2 -0.05 0.0 0.2 0.4 r*


Figure 5.14: Radial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), Yu and Scheele (YS), and Gospodinov et al. (GRP) at axial position z = 2 for system 8. Interface positions are indicated by circles.

0.6

0.8

1.0

successfully simulated high Reynolds number, high buoyancy number, and low Weber number ows, up to the limit of physical instability. The use of this direct numerical simulation allowed us to ascertain the validity of the other approximate steady-state numerical schemes. We have also shown that the present simulation

104

0.0

-0.01

-0.02 u* -0.03 YS

-0.04 GRP -0.05 0.0 0.2 0.4 r*


Figure 5.15: Radial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), Yu and Scheele (YS), and Gospodinov et al. (GRP) at axial position z = 4 for system 2. Interface positions are indicated by circles.

System 2 z* = 4 0.6 0.8 1.0

approaches the steady-state in a time-dependent fashion, thus allowing it to be used in chapter 6 for dynamic studies. We have also introduced a new macroscopic momentum balance predicting interface position that shows qualitative agreement with the experiments.

105

0.0 R

-0.05

u* -0.10 YS

-0.15 System 13 z* = 2 -0.20 0.0 0.2 0.4 r*


Figure 5.16: Radial velocity prole data of Richards and Scheele (triangles) compared to the present model (R), and Yu and Scheele (YS) at axial position z = 2 for system 13. Interface positions are indicated by circles.

0.6

0.8

1.0

The following four conclusions can be drawn about the data and model comparisons: (1) The present simulation, the GRP and YS models, and the RS data are almost indistinguishable from each other and agree well with the experimental

106 data at Reynolds numbers greater than about 2000 and with low contraction. This is consistent with the assumptions (that is the boundary layer limit) made in the models. (2) The present simulation and the GRP models are within experimental error and are within good agreement with each other for all the cases studied. All cases studied, which have Reynolds numbers greater than about 1000, are expected to be within the range of the assumptions of boundary layer theory. However, even the GRP model is not expected to remain accurate as the Reynolds number is decreased. The present program has been independently tested to Reynolds numbers as low as 1, and no limiting high Reynolds number assumptions are made. (3) The YS model diers signicantly from the data and the present simulation results at high interface contraction, with the dierence becoming more pronounced the greater the contraction. This can be explained by the use of parabolic forms in the YS model for the velocity proles as used in the momentum integral method. (4) Interface positions predicted by macroscopic momentum and energy balances give quantitative agreement with the data, but cannot be expected to provide more accurate results than the more comprehensive numerical simulations. In conclusion, we have applied our extended VOF and CSF methods to the liquid-liquid jet problem. This problem has indeed been challenging for several reasons. It involves a moving complex free surface with high surface tension, at high Reynolds number, in the presence of gravity, a dicult problem even for one-phase calculations. In addition, the experiments bear out the fact that the operating region for the jets in this study places them at the limit of physical stability (Richards and Scheele, 1985).

Вам также может понравиться