particle aerosolisation and break-up in dry powder inhalers: evaluation and modelling of impaction...

Particle Aerosolisation and Break-Up in Dry Powder Inhalers:Evaluation and Modelling of Impaction Effects for AgglomeratedSystems

WILLIAM WONG,1 DAVID F. FLETCHER,2 DANIELA TRAINI,1 HAK-KIM CHAN,1 JOHN CRAPPER,3 PAUL M. YOUNG1

1Advanced Drug Delivery Group, Faculty of Pharmacy, University of Sydney, Sydney, New South Wales 2006, Australia

2School of Chemical and Biomolecular Engineering, University of Sydney, Sydney, New South Wales 2006, Australia

3Pharmaxis Ltd., Unit 2, Frenchs Forest, Sydney, New South Wales 2086, Australia

Received 18 July 2010; revised 5 January 2011; accepted 10 January 2011

Published online 1 March 2011 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/jps.22503

ABSTRACT: This study utilised a combination of computational fluid dynamics (CFD) andstandardised entrainment tubes to investigate the influence of impaction on the break-upand aerosol performance of a model inhalation formulation. A series of entrainment tubes,with different impaction plate angles were designed in silico and the flow characteristics, andparticle tracks, were simulated using CFD. The apparatuses were constructed using three-dimensional printing. The deposition and aerosol performance of a model agglomerate sys-tem (496.3–789.2 :m agglomerates containing 3.91 :m median diameter mannitol particles)were evaluated by chemical analysis and laser diffraction, respectively. Analysis of the manni-tol recovery from the assembly and CFD simulations indicated that mass deposition on theplate was dependent on the impactor angle (45◦–90◦) but independent of the airflow rate(60–140 L·min−1). In comparison, wall losses, perpendicular to the impactor plate were de-pendent on both the impactor angle and flow rate. Analysis of the particle size distributionexiting the impactor assembly suggested mannitol aerosolisation to be independent of impactorangle but dependent on the air velocity directly above the impactor plate. It is proposed thatparticle-wall impaction results in initial agglomerate fragmentation followed by reentrainmentin the airstream above the impaction plate. Such observations have significant implications inthe design of dry powder inhaler devices. © 2011 Wiley-Liss, Inc. and the American PharmacistsAssociation J Pharm Sci 100:2744–2754, 2011Keywords: CFD; dry powder inhaler; impaction; agglomerate; deagglomeration; aerosols; insilico modelling; pulmonary drug delivery; simulations; particle size

INTRODUCTION

The delivery of dry powder particulates to the respi-ratory tract, for the treatment of local and systemicdisease states, requires the primary drug particlesto have an aerodynamic diameter less than approxi-mately 5 :m.1 Although, there are many formulationvariables available to achieve adequate levels of drugdelivery to the lung, two are regarded as primary for-mulation methods: carrier-based and agglomeration-based systems.2 These methods are used to ensureefficient entrainment of the active pharmaceutical

Correspondence to: Paul M. Young (Telephone: +61-2-9036-7035; Fax: +61-2-9351-4391; E-mail: [email protected])Journal of Pharmaceutical Sciences, Vol. 100, 2744–2754 (2011)© 2011 Wiley-Liss, Inc. and the American Pharmacists Association

ingredient (API) into the airstream, whilst provid-ing a means of sample dilution (when small micro-gram range of doses are required). Despite these ap-proaches, conventional dry powder inhalation (DPI)formulations have relatively low aerosol efficiencies,with fine particle fractions (i.e. the percentage doseof API with an aerodynamic diameter suitable for in-halation therapy) of less than 30% being observedregularly.3 The reason for such poor performance isdue to the high surface area-to-mass ratios of theAPI drug particles, inducing high cohesive–adhesiveforces between contiguous surfaces within the formu-lation. Subsequently, significant research has beenundertaken at both the fundamental and empiricallevel to understand the complex processes drivingparticle deagglomeration and aerosolisation.

2744 JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 7, JULY 2011

EVALUATION AND MODELLING OF IMPACTION EFFECTS FOR AGGLOMERATED SYSTEM 2745

There are many DPI devices on the market or un-der development, employing different approaches todisperse the micron-sized API.3–5 Interestingly, littleresearch has been conducted to study the exact mech-anism of break-up and in general these systems areoptimised through performance modification and em-pirical study design. Furthermore, where fundamen-tal studies have been conducted, they have generallyfocussed on the aerosolisation of micron-sized drugparticles from carrier-based formulations6–8 ratherthan the break-up of agglomerate-based systems, con-taining micron-sized primary particles.

