You are on page 1of 10

Chemical Engineering and Processing 40 (2001) 345 354 www.elsevier.

com/locate/cep

Spray drying of food ingredients and applications of CFD in spray drying


T.A.G. Langrish *, D.F. Fletcher
Department of Chemical Engineering, The Uni6ersity of Sydney, NSW 2006, Australia

Abstract The use of Computational Fluid Dynamics (CFD) packages in the area of spray drying has been reviewed by Bahu (Bahu, 1992, Proc. Eighth International Drying Symposium, IDS 92, Montreal, Canada, pp. 74 91) and Reay (Reay, 1988, Proc. Sixth International Drying Symposium, IDS 88, Versailles, France, KL-1 KL-8), and the movement towards the use of CFD by dryer manufacturers has been outlined by Masters (Masters, 1994, Drying Technology, 12 (1 & 2), 235 237, 1994). Applications include tighter design of spray dryers and reducing operational problems, such as wall deposition. There is still considerable scope for the application of this approach, and here possible future directions for application are reviewed. Particular issues in the use of spray drying for food ingredients are identied and discussed, namely thermal degradation, aroma loss and particle stickiness. 2001 Elsevier Science B.V. All rights reserved.
Keywords: Computational uid dynamics; Numerical simulations; Spray drying; Aroma loss; Thermal degradation; Stickiness

1. Introduction Kerkhof [4] has reviewed the difculties in quantitatively understanding drying processes, identifying the non-linearity of the processes, the complex transfer processes, and the tendency for the dominating phenomena to change during drying. Examples of the non-linearity include the exponential increase in saturated water vapour concentration with temperature, the typical forms of the water vapour isotherms (equilibrium moisture content as a function of relative humidity), and changes in transport properties (such as internal diffusion coefcient) with moisture content and temperature. In addition, material properties, particularly for foodstuffs that may puff, shrink, or both at different stages, are frequently dependent on the previous drying history. Common elements of any drying analysis include mass and energy balances, transport processes, such as convection and diffusion, and the approach to analysing the reaction processes, reaction rates and reaction outcomes. Examples of how these classical chemical engineering analyses can be applied in food
* Corresponding author. Tel.: + 61-2-93514568. E-mail address: timl@chem.eng.usyd.edu.au (T.A.G. Langrish).

processing include assessing the loss of volatile aroma components and the extent of thermal degradation. This review will cover the area of Computational Fluid Dynamics (CFD) and its applications in spray drying, together with the typical ow patterns encountered in these dryers. These ow patterns are particularly relevant to issues associated with the spray drying of foodstuffs, including food ingredients, namely thermal degradation, aroma loss and particle stickiness, and the links between the ow patterns and these food-related issues will be discussed.

2. Computational Fluid Dynamics (CFD) The performance of many types of drying equipment (including spray dryers used for food ingredients) is heavily inuenced by the uid ow patterns inside them. Fluid ow patterns inside process equipment may be predicted by solving the partial differential equations that describe the conservation of mass and momentum, called the Navier Stokes equations. The complex geometry of most drying equipment makes analytical solutions of these equations impossible, but numerical solutions are possible, and these approaches are implemented in Computational Fluid Dynamics (CFD)

0255-2701/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S0255-2701(01)00113-1

346

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

packages. Additional equations are solved to account for the effects of turbulence on the effective gas viscosity. The advent of fast digital computers has improved the accessibility of CFD considerably for designers and operators, and progress in the application of this approach to spray dryers will be described. For those unfamiliar with CFD, the following paragraph constitutes a brief description of the approach. Firstly, the ow domain of interest is split into a large number (typically of the order of 10000) of small volumes. Then Gausss Theorem is used to transform the conservation equations from a set of partial differential equations to a set of coupled algebraic equations by assuming that, for example, the pressure, velocity eld and temperature are uniform over the volume. This is known as the nite volume approach. At this stage boundary conditions at inlets, exits and walls are also applied. Then, an iterative matrix solver is used to invert this set of coupled, non-linear equations, to provide values of all of the variables at each volume. These values then provide a discrete approximation to the oweld. If spray modelling is to be undertaken, the oweld so calculated is used to provide the environment through which particles are tracked in a Lagrangian manner. These particles are subjected to drag forces, buoyancy, etc. and also exchange heat and mass with their surroundings. Iteration is then carried out to couple the particles to the ow, so that the oweld knows about the presence of the particles. Although use of the above techniques has become routine, due mainly to the availability of user-friendly commercially-available packages, there are many pitfalls the inexperienced user must avoid. Simulations must be carried out to demonstrate that the solution is independent of the chosen mesh size, the chosen timestep (in transient calculations) and the number of representative particles used. In addition, appropriate numerical schemes must be chosen to give high numerical accuracy for the given problem. In particular, the approximation of the convective terms is crucial, as low order schemes are stable but diffusive, whereas high order schemes are more accurate but harder to converge. Finally, a good solution can only be obtained if appropriate physical models have been chosen and the calculations properly converged. Spray dryers can be divided into two basic types, short-form and tall-form designs (Fig. 1). Tall-form designs are characterised by height-to-diameter aspect ratios of greater than 5:1, a feature that results in a signicant plug-ow zone inside these dryers [5]. Shortform dryers, in which the height-to-diameter ratios are typically around 2:1, are most common for a variety of reasons, a signicant one being the ease of accommodating the comparatively at spray disk from a rotary atomiser. The implications of these design differences

will be shown in the following review of research on gas and particle ow patterns in spray-drying equipment.

