You are on page 1of 12

Numerical Investigations

Overview
The field of computational fluid dynamics is key to developing a numerical model of a two-dimensional bubble column. As the general equations governing fluid flow are known, the basic restriction in computational fluid dynamics is the restriction in computing power. In order to cope with the problem of limited computer resources, the fluid dynamical models must be simplified so that they can provide a sufficient representation of the physical system but not more. Generally, there are two models that can be used to model two-phase gas-liquid flow, the Euler/Lagrange approach and the Euler/Euler approach. The major difference between the two models is that the bubbles are treated as discrete particles in the Euler/Lagrange model, while in the Euler/Euler model; the bubbles are treated as a continuous gas phase. This causes a big difference in the numerical solution and affects the ease of computability. In the Euler/Lagrange case, the flow field is calculated in a first step from the balance equations of the quasi-homogeneous gas-liquid dispersion where the mean density varies with the local gas hold-up. Secondly, the local gas holdup is determined by tracking all of the individual bubbles in the system. The two steps are then repeated until convergence is obtained. In the Euler/Euler approach, the balance equations of the two phases are solved simultaneously.

Euler/Lagrange vs. Euler/Euler Models


In the Euler/Lagrange representation, individual bubbles are tracked inside a quasi-homogenous gas-liquid field. Newtons law is used to describe bubble motion,

which is advantageous since it allows a simple implementation of the forces acting on the bubbles. It is easy to describe the pressure, drag, virtual mass, and lift forces using the Euler Lagrange model, especially when the bubbles are small and dispersed. Other characteristics, such as bubble-bubble interactions can be described more easily than in the Euler/Euler approach. One advantage of the Euler/Lagrange method is that mass transfer, with or without chemical reaction, bubble coalescence and redispersion can be added directly to the model. Also, no numerical diffusion, which will be discussed later, needs to be introduced into the dispersed phase since each bubble trace can be calculated accurately within a given volume element. Despite their many advantages, Euler/Lagrange simulations also have many drawbacks. Since, it is necessary to track every single bubble individually, large amounts of computer memory are required. For steady state solutions, the calculations are fairly straightforward, but for dynamic calculations, the bubble positions must be stored after each time step. Since this is computationally infeasible, what is normally done is to track a single bubble that represents the center of a bubble cluster. However, then the consideration of bubble dispersion in the cluster poses a problem. In most applications of bubble columns in chemical processes, the gas volume fraction is not small. Thus, high performance computers must be used, which is both expensive and inconvenient. Therefore, the use of the Euler/Euler approach in these situations is both more suitable and practical. Since the Euler/Euler model treats the gas phase a being continuous, it avoids the difficulty of having to record bubble information in the computer memory. This is

obviously an advantage computationally, but causes a lot of the information of the bubbles to be lost. A main difficulty in the Euler/Euler model is developing the averaged equations for dispersed two-phase flows. Two techniques known as volume averaging and ensemble averaging can be used to derive the equations for the two-fluid form. A rigorous development of the time-averaged conservation equations for the Euler/Euler approach was given by Ishii (1975) and later revised by Lahey and Drew (1988). From the relative motion between the discrete bubbles and the continuous liquid, inter-phase transfer terms were derived. These inter-phase transfer terms are modeled as sources and sinks in the momentum and kinetic energy equations and inter-phase transfer are represented by time-volume-averaged equations. In an ensemble averaging approach presented by Zhang and Prosperetti (1994, 1997) the exact Navier-Stokes equations are averaged over all the configurations occupied by the continuous phase by the phase ensemble averaging method. For the disperse (gas) phase, global particle attributes (e.g. velocity, center of mass, etc.) are averaged directly by a method of particle ensemble average. Various other ways of representing the interactions between the two phases have been attempted in the Euler/Euler model. Gasche et al. (1989) took an equation balancing the forces around a bubble. Various groups, including Torvik & Svendsen (1990), used an equation for the turbulent energy (Schwarz and Turner, 1988) and the dissipation caused by the interaction between the two phases. Hjertager and Morud (1993) regarded both phases as space-sharing interdispersed continua and considered the