The investigation of the underlying mechanismsbehind the dispersion of dry powders usually involvedthe use of entrainment tubes incorporating deagglom-eration apparatuses. Although in some cases agglom-erate systems have been studied using this approach,9

most of these studies have been cross-disciplinary(e.g. in the printing or minerals industry); as such,there is a large variation in the materials and particlesize distributions studied. The very different physicalmechanisms acting in these applications from thoseimportant in the application of interest here limit therelevance of such studies.

In order to study the aerosolisation process inagglomerate-based DPI systems, the authors haveundertaken a series of studies to evaluate howa standard formulation behaves with respect tospecific deagglomeration mechanisms (i.e. airflow,turbulence, impaction, etc.) etc.). In a previousstudy, the authors utilised a combination of com-putational fluid dynamics (CFD) and experimentalentrainment tube measurements to study the break-up and aerosolisation of a model agglomerate sys-tem (containing micron-sized mannitol particles) asa function of airflow and turbulence variables.10

The study design utilised a series of venturi tubesto induce turbulent flow with characteristics equiv-alent to commercial DPI devices, while minimis-ing other potential break-up mechanisms (such aswall or grid impaction). Interestingly, however, al-though this study focussed on the effect of turbu-lence on agglomerate break-up, the small amountof impaction, which inevitably occurred in the ven-turi assembly as the core diameter was reducedand the air velocity increased, appeared to dominateagglomerate break-up.10

To further investigate the mechanism of break-up and aerosolisation in agglomerate-based DPI sys-tems, the influence of impaction angle and speedwere studied. A series of entrainment tubes con-taining different impaction plates are designed andtheir flow behaviour was evaluated using CFDanalysis, and subsequently compared with physi-cal aerosol and deposition measurements to ascer-tain the influence of impaction on the aerosolisationmechanism.

MATERIALS AND METHODS

Materials

Primary mannitol particles (spray-dried micron-sized powder) were supplied by Pharmaxis Ltd.(Sydney, New South Wales, Australia). Water waspurified by reverse osmosis (MilliQ; Millipore, Mol-sheim, France). All organic solvents were suppliedby Sigma (Sydney, New South Wales, Australia) andwere of at least analytical grade.

Preparation of a Model Particulate System

Model agglomerates were prepared by mixing the pri-mary mannitol powder in a Turbula mixer (BachofenAG Maschinenfabrik; Basel, Switzerland), at 42 rpmfor 15 min. The mixing vessel was a custom-builtaluminium cylinder of 25 mm diameter and 26 mmlength. The agglomerated powder was post-processedthrough a nest of sieves (ISO 3310–1 test sieves,Endecotts Ltd.; London, UK) to produce a 500–800:m fraction. The agglomerated powders were storedin sealed containers at 45% relative humidity and25◦C for a minimum of 48 h prior to their use.

Physical Characterisation of the Primary MannitolParticles and Agglomerated Systems

The primary mannitol powder and agglomerate sys-tems were characterised in terms of particle size, mor-phology, density, mass, and surface area. The meth-ods and results are reported in detail elsewhere.10

In general, the primary particle size distribution wasdetermined in a chloroform suspension using laserdiffraction (Malvern Mastersizer 2000; Malvern In-struments Ltd., Worcestershire, UK), whereas the ag-glomerate size distribution was determined in air,using optical microscopy and image analysis (CX41microscope; Olympus, Tokyo, Japan and ImageJ soft-ware; National Institute of Mental Health, Mary-land). The density and surface area of the primarymannitol particles was determined using helium pyc-nometry (Accupyc 1340 gas pycnometer; Micromerit-ics, Norcross, Georgia) and nitrogen adsorption(Tristar II 3020; Micromeritics), respectively. In addi-tion, agglomerate mass was measured using a SevenFigure Cahn microbalance, (DVS-1; Surface Measure-ment Systems Ltd., London, UK). From these specificmeasurements, values such as primary particle num-ber and agglomerate density could be calculated. Ulti-mately, these parameters could be used in theoreticalcalculation and in silico simulation.

Construction of the Impactor Assembly

The impactor apparatus was constructed using arapid three-dimensional prototyping technology. Theapparatus was designed to minimise the turbu-lence kinetic energy (TKE), to ensure a fully de-veloped fluid flow prior to impaction and to induce

