Simulation of spray dryers using computational fluid dynamics

1.2 Scope and objectives of the thesis This dissertation addresses the operational aspects of spray drying performance with focus on the effect of various parameters on particle stick

Therefore, Computational Fluid Dynamics (CFD) can be used as a design tool or

as a design guide to compare the drying and hydrodynamic performance of chambers

of different shapes, different arrangements of inlet drying gas (including single entry

or multi-port entry), supplementary drying or cooling air inlets as well as effect of ambient air leakage due to poor sealing which is not uncommon in old spray chambers

It is too expensive to test the effect of all these parameters experimentally Recent

rapid developments in CFD and the ever-increasing computing power at decreasing cost makes it feasible to evaluate spray dryer designs without undertaking expensive experimental pilot or laboratory tests Although simulation of the complex transport phenomena that occur in a spray dryer cannot yet be modeled with high accuracy, the results are nevertheless useful to guide design and operation of spray dryers when coupled with some empirical experience Some commercial CFD codes [Dombrowski, 1993] are also available now, for example(s), PHOENIX, FLUENT, FLOW3D, FIDAP, CFX etc to study the transport phenomena in spray dryers

However, lack of carefully obtained experimental data, primarily due to the often confidential nature of the process and difficulty of making the necessary detailed measurements, is currently hampering the development of CFD-based design and analysis of spray dryers It is quite possible that perhaps the numerical predictions are almost as reliable as experimental data that can be obtained within the spray dryer chamber under operating conditions However, there are still some limitations to the CFD approach since it does not typically include reliable and validated models for quality changes, attrition or agglomeration of particles that can occur within the chamber

1.2 Scope and objectives of the thesis

This dissertation addresses the operational aspects of spray drying performance with focus on the effect of various parameters on particle stick on the wall in drying chamber, heat/mass transfer coefficient in the drying chamber, heat consumption intensity per unit evaporative rate and volumetric effectiveness, and the impact of the geometry design on the airflow and particle motion in drying chamber The goal is to

contribute to a better fundamental understanding of the drying operation to improve the current spray drying technology

This thesis aims to develop a state-of-the-art CFD-based design and analysis methodology for spray drying plants This design methodology enables the optimization of spray dryers in terms of more compact (intensified) plants, greater energy efficiency and higher product yield while maintaining product quality The project results will assist the spray dryer designer to control the drying gas flow pattern, the droplet/particle trajectory and the heat and mass transfer process so as to prevent product degrading and deposits on the chamber walls

The first objective of this project is to develop and validate a CFD-based model to predict flow patterns and overall drying performance of a conventional cylinder-on-cone spray dryer by comparing the results with published results as well as with new data supplied by collaborating researchers in Brazil and Australia The effects of operating parameters, different layouts on the drying performance and airflow patterns are studied as well

The second objective of the project is to evaluate novel spray dryer geometries that yield better volumetric effectiveness and higher heat/mass transfer performance than the conventional designs The geometries being considered are conical, hour-glass shape and lantern shape chamber The axi-symmetric CFD model is used to evaluate those designs

The third objective is to evaluate a spray dryer fitted with different atomizers, such

as, rotating disc atomizer and ultrasonic nozzle The performance is then compared with that of the normal spray dryer fitted with pressure nozzle CFD model is first used

in simulating the spray dryers fitted with rotating disc atomizer and ultrasonic nozzle

The fourth objective is to evaluate a new one-stage and two-stage, two dimensional horizontal spray dryer concept, proposed initially by Prof Mujumdar, which is expected to allow longer drying times needed for heat-sensitive products and large droplet sizes

The fifth objective is to modify the drying model and add it into the commercial code by user-defined-functions It definitely improves the prediction of the drying process in the spray drying chamber Three different drying models are developed based on the characteristic drying curve

The final objective is to evaluate an industrial scale spray dryer The necessary measurements are carried out The comparisons between prediction and measurements are carried out, as well Different pressure nozzle designs are evaluated

1.3 Outline of the thesis

In chapter 2, the literature pertaining to the key processes involved in a spray dryer, i.e., atomization, droplet and its drying, contact of droplet and drying medium, are reviewed The models developed for simulation of spray dryers are reviewed as well The CFD model used is described in chapter 3 A new drying model and newly defined parameters for describing the drying performance of spray dryer are also obtained

From chapter 4 to chapter 9, selected predicted results and experimental data are presented Chapter 4 presents the validation of a CFD model using a published data by Kieviet [1997] Then the parametric studies are carried out, e.g., effects of inlet temperature of drying medium, air leakage, heat loss from the chamber wall, operating pressure in the drying chamber, the environmental drying medium humidity etc

In chapter 5, an industrial spray drying of coffee is used as an example for modeling It seems that this is the first time such a large-scale industrial spray dryer is modeled and tested using CFD Different pressure nozzle designs are investigated Some useful suggestions to improve the performance of this spray dryer are also provided

Chapter 6 presents the predicted results of spray dryers using different atomizers except pressure nozzle, e.g., rotary disc atomizer and ultrasonic nozzle These two types of spray dryers are also being modeled using CFD for the first time, as well Chapter 7 describes the development of a new spray dryer design using CFD, i.e., a horizontal spray dryer An optimal horizontal spray dryer design is obtained using CFD simulations

Chapter 8 evaluates alternative spray dryer designs without undertaking the expensive experimental pilot or laboratory tests The four types of drying chamber, i.e., cylinder-on-cone, hour-glass, pure conical and lantern, are selected studied

Chapter 9 investigates the new drying model that can be incorporated as part of the commercial CFD code to improve the performance of the CFD model already exists in the software

Finally, the conclusions of this study are given in chapter 10 The recommendations for future work in this field are discussed in chapter 11