interaction by some interfacial drag term. Sokolichin et al. (1993) performed dynamical simulations in such a system using an Euler/Euler representation. The averaging of discrete sources of momentum from discrete bubbles is certainly a disadvantage of the Euler/Euler approach. Nevertheless, if an accurate numerical technique is used, Euler/Euler models can provide results comparable to the timeaveraged Euler/Lagrange approach, as has been shown by Sokolichin et al. (1997). Euler/Euler simulations have brought insight to two-phase flow modeling, especially in the area of closure laws. Under steady-state, all dynamic effects even out and the large scale re-circulating flow structures in multiphase flows must be taken into account by further modeling. Dynamic simulations can resolve these re-circulating structures, which eliminates the need for the additional models of the steady-state approach. If, however, dynamic effects are implemented in steady-state models, they can provide excellent agreement with long-time averaged experimental profiles. Steady-state models are usually preferred for industrial applications since computational time is a key factor. They provide fast and accurate tools for future reactor design and scale-up. This is a major reason why the development of reliable steady-state Euler/Euler models for gas-liquid flow in two-dimensional bubble columns is so important.

Euler/Lagrange Models
Different implifications of the Euler/Lagrange model have been made. Lapin and Lubbert (1994) modeled two-phase flow behavior on different scales, which is known as turbulence theory. On the large scale, the container dimensions determine the turbulent flow structure. On the medium scale, the eddy structures are only determined by the energy dissipation rate. Finally, on the small scale, viscous effects become decisive.

When viewed on the scale characterized by the reactor dimensions, or the macro scale, the gas-liquid system is considered as a quasi-single-phase fluid. Since the bubbles are much smaller than the spatial resolution on this macro scale, this assumption is valid. Also, on the large scale, since the density of the gas is much lower than that of the continuous liquid phase, the bubbles do not possess significant momentum. Therefore, the bubbles are regarded as void areas. Bubble motion or local gas hold-up is responsible for varying density in the gas-liquid medium. The variations in density lead to buoyancy forces that in turn lead to convective flows. However, at low buoyancy forces, the freely convecting flows in containers depict simple flow structures at the macroscopic level. On the medium scale, the variations in density can be explained by the motion of clusters of bubbles in a continuous liquid phase. The bubble positions are dependent on the local motion of the liquid phase and the slip velocity of the bubbles. On the small scale, the details of liquid flow around the bubbles dominate the fluid motions. Particularly, the dynamics of the bubble wake determines the bubble motion relative to the continuous liquid phase and the local mixing effects. These local motions are also responsible for effective bubble rise velocity. For simplicity, Lapin and Lubbert (1994) have considered only the large and medium size scales. Even in making the simplification of developing large-scale treat mean of the gas there is still trouble in developing an effective way to describe the gas phase. In the continuous model, the volume elements considered are assumed to be the same as the volume elements of the macro model. The gas phase is described by a bubble distribution density function in the phase space defined by the location vector and the bubble mass. An equation given by Lapin and Lubbert (1994) equates terms which

consider the total change of the density function, W, with time, due to flow of bubbles across the boundaries of the control volume and mass change due to mass transfer, with terms related to the generalized diffusion appearing in gas-phase motion. The density is then given in every volume element as a function of the density function W and the temperature. Another difficultly in the Euler/Lagrange model is determining effective viscosity. Lapin and Lubbert assumed that the effective viscosity is homogenously distributed across the entire column. No adequate theory is available however; so effective viscosity was estimated from laboratory experiment. However, this is not really a problem as numerical experiments have been rather insensitive to viscosity and similar results were found experimentally by Maniskowki et al. (1994). Delnoij et al. also numerically investigated the gas-liquid flow in two-dimensional bubble columns by the Euler/Lagrange method. Both the discrete bubble model and the volume of fluid (VOF) model were used to solve the motion of the gas phase. One extension that was made by Delnoij et al. was that while Lapin and Lubbert (1994) coupled the liquid and gas phases through the effective density of the liquid, but without any momentum exchange term, Delnoij couple the phases by adding a source term, which included all forces imposed on the liquid surrounding the bubbles, into a volumeaveraged momentum equation for the liquid. Results were compared qualitatively with experiments on the partially aerated two-dimensional bubble column of Becker et al. (1994).