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 7, JULY 2011

2746 WONG ET AL.

agglomerate impaction at a set of specific angles. Inaddition, the impactor geometry was modified to en-hance the probability of agglomerate impaction whilstminimising primary mannitol (deagglomerated) par-ticle impaction. A cone geometry, perpendicular tothe air stream, was chosen as the best impactor de-sign because it avoided geometric issues associatedwith the use of curved tubes (i.e. variations in im-pact angle and airflow across the cross-section ofa curved tube). Initially, a three-dimensional modelof the impactor assembly was constructed in silico,using computer-aided design software (ANSYS De-sign Modeler 12.0; ANSYS, Canonsburg, Pennsylva-nia) and analysed using CFD simulations (ANSYSCFX 12; ANSYS). On the basis of these CFD simu-lation results, the structural parameters were modi-fied using an iterative approach in order to minimiseturbulence whilst maintaining controlled impaction.Parameters, such as cone geometry, internal voidspace, and entrance port were studied and a final se-ries of optimised models were designed. A schematicrepresentation of the final design is shown inFigure 1. In general, a 2.2 m entry port (19 mmcore diameter) ensured a fully developed airflowprior to entry into the impaction chamber (validatedpreviously10). The flow then impinges on an impactionplate, set at angles of 90◦, 75◦, 60◦, and 45◦ to the im-pinging flow (note that the 90◦ plate is effectively aflat plate; as shown in Fig. 1). The impaction plateswere constructed atop a 20◦ angle converging conethat tapered towards the exit port. This cone was sup-ported in the main assembly via four mounting bladesdesigned to minimise the disturbance in the airflowcaused by these supports.

The entire impactor assembly was constructed withinterlocking sections so that the experimental appa-ratus could be taken apart and washed after each ex-periment, allowing for “stage-specific” drug deposition

analysis. The entire assembly was constructed fromacrylonitrile butadiene styrene using a rapid proto-type three-dimensional printer (Dimension Elite, Di-mension Inc., Eden Prairie, Minnesota) and the partswere polished using 1200 grit glass paper (FH Prager,Sydney, New South Wales, Australia) to ensure a uni-formly smooth surface throughout the apparatus.

CFD Analysis

The commercially available CFD code, ANSYS CFX12(ANSYS), was used to simulate the flow of air at25◦C through the impactor assembly, to predict tur-bulence properties and to track particles, so thattheir velocity and impact parameters could be ob-tained. To reduce computational expense, the flowfield was modelled for a 90◦ sector of the geometryhaving symmetry boundaries on the circumferentialfaces. The Reynolds-averaged Navier–Stokes equa-tions were used to determine the flow field throughoutthe impactor assembly and the shear stress transportmodel with scalable wall functions was used to modelturbulence. A finite volume method based on a tetra-hedral mesh with inflation at the walls was used tocapture the boundary layer behaviour correctly.

Mesh independence analysis was conducted to en-sure that the computational results were independentof mesh size. This was achieved by studying the ax-ial velocity profiles across the inlet section and im-paction assembly, as a function of increasing meshdensity. Mesh independence was confirmed when themesh size contained 1.9 × 105 nodes. A higher reso-lution mesh (containing 4.5 × 105 nodes) was utilisedfor computational analysis. A cyclical oscillation wasobserved in the position of a recirculation zone down-stream from the impaction zone. However, this wasdeemed to have a minimal effect on the break-up ofthe particles in comparison with the impaction events.Lagrangian particle tracking was conducted at the

Figure 1. Schematic representation of the impaction assembly (X = 90◦, 75◦, 60◦ and 45◦).Impactor components are: (1) inlet port, (2) outlet port, (3) impactor wall, (4) impactor exit cone,and (5) the main impactor plate.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 7, JULY 2011 DOI 10.1002/jps


end of the simulation. A fixed number of particleswith a given size and density specified at the inletport of the venturi tube were tracked and the charac-teristics of particle-wall impactions were determinedby setting the venturi walls to have a zero coefficientof restitution (allowing the location of the initial par-ticle impact to be visualised). All calculations wereconducted at volumetric flow rates of 60, 100, and140 L·min−1.

In-Line Aerosol Particle Size Analysis

The particle size distribution of the mannitol pow-der at the exit port of the impactor assembly wasmeasured using laser diffraction to evaluate thebreak-up of the model agglomerate system at differ-ent flow rates and impaction angles. The apparatus(Fig. 1) was mounted vertically on a scaffold withthe impaction assembly exit port connected in-lineto a Spraytec particle sizer (Malvern InstrumentsLtd.). A Gast Rotary Vane pump (Erweka GmbH,Heusenstamm, Germany) was connected to the out-let of the particle sizer and the flow was calibratedusing a flow meter (TSI 3063; TSI instruments Ltd.,Buckinghamshire, UK). For each measurement, 50mg of the agglomerated powder was introduced intothe centre of the airstream at the top of the impactorassembly using a funnel.