Hence, this study is believed to present some novel approaches to modeling spray dryers and to the understanding of the performance of a spray dryer as well Due to space limitations all the results obtained during the course of this study could not be accommodated in this thesis Additional results are, however, available in a number of publications

Chapter 2 Literature review

2.1 Fundamentals of spray drying

Spray drying is a suspended particle processing (SPP) technique that utilizes liquid atomization to create droplets which are dried into to individual particles while moving

in a hot gaseous drying medium, usually air [Masters, 1991] Over 20,000 spray dryers are estimated to be presently in use commercially around the world to dry products from agro-chemical, biotechnology products, fine and heavy chemicals, dairy products, dyestuffs, mineral concentrates to pharmaceuticals in sizes ranging from a few kg per hour to 50 tons/h evaporation capacity [Mujumdar, 2000] Liquid feedstocks, e.g., solutions, suspensions or emulsion, can be converted into powder, granular and agglomerate form in an one-step operation in the spray dryer [Masters,

1991 & 2002; Filkova & Mujumdar, 1995; Huang and Mujumdar, 2003]

Among the advantages of spray dryers, one may cite the following It can:

• Handle heat sensitive, non-heat sensitive as well as heat-resistant pumpable fluids as feedstocks from which a powder is produced

• Produce dry material of controllable particle size, shape, form, moisture content and other specific properties irrespective of dryer capacity and heat sensitivity

• Provide continuous operation adaptable to both conventional and computer (PLC) control

• Provide extensive flexibility in its design, such as, drying of organic solvent-based feedstocks without explosion and fire risk; drying of aqueous feedstocks (where the resulting powders exhibit potentially explosive properties as a powder cloud in air); drying of toxic materials; drying of feedstocks that require handling in aseptic/hygienic

drying conditions; drying of feedstocks to granular, agglomerated and agglomerated products

non-• Handle a wide range of production rates, i.e., almost any individual capacity requirement can be designed by spray dryers

However, they also suffer from some limitations, such as:

• High installation costs; high electrical requirement

• A lower thermal efficiency

• Product deposits on the drying chamber walls may lead to degraded product or even fire hazard

Examples of spray-dried products on industrial scale include the following:

• Chemical industry, e.g., Phenol-formaldehyde resin; catalysts; PVC emulsion-type; Amino acid etc [Kim et al., 2001]

• Ceramic industry, e.g., Aluminium oxide; carbides, Iron oxide; Kaolin etc [Yokota

et al., 2001]

• Dyestuffs and pigments, e.g., Chrome yellow; food colour; titanium dioxide; paint pigments etc

• Fertilizers, e.g., Nitrates; Ammonium salts; phosphates etc

• Detergent and surface-active agents, e.g., detergent enzymes; bleach powder; emulsifying agents etc

• Food industry, e.g., milk; whey; egg, soya protein etc [Furuta et al., 1994]

• Fruit and vegetables, e.g., banana; tomato; coconut milk etc [Huang and Mujumdar, 2003a and 2003b]

• Carbohydrates, e.g., glucose; total sugar; maltodextrine etc [Watanabe et al., 2002]

• Beverage, e.g., coffee, tea etc

• Pharmaceuticals, e.g., penicillin, blood products, enzymes, vaccines etc [Newton et al., 1966; Broadhead et al., 1994]

• Bio-chemical industry, e.g., algae; fodder antibiotic; yeast extracts; enzymes etc [Huang et al, 2001]

• Environmental pollution control, e.g., flue gas desulfurization; black liquor from paper-making; etc [Hill et al., 2000; Huang et al, 2001]

• Other new applications, e.g., nano-materials, spray freeze drying etc [Li et al., 1999; Lo et al., 1996; Maa et al., 1999]

A normal spray drying process usually consists of the following four stages:

1 Atomization of feed into droplets: The liquid feed which can be solutions, suspensions, emulsions or slurries is atomized to a spray that consists of fine droplets Ordinarily there are two different atomization methods: Rotary disc or centrifugal atomizers and nozzle atomizers including pneumatic nozzle and pressure nozzle

2 Heating of hot drying medium: The drying medium, such as air, is heated by steam, electricity, oil combustion or coal combustion etc Then it will be sent into the drying chamber through the hot air dispenser

3 Spray-air contact and drying of droplets: The liquid spray is mixed with the hot drying medium in the drying chamber Then the volatile is evaporated into drying medium and dried into particles or powder

4 Product recovery and final air treatment: Separation of the dried particles from the exhaust drying air The cleaned air will be drawn into the atmosphere or recycled

A typical spray drying process flowsheet is shown in Figure 2.1 Although the design, operation mode, handling of feedstock and product requirements are diverse, each stage must be carried out in all spray dryers The formation of a spray and effective contact of the spray with air are the characteristic features of spray drying

This study mostly focuses on these two parts since the other two parts only satisfy the process requirements, e.g., air inlet temperature and mass flow rate through the drying medium heating system and the product recovery efficiency by the collecting system

Figure 2.1 Typical spray dryer layout

2.2 Atomization

Since the choice of the atomizer is very crucial, it is important to note the key advantages and limitations of different atomizers (centrifugal, pressure and pneumatic atomizers) Other atomizers can also be used in spray dryers, such as, the ultrasonic nozzle [Bittern & Kissel, 1999], but they are expensive and have low capacity Although different atomizers can be used to dry the same feedstock, the final product properties (bulk density, particle size, flowability etc.) could be quite different and hence proper selection is necessary

2.2.1 Mean droplet diameter

The mean droplet diameter is the first important parameter to characterize the distribution of a spray There are several measures to define it, e.g., number median droplet diameter, d0.5, number mean diameter, d10, the surface mean diameter, d20, the volume mean diameter, d30, and the Sauter mean diameter, d32 The number median droplet diameter, d0.5, is defined as that diameter, in a sample of droplets, for which half of the droplets are smaller than d0.5 and the other half larger than d0.5 The general expression for the other four mean diameters is presented as follows:

) ( 1

N

i

a i i

ab

d n

d n

∑

∑

= (2.1)

where di represents a drop of diameter di, ni the number of droplets with diameter di,

and N is the sample number of droplets (∑

i i

n ), and a and b are integers designating

the specific type of mean For example, the number mean diameter is given by

N

d n d

N

i i i

∑

=

10 (2.2) The surface mean diameter, d20, is expressed by

N

i i i

∑

= (2.3)

The volume mean diameter, d30, is given by

3 3

N

d n d

N

i i i

∑

= (2.4)

The Sauter mean diameter, d32, is probably the most widely used of the various mean diameter in spray drying It takes the following form:

∑

= N

i i i

i i i

d n

d n d

2.2.2 Droplet size distribution

One of the most important topics in atomization studies is the distribution of droplet size in the spray It is an essential element to understand the atomization process Some of the most widely used droplet size distribution functions in spray drying are discussed below

2.2.2.1 The log-normal distribution function

The log-normal distribution function describes the distribution of diameters which follows a normal or Gaussian curve, with the horizontal axis being the range of droplet diameters in a logarithmic rather than a linear scale The distribution function can be expressed mathematically in the form

)}

2/(

)]

([logexp{

2

1)

πσ

η

m p

p

d d

d = − (2.6)

where, σ = the standard deviation and d m= the mean droplet size (µm ) This distribution presents a straight line on a plot of size d against the number percent oversize on a log probability graph

2.2.2.2 The Nukiyama-Tanasawa distribution function

The Nukiyama-Tanasawa distribution function is mathematically given by

)exp(

)

p p

d =

η (2.7)

where, B and C are constants, and q = the dispersion coefficient This function has

been empirically derived for the droplet size distribution from pneumatic nozzle [Masters, 1991]

2.2.2.3 The Rosin-Rammler volume distribution function

The Rosin-Rammler volume distribution function is widely used to express the droplet size distribution from nozzle, especially for pressure nozzle It is empirical, and relates the volume percent oversize to droplet diameter Its mathematical form is given

100ln( =

d = the Rosin-Rammler mean droplet diameter which is the droplet

diameter above which lies 36.8% of the entire spray volume [Masters, 1991]

2.2.3 Rotating disk atomizer

The rotating disk atomizer is sometimes called a rotary wheel or centrifugal atomizer The liquid is continuously accelerated to the disk rim by the centrifugal force produced by the disk rotation Thus the liquid is spread over the disk internal surface and discharged horizontally at a high speed from the periphery of the disk The droplet size distribution depends on the rotating speed, feed physical properties and feed spray

rate Since the centrifugal force can be adjusted within a wide range, this type of atomizer is extremely versatile and can handle a wide range of liquids Its advantages are summarized below:

• Handles large feed rates with a single wheel or disk

• Suited for abrasive feeds with proper design

• Has negligible clogging tendency

• Change of wheel rotary speed to control the particle size distribution

• More flexible capacity (but with changes powder properties)

In spite of intensive investigations into the mechanism of atomization from the rotating disk, the prediction of spray characteristics still remains uncertain [Westergaad, 1994] The main factors that control the spray characteristics are liquid feed rate; peripheral rotating speed; viscosity of the liquid and disk design

The tangential velocity is written as follows:

601000

2 2 2

)(

0024.0

n h

Q D

N o

l

πρ

=

u (2.10)

where, Q= feed rate (m3/h), µ =feed viscosity (cP), n= number of vanes of holes, and h= height of vanes or diameter of holes Thus, the total velocity for droplets from the periphery of the disk is obtained by

r t

u (2.11) and the angle of the droplet release from the disk edge is

The most quoted correlations about the mean droplet size by the rotating disk atomizer are summarized in Table 2.1

Table 2.1 Correlations for prediction of mean droplet diameter using rotating disk

atomizer [Masters, 1991]

1 0 0

2 0 5

0 6 0 4

nh D

M N

d

o

L l

σµ

0 6

0 2

5

p l

p l

p

M

nh M

Nr

M r

24 0 4 32

)()

(

)(10

4

1

nh ND

M d

2.2.4 Pressure nozzle

In a pressure nozzle, the liquid is forced by pressure through a small orifice The pressure energy within the liquid is converted within the liquid into kinetic energy and produces a thin moving liquid sheet [Filkova and Cedik, 1984] The thickness of the film is around 1 to 5 microns [Fraser et al., 1957] This thin film is dispersed very

rapidly into droplets due to the large friction between the thin liquid sheet and surrounding air But the spray characteristic is also determined by the properties of the liquid such as viscosity, surface tension, density and spray quantity per unit time, and

by the medium into which the liquid is sprayed Its advantages are as follows:

• Simple, compact and cheap

• No moving parts

• Low energy consumption

Limitations:

• Low capacity (feed rate for single nozzle)

• High tendency to clog

• Erosion can change spray characteristics

A pressure nozzle is composed of several parts, e.g., nozzle adapter, nozzle holder, whirl chamber, orifice plate and nozzle body etc., assembled into a stainless steel body The various components of a pressure nozzle are shown in Figure 2.2

Figure 2.2 Components of a pressure nozzle

Keey [1991] and Masters [1991] introduced the following correlations to predict the mean droplet diameter produced by pressure nozzle for the commercial conditions

5 0

25 0 32

2.2.5 Two-fluid nozzle (Pneumatic nozzle)

The mechanism of pneumatic nozzle atomization is one of high-velocity gas creating high frictional forces over liquid surfaces causing liquid disintegration into spray droplets [Masters, 2002] The schematic diagram of this type of nozzle is shown

in Figure 2.3 The important design parameter for this nozzle type is the mass ratio of gas flow rate to the liquid spray rate Its advantages are summarized as follows:

• Simple, compact and cheap

• No moving parts

• Handle the feedstocks with high-viscosity

• Produce products with very small size particle

Figure 2.3 Two-fluid nozzle

Limitations:

• High energy consumption

• Low capacity (feed rate)

• High tendency to clog

The mean spray size produced by pneumatic nozzle atomization follows the relation [Masters, 1991]

β α

liq gas

a

M B u

A

d (2.17)

where, the exponents α and β are function of nozzle design, and A and B are constants involving nozzle design and liquid properties, urel is the relative velocity between gas and liquid (m/s) Mgas and Mliq are mass flow rate of compressed air and feed, respectively [Masters, 1991]

2.2.6 Ultrasonic atomizer

The use of ultrasound for atomization has diversified since its discovery [Berger et al., 1998] in the 1920s In ultrasonic atomization, a liquid is subjected to sufficiently high intensity of ultrasonic field that splits the liquid into droplets, and then the droplets are ejected from the liquid-ultrasonic source interface into the surrounding air

as a fine spray [Morgen, 1993; Rajan and Pandit, 2001] A sketch of this atomizer is shown in Figure 2.4

Figure 2.4 Structure of an ultrasonic nozzle [Berger, et al., 1998]

Since the frequency of ultrasonic vibration determines the droplet size (aside from other parameters) in ultrasonic atomization, very small droplets (1 to 5microns) can be produced by increasing the frequency Although the ultrasonic atomizer is still a newcomer to atomization technology, it is already applicable in many processes, e.g., spray drying, spray freeze drying [sonner, 2002], spray combustion etc Its main advantages are

• Simple and compact

• No moving parts

• Produce products with narrow size particle distribution and low initial velocity

Limitations of the ultrasonic atomizer include:

• Cost

• Low capacity (feed rate)

There are two hypotheses to explain the mechanism of the liquid disintegration during ultrasonic atomization, i.e., cavitation and capillary wave hypotheses In caviation hypothesis, the liquid is sonicated and cavitation bubbles are formed When the bubbles are impulsively collapsed, high intensity shocks are generated These shockes disintegrate the liquid film In capillary hypothesis, the atomization occurs predominantly from the effects of the surface capillary waves produced in a thin film

of liquid as it wets an oscillating surface The most widely used ultrasonic atomizer is the capillary wave atomizer [Zinovev and Margulis, 1993] These atomizers operate with low frequency, i.e., below 100HZ, and require an amplitude in excess of 3mm, corresponding to a power of 10-100W [Topp, 1970]

Hunter [1969] reported that the ultrasonic atomizer gave a fountain spray shape Topp and Eisenklam [1972] listed the effects of different frequencies, but no information about the effects of liquid viscosity, density and flow rate was provided

Rajan and Pandit [2001] assessed the impact of various physico-chemical properties of liquid, its flow rate, the ampliture and frequency of ultrasonic and the area and geometry of the vibrating surface on the droplet size distribution A correlation was proposed to predict the droplet size formed using an ultrasonic atomizer taking into consideration the effect of liquid flow rate and viscosity The correlation is given as follows:

consta

d p = nt(f)−0 66(M L)0 207(σ)0 11(ρl)−0 274(µ)0 166(power/area)−0 4 (2.18) The droplet size distribution from an ultrasonic nozzle follows a log-normal distribution [Berger, 1998] Sonner [2002] used ultrasonic atomizer to spray the protein The fountain shape was also found in its experiments The droplet size distribution in its experiment was shown in Figure 2.5

Figure 2.5 Droplet size distribution measured by laser diffraction for a) nebulized pure water, and b) a lysozyme/trehalose/mannitol/PVP Liquid feed rate = 2 ml/min (P1 peristaltic pump) [Sonner, 2002]

2.3 Contact between droplets and drying medium

The contact between the spray and hot air determines the evaporation rate of volatiles in a droplet, the droplet trajectory, droplet residence time in the drying chamber, as well as potential deposits on the chamber walls It also influences the morphology of particle and product quality So, apart from the selection of atomizers,

the drying chamber geometry and the air disperser selection are other important factors

to consider in spray drying They determine the air flow pattern in the drying chamber Based on the type of interaction between air and droplets, there are three alternative patterns which are possible, viz

(a) Co-current flow (Figure 2.6a)

(b) Counter-current flow (Figure 2.6b)

(c) Mixed-flow (Figure 2.6c)

These arrangements are shown in Figure 2.6

(a) Co-current flow (b) Counter-current flow (c) Mixed-flow

Figure 2.6 Contact between droplets and air

For the co-current flow pattern shown in Figure 2.6a, the droplet and gas pass through the drying chamber in the same direction Such an arrangement is applied in industries, especially for drying of heat-sensitive materials [Masters, 2002; Huang and Mujumdar, 2003], such as, coffee, milk and pharmaceuticals etc On the other hand, if the droplet and gas pass through the drying chamber in opposite directions, it is termed counter-current flow pattern, as shown in Figure 2.6b This arrangement is not suitable for drying heat sensitive materials as the driest product would meet the hottest gas

Finally, if the contact of the droplet and gas is a combination of co-current and counter-current flow, it is termed as mixed-flow pattern, as shown in Figure 2.6c This arrangement is usually used in the production of agglomerated powders or coarse products However, it is also not suitable for drying of heat sensitive materials There are three types of air dispersers that are used in industrial spray dryers, e.g., top air disperser, central air disperser and middle-side entry disperser Central air disperser cannot be used for nozzle spray dryers These three air dispersers combined with the different location of atomizers lead to different spray and air contacts, i.e., co-current, counter-current and mixed flow For example, when the centrifugal atomizer is installed at the top of drying chamber and the top air disperser is used, it becomes a co-current air-spray dryer