Euler/Euler Models
Both stationary and dynamic Euler/Euler simulations have been carried out. Various authors (Torvik & Svendsen, 1990; Pan & Dudukovic 2000) have carried out simulations that differ in the use of the inter-phase forces and the treatment of turbulence. Attempts have been made to understand the physical background for the intensive bubble lateral migration. Several suggestions have been made including the liquid shear (Drew & Lahey, 1987), a wall-lubrication force (Antal, Lahey, & Flaherty, 1991), a forced non-uniform distribution of the Reynolds stresses (de Bertodano et al., 1990), and a Bernoulli like force (Ranade, 1997). Pan et al. (2000) have presented an Euler/Euler dynamic simulation of the twodimensional bubble column. The ensemble-averaged equations were used in solving the velocity and volume fraction field for both the liquid and gas phases. Also incorporated into the models liquid momentum equation is bubble-induced turbulent viscosity and the effects of gas volume fraction on the interphase momentum exchange term. Comparisons were made between the numerical predictions of the model and experimental data provided by Mudde et al. (1997) and Lin et al. (1996). The momentum equations contain interphase momentum-exchange terms that account for the drag and virtual mass force between the bubbles and liquid. The liquid momentum equation also has a term that accounts for the bubble induced turbulent stresses. The turbulent stress in the liquid phase can be subdivided into two components. The first component is independent to the relative motion of bubbles and is due to shearinduced turbulence. The second is caused by bubble-induced turbulence. The shear stress term of the gas phase is not included in the gas momentum equation since each

bubble moves as one entity, thus the rotational and internal motion of the bubbles may be neglected. There is no need to solve the momentum equation within the gas since the pressure inside of the bubble cannot affect the motion of the bubbles directly. Thus it may be assumed that pressure is uniformly distributed within each bubble. The drag and virtual mass coefficients used in the model were obtained by empirical means. Using an Eulerian/Eulerian model, Sokolichin and Eigenberger (1994) presented a laminar, dynamic two-dimensional simulation of gas-liquid bubble flow in a flat and uniformly aerated bubble column. No turbulence model was used in solving the velocity field of the liquid phase and the drag coefficient was assumed constant. In an Euler/Euler model developed by Michele & Hempel (2002) following a similar derivation as that to Pan et al. (2000). One major difference was that Michele & Hempel did not include a term accounting for the virtual mass force. A solution of the system of partial differential equations was carried out using a finite volume scheme with an algebraic multigrid solver (AMG) in the CFD code CFX-4.3. Due to the large size of their bubble column, which was 5 m in height and .63 m in diameter, a coarse grid was used with an average cell length of .059 m, with a total of 13,600 cells. A time step length of .4 s was chosen and a typical solver run of 320 s took about 48 hr to complete with an 800 MHz AMD-K7 PC workstation.

Developments of the Euler/Euler Model


An attempt to explain phase distribution in bubble columns operating in heterogeneous flow was introduced by Krishna et al. (1999). They solved three sets of momentum equations, the liquid phase, the gas phase consisting of small bubbles, and the third for the gas phase consisting of large bubbles. Different slip velocities and

acceleration factors were associated with the small and large bubble phases due to wake interactions. Attention has also been given to the modeling the liquid phase turbulence. Both a standard - model (Pfleger et al., 1999; Sokolichin & Eigenberger, 1999) and a modified - model that accounts for the bubble-induced turbulences (Svendsen et al., 1992; de Bertodano et al., 1994) were used. De Bertodano et al. (1990) used the anisotropic Reynolds stresses model instead of the - model. They demonstrated the effect of Reynolds stresses in creating an additional lift force that pushes bubbles to zones of higher turbulence, which can control the lateral gas volume fraction distributions. As a two-equation turbulence model, the - model introduces two more variables into the calculations and two more sets of equations. After computing the local liquid turbulent kinetic energy and the liquid turbulent dissipation rate , the turbulent liquid viscosity and effective liquid viscosity can be calculated according to equations given by Michele & Hempel (2002). Pfleger et al. notes the differences between a laminar model versus a turbulent approach with k-e model and a turbulent model with additional turbulent dispersion. Overall, the three simulations show a qualitatively correct picture of the overall fluid circulation. However, the laminar result shows exaggerated values of the fluid velocity at nearly all points, especially at the centerline of the column. The velocity gradient also shows a steep increase to the velocity maximums, which is unusual and not observed in experimentation. The velocity profile of the turbulent description more correctly approximates results found from experimentation at upper and medium heights, however, in the neighborhood close to the sparger, better agreement is necessary. The

