You are on page 1of 8

Chemical Engineering Science 56 (2001) 1049}1056

Phase distribution phenomena in two-phase bubble column reactors


Hugo A. Jakobsen
Department of Chemical Engineering, The Norwegian University of Science and Technology, NTNU Sem Saelands vei 4, N-7491 Trondheim, Norway

Abstract In a recent paper, Jakobsen, Sannvs, Grevskott and Svendsen (Ind. Eng. Chem. Res. 36(10) (1997) 4052}4074) presented a review of the present status on #uid dynamic modeling of vertical bubble-driven #ows. Special emphasis was placed on two-phase #ows in bubble column reactors. For these multiphase reactors, the averaged Eulerian multi#uid models have been found to represent a trade-o! between accuracy and computational e!orts for practical applications. Unfortunately, in such multi#uid models constitutive relations are needed to describe the phase interaction processes. It was concluded that the general picture from the literature is that time-averaged liquid velocity "elds are reasonably well predicted both with steady-state and dynamic models of this type. The prediction of phase distribution phenomena, on the other hand, is still a problem, in particular at high gas #ow rates. The present paper gives an overview of the pertinent constitutive relations presented in the literature aiming at a "rm mechanistic prediction of the phase distribution phenomena. This includes transversal forces, steady drag forces, surface tension e!ects, and hydrodynamic bubble}bubble and bubble}wall interactions. Several interaction mechanisms in the turbulence "elds like the so-called turbulent mass di!usion, turbulent migration, turbulent drift velocities, anisotropic turbulent drag forces, as well as the interactions between these mechanisms, and the impacts of variations in bubble size and shape distributions are discussed. Various aspects of these relations have been questioned. It is therefore the aim of this paper to compare the capabilities of the existing Eulerian multi#uid modeling concepts and parameterizations. The various approaches are evaluated using an in-house 2D Euler/Euler steady-state code. There are several reasons why we choose a steady 2D model for evaluation. First, the model has the advantage of being relatively simple, thus the computational e!ort required for practical applications involving chemical reactions, and interfacial heat and mass transfer is feasible. Second, the dynamic axisymmetric 2D models do not give much improvement as the #ow phenomena possibly missing are believed to be 3D. That is, they do not resolve the swirling motion of bubble swarms. Third, most parameterizations presented in the literature are based on, and have so far only been applied to, 2D models. The results obtained indicate that the various phase interaction parameterizations available in the literature predict very di!erent phase distributions in bubble columns. For operating reactors these deviations will signi"cantly in#uence the predicted process performance. The results presented here thus con"rm the demand for improved modeling including more accurate and stable numerical solution algorithms. Low-accuracy algorithms may totally destroy the physics re#ected by the models implemented. 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Multi#uid model; Forces; Turbulence closure; Bubble size and shape; Numerical methods

1. Introduction There are several ways to formulate averaged Eulerian #uid dynamic models intending to describe the bubble column reactor performance. The traditional concept is "rst to formulate integral balances for mass, momentum and energy for a "xed control volume containing all phases in question. These balances must all be satis"ed at any time and point in space, and thus reduces into two types of local equations, one being the local instantaneous equations for each phase and the other an expression of the local instantaneous jump conditions. The next step in this fairly general procedure is to form the
E-mail address: jakobsen@chembio.ntnu.no (H. A. Jakobsen).

average of the local instantaneous transport equations and the corresponding jump conditions. For this purpose, many di!erent averaging procedures have been presented in the literature. These are the volume-averaging, time-averaging, and the ensemble-averaging procedures, and combinations of these basic singleaveraging operators. Unfortunately, the resulting averaged equations cannot be solved directly, as they contain averages of products of the dependent variables. The third step is thus to obtain a solvable set of equations by relating the averages of products to expressions containing products of averaged variables only. This has commonly been done either by employing procedures similar to conventional single-phase Reynolds decomposition, or procedures similar to single-phase Favre

0009-2509/01/$ - see front matter 2001 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 3 2 1 - 3

