Powder Technology 152 (2005) 107 117 www.elsevier.

com/locate/powtec

Modelling the mechanical behaviour of pharmaceutical powders during compaction

C.-Y. Wua,T, O.M. Ruddyb, A.C. Benthamb, B.C. Hancockc, S.M. Besta, J.A. Elliotta
a

Pfizer Institute for Pharmaceutical Materials Science, Department of Materials Science and Metallurgy, University of Cambridge, Pembroke Street, Cambridge, CB2 3QZ, UK b Pfizer Global Research and Development, Sandwich, Kent, CT13 9NJ, UK c Pfizer Global Research and Development, Groton, CT 06340, USA Received 23 June 2004; received in revised form 26 October 2004; accepted 20 January 2005 Available online 25 April 2005

Abstract The mechanical behaviour of pharmaceutical powders during compaction is analysed using Finite Element Methods (FEM), in which the powder is modelled as an elasticplastic continuum material. The DruckerPrager Cap (DPC) model was chosen as the yield surface of the medium, which represents the failure and yield behaviours. Uniaxial compaction experiments were also carried out using a compaction simulator with an instrumented die. The objectives of these experiments were two-fold: (1) to investigate the pharmaceutical powder behaviour during compaction, for which the variation of relative density of the powder bed with applied pressure is analysed; and (2) to calibrate the DPC model with the experimental measurements, from which realistic powder properties are generated and fed into finite element analysis (FEA). The relationship between relative density of powder bed and applied pressure is also obtained from FEA and compared with the experimental data. Good agreement between the experimental and FEA results is observed, which demonstrates that FEA can capture the major features of the powder behaviour during compaction. Furthermore, close examination of the evolution of the stress distribution during unloading reveals that there is a narrow band existing from the top edge towards the bottom centre of the tablet, in which there are localised, intensive shear stresses. It is in this band that potential failure regions, such as cracks, can initiate. This has been demonstrated with experimental evidence from X-ray microtomographical images and photography of fractured tablets. It is therefore demonstrated that FEA can predict the possible mechanism of failure, such as capping, during compaction. D 2005 Elsevier B.V. All rights reserved.
Keywords: Compaction; Failure; Tablet; Finite element modelling

1. Introduction Powder compaction is a process widely used in many industries. For instance, in the powder metallurgy and ceramic industries, powders are generally compacted into a green body before being sintered at high temperature to manufacture engineering components. In chemical engineering, detergent tablets are made of dry powder through a powder compaction process. In the pharmaceutical industry, billions of tablets are produced on a daily basis by

T Corresponding author. Tel.: +44 1223 767059; fax: +44 1223 767063. E-mail address: cyw22@cam.ac.uk (C.-Y. Wu). 0032-5910/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.powtec.2005.01.010

compacting pharmaceutical powders. The use of compaction for pharmaceuticals is mainly due to various advantages of tablets over other dosage forms, such as chemical and physical stability, acceptable shelf life, accuracy of dosage and ease of controlling drug release. In addition, for patients, tablets are convenient since they are portable and easy to administer, and sweeteners or coatings can be used to mask any unpleasant tastes. From the manufacturing point of view, tablets can be mass-produced with high production rates. Therefore, tablets are the most common and popular dosage form for administering drugs via an oral route. The processing of tabletting is similar to the process of other powder compacts and can generally be divided into

