LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

download LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

of 8

Transcript of LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    1/17

    Local analysis of the tablet coating process: Impact of operation conditions

    on film quality

    Daniele Suzzi a, Stefan Radl a,b, Johannes G. Khinast a,b,n

    a Research Center Pharmaceutical Engineering GmbH, Graz, Austriab Institute for Process and Particle Engineering, Graz University of Technology, Inffeldgasse 21a, Graz, Austria

    a r t i c l e i n f o

     Article history:

    Received 3 August 2009Received in revised form

    15 July 2010

    Accepted 16 July 2010Available online 6 August 2010

    Keywords:

    Multiphase flow

    Simulation

    Mass transfer

    Pharmaceuticals

    Tablet coating

    Spray

    a b s t r a c t

    Spray coating is frequently used in the pharmaceutical industry to control the release of the active

    pharmaceutical ingredient of a tablet or to mask its taste. The uniformity of the coating is of significantimportance, as the coating usually has critical functional properties. However, coating uniformity is

    difficult to predict without significant experimental work, and even advanced particle simulations need

    to be augmented by CFD models to fully describe the coating uniformity on a single tablet.

    In this study we analyze the coating process by using detailed computational fluid dynamics (CFD)

    multiphase spray simulations. The impact and the deposition of droplets on tablets with different

    shape, as well as the production and evolution of the liquid film on the surface of the tablets are

    numerically modeled. Spray droplets are simulated with a Discrete Droplets Method (DDM) Euler–

    Lagrange approach. Models for multi-component evaporation and particle/wall interaction are taken

    into account. The wall film is treated with a two-dimensional model incorporating submodels for

    interfacial shear force, film evaporation and heat transfer between film, solid wall and gas phase. Our

    simulations show how different physical parameters of the coating spray affect the coating process on a

    single tablet. For example, we analyze for the first time the deposition behavior of the droplets on the

    tablet. The outcome of our work provides a deeper understanding of the local interaction between the

    spray and the tablet bed, allowing a step forward in the design, scale-up, optimization and operation of 

    industrial coating devices. Furthermore, it may serve as a basis for the combination with state-of-the-art DEM particle simulation tools.

    &  2010 Elsevier Ltd. All rights reserved.

    1. Introduction

    Coating is an important step in the production of many solid

    oral dosage forms, such as tablets and granules. The goal of film

    coating is the application of a thin polymer-based film on top

    of a tablet or a granule containing the active pharmaceutical

    ingredients (APIs). In the last years, more than half of all the

    pharmaceutical tablets were coated (IMS Midas Database, 2007).

    Functional coatings are usually adopted for taste masking or to

    alter the tablet’s dissolution behavior, for example by controllingthe rate of dissolution via semi-permeable membranes or by

    making it resistant to gastric juice through enteric coatings.

    Furthermore, active ingredients may be incorporated in the film

    layer. Colored non-functional coatings are commonly used to

    improve visual attractiveness, handling and brand recognition.

    A well-known example is the ‘‘blue pill’’ VIAGRAs by Pfizer Inc.

    Depending on the tablet’s dimension and coating functionality,

    the film thickness varies between 5 and 100 mm. A detaileddescription of the coating process and the different coater devices

    is presented in the book of  Cole et al. (1995).

    Historically, this process was developed by the confectionery

    industry to sugar-coat different types of candies. The

    pharmaceutical industry implemented this technique using open

    bowl-shaped pan. Nowadays, sugar-coated tablets are rarely

    developed due to the intricacy of the process and the high degree

    of operator skill required. Instead, tablets are typically coated

    with a polymer film of various compositions using modernequipment, such as drum and pan coaters.

    The first commercially available pharmaceutical film-coated

    tablet was introduced to the market in 1954 by Abbott

    Laboratories. Tablets were produced in a fluidized bed coating

    column based on the Wurster principle (Wurster, 1953), which

    was further developed by Merck in their US and UK plants. This

    new technique could be realized due to the development of new

    coating materials based on cellulose derivatives, e.g., hydroxy-

    propyl methylcellulose. Nevertheless, in the following decades

    coating columns were substituted by side-vented pans and the

    use of aqueous film solutions, which reduce the use of organic

    solvents and the related costs of the recovery systems.

    Contents lists available at ScienceDirect

    journal homepage:  www.elsevier.com/locate/ces

    Chemical Engineering Science

    0009-2509/$- see front matter  &  2010 Elsevier Ltd. All rights reserved.

    doi:10.1016/j.ces.2010.07.007

    n Corresponding author at: Research Center Pharmaceutical Engineering GmbH,

    Graz, Austria. Tel.: +43 316 873 7978; fax: +43 316 873 7963.

    E-mail address:  [email protected] (J.G. Khinast).

    Chemical Engineering Science 65 (2010) 5699–5715

    http://www.elsevier.com/locate/ceshttp://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.ces.2010.07.007mailto:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.ces.2010.07.007http://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.ces.2010.07.007mailto:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_1/dx.doi.org/10.1016/j.ces.2010.07.007http://www.elsevier.com/locate/ces

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    2/17

    Nowadays, tablet coating is typically carried out in pan coaters

    or fluidized bed systems. Modern production-scale pan coaters

    have batch sizes ranging from 500 g to 2000 kg, have a fully

    perforated cylindrical drum and use two-material nozzles for an

    effective spray generation. Today’s fluidized bed coaters allow

    continuous coating or have special internals to allow for coating

    processes involving coating solutions with high solid content

    (Porter, 2006). In this study we focus on pan coaters.

    Although coating processes have been used for many decades,there are still serious challenges, as there is a lack of under-

    standing of how material and operating parameters impact

    product quality and cause problems, such as chipping (i.e., films

    become chipped due to attrition), blistering (i.e., local formation

    of blisters due to entrapment of gas), cratering (i.e., penetration of 

    the coating solution into the bulk of the tablet causing crater-like

    structures), pitting (i.e., pits occur on the surface due to

    overheating of the tablet and partial melting), picking (i.e., parts

    of the film are removed due to sticking to other wet tablets),

    blushing (i.e., formation of spots due to phase-transitions of the

    polymer film), blooming (i.e., plasticizer concentrates at the

    surface, leading to a change of appearance), film cracking (i.e.,

    cracking of the film upon cooling due to high stresses) and many

    others. Quite often, poor scale-up of the process and/or insuffi-

    cient process understanding is the cause of such production

    problems and batch failures Pandey et al. (2006a). Although the

    reasons for these manufacturing problems are more or less

    understood, it is still a challenge to predict the occurrence of such

    effects for new systems.

    Therefore, in our work we focus on a basic understanding of 

    the film formation process on single tablets, with the goal of being

    able to predict the impact of material and operation parameters

    on the film quality. The current study is a first step in this

    direction. We investigate the spray fluid dynamics and the film

    formation of a glycerin–water mixture on two different tablet

    shapes, i.e., a sphere and a biconvex tablet, held in one position.

    Our analysis is based on a rigorous computational model that uses

    well established physical submodels for momentum, heat and

    mass transfer. Thus, we are able to predict the transient

    development of the mean film thickness of a wetting coating

    solution on arbitrarily shaped surfaces. Our main objective is to

    provide, for the first time, a science-based and quantitative

    understanding of which physicochemical parameters influence

    the uniformity of the coating layer on a single tablet. This

    knowledge is the key for the design, optimization and the rational

    scale-up of such processes and can form the basis for further

    studies on rotating tablets or whole tablet beds.

    2. Background

     2.1. Spray system

    A modern coating system is conceptually shown in   Fig. 1,

    where the coating suspension is sprayed on top of a moving bed of 

    the solid dosage form. The spray guns are usually mounted on an

    arm inside the pan and are directed towards the tablet bed. As the

    bed is moving, a tablet spends a fraction of a second in the

    spraying zone. The wet surface of the tablet needs to be dried to

    avoid sticking of the tablet to neighboring tablets, leading to

    manufacturing problems such as picking. However, too fast drying

    is counter-productive as well, as other problems may occur, such

    as the formation of a heterogeneous film. The drying air is

    directed towards the surface of the tablet bed in order to achieve

    good heat and mass transfer (i.e., for immediate drying of 

    the sprayed solution). The exhaust air can exit the pan through

    side opening, from inside the tablet bed (through an immersion

    tube system) or through a perforated rotary pan. The latter design

    allows the drying air to flow through the tablet bed in co-flow

    with the injected spray, leading to a more efficient coating

    process. Several companies offer this type of equipment, such as

    Glatt, Bohle, Driam, Manesty or Nicomac, each system being

    significantly different to the other systems.

    As shown in Fig. 2, the coating process can be divided into three

    phases, i.e., spraying, wetting and drying. In an ideal process, each

    tablet or granule passes through the spray zone for a predefined

    number of times, where spray particles impact the surface and wet

    the tablet. The adhering film is dried before the next amount of 

    solution is applied. This process continues until the particle is fully

    coated. The final film structure is typically non-homogeneous due to

    the presence of insoluble ingredients, such as pigments, and to the

    discontinuous and statistical nature of the coating process. A typical

    scanning electron microscope (SEM) image of a film-tablet coating

    illustrating the inhomogeneity of the coated layer is also shownin Fig. 2.