1050

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

decomposition before averaging. Anyway, these procedures give rise to extra terms in the transport equations, containing covariances of the #uctuating components, analogous to the Reynolds stress terms obtained in the case of single-phase turbulence modeling. The "nal step in this procedure is therefore to formulate proper closure laws and parameterizations. Similar models have also been formulated based on alternative concepts like kinetic and probabilistic theories. It is beyond the scope of this paper to consider all possible model formulations available. Fortunately, the model formulations most frequently used describing bubble column reactors can roughly be divided into two groups: One group of bubble column models is based on the traditional concept described above including a doubleaveraging operator, the time- (or ensemble-) after volumeaveraged models (see Jakobsen et al., 1997, and references therein). The covariance terms obtained from the volume averaging part of the procedure are usually neglected. In the time- or ensemble-averaging part, the Reynolds or Favre decomposition and averaging rules are normally applied, in agreement with single-phase turbulence modeling procedures, in order to separate the #uctuating components of the variables from the time- (or ensemble-) averaged variables. In case of applying Favre averaging only the velocity variables are treated by the Favre approach, the remaining variables are averaged in accordance with the standard Reynolds procedure. The second group of bubble column models is based on the same modeling concept but includes a singleensemble averaging operator only (e.g. Kumar et al., 1995; Friberg & Hjertager, 1998; Mudde & Simonin, 1999). In contrast to the "rst group of bubble column models, the main purpose of applying the Reynolds or Favre decomposition and averaging procedures is now to separate the averages of products into products of averages. In most cases, applying the Favre procedure several variables have been weighted, not only the velocity as in the other group of models. The resulting transport equations are thus similar, but not identical to the equations obtained in the "rst group of bubble column models. In other words, the closure laws found in the literature are thus valid for the model approach determining the basis for their derivation only, and are not always generally valid for any model formulation.

phase interaction parameterizations commonly used in bubble column modeling are very brie#y described. For further details, the interested readers are referred to the original papers. 2.1. Interfacial momentum transfer The investigation of the fundamental behavior of dispersed gas}liquid #ow systems in bubble columns are performed by modeling the generalized or total drag force as a linear combination of the various drag components. The forces assumed to be important for a fundamental description of the dispersed #ow in bubble columns are: the steady-interfacial drag, the added mass force, the transversal lift forces, the Basset force and the interfacial mass momentum transfer rate. The approach chosen is to separate the various forces as far as possible in order to verify the signi"cance of the di!erent e!ects as they are developed and validated. In this paper, focus will be put on the steady interfacial drag and the transversal lift forces as they are considered the important ones determining the phase distribution in these reactors. The standard interfacial steady-drag force given by Ishii and Mishima (1984) is usually applied. In this formulation, the ratio of the Sauter mean diameter to the drag diameter appears as a shape factor. In this work, the shape factor is set equal to 1. The physical phenomena giving rise to lift on single bubbles in liquids can roughly be divided into three groups. The Magnus lift force is due to initial bubble rotation causing an asymmetric pressure distribution around the bubble. It was early observed that the lift force on a sphere could act in the opposite direction of the standard Magnus force at high Reynolds numbers due to the transition from laminar to turbulent boundary layers on the two sides of the sphere. A second lift force, the Sa!man force, acts on non-rotating bodies due to shear in the continuous phase #ow pattern. It has been found that at low Reynolds numbers, the lateral force due to shear should be an order of magnitude larger than that due to particle rotation. At intermediate to high Reynolds numbers, as in bubble columns, only very idealized theories on these forces based on potential theory have been presented. Auton (1983, 1987), and Drew and Lahey (1987) developed equal formulations for the lift force on a sphere in an inviscid #uid. The surface pressure variations which depend on the #uid viscosity in the immediate vicinity of the bubble were estimated by prescribing the boundary velocity pro"le. The tendency for bubbles to deform under various #ow conditions allows for yet another lift force. Kariyasaki (1987) observed a lift force on deformable particles of opposite sign to that on rigid spheres under linear shear #ow. In a later paper by Tomiyama, Miyoshi, Tamai, Zun and Sakaguchi (1998) this force was expressed as a transversal force caused by a slanted wake behind

2. Closure Several alternative parameterizations are available intending to describe the pertinent phase distribution mechanisms. As the distinction between the di!erent constitutive relationships are not always obvious, care has to be taken to ensure that the resulting model formulation chosen is consistent. In the following, the various