2.1. Basic ow patterns


The ow patterns observed in short-form dryers (which have small height-to-diameter aspect ratios) are more complex than those in tall-form dryers, with many dryers having no plug-ow zone and a wide range of gas residence times. Early attempts at modelling the gas ow patterns concentrated on tting often arbitrary sequences of well-mixed and plug-ow stages together with bypasses to residence time distributions from helium-injection tracer measurements [6,7]. The tted sequences were complex; that of Paris et al. [6] used ve well-mixed stages with complicated bypass connections to t the residence time distribution curves for a 6.7 m diameter, 24 m tall dryer, while Place et al. [7] found that the degree of backmixing in a 7 m diameter dryer of height 15 m corresponded to about six well-mixed stages in series. Both of these dryers were countercurrent designs commonly used for the drying of detergents. The weaknesses of the approaches described above have been highlighted by Reay [2]. They require accurate measurements using helium injection and ow visualisation equipment to be performed in existing dryers, and as such are unsuitable for designing new drying chambers. A variety of zone sequences (wellmixed, plug ow, bypass) may t the data equally well, and it is then difcult to decide which one is most appropriate. These empirical techniques do not enable the effects of varying chamber geometry or operating parameters to be assessed, and both these types of variation are likely to have signicant effects on the ow patterns inside spray dryers. In turn, these changes will affect the dryer performance both in terms of product moisture content and wall deposition rates.

Fig. 1. Schematic diagrams of the main types of spray-dryer geometry. (a) Measurements. (b) FLOW3D predictions. 25 swirl vanes. (c) Measurements. (d) FLOW3D predictions. 30 swirl vanes.

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

347

These weaknesses have led to the utilisation of more fundamental approaches to modelling the ow patterns inside spray dryers, namely the use of Computational Fluid Dynamics packages, and this trend has been made possible by the development of powerful workstations at accessible cost. The rst worker to apply this approach for spray dryers was Crowe [8]. His work demonstrated the central features of these techniques, namely the use of discrete approximations to the timeaveraged conservation equations and the application of a concept known as the Particle Source-in-Cell or discrete droplet model. In this concept, the inuence of the spray is initially neglected in calculating the axial, radial and tangential components of the gas velocities. A large number of droplets are then tracked through the gas inside the chamber. These droplets are chosen to represent the range of droplet sizes leaving the atomiser so that the sum of the owrates of each droplet size equals the total liquid owrate. The heat, mass and momentum transfer rates from the droplets to the gas phase follow from this calculation, and the components of the gas velocities are recalculated with these rates. Other workers [9,10] continued to develop these numerical techniques, usually making assumptions such as isotropic turbulence at a point (with the turbulent stresses being calculated from the product of an isotropic eddy viscosity and the local shear). They also used the k m model for estimating the turbulent eddy viscosity (k =turbulence kinetic energy, o = turbulence dissipation rate). Both these assumptions are considered to be of doubtful accuracy for the strongly swirling ows that are commonly encountered in spray dryers. Reay [2] discussed the work of Goldberg [10], emphasizing the detail that can be extracted from CFD programs. His discussion of the particle trajectories is important, because these trajectories affect the wall deposition behaviour. Small droplets of water (less than 5 mm) evaporate completely in the air ow under the inlet; if solids were present, the small particles would follow the gas streamlines closely. The medium-sized droplets (530 mm) possess sufcient momentum to move through the air inlet zone, but lose considerable mass in doing so, and they get captured by a recirculating eddy under the air inlet. The largest droplets have enough momentum to avoid being entrained in this eddy, and move towards the chamber wall; if solids were present, considerable deposition on the side walls would occur. The CFD program suggested that the most likely areas for wall deposition are an annular area of the roof, corresponding to the small recirculation eddy (medium-sized drops) and a region below the atomiser where large particles are likely to deposit. These predictions correspond to practical experience, and point to an industrial application of CFD programs.

Fig. 2. Velocity vectors in the axial radial plane of a short-form spray dryer at an air owrate though the chamber of 330 m3h 1 (after Oakley et al. [11]).