The geometric particle diameter distribution at theoutlet of the impactor was measured in real-timeat a data collection rate of 2500 sweeps per secondover a range of 0.1 to 2000 :m. Three volumetricflow rates (60, 100, and 140 L·min−1) and four im-paction angles (45◦, 60◦, 75◦, and 90◦) were stud-ied in triplicate. After each experiment, the impactorassembly was deconstructed and the surfaces werewashed with purified water into separate volumetricflasks. Mannitol concentrations recovered from eachstage of the impaction assembly were analysed usinga validated high-performance liquid chromatography(HPLC) method and system described previously.11

In general terms, an LC20AT pump, SIL20AHT au-tosampler, CBM-Lite system controller with a pc-computer running LC solution v1.22 software, andan RID-10A refractive index detector (Shimadzu,Sydney, New South Wales, Australia) hwere utilised.An 8 mm Resolve C18 Radial Pack chromatogra-phy cartridge (Waters Asia Ltd., Singapore) was usedfor separation, at a flow rate of 1 mL·min−1. Pu-rified water was used as the mobile phase. Thedeconstructed impactor assembly consisted of fivecomponents/stages: (1) inlet port, (2) outlet port, (3)impactor wall, (4) impactor exit cone, and (5) main im-pactor plate (corresponding to the components shownin Fig. 1). After sample recovery, each stage of theimpactor assembly was washed with ethanol and air-dried prior to reassembly.

Statistical Analysis

One-way ANOVA analysis (with Tukey’s post hocanalysis) was used to test significance. A differencewas considered significant when p was less than 0.05.The commercial statistical software package, SPSSStatistics 17.0 (SPSS Inc., Chicago, Illinois) was used.

RESULTS

Primary Mannitol Particle Properties

A representative scanning electron microscope imageof the primary mannitol particles is shown in Fig-ure 2a. It can be seen that all the particles have adiameter less than 10 :m and are spherical in na-ture (presumably due to the nature of the particlesproduced via spray drying of droplets). This was con-firmed by laser diffraction in which analysis of thesize distribution of the primary mannitol particles(mean of triplicates ± standard deviation) indicated alognormal distribution with 90% of the particles hav-ing a volume diameter less than and equal to 6.82 ±0.37 :m and 10% having less than and equal to 1.95± 0.02 :m, as shown in Figure 3. The median d0.5particle diameter of the primary particles was 3.91± 0.15 :m. This value was used for theoretical cal-culation of the agglomerate structure as outlined inthe following section. Such observations are in goodagreement with those reported in previous studies.11

Agglomerate Properties

The agglomerate properties used in this study wereinvestigated extensively in a previous paper.10 A rep-resentative optical microscopy image of the mannitolagglomerates is shown in Figure 2b. Using the Im-ageJ software, the particle diameter of 300 agglomer-ates was measured. As shown in Figure 4, analysis ofthe size distribution indicated a normal distributionthat could be fitted to the following equation (R2 =0.99):

U% = 0.3414 dp − 169.4 (1)

where dp is the agglomerate diameter (:m) and U% isthe percentage undersize. Subsequently, it was calcu-lated that the minimum and maximum agglomeratediameters were 496.3 and 789.2 :m, respectively. Inaddition, the mass of a series of agglomerates (n =25) were measured using a Cahn microbalance andreported as 91.1 ± 22.1 :g. Using the median pri-mary mannitol particle diameter, mass and densityvalues, it is thus possible to calculate an agglomeratedensity of 655 kg·m−3.

Because the purpose of this paper was to investi-gate the influence of agglomerate impaction on pri-mary mannitol aerosolisation, the agglomerate sizedistribution, as well as the theoretical agglomerate


2748 WONG ET AL.

Figure 2. (a) Scanning electron microscope images of theprimary mannitol particles and (b) an optical microscopeimage of the mannitol agglomerates.

density, was used to model the agglomerates in silicofor the CFD analysis of the impactor during the designprocess. This ensured that 100% of the agglomerateswould impact the plate assembly for all angles.

CFD Analysis of the Impactor Assembly

Velocity streaklines and TKE distributions in the im-paction assembly at a flow rate of 140 L·min−1areshown in Figure 5a. Analysis of the 140 L·min−1 CFDdata suggests that the approach velocities are simi-lar to the theoretical average fluid flow velocity for a19 mm diameter tube (8.2 m·s−1). Upon enteringthe impactor assembly, the streaklines follow a path

Figure 3. Particle size distributions of the primary man-nitol particles. The solid line shows the volumetric diameterdistribution.

around the main impactor plate and a concurrent de-crease in velocity was observed, due to the increasein cross-sectional area, before acceleration at the exitport. Interestingly, a small recirculation zone was ob-served in the peripheral void space above the im-paction plate; however, the relative velocities werelow. Analysis of the TKE indicated that effect of tur-bulence in the impactor assembly on particle break-up could be eliminated because the average TKE val-ues were lower than those observed in the 19 mmdiameter entrance port, in which previous studieshad indicated no significant effect on d0.1 (only 1.4± 0.4% particles ≤ 10 :m at 140 L·min−1; Ref.10).Figure 5b shows particle tracking data for a repre-sentative sample of 5 :m particles (n = 500; density= 1435 kg·m−3), whereas the insert shows particletracking for the agglomerates (n = 500; density = 786kg·m−3 with size distribution as given in Eq. 1). Anal-ysis of the data showed that all of the agglomerates

Figure 4. Particle size distributions of the primary man-nitol agglomerates.