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

1051

a distorted bubble in a shear "eld (i.e. a turbulent wake force). 2.2. Turbulent mechanisms It seems to be generally accepted that a turbulent mass di!usion * or dispersion force should occur in the averaged multi#uid model equations. In the double-averaging operator procedure, these mechanisms occur as a phasic di!usion term in the continuity equations applying the Reynolds-like averaging procedure, model 1. Alternatively, these physical mechanisms occur as a dispersion force in the momentum equations if a Favre-like procedure is applied (e.g. Laux, 1998), model 2. In contrast, if a single-ensemble averaging procedure is applied, the phase fraction is de"ned as an ensemble average and no dispersion forces occur explicitly in the continuity equation in this model formulation. Consequently, to account for these physical mechanisms, Lahey, Lopez de Bertodano and Jones (1993) formulated a semi-empirical constitutive parameterization which do not originate directly from the single-averaging procedure applied. However, the resulting relation (model 2c) is here considered to be an alternative to the two models mentioned above. Particle dispersion due to transport by #uid turbulence has also been modeled applying the drift velocity concept of Simonin and co-workers (e.g. Mudde & Simonin, 1999), model 3. This term occurs in the single-average model equations due to correlations between the distribution of the particles and the turbulent #uid motion. The parameterization of these correlations which is based on semi-empirical analysis, is another alternative (i.e. based on Favre-averaged equations) to the di!usive terms in the mass balance (i.e. which originates from the Reynolds-averaged equations, model 1). Simonin and co-workers also developed several more sophisticated versions of this model including additional transport equations based on kinetic * and probabilistic theories. To the author's knowledge, these models have so far only been applied to gas-particle #uidization, and will thus not be described in more detail. If the turbulence is anisotropic, the dispersion coe$cient must be substituted with a tensor. Based on the work by Reeks (1992a), Jakobsen (1993) (in co-operation with Johansen, 1993) developed a turbulent drag force to explicitly account for this anisotropic e!ect in the Reynolds-averaged model. This term was found necessary to enable a proper description of the phase distribution phenomena as an isotropic k} model was applied describing turbulence. The spatial distributions of the turbulent normal stresses will also act as forces on the dispersed phase. This mechanism is based on the theory of bulk liquid turbulent structure control (Subbotin, Ibraginor, Bobkov & Tychinskii, 1971). If the liquid-phase turbulence struc-

ture alone controls the void distribution, the void peaking would occur at the region of the largest turbulent kinetic energy of the liquid phase. This model has been used by Lahey and co-workers (e.g. Lahey, 1987, 1988, 1990; Lopez de Bertodano, Lee, Lahey & Drew 1990). Applying the gradient hypothesis in modeling the Reynolds-like stresses, the turbulent migration term is embedded in the model through the turbulent kinetic energy term. 2.3. Axial wall force According to Lahey et al. (1993), there is an axial wall force acting on the bubbles near an axial wall due to the "nite size of the bubbles and the steep velocity gradient near the wall. It is supposed that there is a region of interaction between the bubble and the wall although the bubbles do not wet the wall. Between the wall and bubbles, located in the immediate vicinity of the wall there is a thin liquid "lm of thickness of the order of the laminar sub-layer thickness. By introducing rough assumptions estimating the gas wall shear, the shear area, the friction coe$cient, and the laminar sub-layer thickness, they derived an approximate wall shear force. 2.4. Hydrodynamic interaction forces Antal et al. (1991) proposed a model of a lubrication wall force that pushes bubbles located in the immediate vicinity of the wall towards the column center. Tomiyama et al. (1995, 1998) noted that this model possesses the defect that a bubble located far from the wall is attracted towards the wall. They modi"ed the original model formulation in accordance with experimental data for bubble trajectories for single bubbles released from a near-wall region in a stagnant liquid. Delnoij (1999) considered the relative importance of the hydrodynamic bubble}bubble interaction force compared to the other components of the generalized drag acting on a non-deformable spherical bubble. In the study of Delnoij (1999) an Euler}Lagrangian model was used to describe a bubble column operating in the homogeneous regime at very low void fractions ((5%). The only indirect hydrodynamic interaction e!ect taken into account was the one due to the disturbance of the local velocity "eld around a speci"c bubble induced by the neighboring bubbles. This induced velocity will contribute to the lift force, the steady drag and the added mass force acting on the bubble under consideration. Delnoij (1999) neglected the hydrodynamic interaction contributions to both the lift and the steady drag forces. It was then shown that the hydrodynamic interaction force exerted by the bubbles on each other, being a modi"ed added mass e!ect only, did not signi"cantly in#uence the macroscopic #ow pattern observed.

