Chapter 8 CONCLUSIONS AND RECOMMENDATIONS

In the mathematical investigations I have usually employed such methods as present themselves naturally to a physicist. The pure mathematician will complain, and (it must be confessed) sometimes with justice, of decient rigour. But to this question there are two sides. For, however important it may be to maintain a uniformly high standard in pure mathematics, the physicist may occasionally do well to rest content with arguments which are fairly satisfactory and conclusive from his point of view. To his mind, exercised in a dierent order of ideas, the more severe procedure of the pure mathematician may appear not more but less demonstrative. Lord Rayleigh, Theory of sound (1894) 8.1 Conclusions The overall goal in this dissertation has been to understand certain aspects of complex liquid-liquid ow behavior. Towards that goal, we developed a numerical code for the solution of axisymmetric liquid-liquid ow problems, including high, but still corresponding to laminar ow, Reynolds number ones. We have applied this code to the realistic simulation of aspects of the complex uid mechanical behavior, and developed quantitative insight into the underlying processes involved, such as drop and jet formation, size, shape, and breakup. The problem was numerically and physically complex in that buoyancy, surface tension, viscous, and inertial forces were all important in the experimental systems studied. In this work we have rst developed a robust and exible numerical methodology to solve a fairly general class of axisymmetric free surface liquid-liquid ow problems of relevance to, for example liquid-liquid extraction. We have used this numerical technique to investigate the steady-state and transient liquid-liquid jet 171

172 formation and breakup as well as drop formation. The numerical code can already be used to help design liquid-liquid extraction systems, by obtaining for example, the specic surface area of drops forming from a specic nozzle under pre-specied conditions. The same approach can be used, in principle, for 3-dimensional investigation, although some code eciency issues still remain, as discussed below. Kothe (1993) has indicated that the CSF algorithm has been extended and is capable of solving 3-D problems. The simulation results developed here encompasses and generalizes the results of many workers in the past who used limiting forms of the equations or other signicant approximations. Our simulation predictions compare favorably to available experimental results and dened the domain of applicability of the simplied models of previous investigators. Our research philosophy in this dissertation has been to approach these complex issues in a very basic, generic way. We have taken a numerical approach since the more complex problems defy closed-form or analytical solutions. The emphasis has been on validation of the numerical method with numerous test problems that exercise the various parts of the algorithm as well as obtaining favorable comparisons with available experimental data on liquid-liquid jet and drop systems. From these studies we have located deciencies in the original SOLAVOF algorithm and improved it where necessary. We have used the simulations to help interpret the data and prevailing mechanisms as well as explore the regions of validity of the best existing simplied analytical and numerical models of other workers. We believe this approach has enabled us to lay a rm foundation for future investigations in other liquid-liquid applications as will be discussed further in this chapter. As discussed in chapter 1, the eciency and stability of the separation depend on the uid ow eld, the thermodynamic equilibria, and the mass transfer within and between phases. We have focused on the uid mechanics alone which are dependent on the type of contactor considered. The work here has been concentrated on the continuum level, where velocities and concentrations are local

173 quantities and transport properties such as viscosities, densities, surface tensions, and diusivities, are true state variables. Since actual contactors are too complex for a complete simulation at the present time, we have analyzed idealized liquidliquid systems that possess key aspects of actual extractors. The parametric dependence of the behavior seen in the model systems is expected to be related to that in these more complex systems. The information developed thus far will be needed in the next stage of sophistication, i.e., that of adding solutes and possibly surfactants to solve more general problems of interest. Thus, in this dissertation, we have learned about, and accomplished, two major goals set forth in our original objectives: that of signicant contributions to the numerical methodology and the application of this method to solve a few simplied problems of important practical interest in a more detailed fashion than was possible previously. To accomplish these goals the following procedure was taken (details can be found in the individual conclusion sections at the end of chapters 37): First, we have discussed in chapter 3 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. In chapter 2, we developed the eld equations that are used in this dissertation from a fundamental point of view so that the assumptions involved, and the proofs of the validity of the equations, are evident using framework developed by Scriven (1960) and Edwards et al. (1991). Also, future workers should be able to relax the assumptions in the program quite readily to include, for example, the eects of surfactant on the free surface dynamics. Numerous test problems were used to explore the limits and improve the algorithms, including the new solution of the Young-Laplace equation with constraint discussed in chapter 4. The numerical solution algorithm developed in this dissertation turned out to be intimately linked with the specic problems of interest, so that it proved impossible to solve these problems with any existing CFD code.