108

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

three distinct stages: (1) die filling, where the blends of powder mixes are delivered into a feed shoe which runs over the die opening before depositing the powder into the die cavity under gravity [1,2]; (2) compaction, where the powder is compressed inside a die by two punches; and (3) ejection, where the compact is ejected from the die. The powder behaviour during the above three stages all determine the properties of final compacts. Therefore, understanding the mechanical behaviour of powders during each stage is very important and has attracted significant attention over the past several decades. In this study, we will focus specifically on the compaction process. In particular, we concentrate on a process where the powder is compressed by the upper punch only, while keeping the lower punch stationary: single-ended compaction (SEC). According to the movement of the upper punch, the compaction process can be divided into two phases: compression and decompression. During compression, as the upper punch moves towards the bottom punch, the powder bed experiences intensive densification and the powder particles move together to form aggregates with appreciable cohesive strength due to van der Waals forces, mechanical interlocking and formation of solid bridges [3]. As the distance between two punches continues to decrease, the compaction pressure and the packing density of the powder bed increase dramatically. Many previous studies have concentrated on the relation between relative density and increasing compaction pressure [46]. This relation is material-dependent and is regarded as a measure of compressibility of the materials. A number of models have been developed and modified to characterise the powder compressibility more accurately [3,7]. Although no direct evidence of densification mechanisms has yet been reported, it is often thought that particle rearrangement and collapse of granules are the dominant behaviours at low pressure. As the pressure increases, the elastic and plastic deformation of constituent particles becomes dominant. Further increases in compaction pressure may result in fragmentation of the primary particles, which then causes further densification of the powder bed. Decompression takes place once the upper punch starts to move away from the lower punch. During this stage, the compaction pressure drops quickly as the distance between the two punches increases, and some of the elastic strain induced during compression will recover. This is accompanied by an increase in the volume of the powder bed and a consequent decrease in the relative density. Controlling the elastic recovery rate is crucial in the processing of powder compacts because defects, such as cracks and fracture of powder compacts, are more likely to be induced by faster elastic recovery, or spring-back, during decompression. To elucidate the cause of these problems is a challenging task, which has to be based on an improved understanding of the powder behaviour during the compaction process. Thus, the objective of this study is to provide a fundamental understanding of the mechanical behaviour

of pharmaceutical powders during compaction with a combined investigation including experimental and numerical analyses. The finite element method has been employed, and a brief introduction to the FEA techniques is presented in Section 2. Experiments with uniaxial compaction using an instrumented die have also been conducted for two purposes: on the one hand, experiments were performed to calibrate the material model chosen in the present FEM study so that material properties of realistic powders can be obtained for the numerical analysis. On the other hand, experimental data was used to verify the numerical analysis. A detailed description of the uniaxial compaction experiment with the instrumented die is given in Section 3. In Section 4, the results of the finite element simulation are presented and compared to the experimental results.

2. FEM for powder compaction Finite element methods (FEM) have been widely used for modelling the compaction of metallic and ceramic powders [813], and have recently been used for analysing pharmaceutical powder compaction [14,15]. In FEM, powders are modelled as continuum media and the compaction behaviour is analysed by solving boundary value problems, i.e. partial differential equations representing balance laws of mass, energy and momentum and constitutive laws, such as stressstrain relations and diewall friction. Powders are generally assumed to be elastic plastic materials with appropriate yield surfaces to represent the yield behaviour of the materials. There is a collection of phenomenological models for yield surfaces of powdered materials, such as the DruckerPrager Cap (DPC) model [16], the CamClay model [17] and the DiMaggioSandler model [18]. All of these models were originally developed for application in soil mechanics and geotechnique and have been used to simulate the compaction for powder metallurgy and ceramic powders by various researchers [815]. PM Modnet computer modelling group (organised by EPMA) have recently conducted comparative studies with different research groups using these models independently. They have shown that all these models produce reasonable results and can predict the packing density within 50 kg m3 and the tooling force within 10% for an iron powder. Among these models, the most widely used yield surface in modelling powder compaction is the DruckerPrager Cap (DPC) model [8,9,12,14,15], not only because it can more reasonably represent both the shear failure and the plastic yielding of granular materials (see below), but also because it can be readily characterised with experiments on real powders [8,9,12,14,15]. Therefore, it is also chosen in this study. A detailed description of the DPC model will be presented first in Section 2.1, followed by the method to determine the die-wall friction.

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

109

2.1. DruckerPrager Cap model The DruckerPragerCap model was originally developed to predict the plastic deformation of soils under compression [16]. It consists principally of two intersecting segments: a shear failure segment F s and a cap segment F c. A transition segment F t has been introduced to provide a smooth surface [19]. For the uniaxial compaction considered in this study, these segments are defined in the p q plane (see Fig. 1) in terms of two stress invariants: the equivalent pressure stress p and the Mises equivalent stress q , where p 1 r1 r2 r3 3 i 1h r1 r2 2 r2 r3 2 r3 r1 2 6 1 2

in . This dependent upon the volumetric inelastic strain e vol dependence controls the hardening or softening of the cap segment; volumetric plastic compaction results in hardening, while volumetric plastic dilation develops softening. In order to ensure that the primary feature of the DPC model is not significantly modified by the introduction of the transition segment for the sake of numerical implementation, the transition segment is always relatively small by restricting the parameter a with typical values of 0.01 to 0.05. The transition segment is given by: s !2   a 2 F t p; t p pa t 1 d pa tanb cosb

ad pa tanb 0

q2

where r i (i =1, 2, 3) are the principal stresses. The shear failure segment in the DPC model provides a criterion for the occurrence of shear flow, which is dependent upon the cohesion d and the angle of friction b of the granular materials according to the Mohr Coulomb hypothesis: Fs p; t t ptanb d 0 3