1052

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

3. Model validation The basic model used in this work originates from more than 10 years research on bubble column modeling at our department (e.g. Torvik & Svendsen, 1990; Torvik, 1990; Svendsen, Jakobsen & Torvik, 1992; Jakobsen, 1993; Jakobsen, Svendsen & Hjarbo, 1993; Jakobsen, D Sannvs, Grevskott & Svendsen, 1997). The original model belong to the "rst group of bubble column models described above, derived using a double time after volume-averaging operator approach. In the present work, three versions of the original program code have been developed. Model 1 can be looked upon as an updated version of the previous model (Jakobsen, 1993). In the time-averaging part of this model version a Reynolds-averaginglike procedure was adopted. For model evaluation and sensitivity studies, the previous model has been updated to include the interaction terms suggested in the literature described above. The reference model 1 was de"ned containing the following sub-models: the steady drag coe$cient relation by Tomiyama, Sou, Zun, Kanami and Sakaguchi (1995), Tomiyama et al., (1998), the added mass, the turbulent mass di!usion in the continuity equations, the turbulent migration due to the turbulent normal stresses, and the anisotropic drag force by Jakobsen (1993), and the transversal forces by Tomiyama et al. (1998). Note that the anisotropic drag force coe$cient applied here has been tuned to experimental data. For sensitivity analyses, the results predicted by the model formulations de"ned below are compared with the corresponding data obtained using this reference model (i.e. model 1). Model 1 variations: Model 1a, Reynolds-averaged basis model with no wake force (Tomiyama et al., 1998). Model 1b, Reynolds-averaged basis model with no anisotropic drag force. Model 1c, Reynolds-averaged basis model with no turbulent migration force. Model 1d, Reynolds-averaged basis model with lift force parameter set to 0.35. Model 2 is also based on the previous model (Jakobsen, 1993). However, in the time-averaging part of this model formulation, a Favre-averaging like procedure was adopted instead of the previous Reynolds procedure. For model evaluation and sensitivity studies the same updated interaction terms as used in model 1 have been implemented. The turbulent dispersion force of Laux (1998) is used instead of the turbulent mass di!usion terms in the continuity equations. Model 2 variations: Model 2a, Favre-averaged basis model with the lubrication force by Tomiyama (1998). Model 2b, Favreaveraged basis model with the axial wall force by Lopez de Bertodano (1992) and Lahey et al. (1993). Model 2c, Favre-averaged basis model with the turbulent dispersion force by Lopez de Bertodano (1992) and Lahey et al. (1993). Model 2d, Favre-averaged basis model with the drift velocity model by Mudde and Simonin (1999).

Model 3, Favre-averaged basis model with the drift velocity model by Mudde and Simonin (1999), without the phase dispersion force by Laux (1998). For model validation, experimental data for an ordinary bubble column have been used. The #ow is isothermal and no interfacial mass transfer occur. Air/water system data: "1000 kg/m, "1.0;10\ Ns/m, J J "0.070 N/m, and "1.2 kg/m at the top of the J E reactor. Inner column diameter, D"0.288 m, column height, H"4.25 m, temperature 253C, pressure 1 bar, gas and liquid super"cial velocities at inlet 0.08 and 0.01 m/s, respectively. The void fraction data were measured at axial level 2 m above column inlet, whereas the velocity variables are measured at level 1.6 m above inlet.

4. Results and discussion The in#uence of drag on the phase distribution has been studied. For high gas void fraction #ows, drag correlations based on empirical single bubble data have been found unable to predict gas velocity pro"les with reasonable accuracy (i.e. compared to experimental data given by Jakobsen, 1993 and Yu and Kim, 1991), as can be seen from Fig. 1(g). This may be due to the fact that none of these drag relations do explicitly take into account the hydrodynamic bubble}bubble interaction effects. Bubbles #owing in line or in wakes rise faster than single bubbles, but models intending to describe this e!ect by including the e!ect of the bubble swarm's void fraction into the expression for C remain unsatisfactory, " as observed by Jakobsen (1993). In particular, the void fraction variable used for a bubble cluster does not identify whether it is composed of many small bubbles or a few larger ones. Delnoij (1999) did only consider very low void fraction #ows ((5%) when concluding that these mechanisms did not in#uence the main #ow pattern. In bubble columns operating in the heterogeneous #ow region, having local void fractions up to about 30}40%, these interaction mechanisms are expected to be important. Delnoij (1999) also suggested that the contributions to the steady drag and lift forces can be implemented by introducing steady drag and lift coe$cients that depend on the local gas fraction. For the Eulerian}Eulerian models usually applied for higher void fraction #ows, Jakobsen (1993) found that such a procedure was not su$cient when using a single bubble size model. Bubble size and shape distribution models are needed to enable a good implementation of the hydrodynamic interaction e!ects. In line with this conclusion, Krishna, Urseanu, van Baten and Ellenberger (1999) reported that the large bubble swarm velocity was found to be 3}6 times higher than that of single isolated bubbles. It was stated that the limited success in modeling bubble column #uid dynamics in the past is due to the lack of reliable procedures for estimating the large bubble sizes and the corresponding

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

1053

Fig. 1. Results from the 2D steady Euler/Euler model versus measured data. The graphs in the "gures indicate: (a) Gas void fraction, predicted by models 1 and 2. (b) Gas void fraction, predicted by model 2. (c) Gas void fraction, predicted by models 2 and 3. (d) Axial liquid velocity pro"les, predicted by models 1 and 2. (e) Axial liquid velocity pro"les, predicted by models 2 and 3. (f) Turbulent intensity, predicted by models 1 and 2. (g) Axial gas velocity, predicted by models 1 and 2. (h) Bubble size distribution, predicted by models 1 and 2.