174 Second, with this methodology we examined the liquid-liquid jet formation and breakup and drop formation before and after jetting in great detail. This work is covered in chapters 57. We have successfully simulated low to 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, encompassing all of the physics of the previous models. We have also introduced a new macroscopic momentum balance (5.9) predicting interface position that shows qualitative agreement with the experiments and reduces to the result of Slattery and Schowalter (1964) when Nj = 0. We have compared the predicted interface and drop shapes and average jet lengths with experimental data and numerical results of previous workers (Addison and Elliott, 1950; Meister and Scheele, 1966, 1967, 1969a, 1969b; Duda and Vrentas, 1967; Scheele and Meister, 1968; Yu and Scheele, 1975; Gospodinov et al., 1979; Richards, 1978; Anwar et al., 1982; Richards and Scheele, 1985; Bright, 1985). We conclude that the present simulation agrees satisfactorily with the data, within experimental error. The use of this direct numerical simulation allowed us to ascertain the accuracy of the approximate numerical models and to examine the validity of their approximations. With our method, there is no need to switch between various mechanisms depending on the ow regime, as is done in the simplied models, as the correct physics is embedded in the full equations of motion and continuity. 8.2 Recommendations for Future Work

8.2.1 Physical System Issues The experimental system used in this work displays many aspects of an actual liquid-liquid contactor but it is especially dicult to simulate. Unknown disturbances as well as the nonlinearity of the governing equations lead to further diculties when comparing with actual experiments. Additional issues include such aspects as impure materials, resulting in the possible need to include dynamic

175 surface tension, which has an inuence on the predicted jet lengths. Predominant among the issues that the present study revealed is the sensitivity to the external disturbances. An experimental study to characterize these ambiguities further could be undertaken with pure materials and forced amplitude and frequency perturbations for liquid-liquid jets, analogous to the experiments of Donnelly and Glaberson (1966), who investigated liquid jets in air. The jets in this study were simulated only in the region where axisymmetric disturbances were dominant. To examine stability in the presence of non-axisymmetric disturbances, full 3-D calculations will be required. Also a very wide variety of other liquid-liquid ows mentioned in chapter 1 could be investigated. A key objective would be detailed simulation, then development of more compact approximate representations of pseudo-continuum events. For example, further elucidation of drop breakage and coalescence frequency expressions useful in pseudo-continuum population balance formulations used by Guimaraes et al. (1988, 1990) who studied the hydrodynamics and mass transfer eciency of liquid-liquid continuous-ow stirred tanks. Another example that is an inherently unsteady one and is pertinent to the startup of stirred tank contactors is the phenomenon of phase inversion (Quinn and Sigloh, 1963; Selker and Sleicher, 1965). Here, as discussed in chapter 1 simplied ows such as the Kelvin-Helmholtz problem may be used as a starting point to generate insight into these more complex behaviors. Addition of the convective-diusion equation to the existing model would allow the calculation of concentration proles. A general reference discussing the use and solution of this equation in various applications once the ow eld is known is Cussler (1984). For example, Coyle et al. (1981) performed a linear stability analysis for liquid-liquid viscous jets assuming dilute solute mass transfer, with the undisturbed solute concentration linear within the jet and uniform outside, while the undisturbed velocity prole corresponded to plug ow in the jet. The diculty with this solution is that the base state plug ow prole is inconsistent

176 with the actual prole in viscous liquid-liquid jets (as was the case with Bright, 1985) and the local nature of the analysis. The present model, augmented with the convective-diusion equation, would enable calculation of mass transfer coecients without these diculties. Mass transfer coecients in other liquid-liquid systems, such as a swarm of drops in a given shear eld, could, in principle, be evaluated with this extension to the present numerical algorithm. This would be useful in predicting mass transfer rates in staged column extractors such as the ones described by Tsouris and Tavlarides (1990) and other types of equipment discussed by Jereys (1987). Turbulent systems are also important in liquid-liquid systems and this would mean solving the time-averaged equations of motion and continuity (Bird, 1960) along with the convective-diusion equation. If this were possible, extending the present method, turbulent agitated liquid-liquid extractors could be simulated. For example, Skelland and Moeti (1990) used semi-empirical expressions on the basis of local isotropic turbulence theory, for mass transfer coecients in nine turbine-agitated liquid-liquid systems, in baed vessels. That this may be possible is the fact that the one-phase commercial predecessor to SOLA-VOF, namely FLOW-3D (Hirt, 1988), can perform turbulent simulations at the present time using the k model (Harlow and Nakayama, 1967) of turbulence. The case of the eect of a surface tension gradient due to a temperature gradient has recently been investigated by Sasmal and Hochstein (1993) in the context of the VOF method. Surfactant concentration variations at the interface also produce surface tension gradients. The consideration of these eects couples the surfactant species balances with the momentum equations. Additions to the model to understand this behavior would involve equations of state for surface tension and, perhaps, the addition of surface shear and dilatational viscosities to the present model. These are concepts that have already been discussed

briey in chapter 2 and are discussed more thoroughly in Edwards et al. (1991). Appropriate modications would be made to the surface boundary condition