Several authors [e.g., Katta and Gauvin, 1975; Gauvin, 1981; Huang, Kumar and Mujumdar, 2003] have worked in this area, but the amount of published data on spray-air contact is still limited and is mainly applicable to small-scale spray dryers Experimental measurements are difficult due to the complex and hostile environments

in the drying chamber during the spray drying running

2.4 Drying of the droplets

2.4.1 Experiments and models for droplet drying

Evaporation and drying of droplets occurs as soon as the spray emerging from the atomizer contacts the drying medium The drying history of a droplet can be followed

by a drying rate curve of the type shown in Figure 2.7 It can be separated into two periods, i.e., constant drying rate period and falling drying rate period Performing a dimensionless analysis would results in the following correlations

Pr)(Re,

f

Nu = (2.19a)

(Re, Sc

f

Sh= (2.19b) They can also be written as

33 0 5

Re0

Re0

Sh= +ψ (2.20b) Ranz and Marshall [1952a and 1952b] suggested that ψ =0.6 When the droplet evaporation is in a still air condition, the Reynolds number is equal to 0 Then the above equations are reduced to

Nu = Sh=2.0 (2.21)

Figure 2.7 Drying rate curve for the drying history of a droplet

Most of the models developed up to date for drying of a single droplet are based on one main assumption, i.e., the absence of temperature gradient within the droplet Wijlhuizen et al [1979], Sano et al [1982] and Stevenson et al [1998] assumed a uniform temperature in the droplets and the water to diffuse through the wet solid and evaporate at the surface of the droplet Nesic [1989] and Nenic et al [1990] assumed that a crust forms on the surface of droplet and a receding crust-bulk interface is

formed A uniform temperature is also assumed The necessary experiments were also carried out by them

Cheong et al [1986] used a receding interface model with a droplet core temperature different from that of the surface However, the core temperature was not linked to the air wet-bulb temperature Chen and Xie [1996] describe the model of drying of a single droplet of milk But they considered neither temperature nor water concentration distributions in the droplet Later, Chen et al [1999] modified the model

by incorporating the mass transfer resistance of the crust

Farid [2004] developed a new model to simulate droplet drying assuming either temperature or water concentration variations in the droplet Langrish et al [2001] investigated the drying curve for milk powder using a computational fluid dynamics model

Adhikari et al [2001, 2003] and Tan [2004] carried out several experiments of single droplet drying in which he measured the droplet mass changes by suspending a droplet under a balance Different substances were tested

2.4.2 Morphology of spray dried products

Most of the early important phenomenological studies on particle morphology were carried out by simply collecting dried powders directly from the spray dryer and examining them under a microscope [Charlesworth and Marshall, 1960] Marshall and Seltzer [1950a and 1950b] briefly describe the various particle morphologies produced

by spray drying in relation to feed concentration and the methods of atomization

More recently, Walton [1994, 1999a and 1999b] investigated the samples obtained from various industrial and pilot plant spray dried materials Particle morphology of each powder sample was examined using both optical and Scanning Electron

Microscopy Three distinct morphological types were identified, i.e., agglomerate, skin-forming and crystalline

2.5 Mathematical models of spray drying

The objective of mathematical modeling of spray dryer is to predict the droplet/particle movement and the evaporation/drying of droplets in a spray dryer Two kinds of numerical models, i.e., one-way coupling and two-way coupling, have appeared in the literature [Crowe, 1977 and 1980] In the one-way coupling model, it was assumed that the condition of drying medium was not affected by the spray or evaporating droplets, although the droplet/particle characteristics change due to the evaporation and drying process In order to improve the model to simulate spray drying which take into account the heat and mass transfer between spray and drying medium, the two-way coupling approach was developed This model considered the interaction, e.g., heat, mass and momentum transfer, between the two phases, i.e., droplets and drying medium Arnason and Crowe [1986] summarized them as shown

in Figure 2.8

Figure 2.8 Two –way coupling between discrete and continuum phases

On the other hand, the models also can be categorized in terms of the geometry, i.e., one-dimensional (1D), two-dimensional (2D) and three-dimensional (3D)

2.5.1 One-way coupling

Seltzer and Marshall [1950a and 1950b]] developed a set of equations to compute the evaporation of pure liquid in a heated drying medium A constant temperature difference between droplets and drying medium and no relative velocity are assumed Duffie and Marshall [1953] extended the study of Seltzer and Marshall They developed the governing equations for the drying time of pure liquid droplet under Reynolds number, 0.2-500 The Ranz-Marshall correlation was used in their model Kerkhof and Schoeber [1974] developed the equations of heat and mass transfer to compute water spray traveling in a static air flow They found that their equations were valid for droplets (smaller than 100 microns) Later, the deceleration time for droplets sprayed from an injection port into a hot drying medium was computed by Masters [1976]

Janda [1973] developed a method to calculate the required diameter of a spray dryer with forced air circulation for drying particles of a given size The calculation of the diameter of the chamber is based on the required period of drying of a particle of given dimensions and of the radial course of its trajectory Values obtained by calculations are compared to experimental data Later, Janda [1978] also developed another method to calculate of the minimum required height of a drying tower using a pressure nozzle No comparison with experimental measurements was reported In order to calculate the drying time of the droplets, he assumed that drying of particles

up to 300 microns in diameter occurred entirely within the constant drying rate period

Although many models for simulation of spray drying have been developed, most have a deficiency in neglecting the thermal coupling

2.5.2 Two-way coupling

Herring and Marshall [1955] developed a procedure for simulating the effects of droplet on spray dryer performance The analysis was carried out using a step-wise procedure while the droplet size distribution was varied into very small size increments Each step used an average diameter to represent the droplets Then each size of the droplet will change due to evaporation over a small time increment For the next step, a new droplet size distribution was used Diskinson and Marshall [1968] extended the work [Herring et al., 1955] by including an energy term to account for the change in temperature of humid air due to droplet evaporation Although the one-dimensional models with two-way thermal coupling represent the essential mechanisms in spray drying better, they are inadequate near the atomizer