drag coe$cients. The development of such a procedure was the main object of their paper. However, any "rm validation of the resulting drag relations has not been reported yet. There are thus several reasons for further studies on the hydrodynamic interaction forces. Figs. 1(a), (d), (f)}(h) show that the predicted void, bubble size, velocity and turbulent kinetic energy pro"les are not signi"cantly altered by the lift forces (i.e. comparing models 1 and 1d). The "gures also show that the importance of the turbulent wake force is negligible too (i.e. comparing models 1 and 1a). This latter result is somewhat surprising as Tomiyama et al. (1995, 1998) found that the transversal forces caused by bubble deformation determine the phase distribution in laminar bubble columns. Several reasonable explanations exist. First, there may be other mechanisms determining the phase distribution in bubble-driven turbulent #ows. Second, the numerical solution algo-

rithms applied to multiphase #ow simulations are often chosen among the very stable ones, and less focus has been put on accuracy. It may be that numerical di!usion is a!ecting the solution to a large extend. Third, the lift and wake force formulations used are developed for singlebubble motions and may be ampli"ed incorporating the e!ects of bubble interactions, coalescence and breakup. Further experimental analyses are therefore needed to validate these models. Fourth, applying the bubble-size model given by Jakobsen et al. (1993), the wake force coe$cient given by Tomiyama et al. (1995, 1998) is zero in most cases because Eo(4. However, the bubble-size model predicts an averaged bubble-size distribution in reasonable agreement with the experimental data, as can be seen from Fig. 1(h). An averaged bubble-size model is thus not su$cient for modeling these phenomena. A bubble size and shape distribution model is needed.

1054

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

The anisotropic drag force has a signi"cant impact on the solution, as can be seen from Fig. 1(a), (d), (f)}(h), comparing models 1 and 1b. This should be expected as the model parameter has been tuned to experimental data. However, without this force both the velocity and the void pro"les become almost #at in contrast to the experimentally obtained data which show a marked parabolic pro"le. From Figs. 1(a), (d) and (g), comparing models 1 and 1c, it can be seen that the turbulent migration terms also have a signi"cant in#uence on the predicted pro"les. In fact, the anisotropic drag force and the turbulent migration force dominate the phase distribution in these model simulations (i.e. for the various model 1 formulations). Zun (1990) stated that the principal weakness of this theory is that it ignores bubble size as one of the key variables. The model also relies on the accuracy re#ected by the turbulence models predicting the normal Reynolds stresses. As has been discussed in a number of papers (e.g. Speziale, 1987; Martinuzzi & Pollard, 1989, Pollard & Martinuzzi, 1989; Launder, 1991; Hrenya, Bolio, Chakrabarti & Sinclair 1995), the popular k} model has severe limitations even for singlephase #ows. One of the undesired consequences of this is that the model predicts spuriously high generation rates of turbulent energy and thus too high levels of turbulent viscosity. Fig. 1(f) shows the turbulent intensity pro"le. It can be seen that the turbulent normal stresses are not accurately predicted with the k} model. Furthermore, the simulated results, Figs. 1(f) and (h), show that the largest discrepancies occur in the wall region where the basic turbulence model is strictly not valid and singlephase wall functions are normally applied. As discussed by Bradshaw and Huang (1995), various aspects of the logarithmic wall law for velocity have been questioned. When the wall law fails the current turbulence models are likely to fail too. The accuracy of the predicted pro"les of turbulent kinetic energy are thus often very low for multiphase #ows, however, as for single-phase #ows, the Reynolds shear stress terms have been reasonably well predicted in many #ow situations. The low accuracy re#ected by the turbulent energy pro"les predicted by this type of turbulence model, limits the possibility of predicting the phase distribution phenomena in these reactors. Lopez de Bertodano et al. (1990) and Jakobsen (1993) extended their models to account for the e!ect of anisotropy by adopting a full Reynolds stress and an Algebraic Reynolds Stress model, respectively. However, in both cases, severe problems were encountered due to the lack of numerical convergence and unphysical solutions. In a later paper, Lopez de Bertodano, Lahey and Jones (1994) instead proposed a modi"ed form of the standard k} model valid for very low void fractions and including two-time scales and an anisotropic matrix to account for the bubble induced turbulence. It was also suggested that the single-phase law of the wall might not be valid for high void fraction multiphase #ows, due to