Full calibration of the model described above requires triaxial test data. However, the model can also be calibrated using experimental data from an instrumented die with radial stress sensors, if an appropriate model is assumed for die wall friction. The experimental techniques and the calibrating method will be presented in Section 3. 2.2. Determination of die-wall friction The friction between the powder bed and the die wall plays an important role during the processing of powder compacts. Accurate determination of die wall friction is thus vital in modelling the powder compaction. Two different techniques have been developed to measure the die wall friction: the shear plate method and instrumented die measurement [10]. It has been argued that the calibration test should be close to the actual process condition to represent the interaction behaviour of powder bed during compaction [13]. Therefore, compaction testing with an instrumented die was performed to determine the die wall friction in this study. The testing data are analysed based on JanssenWalker theory [14,20]. In this theory, a cylindrical powder bed is considered (see Fig. 2). By assuming the vertical stresses are uniformly distributed on horizontal cross sections, force equilibrium for an arbitrary elemental slice dz at a distance z from the bottom of the powder bed gives: pD2 pD2 dz g sz pDdz d rz q 4 4 7

The cap segment which intersects the equivalent pressure stress axis is an elliptical curve with constant eccentricity in the pt plane as given by: s  2 Rt 2 Fc p; t p pa 1 a a=cosb Rd pa tanb 0 4

where R and a are parameters determining the shape of the cap segment and the smooth transition surface between the shear failure segment and the cap segment, respectively. p a is an evolution parameter representing the hardening or softening driven by the volumetric plastic strain, and is given by: pa pb Rd 1 Rtanb 5

where p b is the hydrostatic pressure yield surface that defined the position of the cap. p b is generally assumed be

where r z and s z are the axial and shear stresses at a height of z , respectively, D is the diameter of the cylindrical powder bed, q and g are the density of the powder and gravity acceleration, respectively. Since the gravitational force is relatively small compared to the compression stresses, it can hence be neglected and the second item in the left-hand side of Eq. (7) can be omitted. The radial stress on the slice dz , r r, can be related to the vertical stress, r z , by
Fig. 1. The DruckerPragerCap model in ( p , q ) plane.

rr jrz

110

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

Eq. (14) is used in this study to determine the friction coefficient l from the compaction test with an instrumented die, in which r T, r B and r r are measured with stress sensors. This is discussed further in the following section.

3. Experimental calibration of material properties Powder compaction tests were conducted using an instrumented hydraulic press (compaction simulators, ESH, Brierley Hill, West Midlands, UK). A cylindrical instrumented die of 8 mm in diameter with radial stress sensors was used. Flat-faced punches were employed to compress the powders. A variety of powders were considered and tested. A full description of the mechanical behaviour for different powders will be published elsewhere. In this paper, we will only consider lactose powder. The morphology of this powder is illustrated in Fig. 3, which shows a SEM image of lactose powder. This powder, widely used in pharmaceutical industries as a diluent, has a true density of 1548 kg m3. The powder was manually poured into a die cavity of 13 mm in depth until a small heap of powder developed above the top surface of the die. The powder above the top surface of the die was scraped off with a blade so that the die was fully occupied by loosely packed powder. The initial filling density was calculated as the total mass fed into the cavity, measured immediately after ejection, divided by the volume of die cavity. The effect of lubrication on the die wall was also investigated by using either a lubricated die or a clean die without lubrication. For the tests with lubricated die, a 1% concentration of magnesium stearate in ethanol was used and was daubed onto the die wall to form a thin layer of lubricant. The initial solid fractions (the ratio of packing density to the true density) were found to be around 0.266F0.013 for unlubricated die, and 0.277F0.014 for lubricated die, based on the average of six runs for each test. As expected, a slightly higher filling density was produced with lubrication for this particular powder. The powder was compressed with a single-ended compaction profile, in which only the upper punch was

Fig. 2. Diagram of the stresses inside an instrumented die during powder compaction.

where j is the stress transmission ratio. The shear stress due to friction acting on the slice dz can be given by Coulombs law of friction: sz lrr 9

where l is the friction coefficient. Using Eqs. (8) and (9) and neglecting the gravitational force, the equilibrium Eq. (7) can be rewritten as: d rz 4jl dz D rz 10