    Depending on the desired functionality of the tablet film,

    different coating solutions are used in industrial practice. The

    injected spray commonly consists of a carrier solution or vehicle

    (e.g., water, alcohols, ketones, esters or chlorinated hydrocar-

    bons), polymers (e.g., cellulose ethers, acrylic polymers or

    copolymers), plasticizer (polyols as glycerol, organic esters or

    oils/glycerides) and insoluble solid components (e.g., talcum,

    pigments and opacifiers). The used vehicle has to be compatible

    with the chosen polymer, as this is essential for obtaining optimal

    film properties such as mechanical strength and adhesion. As

    pointed out by Hogan (1982), the originally used organic vehicles

    have been steadily replaced by, mainly due to environmental and

    safety concerns. Several authors (e.g.,   Bindschaedler et al., 1983)

    Fig. 1.  Schematic of a modern pan coater (side-vented) and domain for the spray

    analysis.

    Fig. 2.  Conceptual scheme of the coating process.

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155700

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    3/17

    analyzed the complex process of film formation from a water-

    polymer dispersion. Initially, the polymer is dispersed in the

    aqueous solution in the form of discrete particles. The dispersed

    particles have to come into contact, coalescence and finally form a

    continuous film.

    An important factor in the film coating process is the quality of 

    the spray, as the droplets interaction with the tablet surface

    strongly affects the drying behavior and the uniformity of the final

    polymer layer. Two types of spraying devices are commonly usedin the film-coating technology: the hydraulic (airless) atomizer

    and the pneumatic (air-blast) atomizer. The first device requires

    high load pressures in order to produce adequate atomization of 

    the viscous solutions. However, the absence of air to produce the

    spray reduces early droplet evaporation. In case of aqueous

    vehicles this can lead to product overwetting and to poor-quality

    coatings. For this reason pneumatic atomizers are mostly used for

    water-based coating solutions (Muller and Kleinebudde, 2006).

    The liquid jet instability and the atomization processes in these

    atomizers have been discussed by several researchers, e.g.,  Varga

    and Lasheras (2003), as well as   Mansour and Chigier (1995).

    A combined experimental and theoretical analysis of the atomiza-

    tion of highly viscous non-Newtonian liquids can be found in the

    work of  Aliseda et al. (2008). In this study the breakup process is

    modeled through a two-stage instability mechanism, namely the

    primary Kelvin–Helmoltz instability followed by the secondary

    Rayleigh–Taylor instability. This study starts from the work of 

     Joseph et al. (2002), as well as of  Yecko and Zaleski (2005). The

    main result of  Aliseda et al. (2008)  is a correlation between the

    Sauter mean diameter (SMD) of the disintegrating droplets and

    the atomizer geometry, as well as the fluid-dynamical properties

    of the injected liquid (they used a solution of water and glycerol).

    In the absence of direct measurements of the real spray, e.g.,

    through Laser Diffraction (LD) or Phase Doppler Anemometry

    (PDA) systems (Hirleman, 1996), these models may be helpful for

    the initialization of the ‘‘numerical spray’’. This approach is, for

    example, also adopted in our work, i.e., our simulations are based

    on a single mean diameter of the droplets that make up the spray.

    The liquids being atomized are often highly viscous and

    sometimes non-Newtonian fluids, exhibiting complex physical

    mechanisms for primary and secondary breakup. In addition,

    droplet formation is also strongly affected by other physical

    properties of the coating solution, e.g., density and surface

    tension, as well as by the spray gun type. For example,  Aulton

    et al. (1986) investigated the effects of different atomizers, such as

    Binks-Bullows, Walther Pilot, Schlick and Spraying Systems guns,

    showing strong effects of the atomizing air pressure on the

    resulting mean droplet diameter. Typical mass-averaged droplet

    sizes range between 20 and 100 mm. The atomization properties,such as droplet size and velocity distribution, can be experimen-

    tally obtained via captive methods (these are methods in which

    droplets impinge on a flat surface and the diameter of the droplets

    on the surface is measured using a microscope), photographictechniques or laser-light scattering methods (Lefebvre, 1989).

    Clearly, the characterization of the coating spray represents an

    important step in the design of a coating device, as it strongly

    affects the local behavior of the film formation on the tablet

    surface.

    The evaporation of individual species from the liquid phase

    making up the droplets has to be considered as well. It is clear

    that the composition of the droplets affects the mass transfer from

    the spray droplets and the tablet film to the surrounding gas. For

    example,  Chen and Thompson (1970)   investigated the effect of 

    sodium chloride on the vapor–liquid equilibrium of glycerol–

    water solutions. Gaube et al. (1993) studied aqueous solutions of 

    PEG (often used in coating formulations) and dextran. A similar

    system was also studied by  Hammer et al. (1994), using sodium

    sulfate instead of dextran. Eliassi et al. (1999)  focused instead on

    the activity of water in aqueous PEG solutions with different

    molecular weights. Recent experimental work on PEG solutions

    has been extended by Kazemi et al. (2007). The activity of water in

    aqueous sugar solutions has been analyzed in two studies of  Peres

    and Macedo (1996, 1997).

    Finally, the interaction between droplets and surfaces

    represents a key issue in the description of coating processes.

    Experimental analyses and dimensional modeling of drop splash-ing processes can be found already at the beginning of the 20th

    century in the work of  Worthington (1908). The recent review of 

    Yarin (2006)   comprehensively explains the processes leading to

    film formation on thin liquid layers and dry surfaces, i.e., crown

    formation or splashing, drop spreading and deposition, receding

    (recoil), jetting, fingering and rebound.

     2.2. Tablet flow in coaters

    Experimental and numerical studies of the tablet flow in pan

    coaters are gaining increasing interest in the scientific commu-

    nity. Sandadi et al. (2004) characterized the movement of tablets

    at the top of a granular bed in a rotating pan via a digital imaging

    system to measure the velocity distribution on the surface of thetumbling tablet bed. Tobiska and Kleinebudde (2001) investigated

    the mixing behavior in a new coater type (the Bohle BLC pan

    coater). They showed that the mixing behavior can be character-

    ized by a simple temperature measurement, i.e., the temperature

    difference between the spray and the drying zone. In another

    study the same authors characterized the coating uniformity in a

    Bohle lab-coater using standard procedures (mass variance,

    dissolution testing) (Tobiska and Kleinebudde, 2003).

    Pandey et al. (2006b)   tracked a single tracer tablet (white

    colored) in a bed of black tablets using a CCD camera. They

    recorded the centroid location, as well as the exposed area of the

    tracer tablet in the zone of interest, i.e., the spray zone. The

    camera was directly placed in the coater and oriented in the same

    direction as the spray. They analyzed the average surface velocityprofile along the upper layer of the tablet bed. In addition,  Pandey

    et al. (2006a) performed discrete element method (DEM) simula-

    tions confirming the shape of the velocity profiles along the top

    cascading layer of the tablet bed. The range of the velocities

    reported varied between 0.13 and 0.55 m/s. Pandey et al. (2006a)

    proposed a characteristic velocity V  for the purpose of scaling the

    velocity profile at the top of the granular bed:

    V ¼ kRN 2=3  g 

    d

    1=6v1:8 ð1Þ

    Here k  is a constant,  R  is the pan radius,  N  is the pan rotation

    rate, g  is the gravitational acceleration and d  is the tablet size. The

    term  n  represents the fractional fill volume, defined as the ratiobetween the volume occupied by the bed and the total pan

    volume. The relation was verified using experimental databetween  n¼0.10 and 0.17 and rotational speeds between  o¼6and 12 rpm.  Alexander et al. (2002) used a similar approach and

    scaled the maximum velocity at the top of the granular bed to

    obtain a dimensionless maximal velocity  V S max

    . For low rotational

    speeds (o30 rpm), they found that the value of  V S max

      is between

    2.5 and 3.8. All these scaling laws are useful for the estimation of 

    the peak velocity in coaters and consequently for the time

    individual tablets stay in the spray zone.

    Kalbag et al. (2008)  used a single tracer sphere and a digital

    camera to measure the time that the marked tablet remains in the

    spray zone, also called spray residence time   t R. They manually

    post-processed the videos (50 min runtime at 60 fps) to obtain

    consistent experimental results for the spray residence time. The

    authors defined the dimensionless appearance frequency   ai   of 

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5701

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    4/17

    tablet i  as the number of appearances of a tablet in the spray zone

    during one pan revolution. The dimensionless appearance

    frequency averaged over all tablets  a  can be expressed as

    a ¼  2p

    o

      n

      ð2Þ

    Here  Dt R   is the average residence time per pass in the spray

    zone averaged over all tablets. The averaged dimensionless

    appearance frequencies were between   a ¼ 0:1 and   a¼ 1:4.However, this value depends strongly on the coating fraction,i.e., the ratio of the average number of tablets exposed to the

    spray and the total number of tablets in the coater. Therefore,

    they proposed an average residence time of the tablets per pass,

    i.e.  Dt R ¼ L=V . Here  L  is the length of the spray zone and  V   is the

    average velocity of tablets passing through the spray zone. The

    velocity at the top of the tablet bed is essential for the residence

    time in the spray zone, and hence, is expected to impact the film

    quality on the tablet. The average residence time of the tablets per

    pass was found to be between 0.07 and 0.27 s, depending on the

    pan speed. The standard deviation of the average residence time

    per pass was in the order of 0.03–0.24 s and was strongly

    dependent on the chosen pan speed. These experimental results

    were reproduced by discrete element method (DEM) simulations.