the existence of a second length scale imposed by the bubble size. This is in accordance with the conclusions by Jakobsen (1993). Further work is needed to enable "rm predictions of the turbulence quantities in high gas fraction bubble column #ows. This may also indicate that large eddy simulations (LES) are the natural next step towards a proper description of the phase distribution phenomena. However, the high resolution required by this type of models and the inherent system disability of de"ning proper separation of scales, limit the application of LES since the bubble-size distribution normally encountered covers the range from less than 0.5 mm to about 2}3 cm. A suitable LES model has to resolve this dispersed phase, making the computational task prohibitively large. Considering the Reynolds-averaged model version, no solution was obtained without the turbulent mass di!usion terms in the continuity equations. However, as discussed by Gray (1975) and Jakobsen (1993), these dispersion terms have several undesirable properties when considering reactive #ows and should thus be avoided. A Favre-like procedure is recommended. However, for non-reactive #ows, both formulations should produce the same #ow pattern. The simulated results show that this is the case when comparing the basic models (i.e. models 1 and 2, Figs. (a), (d), (f)}(h)). The predicted pro"les for all variables shown are hardly distinguishable applying the basic Reynolds-averaged equations containing a turbulent dispersion term in the continuity equations and the alternative dispersion force suggested by Laux (1998). When implementing the second alternative dispersion force, suggested by Lahey et al. (1993), the model did not converge properly. However, by adding this force on top of the basic model 2 formulation (model 2c, Fig. 1b), this force did not have much in#uence on the predicted gas void fraction pro"le, but produces results that are less consistent with the basic model 1. The drift velocity model of Mudde and Simonin (1999) is very similar to the model of Laux (1998). In fact, both models are based on semi-empirical analyses having a similar basis. Reeks (1992b) discussed the turbulent particle dispersion e!ect and argued that the approach of Simonin and co-workers is a more appropriate model than the alternative gradient hypothesis. The model of Laux (1998) represents a simpli"ed formulation compared to the drift velocity model, as only the void-velocity covariances are considered. The length of the relative velocity vector occurring in the drag formulation is assumed not to be a!ected by turbulence. The anisotropic drag force on the other hand was developed accounting for these mechanisms. The model of Laux (1998) is also based on the assumption that the turbulent viscosity predicted by the model will be the same applying both Reynolds- and Favre-averaged models. In the present simulations, this seems to be a reasonable assumption,

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056

1055

but it is not veri"ed that this assumption holds for any #ow situation. The drift velocity model does not rely on these assumptions. Instead, it is based on a semi-empirical probalistic analysis of an undisturbed turbulent #ow "eld of the continuous phase. As can be seen from Figs. 1 (b), (c) and (e), the predicted pro"les are signi"cantly altered by the drift velocity. In model 3 the dispersion force of Laux (1998) is substituted by the drift velocity formulation. In model 2(d) the drift velocity is added on top of the basic model 2. In these simulations the drift velocity formulation predicts pro"les less in accordance with the measured data compared to the other model formulations. Model 2 seems to be consistent with model 1, whereas the magnitude of the dispersion force inherent in model 3 is larger. This discrepancy may be due to di!erent handling of the so-called &crossing trajectory' e!ect in the various model formulations. The fact that the particles pass from one turbulent eddy to the next more quickly than if the particles translated at the mean #uid velocity, e!ectively reduces the time available for #uid}eddy interaction and, in turn, the particle di!usion. This phenomenon has been discussed in several papers (see Elghobashi, 1994, and references therein). It may be that the drift velocity concept overestimates the bubble dispersion e!ect, which should be reduced by the &crossing trajectory' phenomena. However, no turbulent dispersion force model preference can be given as none of them are able to predict unknown #ow "elds with reasonable degree of accuracy. The lubrication wall force parameterization developed by Antal et al. (1991) and later modi"ed by Tomiyama et al. (1995,1998), as well as the axial wall force of Lahey et al. (1993) have no signi"cant impact on the phase distribution, as can be seen from Fig. 1(b), models 2a and b. The lubrication force represents a wall lift force due to viscous phenomena close to the wall (i.e. in contrast to the other lift forces used derived applying potential theory). The physical origin of this force has been questioned, as all wall e!ects are supposed to be lumped into the law of the wall. Therefore, the wall force may be looked upon as a correction to the single-phase wall law valid for a limited number of multiphase #ow situations. The generality of the wall forces is strictly not validated yet, and detailed analysis of the multiphase wall phenomena in turbulent bubbly #ow is highly recommended. A "rm model cannot be developed before the actual physical mechanisms determining the #ow pattern in the wall region are su$ciently understood. Recall that proper predictions of the signi"cant turbulence pro"les in the wall region can hardly be accurately performed even for the much simpler single-phase #ows. The computational time of these simulations having a very coarse-grid resolution is several hours on a Cray T3E supercomputer. The number of iterations is very high, typically 100 000}500 000. This is far beyond reasonable limits for practical applications in industry. In

addition, in most multiphase codes, "rst-order numerical schemes are used for the convective terms due to the inherent stability. It is however well known that the low accuracy obtained using these algorithms may totally destroy the physics (which may be poor by itself) re#ected by these models. Therefore, there is a severe demand for better numerical solution algorithms, both in terms of computational time, and in terms of numerical accuracy and stability for bubble-driven #ows.

