Scale-up of High-Shear Mixer Granulators †

The mechanics of particle interactions and the prevailing level of compressive stresses and shear strains are affected by the scale of operation, which in turn affects the granule structure, strength and functional properties. This may be the main reason why the current scaling relationships are ineffective from a viewpoint of product engineering. In a research programme supported by the EPSRC and four industrial organizations, i.e. Borax Europe, Hosokawa Micron BV, Pfizer Global Research and Development, and Procter and Gamble, we have addressed the following topics: • Development of methodologies for quantifying the structure of granules in terms of internal voidage and composition distributions, strength, shape, size and density. • Identification of the parameters that af fect the structure of granules by using fundamental theories of microscopic contact mechanics of particles using DEM, macroscopic granular flow dynamics and kinetics of wetting. • Experimental work across several length scales (1 L, 5 L, 50 L and 250 L) to aid the analysis of the process. In this paper, an overview of findings and their implications for granulation practice is presented.


Introduction
Granulation is a key unit operation in many industrial sectors for manufacturing a wide range of intermediates and final products such as food, fertilizers, metalliferous ores, nuclear fuels, ceramics, carbon black, catalysts, pesticides, plastics, cement kiln feeds and detergents.The objective of the granulation process is essentially to improve material properties and behaviour such as flow, handling, dustiness, strength, appearance, structure and composition, rate of dissolution and resistance to segregation [1][2][3][4] . Ganula-tors used in industry can be widely divided into two categories: low-shear mixer granulators (e.g.rotating drums and pans), and high-shear mixer granulators.High-shear mixer granulators are typically found in the pharmaceutical and detergent industries, and are capable of reducing the processing time and producing granules with high strength and density.One of the most important challenges of granulation technology is process scale-up [5][6][7][8] .The ultimate goal in scale-up is to keep product properties constant.The scaling-up of granulation processes is often difficult, costly and problematic due to the complex dynamics of the process.Little effort has been made in the past to systematically study the scale-up of high-shear granulation processes and its effect on product characteristics and functionality.Most of the reports in the literature have focused on the mecha-nisms of granulation at a single scale of operation and under specific operating conditions with the particle size distribution as the product attribute of interest [9][10][11][12][13][14] .Other product properties such as the strength and structure of the granules have not so far been widely addressed in the scale-up.The literature survey shows that two scaling rules have been most frequently used, i.e. constant tip speed 15,16) and constant Froude number 17) , but more recently Tardos et al. [18][19][20] proposed a new rule based on constant shear stress.They considered the conditions for granule growth due to coalescence and granule breakage under shear deformation and related them to a critical level of prevailing shear stress, which was quantified by experimental work.They validated the proposed scaling rule with three scales of Fukae horizontal high-shear mixer granulator with capacities of 2, 7.5 and 25L 20) .The scaling rules have been mostly evaluated by comparing the granule size distribution.This is obviously incomplete, as granules of the same size can have ver y different structures hence different mechanical and physical properties.Although the characterisation of granule strength and structure has been reported by various authors [21][22][23][24][25] , there is no work on the effect of scale-up on the strength and structure.The objective of this work is to investigate the effects of impeller speed at different scales of a high-shear granulator by following the three scale-up rules of constant tip speed, constant shear stress and constant Froude number on the mechanical strength of granules.An overview of the current progress on the effect of scales of granulators on the properties of granules is presented here.The research work consists of experiments and modelling across the length scales as shown in the project flow chart in Fig. 1.In the experimental part, the granules are produced in four different high-shear mixer granulators (1, 5, 50, and 250 L Cyclomix, manufactured by Hosokawa Micron B.V.), and their properties are analysed as a function of scales in terms of strength and structure.The granule strength is determined by single granule compression and cr ushing between two platens known as the side crushing test.The macroscopic flow field of the granulators as a function of scale is also investigated using the Positron Emission Particle Tracking (PEPT) facility of the University of Birmingham [26][27][28] and high-speed video recording and image analysis to measure particle motion at the single particle level.In order to provide a mean to interpret the experimental observations, modelling work across the length scale is carried out at macroscopic and microscopic levels. The ynamics of the granulation process at the vessel scale (macroscopic level) is simulated by the Distinct Element Method (DEM) and also by computational fluid dynamics continuum (CFD) to analyse the velocity and stress