Oakley et al. [11] presented a powerful illustration of the predictive ability of these techniques. Simulating the same cocurrent dryer geometry as studied by Goldberg [10] (0.76 m diameter by 1.44 m high, with an air owrate through the chamber of 330 m3h 1) using a CFD program (FLOW3DTM), they reported comparisons between the model predictions and experimental measurements of the air ow using laser Doppler anemometry in the absence of spray. Dramatic changes in the air ow pattern in the chamber were measured when the angle of the swirl vanes in the inlet annulus in the top of the dryer around the atomiser was increased

348

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

from 25 to 30 to the vertical. These changes were predicted well by the simulation, as shown in Fig. 2. Convergence difculties were encountered for the case of 30 swirl vanes, and the computer program was run in transient mode to show that these problems were due to a periodic oscillation in the size of the recirculation zones. Fluctuations in the experimental data suggested that this was a real effect, and subsequent measurements [12] have conrmed this feature of the ow. This nding further indicates that this numerical technique accurately predicts the real behaviour of the equipment. Using a development of the model used by Crowe [8], Papadakis and King [13] have also shown that the predictions of this CFD technique agree well with measured air temperatures at various levels below the roof of a 0.56 m diameter by 3.6 m high cocurrent spray-drying chamber when water is sprayed through a pressure nozzle. Their work demonstrated clearly the signicant effects of the spray on the gas ow patterns in the chamber. Oakley et al. [11] also found that the predictions of the air ow patterns were sensitive to the values of the turbulence parameters selected at the annular air inlet. They remarked that experimental measurements of these parameters are not normally available and so either they must be treated as tting parameters (which would reduce the predictive power of the model for design purposes) or they must be obtained from a separate numerical simulation of the air inlet. In a subsequent paper, Oakley and Bahu [14] reported the results of a modied numerical simulation that incorporated a separate simulation of the air inlet (as they had previously recommended) and a differential Reynolds stress model for turbulence, a model that is more likely to accurately predict the effects of turbulence under the strongly swirling conditions in most spray dryers. Even though an improvement was noted in the agreement between the model predictions and the experimental measurements, the use of the complex differential Reynolds stress model involved substantially greater computational effort than the previous k m model. This led Livesley et al. [15] to revert to the use of the k m model when predicting particle sizes and mean axial gas and particle velocities inside industrial spray dryers handling solutions and slurries. In the dryers studied in their work, which ranged from 0.7 m diameter by 1.4 m high to 5.4 m diameter by 10 m high, satisfactory agreement was found between the predictions of the numerical simulation and the experimental measurements. Thus, in spite of the limitations of the k m model for modelling turbulence in spray dryers, this model currently represents an acceptable compromise between accuracy and computational effort in many situations. The measurements were carried out using a variety of instruments including diffractometers (Malvern Instruments 2600) to give particle size distri-

butions, and laser Doppler anemometers (AEA Technology Backscatter LDA and PD LISATEK phase Doppler) to give particle velocity distributions, size distributions and uxes. The success in predicting the gas and particle behaviour in a wide range of dryer sizes suggests that this numerical technique can be reliably used for scale-up purposes. These studies have demonstrated the potential of numerical simulations for improving the design and operation of these types of equipment.

2.2. Time-dependent ows in spray dryers


The stability of the ow patterns inside spray dryers is likely to affect the deposition of particles on walls, the uniformity of the residence time distributions, the energy efciency of the dryers and the temperature time history of particles which pass through the equipment. The relative importance of these issues is strongly dependent on the application. However, there remains sufcient uncertainty in the level of predictive capability about this equipment for the comment of Bahu [1], that spray drying is a mature process about which our understanding remains far from complete, to still hold true. Low-frequency oscillations in the ow eld inside a 1.5 m diameter spray-drying chamber, as measured using a hot-wire anemometer, were reported by Langrish et al. [12]. The strongest low-frequency oscillations were predicted by the CFD program (FLOW3DTM) used by Oakley and Bahu [14], when run in time-dependent mode, for conditions where signicant swirl in the inlet air is sufcient to cause vortex breakdown, the formation of a central recirculation zone and a precessing vortex core (PVC). The successful prediction of vortex breakdown is an important step in assessing the effect of a precessing vortex core on wall deposition, and of a central recirculation zone on overheating and overdrying of particulate material, since small and medium-sized particles tend to be entrained in recirculation zones. Recent fully three-dimensional and transient simulations by Guo et al. [16] (unlike the two-dimensional axisymmetric simulations reported by Oakley and Bahu [14]) for turbulent ow in a sudden expansion of large ratio (greater than three) have been carried out using CFX4. The simulations have shown the existence of both swinging and swirling oscillations of a regular period, even with no inlet swirl. These oscillations are self-induced. Fig. 3 shows an instantaneous snapshot of the oweld downstream of a sudden expansion, as calculated in these simulations. The complex nature of the oweld is evident. Southwell and Langrish [17] observed a similar phenomenon to that reported by Langrish et al. [12] in a pilot-scale dryer using a laser light sheet and ow visualisation.