By assuming the friction coefficient l and the stress transmission ratio j are uniformly distributed along the vertical boundary of the powder bed, integration of Eq. (10) gives: ln rz 4jl z D rB 0 V z V h 11

where r B is the axial stress at the bottom surface of the powder bed. The axial stress at the top surface of the powder bed r T can hence be given as: ln rT 4jl h D rB 12

From Eqs. (11) and (12), the stress r z can be expressed in terms of the axial stress at the top and bottom surfaces of the powder bed as following: z  h rT rz rB 13 rB The friction coefficient l can be determined from Eq. (12) by using Eqs. (8) and (13): l D rB 4h rr  rT rB
z h

ln

rT rB

14
Fig. 3. SEM image of the lactose powder considered.

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

111

350 300 Axial stress (MPa) 250 200 150 100 50

C loading

unloading A D
0.1 0.2 0.3 Axial strain 0.4

E 0.5

0.6

Fig. 4. The variation of axial stress with axial strain during compaction.

moving downwards to compress the powder but the lower punch was stationary during the compression. A bVQ shaped compaction profile was employed and the compacted tablets were ejected immediately without any delay. The compression (loading) and decompression (unloading) speed was set to 3 mm s1. During the tests, the axial forces on the upper and lower punches were recorded with load cells of 50 kN capacity. The displacement of the upper punch was measured with a linear variable differential transformer (LVDT) displacement transducer. The radial stress was also measured with a pressure sensor of 200 MPa capacity. The relative packing density at the end of pre-loading (0.1 kN, equivalent to an axial pressure of ca. 2 MPa) was calculated based upon the volume of the gap space between upper and lower punches. The relative packing densities were around 0.68 for the unlubricated die and 0.73 for the lubricated die. It is clear that the powder beds underwent significant densification even though only very low compaction pressure was exerted. The state of the powder beds at the end of pre-loading was regarded as the starting condition for modelling. Fig. 4 shows the variation of axial stress with axial strain during the compaction. It can be seen that during loading the stress increases with the strain at an increasing rate. The stressstrain relation appears to be linear at the earlier stage of unloading, which implies that the unloading phase begins elastically. For the present study of uniaxial compaction in a cylindrical die in the z -direction, the following assumptions can be made [8,9,15]: r1 rz r2 r3 rr We hence have: p 1 rz 2rr 3 17 18 15 16

The stress path ( p , q ) for the compaction of lactose powder is shown in Fig. 5. It can be seen that during the loading phase, the stress path is essentially a straight line AB, which starts from (0, 0) (point A, in Fig. 5) until the maximum compression is reached (Point B). It is widely recognised that Point B is on the cap surface in the DruckerPrager Cap model [8,9,15]. During unloading, the stress path moves toward hydrostatic state (Point C), where r z =r r and q =0. Once the unloading reaches beyond this point, the axial stress is less than the radial stress. Since the deviatoric stress q is non-negative (Eq. (18)), the unloading continues with increasing q , until it reaches the shear failure as F s( p ,q )=0 (point D). The magnitudes of the slopes of BC and CD are essentially equal. Further unloading will be accompanied by dilatancy (DE). It is interesting to notice that the pattern of the loading and unloading paths shown in Fig. 5 is consistent with the analysis reported in [8,9,15]. Therefore we can determine the bulk modulus K and shear modulus G from the slope of BC in Fig. 4 and the slope of BC or CD in Fig. 5 as the following [8,9,15]: 4 r B rC 1 z kBC K G z C 3 eB z ez 2G qB 2 kBC p B p pC 3K 19 20

where the superscripts B and C denote the values at points B and C in Figs. 4 and 5, respectively. Knowing the bulk modulus K and shear modulus G , the Youngs modulus E and Poissons ratio v can be determined by [8,9,15] E 9GK 3K G 3K 2G 2 3K G 21

22

This gives E =3.57 GPa and v =0.12 for the lactose powder considered, which is comparable with the data reported in [15].

1 q p jrz rr j 3

Fig. 5. The variation of deviatoric stress with mean stress during compaction.