Specification of operational window of granulator of several scales
Modelling-microscopic using DEM

Validation
field of the granulator at various scales.These methods were complemented by high-speed motion analysis of particles at the boundaries, and are particularly helpful where PEPT measurements cannot be used due to the scale of the granulator.DEM simulation is also conducted for the breakage of single granules at microscopic level in a shearing bed in order to analyse the microscopic interactions of the granule behaviour.The results obtained by the PEPT technique and macroscopic modelling are evaluated with DEM to investigate the effect of the velocity field of different scales of high-shear mixer granulators on the structure of the evolved granule.The interaction map of the project is shown in Fig. 1.

Experimental set-up 2.1.1 Materials, equipment and method
Calcium carbonate powder (Durcal 65 supplied by Omya UK, Ltd.) was used as primary particles.The particle size of Durcal 65 on the volume basis was d10= 7μm and d90=225μm.The geometric mean size of the powder was 60μm 29) .A 65-wt% aqueous solution of polyethylene glycol (PEG 4000) was used as the binding agent.The liquid/solid ratio was 10 wt% and the density of primary particles was 2750 kg/m 3 .The density of the PEG solution was 1165 kg/m 3 at 27 ℃ 30) .
A type of high-shear mixer granulator, known as Cyclomix and manufactured by Hosokawa Micron B.V., was used for granulation.A schematic diagram of the granulator and its impellers is shown in Fig. 2.
The granulator has an impeller consisting of a central shaft with four sets of blades and a pair of knives on the top.The impeller is enclosed in a bowl-shape in the frustum of a cone.The granulator has a blade (knife) at the top of the bowl to cut/break the loose large granules formed in the granulator.The experiments were conducted in four granulator sizes with nominal capacities of 1, 5, 50 and 250 litres.The liquid binder was added from the top to a bed of moving powder.The sample fill level was 60% of the granulator volume.Three scale-up rules were evaluated as summarised by Equation ( 1). ( where N is the impeller revolution speed, D is the blade diameter, subscripts x and y represent different scales of granulators and n is a constant depending on the rule: (i) for constant tip speed, n = 1.0; (ii) for constant shear stress, n = 0.8; (iii) for constant Froude number, n = 0.5.The scale-up of the granulation time is based on the assumption that the granulation rate is proportional to the wall surface area of the granulator and inversely proportional to the volume of sample.During the granulation process, the wet mass is generally pushed to the periphery of the granulation bowl due to centrifugal effects when the Froude number exceeds a critical value.Granule growth and consolidation are postulated to take place at the high-impact and high-shear zones between the tips of the blades and the vicinity of the wall region of the granulator.Thus, the granulation rate increases (or processing time decreases) with increasing wall surface area.In contrast, the larger the quantity of material available for granulation, the longer the granulation processing time will be needed.Therefore, it is assumed that the granulation rate is inversely proportional to the volume of the sample.The recommended granulation time for different scales is given by Equation ( 2) The 50 L granulator is taken as the reference scale in this work.The impeller speed for the 1, 5 and 250 L granulators are established according to Equations ( 1) and ( 2) based on the conditions of the 50 L and are summarised in Table 1.As can be seen in Table 1, the granulation time increases as a function of scale.The impeller tip speed was calculated from the impeller rotation speed (N) in RPM and the impeller diameter (D).The Froude number, Fr=N 2 D/g, where g is the gravitational acceleration, is also given in Table 1 for each condition.
After the granulation process, the product was dried in an oven at 30℃ and then sieved to the size range of 125−1000μm.The granules in the desired size range of 500-600μm were then subjected to physical and mechanical properties analyses.More details of granulation methods have been reported previously 29,31) .

Strength measurement
The granule strength was characterised by the quasi-static side crushing test method 32) .Granules of the desired product size range of 500-600μm were selected randomly.They were compressed individually between two rigid plates using an Instron 5566 mechanical testing machine.In order to obtain statistically reliable results, at least 100 granules were tested per sample.Details of strength measurement have already been reported 33) .