Figure 5. (a) Turbulence kinetic energy (contour plot) and (streakline plot). (b) Particle track-ing for 10 :m particles (agglomerate tracking shown in inset). Examples are shown for the 60◦

impaction plate at 140 L·min−1.

impacted the plate, irrespective of impact plate an-gle. In comparison, the particle tracking of approx-imately 5 :m primary mannitol particles indicatedimpaction efficiencies less than and equal to 10%, sug-gesting that deagglomerated primary particles wouldpass through the impactor assembly.

Pressure drop across the impactor assembliesranged from 246.2 Pa for the 90◦ assembly at140 L·min−1 to 53.8 Pa for the 45◦ assembly at 90L·min−1. This is significantly lower than the 4 kPaspecified in pharmacopoeia methodology for the test-ing of DPIs; however, it is important to note that theaim of this study was to investigate the influence ofimpaction forces on the break-up and aerosolisationin agglomerate-based DPI systems through the use of

specialised entrainment tubes rather than an actualDPI device.

Velocity, TKE, and particle tracking data for eachimpactor assembly (45◦, 60◦ 75◦, and 90◦ angle plates)at each flow rate (60, 100, and 140 L·min−1) wereexported from the CFD simulation. These data arediscussed in terms of the in vitro aerosolisation per-formance below.

Aerosolisation Performance at Exit Portof the Impactor Assembly

The aerosolisation efficiency of the deagglomeratedmannitol particles may be described by the aerody-namic size distribution measured at the exit portof the impactor assembly. Figures 6a and 6b show


2750 WONG ET AL.

Figure 6. Particle size distribution of the mannitolaerosol exiting the impactor assembly at (a) 60 L·min−1

and (b) 140 L·min−1 (mean values ± standard deviations).

size distributions of mannitol particles after passingthrough the different impactor assemblies at 60 and140 L·min−1, respectively. Analysis of these data sug-gests a multimodal distribution for all data sets. Thehigher value peak may be attributed to the primaryagglomerate size distribution, whereas the lowerpeak corresponds to the primary mannitol particles.The central peak, spanning from around 20–100 :mis likely to represent particle clusters, fractured fromthe main agglomerate during impaction. Interest-ingly, there are a high percentage of agglomerates/large-agglomerate clusters remaining after impactionat any angle and all flow rates; however, the relativepercentages appear to be dependent on the angle andflow rate. It is also important to note here, that thedata were represented as a conventional (pharma-copoeia) volume distribution, and thus, a weightingtowards the larger particle diameters will always ex-ist (because one 500 :m sphere will have the samevolume as one million 5 :m spheres of equivalentdensity).

In general, as the flow rate is increased for a spe-cific angle, the percentage of particles less than andequal to 10 :m in diameter increases (i.e. at a 90◦

impact angle the % of particles ≤ 10 :m increasesfrom 0.87 ± 0.118% to 26.23 ±5.87% between 60 and140 L·min−1). The effect of angle on particle break-upappears to be dependent on flow rate. An increase inagglomerate break-up is observed when the impact

angle is decreased (i.e. 45◦) at low flow rates; how-ever, conversely, particle break-up increased at highflow rates and high angles (i.e. 90◦). Such observa-tions, however, need to be put in context with respectto mannitol retention within the impactor assembly(because particle bounce and reentrainment may bea dominating factor in this system).

Mannitol Deposition in the Impactor Assembly

This study focussed on impactor-related particlebreak-up, therefore, it stands to reason that signif-icant impactor plate and internal component lossesoccur due to wall deposition. To study this, the im-pactor assembly was carefully disassembled aftereach experiment and mannitol deposition was mea-sured using HPLC. The mannitol deposited on eachcomponent of the impactor assembly at 60, 100, and140 L·min−1 is shown in Figures 7a–7c, respectively.In general, particle deposition in the inlet port, cone,and outlet port are small, with the highest depositionbeing observed at 140 L·min−1.

For example, 1.85 ± 0.13 mg was recovered from theinlet port at 140 L·min−1 using the 90◦ angle impactorassembly. This was significantly higher than all otherflow rates and for all other angles, as it representedaround 3.7% of the loaded mannitol mass. Such ob-servations are most likely due to the agglomerate/particle clusters having enough residual momentum,after ricocheting off the impactor plate, to deposit onthe inner surface of the induction port. At smallerangles, the normal impact velocity will be lower andagglomerate/particle reentrainment more likely. Sub-sequently, at higher flow rates, and for smaller angles,an increase in outlet port deposition was observed(e.g. 1.8% was recovered from the outlet port of the45◦ angle impactor assembly at 140 L·min−1).