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

349

Stafford et al. [18] observed an offset central jet, together with unusual up and down owing wall streams on opposite sides, which was attributed to maldistribution of the inlet air. Instabilities in the slow recirculation zones near the walls were observed by Kieviet et al. [19], who were unable to measure sensible mean velocities in many areas. They speculated that this was due to ow instability created by what they referred to as a wiggling core. The stabilisation of gas ows and enhanced mixing of reactants in combustors by strong swirl has been reported. The mechanism for this is the possible formation of a central recirculation zone [20]. However, a central recirculation zone in a spray dryer may also return dried product back into contact with hot inlet gas, increasing residence times, overdrying the particles and possibly leading to excessive thermal degradation. This central recirculation zone has been observed by Oakley et al. [11] in their dryer when the inlet swirl vanes were angled at more than 25.

2.3. Applications of CFD in spray drying


There have been comparatively few reports of the application of these numerical simulation techniques for improving spray-dryer design and operation. For a StorkBowen laboratory-scale dryer with a rotary atomiser and air entering through an annular gap in the roof above the atomiser, Goldberg [10] predicted the trajectories of typical small, medium and large droplets of water. The numerical simulation suggested that small droplets would dry completely within a short distance of the atomiser, while the mediumsized droplets would have enough momentum to penetrate the downwards air blast under the atomiser. However, their reduction in mass due to evaporation would mean that they would get caught in the recirculating eddy surrounding the atomiser. The largest droplets were predicted to possess sufcient momentum

Fig. 3. Instantaneous streamlines for ow transient ow in a sudden expansion. The complex nature of the ow is evident, as is the tendency of the ow to stick to the wall, a feature that would clearly bring droplets into the near wall zone (after Guo et al. [16]).

to avoid being captured by this eddy, eventually evaporating completely close to the outside chamber wall. This suggests that powder deposition from a rotary atomiser on the walls is likely to be greatest in an annular area around the roof corresponding to the small recirculating eddy (for medium-sized droplets) and in a region of the side wall just below the height of the atomiser (for the larger droplets). These predictions correspond to practical experience, demonstrating the potential of the detailed approach involved in these numerical simulations. Oakley and Bahu [14] predicted the trajectories of water droplets from a hollow-cone pressure nozzle in the same chamber as modelled by Goldberg [13] using their numerical simulation. They noted the strong cooling effect of the evaporating droplets on the central gas jet, particularly under the atomiser where the central jet is partially sheltered from the hot-temperature inlet gas stream by the high droplet concentration around the atomiser. Signicant evaporation was predicted in a recirculation zone in the outer regions of the chamber where the mixing with the central gas jet is poor, also giving low gas temperatures. Langrish and Zbicinski [21] demonstrated the use of a CFD program to minimise the wall deposition rate in a spray dryer, with the program using the k m model for turbulence. To validate the model, a solution of sodium chloride containing 20% by mass of the salt was sprayed at the rate of 0.0012 kg s 1 from a two-uid nozzle into a 0.935 m diameter, 1.69 m high cylinderon-cone chamber. The CFD program predicted the wall deposition rate (measured as 0.000044 kg s 1 of dry salt) within 16% at an inlet air temperature of 245C, with the simulation predicting a strongly recirculating airow pattern in the chamber. The CFD program was used to explore methods for decreasing the wall deposition rate, including simple modication to the air inlet geometry (to eliminate swirl) and a reduction in the spray cone angle from 60 to 45. Within the constraints imposed by the experimental equipment, the program suggested that the maximum spray cone angle (60) and the maximum amount of swirl in the inlet air would minimise the wall deposition rate. The trends in the deposition rate caused by decreasing the amount of swirl in the inlet air (0.000093 kg s 1 measured, an increase) and the included angle of the spray cone (0.000099 kg s 1 measured, an increase) have been predicted by the simulation, which suggested deposition rates of 0.000053 kg s 1 for the no-swirl case and 0.000056 kg s 1 for the reduction in spray cone angle. Even though the quantitative agreement for the changes was poor, the trends were reproduced well, demonstrating the usefulness of the technique. For the case of a narrow spray cone angle, the high deposition rate on the bottom conical wall of the chamber (as observed) was well predicted (Fig. 4).

350

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

Fig. 4. Predicted particle trajectories with a narrow spray cone angle (after Langrish and Zbicinski [21]).

The use of CFD to assist with scale up has been reviewed by Oakley [22]. One of his conclusions was that dimensional analysis is of limited use because of the difculty in ensuring dynamic similarity between small and large chambers. Dynamic similarity of airow patterns is possible with a geometrically identical scale model, identical (scaled) inlet conditions, identical chamber Reynolds numbers (although this requirement may be relaxed if the Reynolds number is sufciently large for viscous forces to be unimportant) and negligible effects of buoyancy and gravity on the airow. True similarity of droplet behaviour is more difcult, because droplet behaviour scales with the droplet Reynolds number, not the chamber Reynolds number, and the required control of atomisation is not yet achievable to give sufcient control of both the drop size distribution and the initial drop velocities. Existing empirical rules are limited in their range of applicability, and CFD is potentially a powerful tool which is available to overcome this limitation. However, he stressed that the difculties in modelling the gas turbulence mean that experimental validation of the results from CFD codes is required before these codes can be considered reliable in any specic application. He suggested that validation work on small-scale equipment was probably adequate, given the impracticality of making measurements on large dryers and the consideration that the CFD codes solve general uid ow equations and model the actual uid and particle ow behaviour. He also emphasised the importance of having a good knowledge of the inlet conditions, since small changes in inlet swirl and turbulence levels have large effects on the subsequent ow structure, and