5. Conclusion Constitutive models and parameterizations suggested in the literature for prediction of the phase distribution phenomena in bubble column reactors have been evaluated. The phase distribution in bubble columns are believed to be determined by complex interactions between the bubble size and form distributions, transversal forces, and several turbulence mechanisms. Accurate modeling of these phenomena is still limited, both by physical understanding and numerical accuracy. The present models are still on a level aiming at reasonable solutions with several model parameters tuned to known #ow "elds. For predictive purposes, these models are hardly able to predict unknown #ow "elds with reasonable degree of accuracy. It appears that the CFD evaluations of bubble columns by use of multidimensional multi#uid models still have very limited inherent capabilities to fully replace the empirically based analysis in use today. This type of models are thus yet not very useful for industrial applications. The results presented here con"rm the demand for improved modeling including more accurate and stable numerical solution algorithms.

Acknowledgements This work has received support from The Research Council of Norway (Programme for Supercomputing) through a grant of computing time. The author gratefully acknowledges contributions of Mr. I. Bourg, Mr. J. "vstaas, Ms A. S. H. Flateb+ and Ms I. Berg performing model runs.

References
Antal, S. P., Lahey Jr., R. T., & Flaherty, J. E. (1991). Analysis of phase distribution in fully developed laminar bubbly two-phase #ow. International Journal of Multiphase Flow, 17(5), 635}652. Auton, T. R. (1983). The dynamics of bubbles, drops and particles in motion in liquids. Ph.D. thesis, University of Cambridge. Auton, T. R. (1987). The lift force on a spherical body in a rotational #ow. Journal of Fluid Mechanics, 183, 199}218.

1056

H. A. Jakobsen / Chemical Engineering Science 56 (2001) 1049}1056 Launder, B. E. (1991). Current capabilities for modeling turbulence in industrial #ows. Applied Scientixc Research, 48, 247}269. Laux, H. (1998). Modeling of dilute and dense dispersed #uid-particle #ow. Dr.ing. thesis, Norwegian University of Science and Technology, Trondheim, Norway. Lopez de Bertodano, M., Lee, S. -J., Lahey Jr., R. T., & Drew, D. A. (1990). The prediction of two-phase distribution phenomena using a Reynolds stress model. Journal of Fluids Engineering, 112, 107}113. Lopez de Bertodano, M. (1992). Turbulent bubbly two-phase #ow in a triangular duct. Ph.D. thesis, Rensselaer Polytechnic Institute, Troy, NY. Lopez de Bertodano, M., Lahey Jr., R. T., & Jones, O. C. (1994). Phase distribution in bubbly two-phase #ow in vertical ducts. International Journal of Multiphase Flow, 20(5), 805}818. Martinuzzi, R., & Pollard, A. (1989). Comparative study of turbulence models predicting turbulent pipe #ow. Part I: Algebraic Stress and k} Models. AIAA J., 27, 29}36. Mudde, R. F., & Simonin, O. (1999). Two- and three-dimensional simulations of a bubble plume using a two-#uid model. Chemical Engineering Science, 54, 5061}5069. Pollard, A., & Martinuzzi, R. (1989). Comparative study of turbulence models in predicting turbulence pipe #ow. Part II: Reynolds Stress and k} Models. AIAA J., 27(12), 1714}1721. Reeks, M. W. (1992a). On the continuum equations for dispersed particles in non-uniform #ows. Physics of Fluids, A4, 1290}1303. Reeks, M. W. (1992b). Modelling the constitutive relations for dispersed particle #ows. In: Sommerfeld, M. (Ed.), Proceedings from The Sixth Workshop on Two-Phase Flow Predictions, Erlangen, March 30}April 2, pp. 131}143. Speziale, C. G. (1987). On Nonlinear k}l and k} Models of Turbulence. Journal of Fluid Mechanics, 178, 459}475. Subbotin, V. I., Ibraginov, M. Kh., Bobkov, V. P., & Tychinskii, N. A. (1971). Turbulent channel #ow characteristics of gas}water mixtures. Soviet Physics-Doklady, 16(3), 192}194. Svendsen, H. F., Jakobsen, H. A., & Torvik, R. (1992). Local #ow structures in internal loop and bubble column reactors. Chemical Engineering Science, 47(13/14), 3297}3304. Tomiyama, A., Sou, A., Zun, I., Kanami, N., & Sakaguchi, T. (1995). E!ect of Eotvos number and dimensionless liquid volumetric #ux K K on lateral motion of a bubble in a laminar duct #ow. Proceedings of the Second International Conference on Multiphase Flow +95-Kyoto, April 3}7, Kyoto, Japan. Tomiyama, A., Miyoshi, K., Tamai, H., Zun, I., & Sakaguchi, T. (1998). A bubble tracking method for the prediction of spatial evolution of bubble #ow in a vertical pipe. Third International Conference on Multiphase Flow, Lyon, France. Torvik, R. (1990). Investigation of 2- and 3-phase bubble columns. Ph.D. thesis, The Norwegian Institute of Technology, Trondheim, Norway. Torvik, R., & Svendsen, H. F. (1990). Modelling of slurry reactors. A fundamental approach. Chemical Engineering Science, 45, 2325}2333. Yu, Y. H., & Kim, S. D. (1991). Bubble properties and local liquid velocity in the radial direction of cocurrent gas}liquid #ow. Chemical Engineering Science, 46(1), 313}329. Zun, I. (1990). The mechanism of bubble nonhomogeneous distribution in two-phase shear #ow. Nuclear Engineering and Design, 118, 155}162.