    Clearly, the standard deviation is an important quality

    indicator for the coating uniformity as tablets with a short

    residence time in the spray region will have a thin or imperfect

    coating. Also, in their work Kalbag et al. introduced other metrics

    that characterize the mixing behavior in the bed, i.e., the

    circulation and the fractional residence times. The circulation

    time t C ,i and the average circulation time per pass  Dt C  characterize

    the total time the tablet spends away from the spray zone, and the

    average time interval between successive appearances of the

    tablet in the spray zone, respectively. Note that the sum of the

    t R and  t Ci is the total coating time. The fractional residence time  f Ris defined as the ratio of time spent by a tablet in the spray zone to

    the total coating time  t 0. The average fractional residence time is

     f R  ¼   t Rt 0

    ¼   nN 

      ð3Þ

    where t R  is the average time the tablets spend in the spray zone,

    n   is the average number of tablets in the spray zone and  N   the

    total amount of tablets inside the pan coater. The ratio  n=N  is also

    referred to as the ‘‘coating fraction’’ and can be increased by

    increasing the size of the spray zone or by decreasing the number

    of tablets in the coater.

    Theoretical models for predicting the surface renewal rates of 

    the tablet bed in a rotary coating drum were reported by  Denis

    et al. (2003). They found an excellent agreement between the

    prediction of their model and experimental results for spherical

    tablets and bifluid pneumatic nozzles.

    Different groups are currently working on the numerical

    prediction of tablet flow in coaters (e.g.,   Dubey et al., 2008;Pandey et al., 2006a; Yamane et al. 1995). The coating event in the

    spray zone has up to now been described only with discrete

    element methods (DEM) and statistical deposition models for the

    tablets crossing the droplets region. One of the first attempts to

    couple a DEM solver with the computational fluid dynamics (CFD)

    gas flow in a rotating drum was proposed by   Nakamura et al.

    (2006). However, they simply assumed that a tablet was coated if 

    it was located within the spray region. This approach neglected

    resolving the droplets motion inside the drum and the local

    interaction of impacting drops on the tablet surface. Few

    additional studies have been reported on the CFD simulation of 

    coating processes. The recently presented work of  Muliadi and

    Sojka (2009) analyzed the interaction between coating spray and

    air flow inside a pan coater. However, the authors did neither

    consider the deposition of droplets on the tablets, nor the film

    formation processes. The recent paper by Freireich and Wassgren

    (2010)   examined both analytically and computationally the

    influence of a tablet’s orientation on the coating uniformity,

    leading to a deeper understanding of the intra-tablet film

    variability.

    3. Objectives

    Currently, the optimization of industrial coaters is mostly done

    by means of experimental and empirical analysis. State-of-the-art

    computational approaches include the use of Discrete Elements

    Method (DEM) , which already represents a consolidated practice

    in particle technology. However, current studies lack a detailed

    description of the film formation process on individual tablets or

    granules as only statistical tools for the film deposition on the

    tablet surface are used. Such an approach cannot capture the local

    behavior of the complex particle–gas–liquid system. Clearly, the

    liquid deposition behavior is strongly affected by the interactions

    of the spray and the solid surface of the tablet to be coated. Hence,

    the presented work will focus on the understanding of the basic

    principles of the spraying and deposition processes on a single

    tablet or granule as shown schematically in  Fig. 2.

    In summary, the major objectives of this work are

      to model the spray, deposition on the tablet, the coatingprocess, as well as the evaporation of the spray and the wall

    film in order to estimate the effects of the drying gas flow,

      to numerically analyze the impact and deposition of dropletson particles with different shape,

     to study the production and evolution of the liquid film on thesurface of the tablets and

      to investigate how different process parameters affect thecoating process on a single tablet.

    For this purpose, a variation matrix was set up and the effect of 

    each variation is analyzed in detail with respect to the filmquality. Also, the shape of the coated particle is varied, i.e., by

    considering a sphere and a standard tablet.

    4. Model and numerical solution

    In this section we present the 3D model used for the numerical

    analysis of the spray and the wall film. We adopted the 3D-CFD

    code AVL FIRE v2008 to simulate the dynamics of the coating

    spray and the film evolution on the tablet. We treat the coating

    process as a gas–liquid multiphase flow with deposition of a

    liquid film on the surface. For the description of the gas flow

    around the object to be coated we used the Reynolds averaged

    Navier–Stokes (RANS) equations including an appropriate turbu-lence model (k–e). As these models are well-known they are notdescribed here. The main difficulty of our work is to accurately

    model the motion of individual droplets, i.e., the spray around the

    object, as well as the droplet deposition and the motion of the

    liquid already deposited on the tablet surface.

    4.1. Spray simulation

    In our work the simulation of sprays is performed via the

    Lagrangian DDM (Discrete Droplet Method) approach. This

    approach is also known as Lagrangian Monte Carlo method,

    which was first proposed by Dukowicz (1980). The basic concept

    is to track the paths of statistical parcels of real droplets

    in physical, velocity, radius and temperature space. Further

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155702

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    5/17

    submodels for drag, particle/wall interaction, evaporation, turbu-

    lent dispersion and breakup may be included in the simulation

    approach. In the DDM method each physical phenomenon

    occurring in a parcel, e.g., atomization or coalescence/collision,

    directly involves all the droplets making up the parcel. This allows

    a drastic reduction in the computational effort to simulate liquid

    sprays, which in reality consist of many millions of single drops.

    In our simulations, the effects of secondary atomization,

    collision and coalescence have been neglected. We are awarethat close to the nozzle outlet this assumption is not valid due to

    the high droplet number density and velocity. For example

    Edelbauer et al. (2006)   and   Suzzi et al. (2007)  showed that the

    high liquid volume fraction close to the nozzle compromises the

    basic assumptions of the Lagrangian particle method. In this study

    we circumvent these difficulties by initializing the spray just

    outside the primary breakup region, a few centimeters down-

    stream the nozzle outlet. We then can neglect secondary breakup

    effects, as the Weber number of the droplets is, in our application,

    far away from critical values.

    Mass, momentum and energy conservation equations are

    solved for each parcel i  of the spray. A parcel represents a certain

    number of individual droplets, depending on their radius and

    the spray rate. The continuity equation for each parcel can be

    written as

    dmiddt 

      ¼ _miE    ð4Þ

    where the term on the right hand side represents the mass source

    due to evaporation. In the Lagrangian DDM the momentum

    equation, i.e., Newton’s second law, is directly integrated over

    time for each spray parcel:

    midd u

    !id

    dt   ¼   F 

    !iD þ F 

    !iG þ  F 

    !iP þ F 

    !iEX    ð5Þ

    The terms on the right hand side of Eq. (6) represent the drag

    force F iD, the gravity and buoyancy force  F iG, the pressure force F iP ,

    and the external force F iEX . The drag force acting on the droplets is

    calculated as

    F !

    iD ¼1

    2r g  AdC D9 u

    !rel9 u

    !rel   ð6Þ

    where   r g   is the gas density,   Ad   the cross-sectional area of thedroplet and   u

    !rel   the relative velocity between the gas phase and

    the parcel. The term C d represents the drag coefficient for a single

    sphere and is modeled in our work according to the formulation of 

    Schiller and Naumann (1993):

    C D ¼

    24

    Redð1 þ0:15Red

    0:687Þ,   Redo103

    0:44,   RedZ103

    8><>: ð7Þ

    Here the particle Reynolds number  Red  is defined as

    Red ¼ r g 9 u

    !rel9Dd

    m g ð8Þ

    In order to calculate the temperature  T id  of the droplets, it is

    necessary to calculate the heat and mass transfer rate to account

    for both the convective and latent heat loss of the droplets. The

    energy conservation equation for each parcel of droplets under

    the assumption of a uniform droplet temperature is (AVL, 2008):

    midc  p,ddT iddt 

      ¼ LðT idÞ _miE þ _Q    ð9Þ

    Here, c  p,d is the mean specific heat capacity of the droplets (i.e., an

    average over all components in the droplet),   L(T id) is the latent

    heat of evaporation (assumed to be a function of the droplet

    temperature) and   _Q    is the heat transfer rate between the

    surrounding gas and the droplets. As the spray consists of a

    mixture of components, i.e., glycerol and water, the calculation of 

    the mass transfer rate (i.e., the evaporation process) represents a

    key challenge in the simulation model. The multi-component

    evaporation model used in this work is based on the   Abramzon

    and Sirignano (1988) approach with the extension by Brenn et al.