consideration of turbulent dispersion in the model results in a decrease of liquid velocity and causes the velocity profile to become flattened out at all heights. Pledger et al. have thus concluded that the dispersion is too big in the model. They have also noted that the influence of turbulence modeling cannot only be observed in long-time-averaged velocity and gas void profiles, but also time behavior at fixed position. When comparing the model results to actual experimentation concerning the time series of the horizontal liquid velocity, it was observed that the periodic movement of the bubble hose that is present in experimental data most closely approximates that given by the - model without dispersion. From these results, Pfledger et al. have concluded that the results of the simulations with the - model without additional dispersion terms best reproduce the measured data. Another important problem is the bubble path dispersion according to Sokolichin et al. (1997). When bubbles rise from a point source at the bottom of a bubble column with a high frequency, they interact with one another and do not rise straight upwards even if the mean liquid velocity is zero, mainly due to bubble wake effects. This also occurs in bubble columns that are aerated uniformly across the entire bottom. It is known that small bubbles are accelerated in the wake of larger ones and other a pushed aside. Therefore, on a small scale, a large amount of path dispersion occurs. Sokolichin et al. accounted for this as random motion. The easiest way to account for this random spatially dispersive effect is to add a diffusion-like term to the continuity equation. Sokolichin et al. used a diffusion coefficient derived by Grienberger and Hofmann (1992). Sokolichin et al. conclude that strong diffusional transports may degradate the

density gradients. Thus, numerical diffusion will lead to incorrect driving forces in the simulations. One paper by Sokolichin and Eigenberger describes the importance of the various interphase forces terms. Usually the interactive force term is broken down into three smaller force terms, a friction force term, and added mass force term, and a lift force term. The friction force term is the dominant force contribution and is often the only one considered. It depends on the gas hold-up, relative velocities between the two phases, and a friction coefficient. The friction coefficient is found to depend upon bubble size, although in cases where the bubble size is small (1-10mm), this dependence is weak. The friction force accounts for the interaction of forces between the liquid and bubble in a uniform flow field under non-accelerating conditions. However, if the bubbles are accelerated relative to the liquid, the surrounding liquid must accelerate as well. This force is known as the added mass force as depends upon gas hold-up, liquid density, the relative acceleration difference between the two phases, and a coefficient which corresponds to the volume fraction of liquid which is accelerated with the bubble. Finally, a lift force is necessary as the flow field may induce particle rotation when a particle with a rigid surface moves in a nonuniform flow field. This force of interaction is perpendicular to the main flow direction and depends on the gas hold-up, liquid density, relative difference between the velocities of the two phases, the curl of liquid velocity, and a lift force coefficient.

Computational Difficulties
Problems also arise in the solution of the partial differential equations. To ensure accuracy, it is necessary to base the time step control upon an efficient estimate of the

time integration error. Since error-based time step control is often not available, constant time steps are chosen and if the convergence fails, a smaller time step must be tried. Sokolichin and Eigenberger (1994) have used a second order extrapolation to estimate time error. The procedure implemented required that during each time step, three time integrations would be performed. The necessity of error-based time step control is important in dynamic simulation of buoyancy driven flow systems since small errors caused by time steps that are too large may lead to vastly different solutions. Another factor to be considered is space grid control, since it to has a strong influence on the solution. According to Sokolichin and Eigenberger (1994), in bubble columns, the flow structure changes substantially if a course grid is refined. Typically, the numerical dispersion on a coarse grid is so strong that flow instabilities dampen out and only a homogeneous flow structure remains.

You might also like