Bradshaw, P., & Huang, G. P. (1995). The law of the wall in turbulent #ow. Proceedings of the Royal Society of London, Series A, 451, 165}188. Delnoij, E. (1999). Fluid dynamics of gas}liquid bubble columns. Ph.D. thesis, Twente University, Twente, The Netherlands. Drew, D. A., & Lahey Jr., R. T. (1987). The virtual mass and lift force on a sphere in rotating and straining invisid #ow. Journal of Multiphase yow, 13(1), 113}121. Elghobashi, S. (1994). On predicting particle-laden turbulent #ows. Applied Scientixc Research, 52, 309}329. Friberg, P. C., & Hjertager, B. H. (1998). Simulation of a 3-dimensional large-scale fermenter with four rushton turbines using a two-#uid model. Third International Conference on Multiphase Flow, Lyon, France. Gray, W. G. (1975). A derivation of the equations for multi-phase transport. Chemical Engineering Science, 30, 229}233. Hrenya, C. M., Bolio, E. J., Chakrabarti, D., & Sinclair, J. L. (1995). Comparison of low Reynolds number k} turbulence models in predicting fully developed pipe #ow. Chemical Engineering Science, 50(12), 1923}1941. Ishii, M., & Mishima, K. (1984). Two-#uid model and hydrodynamic constitutive relations. Nuclear Engineering and Design, 82, 107}126. Jakobsen, H. A. (1993). On the modelling and simulation of bubble column reactors using a two-#uid model. Dr.ing. thesis, Norwegian Institute of Technology, Trondheim, Norway. Jakobsen, H. A., Svendsen, H. F., & Hjarbo, K. W. (1993). On the prediction of local #ow structures in internal loop and bubble column reactors using a two-#uid model. Computers in Chemical Engineering, 17S, S531}S536. Jakobsen, H. A., Sannvs, B. H., Grevskott, S., & Svendsen, H. F. (1997). Modeling of vertical bubble-driven #ows. Industrial & Engineering Chemistry Research, 36(10), 4052}4074. Johansen, S. T. (1993). Anisotropic turbulent drag e!ects. Private communication, SINTEF Metallurgy, Trondheim, Norway. Kariyasaki, A. (1987). Behavior of a single gas bubble in a liquid #ow with a linear velocity pro"le. Proceedings of the The ASME-JSME Thermal Engineering Joint Conference, 261}267. The American Society of Mechanical Engineers, New York. Krishna, R., Urseanu, M. I., van Baten, J. M., & Ellenberger, J. (1999). Rise velocity of a swarm of large gas bubbles in liquids. Chemical Engineering Science, 54, 171}183. Kumar, S., Vanderheyden, W. B., Devanathan, N., Padial, N. T., Dudukovic, M. P., & Kashiwa, B. A. (1995). Numerical simulation and experimental veri"cation of the gas}liquid #ow in bubble columns. A.I.Ch.E. Symposium Series, 305, 11}19. Lahey Jr., R. T. (1987). Turbulence and phase distribution phenomena in two-phase #ow. Invited Lecture, ICHMT International Seminar on Transient Phenomena in Multiphase Flow, Dubrovnic, Yugoslavia. Lahey Jr., R. T. (1988). Turbulence and phase distribution phenomena in two-phase #ow. In N. Afghan (Ed.), Transient Phenomena in Multiphase Flow, ICHMT International Seminar, Hemisphere, Washington DC. Lahey Jr., R. T. (1990). The analysis of phase separation and phase distribution phenomena using two-#uid models. Nuclear Engineering and Design, 122, 17}40. Lahey Jr., R. T., Lopez de Bertodano, M., & Jones Jr., O. C. (1993). Phase distribution in complex geometry conducts. Nuclear Engineering and Design, 141, 177}201.

You might also like