    (2003). The main difference to the single-component case is that

    mass transfer of every component is taken into account

    separately, while heat transfer is still globally described. Hence,

    the evaporation rates of each species  j  are calculated and summedup to yield the total mass loss due to evaporation:

    _miE  ¼X

     j

    _miE , j   ð10Þ

    In the multi-component evaporation model used in this work,

    the mass transferred for each component   j   to the gas phase is

    given by

    _miE , j ¼pr g b gjDdSh

     j lnð1 þBM , jÞ ð11Þ

    The overbars in the gas density  r g   and the binary diffusioncoefficient   b gj   of species   j   in the gas phase indicate that these

    values are evaluated at a reference temperature and composition

    (for more details refer to AVL, 2008).  Dd   is the droplet diameter,

    Sh j  is the corrected Sherwood number of species j (defined below)

    and BMj   is the Spalding mass transfer number defined as

    BM , j ¼w j,sw j,1

    1w j,sð12Þ

    Here,   w j,s   is the gas phase mass fraction of species   j   at the

    surface of the drop (to be calculated from the vapor pressure of 

    species   j   at the droplet temperature) and   w j,N   is the bulk gas

    phase mass fraction. The total mass transfer rate can be also

    derived from the energy balance (Eq. (9)) at the surface of the

    drop, as

    _miE  ¼p  k g c  p,d

    Dd Nu lnð1 þBT Þ ð13Þ

    Here, k g  is the heat conductivity at a reference temperature and

    composition, and   Nun is the corrected Nusselt number defined

    below. In order to account for the relative motion between spray

    particles and gas phase, a Nusselt and Sherwood number is first

    computed according to the empirical relations of   Ranz and

    Marshall (1952):

    Nu0 ¼ 2 þ 0:552Re1=2 Pr 1=3 ð14Þ

    Sh0, j ¼ 2 þ0:552Re1=2 Sc  j

    1=3 ð15Þ

    The corrected Nusselt and Sherwood numbers  Nun and Sh j   are

    then calculated taking into account the deviation of the

    streamlines due to the evaporating mass flow:

    Nu ¼ 2 þðNu02Þ

    F T ,   Sh j   ¼ 2 þ

    ðSh0, j2Þ

    F M , jð16Þ

    The temperature and mass correction functions  F T  and  F M,j arecalculated as

    F ðBÞ ¼ ð1 þBÞ0:7lnð1 þBÞ

    B  ð17Þ

    using BT  or  BM,j for  F T  and  F M,j, respectively. In the relation for the

    temperature correction function   F T ,   BT    is the Spalding heat

    transfer number defined as

    BT  ¼ ð1 þBM Þf1   ð18Þ

    f¼c  p,dc  p, g 

    Sh

    Nu1

    Le  ð19Þ

    Here,   c  p, g   is the gas phase specific heat capacity at reference

    conditions and  Le  is the Lewis number. Finally, the heat transfer

    rate  _

    Q  between the droplet and the gas phase for the whole parcel

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5703

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    6/17

    is defined as

    _Q  ¼   _miE c  p,dðT 1T idÞ

    BT LðT idÞ

      ð20Þ

    Mixture fractions and mixture properties for each component  j

    at the drop surface needed in Eq. (12) are calculated using the

    activity coefficients g j:

     x j,s ¼ x j,Lg j pv, j

     p

      ð21Þ

    Here, x j,s   and  x j,L  are the mole fraction of species   j   in the gas

    and liquid phase, respectively. Note, that the mole fraction  x j,s   is

    directly related to the mass fraction   w j,s   that is used in the

    calculation for the mass transfer rate.  pv,j is the vapor pressure of 

    pure species j  and  p  is the total pressure. Instead of using Raoults’

    law, i.e., assuming gi to be equal to 1, the activity coefficients usedin our work have been calculated using a group contribution

    method (UNIFAC method, Peres and Macedo, 1997). This is in line

    with the work of  Attarakih et al. (2001), which described water–

    glycerol mixtures using the UNIFAC method and used the Antoine

    equation to describe the temperature-dependency of the vapor

    pressure.

    In summary, the calculation of the mass transfer rate   _miE , j   for

    each species and the heat transfer rate is performed using thefollowing procedure:

     calculate the mass fraction  w j,s  of each species  j  at the surfaceof the droplet (Eq. (21)),

      calculate all physical properties at the reference conditions,   calculate  Nu0 and  Sh0,   calculate   BM , j,   F M , j,   Sh j   and the mass rate of change for each

    species from Eqs. (12), (16) and (17), as well as the total mass

    transfer rate from Eq. (10),

     evaluate the Spalding heat transfer number  BT  (Eq. (18)), thecorrected Nusselt number   Nun (Eq. (16)) as well as the total

    mass transfer rate from the energy balance (Eq. (13)),

     compare the total mass transfer rates from Eqs. (9) and (13)and correct the heat transfer number  B

    T  until both total mass

    transfer rates are equal,

      evaluate the heat transfer rate from Eq. (20).

    The presented simulations are performed with a two-way

    coupling between the continuous and the discrete phases, i.e., all

    source terms for mass, momentum and energy can be also found

    in the transport equations for the gas phase.

    4.2. Droplet impact 

    The numerical model describing the interaction between

    impacting droplets and the wall (i.e., the tablet surface) is based

    on the work of  Mundo et al. (1995). Splashing or deposition occur

    depending on the dimensionless droplet Reynolds and Ohnesorgenumbers, defined as

    Re ¼ rLvd?Dd

    mL,   Oh ¼

      mL ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffirLsLDd

    p    ð22ÞThe (empirical) critical curve delimiting the splashing and

    deposition regimes is shown in Fig. 3 and can be expressed as

    Ohcrit  ¼ 57:7Re1:25 ð23Þ

    The ratios of the incoming and outgoing tangential and the

    normal velocities are also included in the spray-wall interaction

    model, leading to empirically determined ratios of 1.068 and

    0.208 for smooth walls, respectively. This critical curve is valid for

    the impact of single droplets, i.e., we neglect droplet–droplet

    interactions during the impact. Since the mass loading of droplets

    is relatively low, this assumption is expected to be valid.

    Furthermore, we do not take into account the exact shape of 

    the liquid film and assume a planar film surface on the tablet. This

    assumption is supported by the fact that (i) the characteristic time

    of drop spreading   t s   is of the order of (Rrim/Dd)10mLDd/sL   (Rrim)

    being the characteristic rim radius,   Yarin, 2006), and hence, is

    very small for the small droplets considered in our work and (ii)

    the tablet will be quickly covered by a film with a thickness in theorder of a few droplet diameters (see Fig. 10).

    According to the local properties of the impacting droplets

    either the liquid mass is transferred to the wall film (deposition)

    or new particles are generated (splashing regime), which rebound

    away from the tablet surface. Specifically, the secondary droplets

    could then

     evaporate and not deposit (i.e., spray drying effect),   deposit on the coater wall,  exit the coater with the exhaust air, or  deposit on another tablet.

    In our study we neglect the last option. The flow path of these

    droplets can only be analyzed using a detailed simulation of theair flow inside a coater. This will be part of a future study.

    As we have a binary mixture of glycerol and water, the

    physical properties of the droplets (i.e., density, surface tension

    and viscosity) are a function of the local composition and

    temperature. In our work we have taken this information from

    tabulated values from a manufacturer’s specification (The Dow

    Chemical Company, 2009) using linear interpolation.

    4.3. Wall film model

    The deposition, flow and drying of the coating solution on a

    tablet is critical for the quality of the tablet coating. In order to

    predict the distribution of the coating solution on the tablet, it is

    necessary to model the flow of the deposited fluid film. Somegeneral theoretical models to describe film formation and its flow

    on objects are available in literature (e.g., Yih, 1986; Baumann and

    Thiele, 1990). However, they are still not used in the pharmaceu-

    tical coating technology. In our work, we tried to adopt some of 

    these models for the prediction of film formation on tablets using

    the modeling assumptions described in the next chapter.

    4.3.1. Model assumptions

    Due to the high viscosity of the coating solution compared to

    the surrounding air, the fluid film is only slowly flowing over the

    tablet. In addition, evaporation of volatiles from the film, as well

    as heat transfer from and to the surrounding gas are major factors

    impacting that distribution of the film. In order to obtain a

    detailed but computationally still tractable prediction of the film

    Fig. 3.  Droplets/wall deposition model: critical curve.

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155704

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    7/17

    behavior, we make the assumption of a relatively thin film, i.e.,

    the film thickness is much smaller (o500 mm) than thecharacteristic dimensions of the simulated domain. For film

    coating processes this assumption is valid, as the final coating

    layer thickness is typically in the range of 100mm. It can beexpected that the film thickness will be of the same order of 

    magnitude. Following this assumption, the volume of the film can

    be neglected and no adaptation of the computational grid is

    necessary. Furthermore, the film surface can be assumed to beparallel to the solid wall. Thus, the wall film is modeled as a

    two-dimensional layer with a spatially distributed thickness   d.