Analysis of the impaction plate and the surround-ing wall suggested significantly higher mannitol de-position than for the inlet, outlet, and cone compo-nents. Furthermore, the relative deposition on eitherthe impactor or surrounding wall was dependent onthe impaction angle and flow rate. At high impactangles and low flow rates (Fig. 7a), the majority ofthe mannitol was found to deposit on the impactionplate; however, as the angles approached 45◦, a re-duction in impactor deposition was seen with con-current increase in wall losses. Conversely, at thehighest flow rate, both impactor and wall depositiondecreased with decreased angle (Fig. 7c).

Interestingly, the amount of mannitol deposited ona specific angled impactor plate appeared to be inde-pendent of the flow rate (i.e. approximately 15 mgmannitol is deposited on the 90◦ plate at all flowrates). Such observations suggest that the impactorplate may be overloaded due to the high mass of man-nitol passing through the system (i.e. 50 mg).11 To



Figure 7. Mannitol deposition on the impactor assemblyat (a) 60, (b) 100, and (c) 140 L·min−1 (n = 3; 50 mg agglom-erate samples).

test this hypothesis, the 90◦ impaction assembly wastested at different agglomerate doses (5, 10, 20, 50,and 100 mg) and the stage deposition evaluated. Anal-ysis of the plate deposition versus agglomerate doseover this range showed a dose-dependent response(R2 = 0.995), suggesting the plate was not overloaded.Subsequently, it may be concluded that the mass of

agglomerate remaining on the impactor is due to thenature of the impact event and the geometries of thecontacting surfaces. It is also important to considerthe total amount of mannitol retained within the de-vice. Figure 8 shows the influence of both flow rateand impact angle on overall mannitol retention in theimpactor assembly. As expected, an increase in im-pact angle and flow rate resulted in an increase inoverall impactor losses as the impact force increases.However, analysis of the data suggests that this in-crease occurs on the impactor wall because mass de-posits for any particular impaction angle remainedconstant with respect to the flow rate.

Total and regional deposition, as well as ag-glomerate break-up and aerosolisation, is de-pendent upon three factors: (1) impact velocityand inelastic component of momentum (agglomer-ate fracture/compressive forces), (2) forces impartedby the airstream (to detach/reentrain particles), and(3) the elastic component of the particles momen-tum (leading to particle bounce and reentrainment).These factors and their relationship to the observa-tions made here are discussed in more detail below.

DISCUSSION

Agglomerate Impaction, Break-Up, and Reentrainment

In order to understand the process of agglomeratebreak-up and powder aerosolisation, it is importantto consider the forces acting within the system dur-ing the impact event and to relate these to impactordeposition and aerosol performance at the exit port ofthe assembly.

Figure 8. A three-dimensional plot showing the influenceof flow rate and impact angle on mannitol wall loss withinthe impactor assembly.


2752 WONG ET AL.

Momentum at Impaction

Because the particle mass and velocity are known, itis possible to calculate the maximum momentum ( p)carried by each agglomerate upon impaction (wherep = mass × velocity). Using the minimum and maxi-mum agglomerate diameters (496.3 and 789.2 :m, re-spectively), and a theoretical agglomerate density of655 kg·m−3, the momentum at 9.88 m·s−1 (equivalentto the impact velocity on a 90◦ plate at 140 L·min−1)will be between 0.4 and 1.7 :N·s. Similarly, at60 L·min−1 (5.77 m·s−1) the momentum will be be-tween 0.2 and 1.0 :N·s. For 75◦, 60◦, and 45◦ plate,the linear momentum is equivalent to that of the 90◦

plate; however, the force encountered during the im-paction event will depend upon the impact angle andnormal component of agglomerate momentum.

Reentrainment Forces

Reentrainment of primary drug particles, particlefragments, and unbroken agglomerates will be de-pendent upon the nature of the impaction event (i.e.whether it is inelastic or elastic) and the adhesion be-tween the agglomerate components and the plate sur-face. Assuming an inelastic collision event, the forcerequired to reentrain impacted particles in the airstream (Fair) will increase as the impaction angle be-comes perpendicular to the airflow (i.e. approaches90◦; Eq. 2):12

Fair = µf · Fad

cos2 − µf sin2(2)

where µf is the coefficient of friction, Fad is the ad-hesion force, and θ is the angle of the impinging airstream.

It is also important to note that simultaneouslythere will be an elastic component to the collision,resulting in particle bounce and reentrainment abovethe impactor surface. It is expected that the elasticresponse would increase as the impact angle increasesbecause the normal component of velocity must betaken into consideration.

Whilst the agglomerate particle velocity in the airstream prior to impact may be calculated using La-grangian particle tracking, the normal impact velocity(VN) can be calculated from Eq. 3:

VN = Vi sin(B · 2i

180

)(3)

where Vi is the agglomerate velocity prior to impactand θi is the impaction angle.