different swirl levels alter the turbulence structure and the applicability of various turbulence models. This feature makes the validation of CFD predictions at one level of inlet swirl applicable only to similar swirl levels. An industrial application of a CFD program to spray dryer design has been reported by Masters [3]. Wall deposits were to be avoided in a design which involved a 20% scale-up in capacity on a previous design. A pilot-plant trial was used to conrm the potential of an air curtain, which involved using a secondary air ow around the walls to prevent powder settling on them. A CFD program (FLUENTTM) then enabled this modication to be scaled up into the larger drying chamber which represented the nal design. No details were given of the turbulence modelling approach selected within the package, a feature which Oakley and Bahu [14] have found to be critical to successful prediction of the airow pattern and hence the droplet trajectories. Southwell et al. [23] have used CFD to improve the ow distribution in a pilot spray dryer. This study investigated ways in which a uniform ow in the plenum chamber could be obtained for a conguration with a single, off-axis, inlet pipe. They used the CFX4TM code to perform calculations for a variety of bafe congurations to determine the effect of various options, without the need to carry out extensive experimental studies. The code results were conrmed by experimental data for some congurations.

2.4. Process intensication


Bahu [1] has identied process intensication of spray dryers as a signicant challenge, given their large size (a dryer of 6.5 tonne capacity, fairly typical in the Australian milk industry, is normally 12 m in diameter and 24 m tall). Smaller dryers would result in a signicant decrease in capital cost. Zbicinski et al. [24] conducted a set of experiments on a tunnel dryer of 30 cm diameter to investigate the effect of enhanced turbulence in the inlet air ow on overall drying performance. This could be viewed as one approach to reducing the size of dryers. A specially-designed grid (a turbulence promoter) was inserted in the air stream upstream of the atomiser, which only sprayed water in their experiments. Three pairs of experiments were reported. The conclusions of their study were that, for a given set of inlet air temperatures and water ow rates, increasing the average inlet turbulence level from 2% to approximately 19% could result in an improved evaporative performance by about 16 25%. Southwell et al. [25] simulated this set of data using the CFD package CFX4 and conrmed the trend of increased evaporative performance from increased inlet turbulence, but only by between 5 and 11%. From their simulation, enhancing the inlet turbulence level appeared to increase evaporation by improving spray

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

351

dispersal, by forcing the particles to zig zag through the air ow (increasing the relative air-particle velocity), and by enhancing the exchange of air between the spray envelope and the surrounding hot and relatively dryer air. The same trends were observed between the experimental data and the simulations in terms of the effects of inlet parameters. Spray-generated turbulence was predicted to dominate the near-atomiser region for all the cases, particularly where the inlet turbulence levels were not enhanced, and it was in this region that virtually all the difference between experimental data and simulation was observed. This turbulence from the spray was predicted to decay rapidly downstream of the atomiser, although the residual effect was apparent downstream, particularly where inlet turbulence levels were not enhanced. Cross-stream gradients in temperature and humidity were also predicted to be less than those without enhanced inlet turbulence. Figs. 5 and 6 highlight the effect of the ow conditions on the particle trajectories for changing spray velocities and ow inlet turbulence intensities, respectively.

namic driving forces (chemical potentials) are all automatically included in the equations. The main difculty is the need to estimate or t values for the Maxwell Stefan diffusivities. However, the practical advantage of using the generalised MaxwellStefan equations to model the movement of salt and water in cheese found by these authors was that the technique was able to t salt and water concentration proles as well as the overall salt/moisture ratio as a function of time with three tted coefcients, a result that was not possible using an approach based on Fickian diffusion.

3. The spray-drying of foodstuffs As mentioned in the introduction, signicant issues in the use of spray drying for food ingredients include aroma loss, thermal degradation and particle stickiness. The problem of wall deposition involves understanding both the ow patterns (CFD) and particle stickiness.

3.1. Aroma loss and thermal degradation


Kerkhof [4] has thoroughly reviewed the loss of volatile aroma components under the conditions of spray drying, an example being the overview of Coumans et al. [26]. These analyses are very illuminating. A typical approach involves simplifying the system to a ternary one including water, a trace aroma component and solids. The transport of the aroma component includes considering both a straight aroma diffusion coefcient (with the driving force for one component of the aroma ux being the gradient in aroma concentration) and a cross-diffusion coefcient (with the driving force for another component of the aroma ux being the gradient in moisture concentration). Aroma transport becomes abnormally slow at low moisture contents due to the very low aroma diffusion coefcients compared with those for water at these moisture contents, creating effectively semi-permeable dry layers. When considering the multi-component diffusion of salt and water in cheese, Morison and Payne [27] have recently suggested that the generalised Maxwell Stefan equations have a number of advantages over standard Ficks law methods. Effects such as viscous ow, electrostatic potentials, pressure drops and the thermody-

Fig. 5. Droplet trajectories showing the effect of nozzle droplet velocity on the ow pattern and wall deposition rates (after Southwell et al. [25]).