112

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

Using the values of p on the loading curve AB (in Fig. 5), the hydrostatic pressure p a at the intersection of shear failure surface and cap surface can be calculated from Eq. (4). The corresponding values of p b can then be determined using Eq. (5). Fig. 6 shows the variation of p b with ein vol for the compression tests of lactose powder. The experimental data can be better fitted with the following expression: C2 ein vol pb f ein 24 vol C1 e where C 1 and C 2 are parameters to be determined by fitting the experimental data. For the data shown in Fig. 6, we have C 1=8.118104 MPa and C 2=9.050. The die wall friction can be readily determined from Eq. (14) since all the parameters are known from the uniaxial compaction with the instrumented die. Fig. 7 shows the die wall friction as a function of the upper punch pressure for the tests. In this figure, the results for the test with a lubricated die are also superimposed for comparison. It is clear that, as expected, the die wall friction for the lubricated die is lower than that of the unlubricated die. It was also noted that die wall friction depends upon the compression pressure at very low pressures (say V40 MPa), but no unique trend is obtained for the two cases considered. For the test without lubrication, the die wall friction decreases with increasing compression pressure, which was also observed by others [14,24]. For the test with lubrication, an increase in die wall friction is observed. We attribute this to the lubricating method employed in this study, which produces a layer of magnesium stearate between the powder and die wall. At very low compression pressures, the transmitted radial stress is very low. In addition, the coated magnesium stearate layer resists the pressure being transmitted to the pressure sensors located at the die wall surface. As the compression pressure increases, the resistance of the magnesium stearate layer reduces. As a result the radial stress measured and the determined die wall friction increase. Nevertheless, essentially constant die wall frictions are obtained at higher compression pressures (say N40 MPa) for both cases. A similar feature was also observed in the

in Fig. 6. The variation of p b with e vol for lactose powder during compression.

As shown in Fig. 5, the line DE falls on the shear failure surface of the DruckerPrager Cap model. Thus, by fitting the regime DE with a straight line, the cohesion d and the friction angle b can be determined (d =0.91 kPa and b =41.028). The determination of the geometry parameters (a , R ) of the cap surface normally requires the powder compact sample to be tested under more than two different stress paths, such as hydrostatic compression, and isostatic compaction, etc. These tests may be conducted with sophisticated triaxial cells [2123], which have been widely used in powder metallurgy. The application of triaxial cells in pharmaceutical materials science is rare, mainly because pharmaceutical powders are very soft and loosely packed [14] when compared to metallic and ceramic powders. This makes accurately characterising pharmaceutical powders using triaxial cells a challenging task. Furthermore, Zeuch et al. [23] have shown that the material parameters for ceramic powders calibrated from uniaxial compression tests [8,9] are comparable with the values measured from triaxial cell testing. It appears that the geometry parameters play limited roles in the uniaxial compression of powders. Therefore, following [8,9,15], the geometry parameters have been arbitrarily chosen as a =0.03 and R =0.6. Using the stress state at point B (see Fig. 5), the cap surface can hence be defined by Eqs. (4) and (5). The DPC yield surfaces calibrated are superimposed in Fig. 5. The hardening mechanism in the DPC model is generally referred to as the evolution of the cap surface, defined by the hydrostatic yield stress p b , with the volumetric plastic strain, i.e. p b =f (e in vol). This function can be determined from Fig. 4 by assuming that the unloading responses at intermediate stages of the compression are identical so that the unloading stressstrain curves are parallel. The values of the volumetric plastic strain for any axial stress value on the compression curve AB in Fig. 4 can be determined by: ein vol ez rz K 4G=3 23

Fig. 7. Die wall friction as a function of upper punch pressure.

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

113

literature [14]. For simplicity, we have used constant die wall friction in the simulations reported in the next section. A constant die wall friction is obtained by averaging the data at compaction pressure higher than 40 MPa. For the lubricated die, we have l =0.195F0.015 while for unlubricated die l =0.483F0.007.

were terminated once the compaction pressure returned to zero.

5. Results and discussion The finite element meshes at maximum compression and at the end of compaction are given in Figs. 8b and c, respectively. It is clear that the powder bed springs back during the unloading. This is due to the recovery of elastic strain accumulated during the compression stage. To know how the powder deforms during the compaction under applied pressure, it is worth exploring the packing density under the applied pressure. We start our investigation with an examination of the variation of overall packing density of the powder bed with compression pressure (see Section 5.1), down to a detailed evaluation of the packing density distribution within a powder bed which will be presented in Section 5.2, in which the potential link of the occurrence of cracks with packing density distribution is emphasised. 5.1. Density evolution during compaction In the finite element analysis, the average relative density of the powder bed at a given compaction pressure can be determined using two methods: (1) it can be calculated from the volume of the entire bed that is a function of the displacement of the upper punch, Da m 1 V q0 25