Yuan et al [1985] developed a one-dimensional model for simulation of a current spray dryer which took into account the droplet size distribution by means of a volume frequency function

In 1975, Parti and Palanz [1974] developed a general numerical model to simulate

a co-current and counter-current spray dryer However, they assumed the uniform droplet size and velocity only in the vertical direction Keey et al [1980] developed a program to evaluate the spray dryer using two-way coupling method, but with an assumption of uniform droplet size distribution as well

Katta and Gauvin [1976] and Gauvin et al [1975] developed a dimensional model to study a co-current spray dryer They separated the drying chamber into a nozzle zone and a free-entrainment zone Their model was verified by

quasi-one-making comparison with the experimental observations in a 1.22m diameter co-current spray dryer

In conclusion, it is apparent from the literature earlier in this chapter that there is still need for a model which incorporates complete coupling and accounts for property variations in axial, radial and tangential directions for both the discrete and the continuum phases [Blatas and Gauvin, 1969a and 1969b] Fortunately, recent rapid developments in Computational Fluid Dynamics (CFD) and ever-increasing computing power at decreasing cost make it feasible to evaluate the spray dryer problem With the advent of digital computer technology, it is now feasible to consider three dimensional gas-droplet flows with heat and mass transfer

2.5.3 CFD model for spray dryer

Crowe et al [1977] and Crowe [1980] proposed an axi-symmetric spray drying model called Particle-Source-In-Cell model (PSI-Cell model) This model includes two-way mass, momentum, and thermal coupling In this model, the gas phase is regarded as a continuum (Eulerian approach) and is described by pressure, velocity, and temperature and humidity fields The droplets or particles are treated as a discrete phase which are characterized by velocity, temperature, composition, and the size along trajectories (Lagrangian approach) It incorporates a finite difference scheme for both the continuum and discrete phases More details can be found in the work by Crowe et al [1998]

Goldberg [1987] predicted the trajectories of typical small, medium and large droplets of water in a spray dryer with a 0.76m diameter chamber with 1.44m height using a CFD (FLOW3D) program

Papadakis and King [1988a and 1988b] used this PSI-Cell model to simulate a spray dryer and compare their predicted results with the limited experimental results

However, only a lab spray dryer is used to measure the results They have shown that the measured air temperatures at various levels below the roof of a spray drying chamber are well predicted by the CFD model Negiz et al [1995] developed a program to simulate a co-current spray dryer based on the PSI-Cell model, as well Straatsma et al [1999] developed a drying model, named NIZO-DrySim, to simulate aspects of drying processes in the food industry It can simulate the gas flow in a two-dimensional spray dryer chamber and calculate the particle trajectories

Livesley et al [1992] and Oakley et al [1990] found that numerical simulations using the k−ε turbulence model are useful for simulating the measured particle sizes and mean axial velocities in the industrial spray dryers

Oakley and Bahu [1993] reported a three-dimensional simulation using the CFD code FLOW3D which is an implementation of the PSI-Cell model They proposed that additional research needs to be done to verify the performance of their model But in the open literature, most of the studies were carried out in small scale spray dryers For example, Oakley [1994] carried out an experiment in a 0.453m3 spray dryer and Langrish and Zbicinski [1994] in a 0.779m3 spray dryer

Kieviet et al [1995, 1996 and 1997] carried out the measurement of air flow patterns and temperature profiles in a co-current pilot spray dryer (diameter 2.2m) A CFD package (FLOW3D) was used to model such a spray dryer An industrial application of a CFD program to spray dryer design was reported by Masters [1994] Southwell and Langrish [2000], Harvie and Langrish et al [2001 and 2002] and Lebarbier et al [2001] also carried out the CFD simulation of typical spray dryers with co-current and counter-current flow using commercial code (CFX) Necessary comparison with the limit experimental data was made

Frydman et al [1998 and 1999] and Ducept et al [2001 and 2002] used the commercial code, i.e., FLUENT, to simulate a spray drying using superheated steam as drying medium However, in their model, the elevation of the boiling point for suspension was not considered Wu et al [2001] used the same code to simulate a spray drying with the pulsating flow pattern Reinhold et al [2001] investigated the spray drying with simultaneous chemical reaction

Cukaloz et al [1997] studied a horizontal spray dryer to dry α - amylase However, it is observed that the flow pattern in Cakaloz et al design is not optimal for spray drying since the main air inlet is located at a corner of the chamber This arrangement makes the spray more likely to hit the top wall unless designed carefully Verdurmen et al [2004] proposed an agglomeration model to be included into the CFD model to predict agglomeration process in a spray dryer However, this is still under investigation

Chapter 3 Computational Fluid Dynamic Model

This section gives a brief overview of the CFD solver used in this study which is the FLUENT software from FLUENT Inc For further detailed information, which is not covered in the following sections, the interested reader is referred to the software User’s Guide [FLUENT, 2004]

The FLUENT 6.1 CFD package consists basically of two components: (1) GAMBIT 2.0, the pre-processor and (2) FLUENT 6.1, the solver The pre-processor

“GAMBIT” was used to create the model of the actual fluid-laden geometry in two dimensions (2D) or three dimensions (3D) Using this pre-processor, the geometry was subdivided in a grid (or mesh) consisting of cells which form the calculation domain for the subsequent flow analysis The grids can either be structured where each cell is part of a rectangular block structure and follow the Cartesian co-ordinate system or it could be totally unstructured The cells may be of quadrilateral or triangular shape for 2D-meshes Three dimensional meshes may consist of hexahedral, tetrahedral, wedge-shape elements or pyramids or combinations of them