    Due to the small dimensions and the small velocity of the film,

    interfacial shear stresses and wall friction influence the film much

    more than inertial forces and lateral shear (see   Cebeci and

    Bradshaw, 1977). For this reason, we have neglected these effects

    in the momentum conservation equation of the wall film,

    significantly reducing computational costs. When neglecting

    inertial forces we assume that the film is at a steady state. Thus,

    the velocity profile of the film is instantaneously determined by

    the forces acting on the wall film. In this work, the following

    effects have been taken into account:

      the stress induced by the surrounding gas flow on the liquidfilm, i.e., the interfacial shear stress, as well as the pressuregradient induced by the surrounding gas;

     body forces, i.e., gravitational acceleration;   multi-component evaporation from the film, taking into

    account individual diffusion coefficients of each component

    in the gas phase;

      interaction with impinging droplets, i.e., deposition of thecoating solution on the film, as well as the change of droplet

    size due to splashing on the droplet (Mundo et al., 1995). This

    effect has already been detailed in Section 4.1 of this paper.

    The impact of film deformation on the interaction between the

    gas phase and the film (momentum, heat and mass transfer) is

    taken into account via empirical models for the ‘‘equivalent sandgrain roughness’’ of the film. In addition, we solve the enthalpy

    equation of the wall film in order to predict its temperature, i.e.,

    we take into account conductive and convective heat transfer, as

    well as latent heat effects due to evaporation. In our model we

    assume laminar flow behavior. This hypothesis is acceptable as

    turbulence occurs only at large Reynolds numbers not obtained in

    the film. Film entrainment, i.e., the re-dispersion of the wall film

    into the gas flow via detachment of droplets from the film, does

    not play a significant role in our application and is therefore

    excluded. The droplet spreading after the impact at the tablet

    surface is accounted for in the statistics of the Lagrangian DDM

    method. The hypothesis of parcels containing a certain number of 

    real droplets leads to the assumption that the droplets impacting

    on a tablet mesh face homogeneously distribute on it. The average

    number of real droplets in such a parcel is in the order of a few

    thousand (for the parameters as per   Table 1). Thus, we assume

    that the film created by these droplets is uniform and is well

    described with a mean film thickness. This allows us also to use a

    two-dimensional flow model for the film spreading (see the next

    section). Furthermore, we have assumed the tablet to be non-

    porous, i.e., the coating suspension cannot penetrate into the

    tablet. Also, we take into account the change of the liquid-phase

    density and viscosity due to temperature or composition change.

    The spreading of the wall film around the edges of the tablet land

    is highly important for the coating quality. Anyhow, the high filmcurvature and the deriving surface tension effects would only

    locally affect the transport equation for the wall film in a tiny

    fraction of the total surface area. Thus, we neglect the effects in

    this area.

    4.3.2. Governing equations

    In this section, the governing equations that are used to model

    the above effects are described. Other aspects, e.g., such as the

    numerical discretization or alternatives to the models used in

    our work, can be found in the user guide of the software used

    (AVL, 2008).

    Here we introduce the film thickness equation, which

    represents the basic governing law for the wall film flow. It is a

    modified formulation of the continuity equation for the liquidphase on the tablet and is presented here for a Cartesian

    coordinate system:

    @d

    @t  þ

    @du1@ x1

    þ@du2@ x2

    ¼  1

    rsm   ð24Þ

    The terms d  and r  represent the thickness and the density of thewall film,  sm is the area-specific mass source term for the liquid in

    the wall film. Since in our case the wall (i.e., the tablet surface) is a

    closed surface, no boundary conditions (BCs) but only initial

    conditions (ICs) are needed, i.e., zero film thickness at time zero.

    Eq. (24) can be solved in a straightforward manner once the source

    term sm (due to deposition of droplets on the tablet and evaporation

    from the film) and the mean velocity components   u1   and   u2   are

    known. The source term   sm

      is known from the spray solution as

    described in Section 4.1 of this paper. The mean velocity

    components   u1   and  u2  are calculated from a momentum balance

    of the liquid film. In our work we use an analytical solution for the

    wall film’s momentum equation, which is motivated by the

    assumptions made above. Thus, the momentum equation reduces

    to a balance of the shear stress imposed on the film   t!I   and theviscous and turbulent dissipation within the film (see  Fig. 4)

    t!ð yÞr

      ¼ ðnþemÞ@ u

    !

    @ y  ð25Þ

     Table 1

    Basis set (B) of the simulation parameters.

    Parameter Symbol Value

    Droplet diameter   Dd   20 mmDroplets injection velocity   vd   15 m/s

    Gas temperature   T  g    298.15 K

    Droplets temperature   T d   298.15 K

    Tablet temperature   T TAB   298.15 K

    Total injected mass   M inj   0.1 g

    Injection time   t inj   0.1 s

    Mass fraction of glycerol in water   w   20 wt%

    x

    y

    wallfilm’s

    surface

    wall(tablet)

    Fig. 4.  Stress and velocity distribution in the wall film.

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5705

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    8/17

    Here, P

    m  denotes the turbulent eddy viscosity within the frame-

    work of Boussinesq’ hypothesis for the description of turbulent

    dissipation.   t!ð yÞ  and   u!

    represent two-dimensional vectors in the

    plane of the wall for the shear stress and the velocity, respectively.

    Both depend on the wall-normal coordinate   y. Clearly, the local

    distribution of the shear stress   t!ð yÞ   uniquely defines the shape of the velocity profile in the film once the turbulent eddy viscosity

    Pm

    is known.

    The interfacial shear stress   t!

    I   induced by the gas flow, thecomponent of the gravitational force   g 

    !99  parallel to the wall, as

    well as the longitudinal pressure gradient   @ p=@ x   determine the

    distribution of shear stress across the film, given by

    t!ð yÞ ¼   r g !

    99@ p

    @ x

    ðd yÞ þ t!I    ð26Þ

    The calculation of the interfacial shear stress   t!I  is non-trivial,as the interfacial stress itself influences the flow of the gas over

    the film due to the deformation of the phase boundary. In our

    work, this effect is taken into account by calculating   t!I  using anapproximation of the velocity profile in the gas phase, i.e., the so-

    called wall functions, with coefficients that depend on the wall

    shear stress and on the film thickness. In essence, we model the

    deformation of the film surface by a correlation for the equivalent

    sand grain roughness ks as a function of wall shear stress and film

    thickness.   ks   is then used to calculate a characteristic Reynolds

    number, and finally we can correlate this Reynolds number with

    the coefficients in the wall function. Due to this complex

    interaction between film flow and interfacial shear stress, it is

    necessary to iteratively solve for the mean film velocity, as

    detailed below. Details of this calculation can be found in   AVL 

    (2008).

    To obtain the velocity profile in the film, we first transform   u!

    and  y  in Eq. (25) into dimensionless coordinates by introducing

    the friction velocity   ut ¼ ffiffiffiffiffiffiffiffiffiffiffiffitW =r

    p   (see, for example,   Holman,

    1989). Here,   tW   is the wall shear stress, i.e., the stress at   y ¼0.Thus, we define the dimensionless wall film velocity   u

    !þ¼   u

    !=ut

    and the dimensionless wall distance  y þ ¼ yut=n.

    Hence, we obtain

    @ u!þ

    @ y þ  ¼

      t!ð y þ Þ=tW 1 þem=n

      ð27Þ

    This equation represents a general formulation of the film flow,

    both for turbulent and laminar films, with or without gravity,

    interfacial shear or pressure gradients. In case of laminar flow,

    where em is equal to zero, the integration of Eq. (27) leads to ananalytical solution for the dimensionless velocity profile (see, for

    example Prandtl et al., 1990). In our case, we assume a laminar

    flow of the wall film. This assumption is justified by the fact that

    in our simulations the wall shear stress and the film thickness are

    significantly below 1.2 Pa and 0.2 mm, respectively, values for

    which a transition from laminar to turbulent flow has been

    observed in the literature for internal combustion enginesapplications (AVL, 2008). By integrating over the film thickness,

    we obtain the mean film velocity  u  for laminar flow:

    u!

    ¼  d

    6m  2d   r g 

    !99

    dp

    dx

    þ3 t!I 

      ð28Þ

    Eq. (28) is used in the film thickness equation (Eq. (24)) to

    solve for the time evolution of the film thickness. We stress once

    more, that due to the assumption of negligible inertial forces the

    mean film velocity adapts instantaneously to the stresses acting

    on it. The film velocity is, however, transient due to the inherently

    instationary flow of the surrounding gas flow resulting in an

    instationary interfacial stress. Also, the mass, and consequently

    the thickness of the wall film, change with time due to droplet

    deposition and evaporation of the coating solution.

    The integration of Eq. (27) for turbulent flows (i.e.,   ema0)requires the definition of the eddy viscosity em as a function of thedimensionless wall distance   y+ . However, since the film flow

    remains laminar in our work, this is not discussed here.