352

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

d zi = kizi dt

(1)

where zi is the concentration of the heat-sensitive component. The degradation rate constant ki depends on the temperature and moisture content of the material, typically decreasing at lower moisture contents and increasing at higher temperatures. The material temperature and moisture content are dependent on the gas temperature and humidity, and the owrates of gas and solids, through mass and energy balances, and the rates of heat and mass transfer processes. Liou and Bruin [28] assumed that the heat-sensitive material was immobile. All of the works reviewed by Kerkhof were carried out before the uid ow patterns were understood in great detail. Hence, the predictions were carried out using rst approximations to the particle and uid ow patterns and, in some cases (including a simulation of the thermal degradation of skim milk particles), there was no comparison with experiments. In the case of the thermal degradation of the milk particles, the droplet diameter was predicted to have a signicant inuence on the nal milk activity, due to the effect of diameter on residence time. Using the assumption that the gas and the particles moved through a spray dryer in cocurrent plug ow, Rulkens and Thijssen [29] found that the experimentally-measured aroma retention, using methanol, n-propanol and n-pentanol as aroma model components, was around 70% of that predicted using the model assumptions. Apart from the uncertainty in the assumption about the ow pattern, another reason for the discrepancy was uncertainty about the particle morphology (whether the particles were hollow or solid). The model predicted that the aroma retention should increase with increasing solids content, due to the lower aroma diffusion coefcient, and this trend was also found in the experiments.

3.2. Wall deposition


Fig. 6. Droplet trajectories showing the effect of airow inlet turbulent intensity on the ow pattern and wall deposition. The inlet turbulence intensity is 2% in the top gure and 19% in the bottom one (after Southwell et al. [25]).

Regarding thermal degradation, important issues in the drying of heat sensitive products, such as milk and vitamin C, include residence time and the maximum exposure temperature. A large range of residence times will lead to non-uniform moisture contents even with thermally-insensitive products. For many spray-drying problems, Kerkhof [4] suggests that the thermal degradation of heat-sensitive components (such as enzymes) follow rst-order reaction kinetics:

Particles build up on the walls of spray dryers due to the adhesion of particles to initially clean walls. Subsequent layers of particles become attached to this initial layer (cohesion). On the other hand, particles may be removed from the wall deposits by the shear stress created from the gas ow past the wall. Eventually, a dynamic equilibrium is established between newly attached particles and detaching layers [30]. The interaction of substantial wall deposition with the residence time distribution was noted by Kieviet et al. [19], who concluded that the time taken to slide down the conical wall to the outlet was the most important factor in determining residence times with high wall deposition rates. Not only may wall deposition reduce product quality and yield, but it may also pose a potential re risk and compromise hygiene re-

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354

353

quirements [31]. Such hazards include ignition of explosible dust clouds, dust deposits, bulk powder deposits and ammable vapour. Chen et al. [32] suggested that a combination of modied near wall airow patterns and inlet temperature distribution could be used to signicantly reduce wall deposition in industrial-scale spray drying of milk in New Zealand. Although they were unable to reach quantitative conclusions about the effects of dryer operating parameters, such changes are likely to be substantially less costly than making major modications to the design of an existing dryer. The history of ideas and designs proposed for the reduction of wall deposition has been reviewed by Masters [33], who concluded that CFD techniques offered assistance in nding practical solutions.

4. Current and future developments There appears to be considerable scope for using these numerical simulations for other purposes, such as: 1. investigating methods for reducing wall depositions rates and thermal degradation of particles by modifying the air ow patterns in the chamber through small changes in the air inlet geometry for existing dryers; 2. carrying out performance (rating) and design calculations to get the greatest throughput for a given dryer, or designing a dryer for a given duty. The constraints include maximum allowable wall deposition rates and maximum product moisture contents, both of which may be estimated using these techniques. 3. degradation kinetics need to be measured under similar conditions to those in spray dryers. The numerical simulations should be valuable supplements to pilot-scale testing, enabling more extensive and accurate optimisation to be carried out than hitherto possible. The simulations can be performed using commercially-available Computational Fluid Dynamics programs, examples of which include CFXTM, PHOENICSTM, FIDAPTM and FLUENTTM (this list is not exhaustive) as reviewed by Dombrowski et al. [37].