Furthermore, GAMBIT allows examining the quality of the mesh by assessing the various aspects of cell quality (e.g the aspect ratio of a quadrilateral cell) In general, to obtain a mesh-independent solution, the mesh resolution needs to be fine For the 2D-calculations, the mesh was therefore refined until the solution became mesh-independent This could also be done to a certain extent with 3D-meshes The

meshes also can be refined locally by grid adaptation, where necessary After

completing the pre-processing, the mesh is then exported to the solver FLUENT 6.1

The solver FLUENT 6.1 is used to select the appropriate physical models specifying the current problem For example, (1) the flow regime: laminar/turbulent conditions; (2) the problem type: steady state/transient conditions; (3) the fluid properties: incompressible/compressible; (4) the detailed problem specification such as boundary conditions, materials and injectors

In addition to this, a solver formulation and solution parameters have to be chosen Finally, an initial solution for all cells must be provided The underlying governing equations will then be solved in an iterative process for each individual cell

of the calculation domain

The post-processing, i.e the graphical illustration of the results, can be done by FLUENT 6.1

In this study, all geometry and meshing works were carried out with GAMBIT 2.0 All calculation works have been carried out with FLUENT 6.1 on UNIX platform

3.1 Governing equations for the continuous phase

For any fluid, its flow must obey the conservation of mass and momentum These conservation equations can be found in standard fluid dynamic literature (for incompressible gas) [Bird, Stewart and Lightfoot, 1960; Ferziger & Peric, 1999]

The general form of the continuity equation for mass conservation is

j i

i j i

j i i

M g u

u x

u x

u x

x

P x

u u t

∂

∂+

∂

∂

∂

∂+

µρ

ρ

])

([)

increase of momentum per unit volume and the second term is the momentum increase/decrease per unit volume due to convection The first term on the right-hand side of equation (3.2) is the pressure force on a fluid element per unit volume, the second term is the viscous force on a fluid element per unit volume and the third term

is the gravitational force on a fluid element per unit volume The last term is the momentum source term

x

u x

u x

u

ij i

j fori

i i i

i p p

M T u x

T k x x

T u c t

T c

∂

∂

][

)(

)

ρρ

r r

u

∂

∂+

∂

∂

)(

1)( ρ ρ (3.4)

* Momentum equation

Axial momentum:

])[(

1])[(

)(

r x

u x

uv r r r

u

∂+

∂

∂+

∂

∂+

u T

L T

x

v r r

r r

u

∂

∂+

∂

∂+

∂

∂+

∂

∂

])[(

1])

[(µ µ µ µ (3.5a)

Radial momentum:

])[(

1])[(

)(

r x

v x

v r r r

uv

∂+

∂

∂+

∂

∂+

v g

T L T

L T

r

w r

v r

v r r

r r

u

∂

∂+

∂

∂+

∂

∂+

∂

2

)(

2])[(

1])

µµµ

Tangential momentum:

])()[(

1])()[(

)(

1)

(

r

rw r r

r x

rw x

vrw r r r

urw

∂+

∂

∂+

∂

∂+

∂

w T

r x

q x

vq r r r

uq

T L h

T L g

g

∂

∂+

∂

∂+

∂

∂+

∂

∂

σ

µµσ

µµρ

where, u,v,w are the average axial velocity, radial velocity and tangential velocity of gas, respectively; q is the enthalpy of gas; Mh is the rate of heat transfer between the droplets and the gas Also, are the rate of momentum between droplets and gas in the axial, radial and tangential directions, respectively the laminar viscosity

of the fluid is

w v

M , ,

L

µ and the turbulent viscosity is described by µT

3.2 Governing equations for the particle

Based on the solution obtained for the flow field of the continuous phase, using an Euler-Lagrangian approach we can obtain the particle trajectories by solving the force balance for the particles taking into account the discrete phase inertia, aerodynamic drag, gravity g i and further optional user-defined forces F xi

g

g i pi i p

p D pi

F g

u u d

C dt

du

+

−+

−

=

ρ

ρρρ

µ

)(

24

Re18

2 (3.7)

with particle velocity u and fluid velocity u in direction, particle density pi i i ρp, gas density ρg, particle diameter d p and relative Reynolds number

a a a

C D = + + (3.7b) where a1, a2 and a3 are constants [Fluent, 2004]

Two-way coupling allows for interaction between both phases by including the effects of the particulate phase on the fluid phase Further, the particles are assumed to

be fully dispersed, i.e they are not interacting with each other

The particle trajectory is updated in fixed intervals (so-called length scales) along the particle path Additionally, the particle trajectory is updated each time the particle enters a neighboring cell FLUENT in general interpolates the gas velocity to the particle position assuming linear interpolation

3.3 Turbulence models

3.3.1 Governing equations for different turbulence Models

In turbulent flows, the instantaneous velocity component is the sum of a

time-averaged (mean) value

u (3.9) These fluctuations need to be accounted for in the above illustrated NAVIE-STOKES equation (3.10)

)(

)]

3

2(

[)