177 (2.50) and momentum equations (2.51). The more general interfacial boundary condition, rst derived by Scriven (1960), can be found in chapter 4 and the surface equations of state and the species convective-diusion equations can be found in chapter 5 of Edwards et al. (1991). A recent example of the FEM calculation of a coating ow using surface viscosities can be found in Giavedoni and Saita (1992). An example of surfactant eects on the motion of drops at low Reynolds number using an Oseen-type solution has been performed by O guz and Sadhal (1988). 8.2.2 Numerical Method Issues The main drawback of the current numerical method is the work required to carry out a full realistic free surface calculation. This will become particularly critical if 3-D calculations are desired. A stringent numerical stability criterion (Hirt and Nichols, 1981) is used to adjust the time step automatically to a small value (typically 1 105 s), and combined with the large number of mesh cells, this results in about 10 teraop (10 1012 op) of work for a typical full jet breakup calculation reported in this work. However, massively parallel teraop machines involving hundreds or thousands of processors will become reality in the next few years (Orszag et al., 1993). Parallel machines have been exploited to solve, for example, 3-D time-dependent incompressible ows using the FEM (Salinger et al., 1993); parallelization of the code is one possibility, but, it is not known at this time if this will be possible. Most of the work in the present code is being performed in the SOR iteration to satisfy continuity. The RIPPLE program (Kothe et al., 1991) uses a pre-conditioned conjugate gradient method to solve a Poisson equation for pressure pi,j , designed to cut down on the number of iterations during a time step. However, if the Poisson equation for pi,j is used instead of the iterative method for pi,j described in chapter 3, pressure boundary conditions on p are required to be supplied, and for exibility, we have chosen not to implement this method. Moreover, RIPPLE can not be used directly since it has a serious limitation in that

178 it has only been implemented to incorporate free-slip boundary conditions on the pressure p. A new method designed to solve the Poisson-like equation (3.38) for pi,j along with equations (3.45)(3.49) for ui,j , vi,j could be implemented with the conjugate-gradient method, while still retaining simple boundary conditions on pi,j . An example of this type of approach has been implemented by Wilkes et al. (1985) and Kightley (1985) with the SIMPLE algorithm of Patankar (1980). An alternative is to use the Alternating Direction Implicit (ADI) method as does FLOW-3D (Hirt, 1988). Another way to address this issue is to decrease the number of variables that need to be solved for during the pressure iteration by reducing the mesh size. Adaptive multigrid methods have been used by Thompson and Ferziger (1989) to solve steady-state incompressible Navier-Stokes equations and may be one possibility. It has been suggested to us by Eggermont (1993) that the VOF advection equation (3.8) could be solved on a ner grid using the velocity elds interpolated from a coarser grid. Another, perhaps more promising, method is local renement of the mesh, particularly at the interface. Locally rened grids have been developed, for example, by Young et al. (1991) and Chen et al. (1990) to simulate geometrically complex aircraft designs, by solving the velocity potential equation using the FEM method. Here local renement is implemented by dividing a Cartesian grid cell into a number of smaller cells. Several possibilities exist for accuracy improvement beyond simple mesh renement. First, the current Hirt-Nichols VOF function advection algorithm advects the free surface assuming it to be either near-horizontal or near-vertical. This advection algorithm can be replaced with the more accurate Youngs (1982) algorithm where the interface is approximated by a straight line cutting through the cell (Kothe, 1993). A related algorithm is FLAIR (ux line-segment model for advection and interface reconstruction) by Ashgriz and Poo (1991). Second, the advection terms in the momentum equations use a linear combination of centraldierence and upwinding by making use of upstream velocities for stability. Kothe

179 et al. (1991) have implemented in RIPPLE an accuracy improvement on the upwinding procedure by retaining second-order terms in the Taylor expansion for the upwind velocities, as originally developed by van Leer (1979). Third, as suggested by Kothe et al. (1991), implicit treatment of the surface tension terms (e.g., as is the pressure in the current algorithm) it is possible to eliminate the sometimes restrictive surface tension stability criterion. Fourth, again as suggested by Kothe et al. (1991), a fully second-order solution procedure may be implemented in time as well as space as proposed by Dukowicz and Dvinsky (1992). In summary, the program and numerical techniques developed in this dissertation have been used to solve free surface problems involving two immiscible liquid phases, high Reynolds number laminar ows and complex, time-dependent interfaces, which is still a topic of much current research. We have contributed to the understanding of liquid-liquid interface issues by investigating the liquid-liquid jet and drop system. We expect this research to inuence the way complex liquidliquid contactors are designed in the future, with extensions to other technologies such as ink jet printing and low-gravity ows in space vehicles where complex interface formation must be addressed.