    In the momentum equation (see Eq. (25)) for the film flow,

    surface tension effects near the front of the film have been

    neglected, since these effects (i) will be limited to the front of the

    film, as the curvature of the film is significant only in this region,

    (ii) we assume that the coating solution wets the surface, andthus, spreading is governed by viscous flow as the curvature at the

    edge of the film is small. Furthermore, a rough estimate of the

    capillary number (mLV /sL) (for  mL¼102 Pa s,  sL¼7 10

    2 N/m,

    V ¼30 m/s) yields a quantity b1, indicating small surface tension

    effects. However, in regions where the characteristic film velocity

    V  is low, the capillary number is small and surface tension may

    influence film spreading. In order to take into account

    surface tension and contact angle effects, a detailed resolution

    of the front of the film is necessary. This will be considered in

    further studies.

    Next, we describe the enthalpy equation for the wall film to

    obtain the film temperature distribution on the tablet. The

    simplest approach would be to assume that the film has the

    same temperature as the tablet. This is only valid, if the film is

    very thin and heat transfer between tablet and film is very fast. In

    our work, we assume that the gas phase, the wall film and the

    wall (i.e., the tablet) have different temperatures. Assuming a

    homogeneous film temperature over the film thickness, the

    enthalpy equation for the film can be written as

    rd  @h

    @t   þr Uðh u

    ¼ ð _hS , fw

    _hS , fg  _mE hE þ_hS ,imp þ

    _hS ,ent Þ ð29Þ

    The first two terms on the right hand side of Eq. (29) represent

    the heat fluxes in W/m2 between film and wall, and between film

    and the gas phase, respectively. In our work, these terms are

    modeled using appropriate correlations for the Nusselt number,

    i.e., predictions for the heat transfer coefficient based on

    experimental data have been used. Also, the temperature of thewall, i.e., the tablet, has been assumed to be uniform. The third,

    fourth and fifth term on the right hand side of Eq. (29) denote the

    enthalpy change due to evaporation (  _mE   is the evaporation mass

    flux in kg/(m2 s)), the area-specific enthalpy transfer from spray

    droplet via impingement and the area-specific enthalpy loss from

    droplet entrainment, respectively.

    Similar to the film thickness equation, the enthalpy equation is

    solved by using the result of the simplified momentum equation

    for   u!

    , i.e., Eq. (28).

    Finally, the total evaporation mass flux   _mE    from the film

    has to be modeled. The evaporation process can be described by

    Stefan’s law of unidirectional diffusion, which is used in our

    work, i.e.,

    _mE , j ¼ r g b g , j1wI , j

      @w j

    @ y

    ð30Þ

    Here,  r g   is the density of the gas phase,  D j,2  is the moleculardiffusion coefficient of the evaporating species  j  in the gas,  wI,j is

    the mass fraction of each evaporating species j at the interface and

    ð@w j=@ yÞI   is the gradient in wall-normal direction of the mass

    fraction at the interface. r g , D j,2 and wI,j can be calculated from theideal gas law, empirical correlations and the saturation pressure,

    respectively. However, the gradient of the mass fraction at the

    interface depends on the local flow conditions and is therefore not

    known. In our work we use the analogy to the turbulent velocity

    profile to approximate this gradient, taking into account the

    rough surface of the wall film. Details of this model can be found

    in AVL (2008).

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155706

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    9/17

    5. Results

    5.1. Base case definition

    As mentioned above, our goal was to investigate the influence

    of different operating parameters on the film formation on coated

    tablets. In order to define a realistic base set of parameters for our

    simulations, experimental investigations of a spray gun via Phase

    Doppler Anemometry (PDA) technique have been performed (forthe technique refer to  Hirleman (1996), the measurements have

    been performed by us at Duesen-Schlick GmbH, Germany).

    This experimental method is capable of simultaneously measur-

    ing diameter, velocity and mass flux of the injected spray droplets.

    The chosen nozzle was a Schlick 930Form 7-1 S35 ABC, typically

    used for pharmaceutical coating processes. Atomizing air (AA) and

    pattern air (PA) were both set equal to 1.2 bar, leading to an

    injected mass flow of approximately 60 g/min. The distribution of 

    droplet diameter and velocity at a distance of 200 mm from the

    nozzle tip are shown in Figs. 5 and 6, respectively. These average

    values were obtained by scanning the spray along a lineperpendicular to the spray axis. As well known in the literature,

    real sprays have a range of drop sizes and velocities, which will

    greatly influence their trajectories, their interaction and influence

    on the turbulent gas flow, evaporation time, likelihood of 

    bouncing, and degree of coverage on the tablet’s surface. In

    order to understand in detail the behavior of different droplet

    sizes and velocities, we considered variations of mono-disperse

    droplet population in order to quantify the singular effects of 

    diameter and velocity variations. Mono-dispersed droplets size of 

    20 mm, as well as an initial velocity of 15 m/s was selected as abase case, which is a good compromise between the volume- and

    number averaged data in Figs. 4 and 5. The temperature in the

    computational domain was initially set to room conditions

    (i.e., 298.15 K). The base set of parameters is defined in  Table 1.

    As already discussed in Section 4.1, these values are the initial

    conditions of a few centimeters downstream the nozzle outlet

    where secondary atomization, collision and coalescence become

    insignificant. Standard values for the physical properties of the air

    and water have been used. The physical properties (viscosity,

    density) of the glycerol–water mixture have been taken from the

    manufacturer’s specifications (The Dow Chemical Company,

    2009).

    A hybrid three-dimensional computational grid has been

    generated with a structured wall layer around the tablet (see

    Fig. 7). This structured wall layer is three cells in depth in order to

    sufficiently resolve the wall-near region. The computational grid

    consisted of a rectangular box with a cross section of 

    0.18 m 0.18 m and a length of 0.25 m. The spray nozzle was

    located at the upper part and in the center of the box. The distance

    between spray nozzle and the object (granule, tablet) to be coated

    has been set to 15 cm, which is a realistic value in industrial

    practice. Typically, round convex tablets were used in our work.

    The tablet’s main diameter was chosen to be 10 mm, and the

    Fig. 5.   Phase Doppler Anemometry (PDA) measurements of droplets size

    distribution (average values 200 mm from the nozzle tip).

    Fig. 6.   Phase Doppler Anemometry (PDA) measurements of droplets velocity

    distribution (average values 200 mm from the nozzle tip).

    150 mm

    Tablet

    Spray nozzle

    Droplets

    g

    Fig. 7.  Simulation domain and section of the 3D-hybrid computational mesh.

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5707

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    10/17

    height-to-diameter ratio was set to 0.67:1. The band thickness

    and the cap radius of curvature were equal to 3 and 7.6 mm,

    respectively. The box was modeled to be open on top and bottom,

    in order to allow for a gas flow induced by the injected liquid

    spray. In our work we oriented the top surface of the tablet

    perpendicular to gravity and to the incoming droplets. The effects

    of different tablet orientations with respect to the spray, as well as

    the impact of different tablet bed angles (and thus gravity) will be

    part of future work.In order to test the mesh quality and the convergence of the

    numerical simulation, preliminary test runs have been performed

    for pure water droplets and continuous spray injection. Based on

    these results a computational time step of 1 104 s was shown

    to be adequate in order to describe all the important scales of the

    process. The convergence criteria for the residuals have been

    chosen as 1 104 for momentum, turbulence and species

    conservation equation, and 1 106 for the energy conservation

    equation. A grid dependency study has also been performed to

    assess the quality of the computational mesh. For this purpose we

    used the well-accepted Grid Convergence Index (GCI) from

    Roache et al. (1986). Three meshes made of 15,527, 3884 and

    1205 face cells on the biconvex tablet surface, respectively, called

    mesh 1, mesh 2 and mesh 3, were used. The average film

    thickness f after 0.25 s has been chosen as the key variable for the

    GCI study. The face cells numbers  N , the grid refinement factors

    r 21   and   r 32, the values of the key variables  f, the approximate

    relative error   ea21, the extrapolated relative error   eext 

    21 , the

    extrapolated solution  fext 21 and the fine-grid convergence index

    GCI  fine21 are shown in Table 2. According to these results, the film

    thickness appears nearly equal for the finest and the middle mesh,

    leading to a deviation of only 0.020%. Thus, mesh 2 was used for

    further simulations. Note that the  GCI  method accounts only for

    discretization errors and not for modeling errors.

    A typical example of our results is presented in Fig. 8. The initialchoice of pure water leads to increased wall film evaporation

    compared to realistic cases. The evaporation process mainly takes

    place in the upstream part of the tablet, indicated by the low film

    thickness and the significant accumulation of water vapor near the

    upper edge of the tablet. Clearly, evaporation as well as the flow of 

    the film induced by the interfacial shear stress seems to surpass the

    accumulation of water by droplet deposition in this region of the

    tablet. Furthermore, it can be seen from  Fig. 8 that the evaporated

    water is transported along the cylindrical part of the tablet into the

    wake region of the flow. Accumulation of water vapor is highest

    near the cylindrical part of the tablet, whereas water vapor

    accumulation in the wake region is less pronounced. In both

    regions, i.e., in the wake and the cylindrical part, the high vapor

    concentration leads to a decreased evaporation rate, resulting in

    locally higher film thicknesses. It should be noted that in a ‘‘real’’

    tablet bed a single tablet is not suspended in space and the wake

    would be significantly different (or even not present). Hence, the

    results for the rear part of the tablet may change significantly.