Obtaining a physical value for the elastic and in-elastic components of impaction of the particle mo-mentum is difficult because neither the Young’s mod-ulus nor the yield strength of the agglomerate and

primary particles can be measured (allowing predic-tion of the elastic and inelastic components relatingto the conservation of momentum). Also, because theinelastic deformation component is not known, it isnot possible to predict the contact area and thus Fad.However, the relationships between momentum orimpaction velocity and agglomerate break-up may bestudied.

Relationship Between Impaction/Flow Parameters andAgglomerate Aerosolisation

The relationship between normalised (impaction) air-flow or linear airflow (above the impaction plate) andthe 10th percentile particle diameter (d0.1), as a func-tion of impaction angle is given in Figures 9a and 9b.Furthermore, the relationship between the d0.1 andimpact angle as a function of linear airflow is given inFigure 9c.

From Figure 9 it can be seen that a decrease inthe d0.1 is observed as both impact and air velocityare increased, indicating more efficient agglomeratebreak-up and primary particle aerosolisation. Inter-estingly, however, analysis of the normalised impactvelocity data indicates that there is not a direct re-lationship between angle of impact and d0.1 (Fig. 9a).For example, the d0.1 for the 45◦ angle plate at an im-pact velocity of 4 m·s−1 is not significantly differentthan the d0.1 for the 90◦ plate at an impact velocity of7 m·s−1. Furthermore, when plotting the airflow rate(directly above the impaction plate) as a function ofd0.1 (Fig. 9b) or the d0.1 as a function of impact angle(Fig. 9c), no change in particle break-up is observed asangle is increased for any specific linear flow velocity.

Conversely, the direct relationship between the lin-ear air velocities directly above the impaction plate isand d0.1 was observed to be independent of angle. Thisis further exemplified in Figure 10 when the percent-age of particles less than and equal to 5 :m is plottedas a function of air velocity above the plate. Analysisof the data for all impaction plates at all flow veloc-ities indicated an exponential relationship betweenvelocity (v; m·s−1) and percentage of particles lessthan and equal to 5 :m, as shown in Eq. 4:

% ≤ 5:m = 0.0205 × e0.6936v (4)

where an R2 of 0.978 was observed.

Mechanism of Break-Up and Aerosolisation

In general, analysis of the data indicate that thereis not a direct relationship between impaction platelosses and aerosol performance, inferring that theevent is a lot more complicated than simply dependingon the impact force. Subsequently, in order to studythe process of agglomerate break-up and aerosolisa-tion in the impactor assembly it is important to high-light the factors influencing (1) impactor deposition



Figure 9. d0.1 versus (a) normal impact velocity and (b)air velocity as a function of angle and (c) angle as a functionof air velocity.

and (2) aerosolisation performance, independently.These observations are summarised below:

(1) Impactor and wall deposition:(a) particle mass deposited on the impactor

plate is dependent on impactor angle;(b) particle mass deposited on a specific im-

pactor plate is independent of airflow rate;(c) wall losses, perpendicular to the impactor

plate, are dependent on both impactor angleand airflow rate.

(2) Aerosol performance:(a) particle size is independent of impactor an-

gle for a given flow rate;

(b) particle size is dependent on the air velocitydirectly above the impactor plate and not onimpact velocity.

From such observations, it may be inferred that thebreak-up and aerosolisation events are due to two dis-tinct processes: bulk agglomerate fracture on impactfollowed by powder dispersion and aerosolisation inthe airstream.

It is envisaged that, in this system, a criticalthreshold for agglomerate fracture has been exceeded.As such, regardless of the flow rate, particles will frac-ture and the difference in the amount deposited on theplate at different angles is due to the difference in thegeometry of the plate (satisfying observations 1a and1b above). During the collision, a significant amountof fractured agglomerate and particulate clusters re-bound into the airstream directly above the impactionplate where they are reentrained. The velocity of theair directly above the plate disperses the already frag-mented agglomerate and the degree of aerosolisationis directly proportional to the linear airflow velocity(satisfying observations 2a and 2b above).

Interestingly, small variations in aerosolisationperformance with respect to impact angle at high andlow velocities were observed (see Fig. 10), and this canbe attributed to the variation in wall deposits adjacentto the impactor plate due to secondary impaction oflower momentum fragments (Fig. 7; satisfying obser-vation 1c).

Previous studies using discrete element modelling(DEM), to some extent, correlate with the experimen-tal data described here. For example, work by Ninget al.13 showed that lactose agglomerates (rangingin size from 9–11 :m) underwent a ductile deforma-tion upon impact, fragmenting into smaller clusterswhich could be described by a damage ratio relating tothe number of broken particle contacts after impact.Interestingly, they found that the degree of damage

Figure 10. Percentage of particles less than 5 :m plottedas a function of air velocity.