4. Finite element analysis of the compaction process The material properties, calibrated from experiments with the instrumented die described in Section 3, were input into a commercial package, ABAQUS [19], in which the DruckerPrager Cap model was implemented. User subroutines were developed to determine the relative density distribution and the average density of the compact at various compression pressures. In this study, we analyse the same cases as the calibrating experiments with the instrumented die, in order to fully validate the simulations. The simulated powder bed was 8 mm in diameter and 5.027 mm in height, and initial relative density was set to be 0.68. The powder bed was compressed using a single-ended compression (SEC) profile. Since it is an axisymmetric case, it can be analysed with a 2-dimensional finite element model. Only half of the vertical section is discretised with four node axisymmetric continuum elements. The FE mesh before compaction is shown in Fig. 8a, which consists of 150 elements. The die wall and upper punches were modelled as rigid bodies. The interaction between the powder, die wall and upper punch was modelled by master-slave contacts with finite sliding. The friction in the contact was set to the values determined from experiments. The nodes on the symmetrical axis (AB) are restricted to move in horizontal direction and the nodes at boundary BC are only allowed to move in horizontal direction. The upper punch, as modelled as a rigid wall (AD), is only allowed to move vertically with a specified compression speed. The same compaction speed as employed in experiments was used, i.e. both loading and unloading speeds were set to 3 mm s1. The simulations

where m is the mass of the powder bed, and q 0 is the true density of the powder measured by a gas pycnometer. V is the volume of the powder bed under pressure and is determined by: V hAp 26

where h is the distance between the upper and lower punches and A p is the area of the cross section of the cylindrical powder bed.

Fig. 8. Finite element meshes: (a) before compaction, (b) at the end of compression and (c) during decompression.

114

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

Fig. 9. Compaction pressure versus relative density.

(2) The average relative density can be calculated from the relative density of each element in the powder bed, i.e.:
m X

Di Vi 27 Vi

De

i 1 m X i1

methods produce identical results in FEA. In addition, the FEA results are consistent with experimental data, except that they depart from each other at the end of decompression, whereupon the density obtained from FEA reduces faster than experimental measurements. This is due to the rapid dilation occurring inside the simulated powder bed once the shear failure takes place. Nevertheless, it is clear that FEA can capture the general feature of the compression behaviour during compaction. It should be noted that the relative density shown in Fig. 9 is based upon the in-die measurements that include recoverable elastic strains with the removal of compaction pressure, and can therefore take values larger than unity. It can be seen that during compression the relative density of the powder bed increases with the increasing compression pressure at an increasing rate. Upon the unloading of the compaction pressure, the relative density decreases. It is clear that the decreasing rate increases sharply when the compaction pressure is reduced to less than 50 MPa. This is mainly due to the dilation of the powder bed once its stress state reaches the shear failure surface as shown in Fig. 5. 5.2. Stress distribution during compaction Since Eqs. (25) and (27) give identical results to those shown in Fig. 9, Eq. (28) can hence be used to accurately determine the density of each element. The density distribution within the powder compacts can hence be produced by a contour mapping of the density of every element. The relative density distribution at the maximum compression is shown in Fig. 10. It can be seen that the density distribution is not uniform although the initial density was uniform. During compression, a high density zone is developed around the top edge of the powder

where Vi is the volume of element i , m is the total number of elements and D i is the relative density of element i and is determined from [8] D i D 0 ee v
i

28

where D 0 is the initial relative density of powder bed as it is assumed that the powder bed is uniformly distributed i before compaction. e v is the volumetric strain of the element i. The compaction pressure, P c, can also be determined by two different approaches. At first it can be calculated from the total contact force, F c, between the powder bed and the upper punch (rigid wall), Pc Fc Ap 29

Alternatively, it can be calculated by averaging the contact pressure at each contacting node, i.e.: Pn
n 1 X pi n i 1

30

where n is the number of contacting nodes. It has been found that both methods give identical results (not shown here), therefore Eq. (29) is used to calculate the compaction pressure herein. The variation of the compaction pressure with the average relative density is shown in Fig. 9, in which both FEA and experimental results are shown. It can be seen that the curves for P n versus D a , P n versus D e and P c versus D e coalesce, which indicates that different

Fig. 10. Density distribution at maximum compression.