    Nevertheless, the goal of our study was to compare the effects of 

    different process parameters on the film formation process. Thus,

    the major aim was to analyze the droplets collision and the film

    spreading on the surface of the singular tablet. A more detailed

    reproduction of the tablet bed environment was included in further

    analysis.

    Furthermore, our analysis describes only one pass of a tablet

    through the spray zone at a defined angle with respect to the

    spray. In a real system, tablets will enter the spray zone multiple

    times at different angles, thus resulting in a statistical distribution

    of the coating layer. Nevertheless, the presented analysis is

    important as it details under which conditions a uniform layer can

    be achieved and how the operating conditions impact the coating

    process.

     Table 2

    GCI calculation of discretization error.

    Parameter Symbol Value

    Face cells number   N 1; N 2; N 3   15,527; 3884; 1205

    Grid refinement factor (mesh 2 to 1)   r 21   2.0

    Grid refinement factor (mesh 3 to 2)   r 32   1.8

    Average film thickness (mesh 1)   f1   1.606e4 m

    Average film thickness (mesh 2)   f2   1.596e 4 m

    Average film thickness (mesh 3)   f3   1.370e4 m

    Extrapolated solution f21ext    1.60626e 4 m

    Approximate relative error   e21a   0.62%

    Extrapolated relative error   e21ext    0.016%

    Fine grid convergence index   GCI21fine  0.020%

    Fig. 8.  Simulation with water droplets. Color code in the gas phase denotes water vapor mass fraction. Color code on the tablet denotes film thickness.

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155708

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    11/17

    5.2. Variations

    In a second step, the main parameters of the base set have

    been modified and a glycerol–water mixture has been used in

    order to mimic a realistic coating process. Two different tablet

    shapes have been considered, i.e., a sphere and a convex tablet

    with the main diameter equal to 10 mm. Furthermore, following

    parameters has been varied:

     droplets diameter Dd,  environmental gas temperature  T  g ,  droplets injection velocity  v g ,   glycerol mass fraction in the coating solution  w.

    In order to reduce the amount of simulations, only one variable

    has been varied at once, resulting in the variation stars shown in

    Fig. 9. Simulations have been performed for a total time span of 

    0.5 s, whereas the injection of droplets stopped after 0.1 s. This

    choice was motivated by typical tablets velocities and residence

    times in the spray zone of industrial coaters, as described by

    Kalbag et al. (2008) and also discussed in Section 2.2 of this paper.

    5.3. Analysis of the results

    Fig. 10   shows the transient behavior of the film formation

    process for both the sphere, as well as the tablet. Clearly, during

    the injection of droplets for 0.1 s they primarily deposit at the

    front of the surface to be coated. However, after the injection has

    stopped (t ¼0.1 s), the film is more or less uniformly distributed

    over the sphere and the tablet. Thus, the transport of the liquid

    phase on a tablet to be coated is substantial, and it is important to

    model this part of the process. In our simulations the spreading of 

    the film is mainly influenced by the stress from the gas phase, the

    momentum introduced by the impacting droplets and by gravity

    to a smaller extent. The eventual tumbling of the tablet is not

    taken into account in the current work.

    The results in   Figs. 8   and   10   show that the film thickness

    reaches 70 and 100 mm already after a single ‘‘pass’’ in the spray.However, as discussed in the introduction, typical film thick-nesses after an entire coating operation are less than 100 mm. Theexplanation is that the typical film thickness refers to a solid film,

    whereas in our simulations the film thickness in a single ‘‘pass’’

    refers to a liquid film with suspended polymers. Thus, the drying

    process is not completed and the film mainly consists of the liquid

    components.

    According to the results in  Fig. 10, the spreading of the film

    seems to be completed after approx. 0.4 s for the base conditions

    in our study. As can be seen, even the shape of the coated object

    strongly influences the film thickness distribution as well as the

    total mass deposited. For example, in case of the tablet, a

    significant higher amount of droplets deposit on the surface

    leading to a substantially higher film thickness. Also, the location

    of the maximum film thickness after 0.5 s is different for the

    sphere (hmax   at a polar angle of approx. 1201) and the tablet

    (hmax at the backside of the tablet).

    In the following section, results for different cases are

    presented at a spray time of 0.5 s. The curves shown in Figs. 11

    and 12 represent the cumulative frequency distributions of 

    the local film thickness for all the cases in both variation stars.

    Fig. 9.  Variation star for sphere (left) and tablet (right). The base case conditions (B) are specified in Table 1.

    Fig. 10.  Time evolution of the film at different time steps for a sphere (top) and the tablet (bottom) for the base conditions ( Table 1).

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5709

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    12/17

    These plots have been derived from the simulation results by

    defining classes for the film thickness and allocating the fraction

    of the surface that fits into these classes. Thus, the cumulative

    frequency distribution represents the fraction of surface area that

    is covered with a film with a thickness lower or equal than a

    certain value. A wide distribution of the film thickness on the

    surface, i.e., poor coating uniformity, is indicated by a small slope

    of the curve. On the contrary, a narrow distribution, i.e., a good

    coating uniformity, is indicated by a steep increase of the curve.

    Concentrating on the case for a sphere (Fig. 11), we see that the

    base case, as well as the cases 2–5, behave similarly and do not

    show significantly different mean wall film thicknesses. In

    contrast, case 1 (droplet size increase from 20 to 50 mm) showsa significantly lower mean wall film thickness. A significant

    fraction of the surface does not seem to be covered by liquid at all.

    This is indicated by the fact that the first class of the cumulative

    frequency distribution has a value of approx. 0.37, i.e., 37% of the

    surface have a lower film thickness than the first class that has

    been analyzed. Looking at the shape of the distribution, it can be

    seen that for the base case, as well as for the cases 3–5, the

    distribution is bimodal, i.e., the distribution shows two regions

    with a local maximum in the slope. Such a bimodal distribution

    indicates that there exist zones with substantial different film

    thicknesses. In summary, only the increase in droplet size (case 1)

    results in a significant decrease in spray deposition, consequently

    leading to uncoated spots on the surface. The best conditions with

    respect to surface coverage by the film are realized in case 2,

    because the cumulative frequency distribution has the smallest

    initial slope.

    For the tablet (results shown in   Fig. 12), the situation is

    different and cases 1, 4 as well as 5 indicate significant effects on

    the film thickness distribution. Same as for the sphere, the larger

    droplet diameter (case 1) results in a much lower deposition of 

    droplets on the surface. This is caused by splashing and rebound

    of the droplets from the tablet’s surface. In addition, a significant

    part of the tablet’s surface is not covered, indicated by a value of approx. 0.26 for the first class of the cumulative frequency

    distribution. Case 4 (i.e., a droplet velocity of 30 m/s compared to

    15 m/s of the base case) shows mainly two effects on the film

    thickness distribution: (i) the film thickness after 0.5 s is only a

    fraction of that obtained in the base case as indicated by the shift

    to the left of the distribution. This is due to splashing of droplets

    on the tablet; (ii) the coating quality decreases as there exist

    regions that are completely free of liquid. This is again indicated

    by a high value for the first class in the cumulative frequency

    distribution for case 4 in  Fig. 12. Finally, case 5 (higher glycerol

    content of the liquid phase) shows similar trends, i.e., a slightly

    reduced film thickness, as well as decreasing coating uniformity.

    The decreased film thickness for the higher glycerol content can

    be attributed to a change in physical properties of the droplets

    (density, viscosity and surface tension) resulting in a reduced

    deposition on the tablet. The decreasing coating quality is due to

    the higher viscosity of the coating film, resulting in lower mean

    film velocities on the film. Consequently, the film cannot spread

    as quickly as in the case of lower glycerol concentration. The

    optimal conditions with respect to the coverage of the surface

    with the film seem to be case 3, as here the initial slope is

    smallest. However, also the base case, as well as case 2, indicate

    acceptable coverage of the surface with the coating solution.

    The time evolution of the total film mass on the tablet is shown

    in   Fig. 13. A total of 100 mg has been injected, of which only a

    fraction impacts on the tablet’s surface. 80 mg of the total mass

    are water and 20 mg are glycerol, the latter having a very low

    vapor pressure, leading to a significantly lower evaporation rate

    compared to water. In the base case approximately 9.4 mg of the

    coating liquid are deposited after 0.15 s. In the following, the film

    mass starts to decrease, due to the fact that the injection of 

    droplets is stopped and evaporation of the film starts. After 0.5 s

    the film mass has nearly linearly decreased to 8.4 mg ( 10%),

    indicating an almost constant mean evaporation rate. Compared

    to the base case, the bigger droplets (case 1) appear to deposit

    consistently less than the smaller ones, resulting in a peak value

    of only 0.85 mg for the total film mass after 0.12 s. This can be

    explained by the significantly higher Reynolds number of the

    impacting droplets (case 1 leads to a 2.5-fold increase in the

    Reynolds number, but only to a 37% decrease in the Ohnesorge

    number), which leads to the occurrence of splashing. Also, the

    Fig. 11.   Cumulative frequency distribution of the local film thickness of the coated

    sphere at   t ¼0.5 s (for the base case B defined in   Table 1  and the variations in

    Fig. 9).