3.3. Particle stickiness


Closely linked to the issue of wall deposition is that of particle stickiness. CFD allows the particle trajectories to be predicted, so that it is possible to estimate whether the particles hit the walls and the temperatures and moisture contents of such particles. However, whether the particles bounce off the walls, adhere to them, or cohere to other particles that are already adhering to the walls, is closely related to whether the particles are sticky or not. Above a certain temperature, foodstuffs containing sugars and acids undergo structural changes towards a more sticky (rubbery) state [34]. This transition can be detected mechanically in a stirred laboratory-scale beaker arrangement, giving information on the temperature and moisture content required. Pasley et al. [35] found this method to give reproducible results. The transition may be important both for cohesion of particles to one another in agglomeration (the situation that was strictly involved in this test) and adhesion of particles to walls, if the deformability of the particles, as assessed in this test, is important. Other workers [36] have also noted this characteristic sticky behaviour. When the sugar content is signicant, the material property stickiness is strongly related to the glass transition temperature, which is measured calorimetrically, in a Differential Scanning Calorimeter, by detecting the change in specic heat capacity associated with the transition from the amorphous to the rubbery phase. It is often used synonymously for sugary (carbohydrate) materials. Bhandari et al. [36] stress that stickiness can begin 1020 K above the glass transition temperature, since the critical viscosity associated with stickiness does not necessarily occur exactly at the phase transition point. They gave formulae to predict the glass transition temperature for sugars and acids as a function of the moisture content.

5. Conclusions CFD techniques have been demonstrated to give useful predictions of the performance of short-form spray dryers where the ow patterns are complex. Some industrial applications have been described, particularly in the area of predicting the deposition of particles on walls, and using this knowledge to design preventative measures. Other design and performance analysis applications are possible, ranging from predicting the thermal degradation of particles in these dryers, to predicting aroma loss and to facilitating the start-up and shut-down of spray dryers. Most of these applications are relevant to the use of spray drying for food ingredients, as well as to conventional dryer applications.

References
[1] R.E. Bahu, Spray drying maturity or opportunities, in: A.S. Mujumdar (Ed.), Proc. Eighth International Drying Symposium IDS 92, Montreal, Canada, Drying 92, Elsevier, Amsterdam, 1992, pp. 74 91. [2] D. Reay, Fluid ow, residence time simulation and energy efciency in industrial dryers, in: M. Roques (Ed.), Proc. Sixth International Drying Symposium IDS, Unknown, Versailles, France, 1988, pp. KL.1 KL.8.

354

T.A.G. Langrish, D.F. Fletcher / Chemical Engineering and Processing 40 (2001) 345354 [20] A.K. Gupta, D.G. Lilley, N. Syred, Swirl Flows, Abacus, Turnbridge Wells, 1984. [21] T.A.G. Langrish, I. Zbicinski, The effects of air inlet geometry and spray cone angle on the wall deposition rate in spray dryers, Trans. I. Chem. E. 72 (A) (1994) 420 430. [22] D.E. Oakley, Scale-up of spray dryers with the aid of computational uid dynamics, Drying Technol. 12 (1 2) (1994) 217 233. [23] I. Zbicinski, J. Grad, C. Strumillo, Effect of turbulence on heat and mass transfer in the atomisation zone, Drying Technol. 14 (2) (1996) 231 244. [24] D.B. Southwell, T.A.G. Langrish, D.F. Fletcher, Use of computational uid dynamics techniques to assess design alternatives for the plenum chamber of a small spray dryer, in: K. Abdullah, A.H. Tambunan, A.S. Mujumdar (Eds.), Proc. First Asian-Australian Drying Conference, Bali, Indonesia, 1999, pp. 626 633. [25] D.B. Southwell, T.A.G. Langrish, D.F. Fletcher, Process intensication in spray dryers by turbulence enhancement, Trans. I. Chem. E. 77 (A) (1999) 189 205. [26] W.J. Coumans, P.J.A.M. Kerkhof, S. Bruin, Theoretical and practical aspects of aroma retention in spray drying and freeze drying, Drying Technol. 12 (1 2) (1994) 99 149. [27] K. Morison, M.R. Payne, Multi-component diffusion of salt and water in cheese, in: G.D. Rigby (Ed.), Proc. Chemeca, Unknown, Newcastle, Australia, 1999, pp. 791 798. [28] J.K. Liou, S. Bruin, An approximate method for the nonlinear diffusion problem with a power law relation between diffusion coefcient and concentration. 1. Computation of desorption times. 2. Computation of concentration proles, Int. J. Heat & Mass Transfer 25, (1982) 1209-1220, 1221-1229. [29] W.H. Rulkens, H.A.C. Thijssen, Numerical solution of diffusion equations with strongly variable diffusion coefcients. calculation of avour loss in drying food liquids, J. Food Technol. 7 (1973) 95 101. [30] H. Masuda, S. Matsusaka, Particle deposition and reentrainment, in: K. Gotoh, H. Masuda, K. Higashitani (Eds.), Powder Technology Handbook, 2nd edn, Dekker, New York, 1997, pp. 143 154 Ch. II.7. [31] J.A. Abbott, Prevention of Fires and Explosions in Dryers A User Guide, 2nd edn, Institution of Chemical Engineers, Warwickshire, England, 1990. [32] X.D. Chen, R. Lake, S. Jebson, Study of milk powder deposition on a large industrial dryer, Trans. I. Chem. E. 71 (C3) (1993) 180 186. [33] K. Masters, Deposit-free spray drying: dream or reality?, in: C. Strumillo, A.S. Mujumdar (Eds.), Proc. Tenth International Drying Symposium (IDS 96), Drying 96, Krakow, Poland, vol. A, 1996, pp. 52 60. [34] B.R. Bhandari, N. Datta, T. Howes, Problems associated with spray drying of sugar-rich foods, Drying Technol. 15 (2) (1997) 671 684. [35] S.E. Papadakis, R.E. Bahu, The sticky issues of drying, Drying Technol. 10 (4) (1992) 817 837. [36] H. Pasley, P. Halous, S. Ledig, Stickiness a comparison of test methods and characterisation parameters, Drying Technol. 13 (5-7) (1995) 1587 1601. [37] N. Dombrowski, E.A. Foumeny, A. Riza, Know the CFD codes, Chem. Eng. Progress 89 (9) (1993) 46 48.