2754 WONG ET AL.

was directly scalable with the impact velocity. 13,14

This group went on to show high-speed images of thisbreak-up process, in which the agglomerate clearlyfragmented and were deflected from the impact sur-face. More recently, Tong et al.15 used DEM to studythe impact of agglomerates containing 5000 monodis-persed 5 :m particles (that had a similar density of1450 kg·m−3 and packing density of 0.55 to those stud-ied here). As with the study by Ning et al.13, this groupreported a ductile transformation and the formationof particle clusters. In addition, the group reportedthat the damage ratio scaled with velocity, whereasimpact angle had little effect for angles more than 30◦.Ultimately, these simulations go some way to corrob-orate what is reported here; however, it is importantto note a limited number of particles were used withconfined physical parameters (such as fixed interfa-cial energies and narrow size distributions). More im-portantly, these previous models do not describe thedynamic event of aerosolisation directly above the im-pactor plate. These should be considered in futurestudies.

CONCLUSIONS

This study focussed on the influence of impaction ge-ometry effects on the aerosolisation and break-up ofpharmaceutical agglomerates for inhalation. The im-paction assemblies were designed to minimise otherpotential powder deagglomeration mechanisms (suchas turbulence), and the influence of velocity and im-paction was studied. It appears that for agglomer-ated inhalation powders, particle-wall impaction re-sults in initial agglomerate fragmentation followedby deagglomeration in the airstream above the im-paction plate. Direct visualisation of this event, aswell as the evaluation of turbulent aerosolisation, af-ter impaction will be considered in further studies.

ACKNOWLEDGMENTS

This research was supported by the Australian Re-search Council’s Linkage Projects funding scheme(project LP0776892). The views expressed herein are

those of the authors and are not necessarily those ofthe Australian Research Council.

REFERENCES

1. Pritchard JN. 2001. The influence of lung deposition on clinicalresponse. J Aerosol Med 14:S19–S26.

2. Young PM, Traini D, Edge S. 2007. Advances in pulmonarytherapy. In Advanced drug formulation design to optimizetherapeutic outcomes; Williams RO, Taft DT, McConville JT,Eds. London: Informa Healthcare, pp 1–51.

3. Smith IJ, Parry-Billings M. 2003. The inhalers of the future?A review of dry powder devices on the market today. PulmPharmacol Ther 16:79–95.

4. Newman SP, Busse WW. 2002. Evolution of dry powderinhaler design, formulation, and performance. Resp Med96:293–304.

5. Islamand N, Gladki E. 2008. Dry powder inhalers (DPIs)—Areview of device reliability and innovation. Int J Pharm360:1–11.

6. Voss A, Finlay WH. 2002. Deagglomeration of dry powder phar-maceutical aerosols. Int J Pharm 248:39–50.

7. Wang ZL, Lange CF, Finlay WH. 2004. Use of an impinging jetfor dispersion of dry powder inhalation aerosols. Int J Pharm275:123–131.

8. Louey MD, Van Oort M, Hickey AJ. 2006. Standardized en-trainment tubes for the evaluation of pharmaceutical dry pow-der dispersion. J Aerosol Sci 37:1520–1531.

9. Calvert G, Ghadiri M, Tweedie R. 2009. Aerodynamic disper-sion of cohesive powders: A review of understanding and tech-nology. Adv Powder Technol 20:4–16.

10. Wong W, Fletcher DF, Traini D, Chan HK, Crapper J, YoungPM. 2010. Particle aerosolisation and break-up in dry pow-der inhalers 1: Evaluation and modelling of venturi effects foragglomerated systems. Pharm Res 27:1367–1376.

11. Wong W, Crapper J, Chan HK, Traini D, Young PM. 2010.Pharmacopeial methodologies for determining aerodynamicmass distributions of ultra-high dose inhaler medicines. JPharm Biomed Anal 51:853–857.

12. Bhattacharya S, Mittal KL. 1978. Mechanics of removingglass particulates from a solid-surface. Surf Technol 7:413–425.

13. Ning Z, Boerefijn R, Ghadiri M, Thornton C. 1997. Distinct el-ement simulation of impact breakage of lactose agglomerates.Adv Powder Technol 8:15–37.

14. Boerefijn R, Ning Z, Ghadiri M. 1998. Disintegration of weaklactose agglomerates for inhalation applications. Int J Pharm172:199–209.

15. Tong ZB, Yang RY, Yu AB, Adi S, Chan HK. 2009. Numer-ical modelling of the breakage of loose agglomerates of fineparticles. Powder Technol 196:213–221.


particle aerosolisation and break-up in dry powder inhalers: evaluation and modelling of impaction...

Documents