Positron emission particle tracking (PEPT)
The positron emission particle tracking (PEPT) facility of the University of Birmingham (Birmingham, UK) was used to track particle motion.The principles of the PEPT technique and its capabilities can be found elsewhere [26][27][28] . I brief, the PEPT technique makes use of a single radioactive tracer that carries positrons.Positrons annihilate with local electrons, resulting in the emission of back-to-back 511 keV γ-rays.Detection of the pairs ofγ-rays enables the tracer location to be found as a function of time by triangulation.
In this very initial attempt to study the scale-up of high-shear mixer granulators, the solids motion in two scales of 1 and 5 L were investigated and compared by using the positron emission particle tracking (PEPT) technique.The granular materials used were calcium carbonate (Durcal 65).While keeping all other operating conditions identical, the operating shaft speeds were set to follow the constant tip speed criterion, which corresponded to the top impeller tip speed of 4.1 ms-1 for both machines.In a typical experiment, calcium carbonate particles were loaded into the vessel of the mixer granulator, which was then started and run for a couple of minutes to ensure that the steady-state was reached before starting the data acquisition process.Resin beads of 60μm diameter and 1050 kg/m 3 density were used as tracers, which were activated by an ion exchange method with radioactive water produced in a cyclotron [26][27][28] .
The data acquisition was performed for 12 minutes for each run, which gave at least 20000 data points in the form of spatial locations in the Cartesian coordinate as a function of time.

Analysis of par ticle motion and granule breakage by distinct element method (DEM)
The distinct element method (DEM) models the interaction between constituent particles as a dynamic process, and the time evolution of the particles is advanced using an explicit finite difference scheme.The interactions between the constituent particles are based on theories of contact mechanics.More details on the methodology of DEM are given by Cundall 34) .For analysis of the macroscopic flow field of granulators, the PFC3D computer code developed by Itasca was used as it is capable of generating complicated geometries, i.e. vessel with knife and impellers.Particle motion in the granulator has been analysed to determine the granular dynamics, and in particular to identify the velocity gradient and stress field responsible for granule breakage in various scales.For granule simulation at a microscopic level, the TRUBAL code originally developed by Cundall and Strack 35) and further modified by Thornton and co-workers 36) was used.This code cannot easily handle complicated geometries, however, it used more realistic contact models for particle-particle interactions 36) , which is more suitable for the microscopic modelling of granule behaviour within a shearing bed.The granule deformation and breakage under microscopic interactions within a shearing bed was analysed in order to quantify the effect of the velocity gradient on the breakage and deformation.

Single particle crushing
The mean crushing strength of granules produced in the different scales of Cyclomix granulators under the operating conditions specified in Table 1 was reproduced by Rahmanian et al. 33) in Fig. 3.The error bars show one standard deviation.For the scales smaller than the reference scale, i.e. 1 and 5 L scales, granules with the highest mean crushing strength are those prepared at the constant tip speed, i.e. 4.13 m/s.The constant shear stress rule produces weaker granules than the constant tip speed rule.The Froude number rule produces the weakest granules in 1 and 5 L scales, but the opposite for the 250 L scale.For a given tip speed, the average crushing strength does not change much with the scale of the granulator.This is also the case, albeit to a lesser extent, for the constant shear stress rule.In contrast, the crushing strength clearly depends on the scale of manufacture for the constant Froude number case.For the given Froude number, i.e.Fr=0.347, the mean crushing strength of granules made in the 1 L scale is higher than that in the 5 L scale but less than those in the 50 and 250 L scales.It seems that the Froude number is not a suitable criterion for scaling-up high-shear granulators.This was also concluded by Hooijmajers et al. 37) , who recently used a different methodology for scale-up and a different granulator, namely a Gral high-shear granulator using lactose as the primary powder and HPC solution as the liquid binder.
The trend of the variation of strength for different scales using the constant Froude number is surprising, as it is intuitively expected that the 5 L scale should produce stronger granules as compared to the 1 L scale.This point was investigated by identifying the velocity field in the Cyclomix [38][39] . I was found that under a constant Froude number, the velocity gradient in the 1 L is higher than the 5 L scale.This produces higher shear stresses and hence stronger granules.At constant shear stress conditions, the mean crushing strength gradually increases as the Cyclomix is scaled-up.For all the samples tested, granules with the weakest and strongest crushing strengths are those produced by following the constant Froude number rule in the 5 and 250 L scale, respectively, as demonstrated in Fig. 3.The coefficient of variations, i.e. the ratio of the standard deviation to the mean crushing strength, has the lowest (0.21) and highest (0.54) values for granules made at constant tip speed and Froude number, respectively.This indicates that the constant Froude number criterion (n=0.5)produces a wider strength distribution of the granules as compared to the other two scale-up rules.The strength distribution and associated statistical analysis for these granules have already been reported 33) .It is shown that the Weibul statistical function is the best fit to the strength distribution of the granules produced under the constant tip speed rule.