()

j i j l

l ij i

j

j i

j i j

i j

x x

u x

u u

u x

x

P u

u x

u

∂

∂+

∂

∂

∂

∂+

Compared with equation (3.2), equation (3.10) contains an additional term

− , the so-called Reynolds stress which represents the effect of turbulence and

must be modeled by the CFD code Limited computational resources restrict the direct simulation of these fluctuations, at least for the moment Therefore, the transport equations are commonly modified to account for the averaged fluctuating velocity components Two commonly applied turbulence modeling approaches have been used

in the present studies: k−ε model [Launder et al., 1972 and 1974]/a RNG k−ε

model [Yakhot and Orszag, 1986; Choudhury, 1993] and a Reynolds stress model (RSM)[ Launder, Reece and Rodi, 1975; Gibson and Launder, 1978; Launder, 1989] The standard k−ε model focuses on mechanisms that affect the turbulent kinetic energy Robustness, economy, and reasonable accuracy over a wide range of turbulent flows explain its popularity in industrial flow and heat transfer simulations Two variants of this model are available in FLUENT: the RNG k−ε model and the realizable k−ε model [Shih, Liou, Shabbir and Zhu, 1995] The RNG k−ε model was derived using a rigorous statistical technique (called Re-Normalisation Group theory) It is similar in form to the standard k−ε model, but the effect of swirl on turbulence is included in the RNG mode enhancing accuracy for swirling flows The realizable k−ε model contains a new formulation for the turbulent viscosity, and a new transport equation for the dissipation rate is derived from an exact equation for the transport of the mean-square vorticity fluctuation [FLUENT, 2004]

The standard k−ε model and its variants (k−ε RNG model etc.) have become popular turbulence models for calculating typical engineering problems with CFD These models apply the so-called Boussinesq hypothesis [Hinze, 1975] to describe the Reynolds stresses as a function of the velocity gradients and the turbulent viscosity µt

as follows:

t ij

i j

j

i t j

i

x

u k

x

u x

u u

3

2)(

1

1 '

'

∂

∂+

−

∂

∂+

ερ

k M b

k j k t

j i

i

S Y G

G x

k x

ρ

])[(

)(

ε ε

ε

ερε

εσ

µµρε

ρε

S k C G C G k

C x x

∂

∂+

])[(

)(

)

(

(3.14)

The turbulence viscosity,µt, is computed using equation (3.12)

The model constants values, i.e., C1ε,C2ε,Cµ,σk andσε, are set to the following values [Launder and Spalding, 1972]:

3.1,0.1,09.0,

92.1,

k j eff k j i

i

S Y G

G x

k x

ε ε

αρε

ρε

S k C G C G k

C x x

][

)(

)

(

(3.16)

The turbulence kinetic energy, , and its dissipation rate,k ε , for Realizable k−ε

turbulence model are

k M b

k j k t

j i

i

S Y G

G x

k x

ρ

])[(

)(

ερ

ρ

εσ

µµρε

ρε

S G C k

C k

C S C x

x x

u

t

j i

+

−+

∂

∂+

2 2 1

])[(

)()

(

(3.18) The governing equations for the RSM model are as follows [FLUENT, 2002]

user jkm

m i ikm m j k

k j

k i

i j

j

i i

j j i k

i k j k

j k

i

j i k k j ik i kj k

j i k j

i k k j

i

S u

u u

u

x

u x

u x

u x

u p u g u g x

u u u x

u u

u

u u x x u u

p u u u x u

u u x u

u

t

++

∂

∂++

−

∂

∂+

∂

∂

)(

2

2)(

)(

)(

)]([

](

[)

()

(

' ' '

'

' ' '

' '

' '

' '

'

' ' '

' '

' ' '

' '

'

εε

ρ

µθ

θρβρ

µδ

δρ

ρρ

(3.19)

3.3.2 Effect of turbulence on particle motion

Turbulent dispersion of particles can be modeled using either a stochastic particle approach or a "cloud" representation of a group of particles about a mean trajectory In the current work, the stochastic tracking approach is used

In Fluent, the Discrete Random Walk (DRW) model is applied to account for the impact of turbulent fluctuations on the particle motion Particle trajectories are hereby predicted by integrating the trajectory equations for individual particles based on the instantaneous fluid velocity, '

i i

u = + , along the particle path The repeated calculations of a single particle trajectory will show different particle trajectories

However, with sufficient number of calculations (the so-called number of tries in

FLUENT), a representative turbulent dispersion of the particles can be obtained To compute a particle trajectory, the random fluctuating velocity component (obtained from a GAUSS distribution) has to be kept constant for a certain time interval This time interval, the so-called Lagrangian integral time, can be approximated with

when the RSM approach is applied

3.4 Heat and mass transfer models

In general, there are two drying rate periods, i.e., constant drying rate period (CDRP) and falling drying rate period (FDRP) during droplet drying CDRP is controlled by mass transfer between the drying medium and the droplet But FDRP is controlled by the mass diffusion within the droplets/particles In the following section, the built-in models available in FLUENT will be described and followed by models developed in this study and incorporated into FLUENT

3.4.1 Built-in heat and mass transfer models

Because the heat and mass transfer between droplet and drying medium is very complex, several heat and mass transfer relationships are employed in this thesis Some changes were made as needed using the User Defined Function (UDF) option While the particle temperature Tp is less than the vaporization temperature Tvap(defined as the temperature at which the droplet/particle will start to evaporate),

T p <T vap (3.23) the droplet is only heated and no evaporation occurs The heat balance is as follows

m = − (3.24)

where m p is mass of the particle (kg); c is heat capacity of the particle (J/kg.K); A p

g

p is the surface area of the particle (m2); T is the local temperature of the hot medium (K);

h is the convective heat transfer coefficient (W/m2.K) The heat transfer coefficient, h,

is evaluated using the correlation of Ranz and Marshall [1952a and 1952b]:

Ji = kc( Cd,s − Cg) (3.26) where is the molar flux of vapor (kmol/mJ i 2.s); k c is the mass transfer coefficient

(m/s); Cd,s is the vapor concentration at the droplet surface (kmol/m3); C is the vapor

concentration in the bulk gas (kmol/m

RT

p X

=

C (3.28)

where is the saturated vapor pressure at the particle droplet temperature (Pa);

R is the universal gas constant ( ); X

)( p

p

K mol

J/ i is the local bulk mole fraction of species I; pop is the operating pressure (Pa) The mass transfer coefficient in equation