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

115

Fig. 11. The distributions of (a) von Mises stress and (b) shear stress at instant A.

compact, while a low density zone is located in the bottom rim. The simulated density distribution is consistent with the experimental results obtained by others [2527]. The non-uniform density distribution is mainly induced by the friction along the die wall. The presence of friction inhibits the powder moving downwards as the upper punch moves downwards, and consequently the powder around the bottom edge is less compressed while

the powder around the top edge is highly compressed [28]. 5.3. Density distributions and failure patterns During unloading, the retreat of the top punch leads to the relaxation of the compressed powder bed as a result of the recovery of elastic strain. Typical powder behaviour

(a)
0mm 2.0 Z=3.570mm 6.0 8.0 0mm 2.0 Z=3.869mm 6.0 8.0 0mm 2.0 Z=4.458mm 6.0 8.0

(b)
0mm 2.0 z=4.701mm 6.0 8.0

2.0

2.0

2.0

2.0

4.0

4.0

4.0

4.0

6.0

6.0

6.0

6.0

8.0 SkyScan_1072

8.0 SkyScan_1072

8.0 SkyScan_1072

8.0 SkyScan_1072

(c)

(d)

(e)

(f)

Fig. 12. X-ray computed tomographic images of ejected lactose tablets: (a)(b) vertical slices; (c)(f) cross sections at different height.

116

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117

6. Conclusion The mechanical behaviour of lactose powder during compaction was analysed using FEM. The FE materials model and die wall friction were determined from experiments, in which a compaction simulator with an instrumented die has been used. In FE simulations, the compaction pressure and the evolution of average relative density of powder compacts were determined and compared with experimental data. It has been found that the FEA can reproduce the dependence of relative density on the compaction pressure of powder beds measured from experiments. The density distribution is also obtained from FE simulations. It has been shown that the density distribution patterns are comparable to the experimental results of others [2527]. Close examination of stress distribution during unloading reveals that there is an intensive shear band which runs from the top edge towards the mid-centre. The patterns of this intensive shear bands are identical to the failure patterns of capping observed from the experiments. It has hence been demonstrated that FEA can capture the essential aspects of the behaviour of pharmaceutical powders during compaction and can predict the possible failure patterns, such as capping, during tabletting.

Fig. 13. Photography of broken tablets upon ejection.

during unloading is illustrated in Fig. 11, which shows the von Mises stress and shear stress distributions at instant A (see Fig. 9). It is clear that the effective stress at the top is generally lower than that at the bottom of the compressed powder bed, which is consistent with the analysis of Michrafy et al. [15]. This indicates that the powder near the top will generally relax more than that at the bottom. Close examination of the shear stress distribution (see Fig. 11b) reveals that there is an intensive shear band which runs from the top edge towards the mid-centre. Within this shear band the shear stress changes from positive to negative, indicating the change in the direction of the shear stress. Shear deformation in granular materials generally induces dilation of the materials. Consequently, the strength in this zone is relatively low. This is demonstrated by the experimental observation of the failure patterns for the tablets produced in this study, as shown in Figs. 12 and 13. Fig. 12 shows the X-ray tomographic images of the ejected tablet. It is clear that cracks develop from the top edge towards the bottom centre, similar to the pattern of shear band shown in Fig. 11b. It can also be seen that the main crack is essentially axisymmetric and gradually shrinks towards the bottom centre. This is further illustrated in Fig. 13, which shows the photography of three broken tablets upon ejection in the same series of tests. It is clear that the compacts are broken into two parts: a top cap of an up-side-down cone shape and a base with a concave dip. It has been noticed that the failure patterns shown in Figs. 12 and 13 are a common feature during the tests with this particular powder, even for those tests with the lubricated die. This is not surprising as the same density distribution patterns are produced with different die wall frictions, even though high friction amplifies the density gradients [28]. By comparing the shear stress distribution during unloading shown in Fig. 11b with the failure patterns shown in Figs. 12 and 13, it suggests that the diagonal shear band running from the top edge towards the mid-centre (Fig. 11b) is believed to be responsible for the occurrence of cracks. FEA can hence give an indication of the possible failure modes during the processing of powder compacts.

Acknowledgement This work is supported by Pfizer Global Research and Development.