[3] K. Masters, Scale-up of spray dryers, Drying Technol. 12 (1 2) (1994) 235 257. [4] P.J.A.M. Kerkhof, The role of theoretical and mathematical modelling in scale-up, Drying Technol. 12 (1 2) (1994) 1 46. [5] R.B. Keey, Q.T. Pham, Behaviour of spray dryers with nozzle atomizers, Chem. Eng. (1976) 516-521. [6] J.R. Paris, P.N. Ross, S.P. Dastur, R.L. Morris, Modelling of the air ow pattern in a countercurrent spray-drying tower, Ind. Eng. Chem. Process Des. Develop. 10 (2) (1971) 157 164. [7] G. Place, K. Ridgway, P.V. Danckwerts, Investigation of airow in a spray-drier by tracer and model techniques, Trans. I. Chem. E 37 (1959) 268 276. [8] C.T. Crowe, Modelling spray air contact in spray-drying systems, in: A.S. Mujumdar (Ed.), Advances in Drying, vol. 1, Hemisphere, New York, 1980, pp. 63 99. [9] P.J. ORourke, W.R. Wadt, Los Alamos National Laboratory Report, LA-9423-MS, 1982. [10] J.E. Goldberg, Prediction of spray dryer performance, DPhil. Thesis, University of Oxford, 1987. [11] D.E. Oakley, R.E. Bahu, D. Reay, The aerodynamics of cocurrent spray dryers, in: M. Roques (Ed.), Proc. Sixth International Drying Symposium IDS, OP 373-OP, Unknown, Versailles, France, 1988, p. 378. [12] T.A.G. Langrish, D.E. Oakley, R.B. Keey, R.E. Bahu, C.A. Hutchinson, Time-dependent ow patterns in spray dryers, Trans. I. Chem. E. (A) 71 (1993) 355 360. [13] S.E. Papadakis, C.J. King, Air temperature and humidity proles in spray drying. 1. Features predicted by the particle sourcein-cell model. 2. Experimental measurements, Ind. Eng. Chem. Res. 27 (11) (1988) 2111-2116, 2116-2123. [14] D.E. Oakley, R.E. Bahu, Spray/gas mixing behaviour within spray dryers, in: A.S. Mujumdar, I. Filkova (Eds.), Drying, Elsevier, Amsterdam, 1991, pp. 303 313. [15] D.M. Livesley, D.E. Oakley, R.F. Gillespie, B. Elhaus, C.K. Ranpuria, T. Taylor, W. Wood, M.L. Yeoman, Development and validation of a computational model for spray-gas mixing in spray dryers, in: A.S. Mujumdar (Ed.), Proc. Eighth International Drying Symposium IDS 92, Montreal, Canada, Drying 92, Elsevier, Amsterdam, 1992, pp. 407 416. [16] B. Guo, T.A.G. Langrish, D.F. Fletcher, Time-dependent simulation of turbulent ows in axisymmetric sudden expansions, in: M.C. Thompson, K. Hourigan (Eds.), Proc. 13th Australasian Fluid Mechanics Conference, Melbourne, Australia, vol. 1, 1998, pp. 283 286. [17] D.B. Southwell, T.A.G. Langrish, Observations of ow patterns in a spray dryer, Drying Technol. 18 (3) (2000) 661 685. [18] R.A. Stafford, O. Fauroux, D.H. Glass, Flow visualisation and instantaneous velocity measurements of spray dryer gas and spray ows using particle image velocimetry, in: C. Strumillo, A.S. Mujumdar (Eds.), Proc. Tenth International Drying Symposium (IDS 96), Drying 96, Krakow, Poland, vol. A, 1996, pp. 555562. [19] F.G. Kieviet, J. Van Raaij, P.P.E.A. De Moor, P.J.A.M. Kerkhof, Measurement and modelling of the air ow pattern in a pilot-plant spray dryer, Trans. I. Chem. E. 75 (A3) (1997) 321328.

You might also like