Positron Emission Particle Tracking (PEPT)
The flow fields obtained by PEPT for both scales are shown in Fig. 4. The velocity fields obtained for the two scales are different.There are two clockwise swirl flows bounded at the top impeller in Cyclomix 5L, while there is one clockwise and one anti-clockwise swirl flow in Cyclomix 1L.Unlike the stagnant bed in Cyclomix 1L, the lower swirl in Cyclomix 5L involves the entire bed below the top impeller.The results imply that the scale of the toroidal flow structure is restricted by the limited space in the Cyclomix 1L.The overall velocity distributions in axial, radial and tangential directions are shown in Fig. 5.The difference in velocities can be seen clearly from Fig. 5(a), which shows that Cyclomix 5L has a higher average magnitude in tangential direction.This is due to the larger area of highly agitated region around the top impeller (Fig. 4(b)).The maximum tangential velocities of both machine scales are close to each other (Fig. 5(b)) as a result of constant tip speed.
The axial, radial and tangential velocities are compared quantitatively at elevations of geometrical similarity (Fig. 6).The results are compared at 70 and 105 mm, which corresponds to the height of the third impeller in Cyclomix 1L and 5L, respectively.Generally, the radial and tangential velocities are greater while the axial velocity magnitude is less for Cyclomix 1L as compared to that of Cyclomix 5L.The trends of axial and tangential velocity are broadly similar.The correlation coefficients between the velocity distributions of different scales in Figs.7(a-c) are 0.893, 0.6152 and -0.796, respectively.The results indicate that a similar flow field can be obtained at different machine scales by using the constant tip speed criterion while keeping other conditions identical.Fig. 3 Mean crushing strength of the granule as a function of scale-up for different scale-up criteria, after Rahmanian et al. 33) .
In both scales, the granular material forms a thick layer attached to the vessel wall (Fig. 7).The normalised tracer occurrence for Cyclomix 1L and Cyclomix 5L is similar, indicating that both machine scales have similar dispersion characteristics with constant tip speed criterion.
In addition to the flow structure, the vertical circu-lation period and number of circulation are used to compare the vertical mixing power between the two scales.The upper boundary of Cyclomix 5L is about 10 mm above the top impellers, and the lower boundary is at the level of the second impellers.The selection of the boundaries of the Cyclomix 1L follows the geometrical similarity where the upper boundary is chosen to be 10 mm above the top impellers (Y120) and the lower boundary to be the level of the second impellers (Y43).There is no obvious difference between the Cyclomix 1L and 5L in the vertical circulation period and number of cycles (Fig. 8), although the number of cycles in Cyclomix 1L is slightly smaller than that of Cyclomix 5L.The results imply that, in so far as the mixing is concerned, geometrically similar mixer granulators can be scaled by applying the constant tip speed criterion.This is attributed to the finding that the constant tip speed criterion gives comparable vertical mixing, both in period and number of cycles.