    Fig. 12.   Cumulative frequency distribution of the local film thickness of the coated

    tablet at t ¼0.5 s (for the base case B defined in Table 1 and the variations in Fig. 9).

    Fig. 13.   Time evolution of the film mass on the surface of the coated tablet (for the

    base case B defined in Table 1 and the variations in  Fig. 9).

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–57155710

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    13/17

    time profile of evaporation for case 1 is significantly different

    from that of the base case. As can be seen from  Fig. 13, after the

    peak value of the film mass has been reached in case 1,

    evaporation takes place at a relatively high rate, until the film

    mass has been reduced to 0.39 mg, i.e., half of the peak value, after

    approx. 0.2 s. At this point the evaporation rate reduces

    significantly due to the fact that glycerol mass fraction

    increased (water is evaporating first due to the higher vapor

    pressure from the glycerol–water mixture). This results in adecrease of the vapor pressure of the film liquid, causing a

    pronounced decrease in the evaporation rate. The final film mass

    is 0.32 mg, i.e., the total loss of film mass in the second phase of 

    the evaporation of the film is marginal. Comparing the base case

    with case 4, i.e., a higher droplet velocity, we observe a similar

    shape of the time profile for the total film mass as in case 1: In

    case 4 the peak value of the film mass is significantly less (4.4 mg

    after 0.14 s) compared to the base case. This is again due to

    splashing, as the Reynolds number of the droplets is again higher

    than in the base case. Also the evaporation rate of the film on the

    tablet is significantly higher due to the higher gas velocity

    induced by the higher droplet velocity. This leads to a nonlinear

    time profile of the film mass caused by the accumulation of 

    glycerol in the film, because a significant fraction of the water has

    already evaporated. In summary, the total loss of film mass after

    0.5 s for case 4 is 2.3 mg or 52% due to evaporation, which is

    significantly more than in the base case. Thus, the tablets are

    already relatively dry after 0.5 s.

    The presence of more glycerol in the coating solution (case 5)

    leads to (i) a lower level of film mass on the surface, as well as to

    (ii) a significantly lower mean evaporation rate from the film. The

    initially deposited droplet mass is 7.3 mg after 0.15 s, whereas the

    final film mass after 0.5 s is 7.1 mg ( 2.7%). The first effect, i.e.,

    the lower level of film mass, can be explained by the change of the

    physical properties (i.e., density, surface tension and viscosity) of 

    the spray droplets, such that the deposition rate is decreased. The

    second effect, i.e., the reduced evaporation, is again due to the

    lower vapor pressure in case of a higher glycerol mass fraction in

    the film liquid.

    Case 2 (higher temperature) does not show a strong effect on

    the total film mass time profile. Obviously, the coating process is

    not very sensitive with respect to small changes in the gas

    temperature, i.e., the evaporation rate seems unaffected. Also, for

    case 3 (significantly higher gas temperature) the evaporation rate

    is only slightly increased (evaporation loss of 1.3 mg compared to

    1.0 mg in the base case after 0.5 s). Thus, even the wide range of 

    gas temperatures does not significantly alter the time evolution of 

    the total film mass present on the tablet.

    5.4. Assessment of the coating quality

    In order to analyze the coating quality, i.e., the homogeneity

    and the uniformity of the obtained film, the following quality

    indicators have been analyzed:

     mean film thickness (hmean),  variance of the film thickness on the surface (s 2),  delta (d), defined as the quotient of the maximum ( hmax) and

    the mean (hmean) film thickness value:

    d ¼  hmaxhmean

    ð31Þ

    According to this definition a perfectly homogeneous film

    would have a  d  value equal to 1:

      zero-thickness surface fraction ( Z ).

    This factor is the fraction of tablet surface that are not covered

    by the coating film.

    Based on these indicators, other parameters may be derived to

    assess the coating quality. For example, the relative standard

    deviation of the coating thickness can be easily obtained by

    dividing the mean value by its variance. In Fig. 13 we have already

    discussed the rate of change of the total film mass for a tablet, a

    quantity which is directly proportional to the mean film thickness

    introduced in this section. Here we focus once more on the

    comparison of the mean film thicknesses obtained for different

    cases. However, we also include the results for the coated sphere

    (see Fig. 14). As can be seen, in the base case, as well as in cases 2,

    3 and 5, the mean film thickness is significantly lower for the

    sphere compared to the tablet. This indicates that under the

    droplet deposition parameters defined in the base case (which

    essentially do not change in the cases 2, 3 and 5), the sphere

    receives consistently a lower amount of coating liquid, i.e., sphere

    and tablet behave similar and are nearly unaffected by

    temperature and viscosity of the coating solution. This indicates,

    as already mentioned in the discussion of   Fig. 13,   that theincreased evaporation rate due to a higher temperature does not

    play a significant role under the conditions used in this work.

    However, when changing droplet size (case 1) or droplet velocity

    (case 4), the sphere receives more coating solution compared to

    the tablet. This change is thought to stem from a regime change

    from droplet deposition to splashing. Obviously, in the case of 

    spheres the deposition is significantly less reduced in the

    splashing regime compared to tablets. We believe that this

    behavior is due to the differences in the separation behavior of 

    the gas flow. The gas flow is aligned longer with the sphere’s

    surface, and droplets generated by splashing have a second

    chance to deposit. For the tablet, the flow separates early, i.e., at

    the beginning of the cylindrical region, and droplets are less prone

    to impact a second time. Hence, we conclude that the mean filmthickness deposited on a given surface depends mainly on its

    shape (e.g., we observe an almost 50% decrease in the case of a

    sphere compared to the tablet for case 3) as well as the impact

    parameters (Re,   Oh) of the droplets. The solution’s viscosity, as

    well as the air temperature, show only minimal effects on the

    mean film thickness.

    In Fig. 15  we analyzed the film thickness variance on coated

    sphere and tablet. We observe that we have a similar situation as

    for the mean film thickness. Thus, the variance is lower for the

    sphere compared to the tablet for cases B, 2, 3 and 5, i.e., in the

    case where almost no splashing occurs. This clearly indicates that

    the coating solution can flow more easily over the regularly

    shaped sphere. In contrast, the edges on the tablet make it more

    difficult to obtain an even distribution of the film. In the other

    Fig. 14.  Mean film thickness on coated sphere (left bars) and tablet (right bars) at

    t ¼0.5 s (for the base case B defined in  Table 1 and the variations in  Fig. 9).

    D. Suzzi et al. / Chemical Engineering Science 65 (2010) 5699–5715   5711

  • 8/19/2019 LocalanalysisofthetabletcoatingprocessImpactofoperationconditionsonfilmquality.vv

    14/17

    cases (i.e., 1 and 4), the sphere shows a slightly higher variance

    compared the tablet. This indicates that the change to a splashing

    regime reduces the importance of the flow of the liquid on the

    object to be coated. In contrast to the mean film thickness where

    there was almost no effect of the gas temperature, this parameter

    seems to have a pronounced effect on the film thickness variance

    (see cases 2 and 3, note that the y-axis has a logarithmic scale and

    that case 3 has an almost 18-fold higher  s2 value than the basecase!). This strong sensitivity with respect to the temperature can

    be explained by the significant change of the film’s viscosity,

    which strongly decreases with temperature. Consequently, the

    film can flow more easily over the tablet and can accumulate at

    the rear part of the tablet’s surface, resulting in an uneven

    distribution of the coating solution. However, the glycerol

    content, i.e., the change of the viscosity with glycerol content at

    the temperature of the base case, shows only a negligible effect. Inthe cases where splashing occurs (cases 1 and 4), the variance

    drops below the value for the base case. The relative variance, i.e.,

    s2=h2mean  (data not shown), however, is still higher as in the basecase. This is especially true for the case of a larger droplet size,

    where we observe an almost 9-fold increase of the relative

    variance compared to the base case. Thus, the occurrence of 

    splashing seems to decrease the quality of the coating

    significantly.

    Similar trends are observed in Fig. 16, in which we focus on the

    d value, i.e., on the ratio of maximum to mean film thickness. For

    the tablet, the gas temperature has again a strong effect on the

    coating quality and the influence of the glycerol content is small.

    This can be interpreted by the fact that there will be an

    accumulation of coating solution at the rear part of the tablet,

    resulting in a spot with an extremely high film thickness. Also, the

    droplet size seriously influences the coating quality of the tablet,

    whereas droplet velocity does not significantly alter the  d  value.

    This is also true for the sphere.

    Finally, we present   Fig. 17  which shows the fraction of the

    surface to be coated that has received no coating. Clearly, in those

    cases where splashing occurs (cases 1 and 4)  Z  is between 27% and

    67%, whereas for all other situations   Z   is below 12%. Thus,

    splashing results in a significant reductio