References
[1] C.Y. Wu, L. Dihoru, A.C.F. Cocks, The flow of powder into simple and stepped dies, Powder Technology 134 (2003) 24 39. [2] C.Y. Wu, A.C.F. Cocks, O.T. Gillia, Die filling and powder transfer, International Journal of Powder Metallurgy 39 (2003) 51 64. [3] G. Alderborn, C. Nystrom, Pharmaceutical Powder Compaction Technology, Marcel Dekker Inc., New York, 1996. [4] R.W. Heckel, Densitypressure relationship in Powder Compaction, Transactions of the Metallurgical Society of AIME 221 (1961) 671 675. [5] R.W. Heckel, An analysis of powder compaction phenomena, Transactions of the Metallurgical Society of AIME 221 (1961) 1001 1008. [6] K. Kawakita, K.H. Ludde, Some considerations on powder compression equations, Powder Technology 4 (1970/71) 61 68. [7] J.T. Carstensen, J.M. Geoffroy, C. Dellamonica, Compression characteristics of binary mixtures, Powder Technology 62 (1990) 119 124. [8] I. Aydin, B.J. Briscoe, K.Y. Sanlitqrk, The internal form of compacted ceramic components: a comparison of a finite element modelling with experiment, Powder Technology 89 (1996) 239 254. [9] I. Aydin, B.J. Briscoe, N. Ozkan, Modelling of powder compaction: a review, MRS Bulletin (1997) 45 51. [10] PM MODNET Research Group, Numerical simulation of powder compaction for two multilevel ferrrous parts, including powder characterisation and experimental validation, Powder Metallurgy 45 (2002) 335 343.

C.-Y. Wu et al. / Powder Technology 152 (2005) 107117 [11] PM MODNET Computer Modelling Group, Comparison of computer models representing powder compaction process, Powder Metallurgy 42 (1999) 301 311. [12] O. Coube, H. Riedel, Numerical simulation of metal powder die compaction with special consideration of cracking, Powder Metallurgy 43 (2000) 123 131. [13] A. Zavaliangos, Consitutive models for the simulation of P/M process, The 2002 Internation Conference on Process Modelling in Powder Metallugy and Particulate Materials, Princeton, NJ, MPIF, 2002. [14] I.C. Sinka, J.C. Cunningham, A. Zavaliangos, The effect of wall friction in the compaction of pharmaceutical tablets with curved faces: a validation study of the DruckerPrager cap model, Powder Technology 133 (2003) 33 43. [15] A. Michrafy, D. Ringenbacher, P. Techoreloff, Modelling the compaction behaviour of powders: application to pharmaceutical powders, Powder Technology 127 (2002) 257 266. [16] D.C. Drucker, W. Prager, Soil mechanics and plastic analysis or limit design, Quarterly of Applied Mathematics 10 (1952) 157 165. [17] A.N. Schofield, C.P. Wroth, Critical State Solid Mechanics, McGrawHill, London, 1968. [18] F.L. DiMaggio, I.S. Sandler, Material Model for Granular Soils, Journal of Engineering Mechanics, ASCE 96 (1971) 935 950. [19] ABAQUS Theory Manual, Ver. 6.2, HKS Inc., 2001. [20] R.M. Nedderman, Statics and Kinematics of Granular Materials, Cambridge University Press, New York, 1992.

117

[21] L.C.R. Schneider, A.C.F. Cocks, Experimental investigation of yield behaviour of metal powder compacts, Powder Metallurgy 45 (2002) 237 245. [22] E. Pavier, P. Doremus, Triaxial characterisation of iron powder behaviour, Powder Metallurgy 42 (1999) 345 352. [23] D.H. Zeuch, J.M. Grazier, J.G. Arguello, K.G. Ewsuk, Mechanical properties and shear failure surfaces for two alumina powders in triaxial compression, Journal of Materials Science 36 (2001) 2911 2924. [24] B.J. Briscoe, S.L. Rough, The effects of wall friction on the ejection of pressed ceramic parts, Powder Technology 99 (1998) 228 233. [25] D. Train, Transmission of forces through a powder mass during the process of pelleting, Transactions of the Institution of Chemical Engineers 35 (1957) 258 266. [26] H.S. Kim, Densification modelling for nanocrystalline metallic powders, Journal of Materials Processing Technology 140 (2003) 401 406. [27] B. Eiliazadeh, K. Pitt, B. Briscoe, Effects of punch geometry on powder movement during pharmaceutical tabletting processes, International Journal of Solids and Structures 41 (2004) 5967 5977. [28] C.Y. Wu, J.A. Elliott, A.C. Bentham, S.M. Best, B.C. Hancock, W. Bonfield, A numerical study on the mechanical behaviour of pharmaceutical powders, Proc. Int. Conf. on Pharmaceutics, Biopharmaceutics and Pharmaceutical Technology, 15th18th March, Nuremberg, Germany, 2004, pp. 17 18.