DEM modelling 3.3.1 Macroscopic flow field of granulator
For simulation of the macroscopic flow field of the granulator, the vessel and the impeller were geometrically generated using the PFC3D DEM code (Fig. 9).Particles with the properties given in Table 2 were generated randomly within the vessel.Gravity was then introduced to settle the assembly.The current computer power is unable to simulate the actual number of particles inside the granulator in reasonable computing time.Smaller numbers of particles, but of greater size, were therefore used for the simulation work.The approach here is to take the largest number of particles which can be simulated in a reasonable time and attribute to them a density (smaller than true density) such that the mass of the assembly is equal to the mass of fine powder (Table 3).More details on the method can be found in Hassanpour et al. 39) .
After generating a stable assembly of particles in terms of contact and coordination number, the impeller was rotated at various rotational speeds according to the conditions of scale-up (n = 0.5, 0.8 and 1 as defined in Table 1.A snapshot of the velocity field of particles during agitation (7.47 Hz) is shown for 1 L granulator in Fig. 10 as an example.The velocity field obtained from DEM analysis is qualitatively compared with PEPT measurements (Fig. 7a).
The shear stresses within the granulator in various regions were calculated for various granulator sizes and all scale-up rules.The results were compared to identify how the shear stress changes as a function of granulator size and scale-up rule.As an example, the results of variation of shear stress for 1 L granulators are shown in Fig. 11.In this figure, it can be seen that at the beginning of agitation, the highest shear stresses are in bottom part of the vessel (region 4).This is due to the stationary mass of powder sitting above region 4 which creates a high pressure.The stress in region 4 decreases rapidly as the bed of powder becomes dynamic.After a certain period of time (around 0.2 s) the shear stresses within all regions have peaks, corresponding to the times when the impellers pass through the stress calculation area.From Fig. 11, it can be seen that the highest level of shear stress exists in region 2, where the top impeller is placed.
The qualitative agreement between PEPT and     Fig. 9 Geometry of the Cyclomix granulator generated using DEM, after Hassanpour et al., 39) .
DEM indicates that DEM analysis provides useful information on the macroscopic flow and shear stress field of the granulator.This approach can be used for analyses where PEPT measurement cannot be carried out due to the large scale of the granulator (larger than 5 L).

Microscopic interactions
In order to simulate the behaviour of granules at microscopic level during shearing, a small segment of the granulator is simulated.A granule of about 1 mm in diameter is embedded in a cubic assembly of primary particles (Fig. 12).The granule was made up of 500 primary particles with the properties given in Table 4.The assembly was then subjected to shear deformation by superimposing a velocity profile on the particles as shown schematically in Fig. 12.The results of the velocity gradient obtained from PEPT analysis (Fig. 4) are used as boundar y conditions for superimposing a velocity profile on the particles (Fig. 12).During shearing, the normal pressure was kept constant by allowing movement of the horizontal platen on the bed.The evolution of shear stress and behaviour of the granule is then monitored.More details on the method of creation of the granule and the assembly can be found in Hassanpour et al., 25,39) .
In this simulation, the shear stress on the granule is calculated from the principal stresses on the granule.As an example, the results for the two velocity fields corresponding to 1 and 5 L granulator scales (scaled-up according to n=0.8) are shown in Fig. 13.
It can be seen that for the case where the conditions of 1 L are used, the shear stresses are larger than Fig. 12 The granule within an assembly of primary particles and the velocity profile superimposed on the surrounding particles for shearing of assembly.

L
Fig. 13 Simulation of evolution of shear stress on the granule during shearing according to different conditions, after Hassanpour et al., 39) .The elongation factor expressed as the ratio between the maximum (L in Fig. 14) to the minimum dimensions of the granule (W in Fig. 14) and the packing fraction have been calculated for both granules and the results are shown in Table 4.It can be seen that for the case where the conditions of the 1 L granulator are used, the granule is more elongated and has a smaller packing fraction.According to the models of Rumpf 40) and Kendall 41) , granules sheared with the conditions of 1 L would have a lower strength.This agrees well with the experimental results obtained in Fig. 2, where granules produced in 1 L have a lower strength if the granulator is scaled up using the condition of constant shear stress (n=0.8).

Conclusions
The granulation process operated in several scales of Cyclomix high-shear granulators is influenced by different scale-up rules.Three different scale-up rules of constant tip speed, constant shear stress and constant Froude number have been evaluated.The effect of scale of operation on the granules' strength distribution, characterised by side crushing tests, was investigated.
The constant tip speed rule produces agglomerates of comparable strength.The difference in agglomerate strength distribution produced in different scales becomes progressively larger for the constant shear stress and constant Froude number rules, respectively.Hence, it can be concluded that the constant Froude number rule is unsuitable for the scale-up of high-shear granulators in so far as the interest is on keeping the crushing strength constant.However, when the conditions of constant shear stress are used to scale up the granulator, the granules produced in 1 L are weaker than those of 5 L. This behaviour has been examined by detailed work to characterise the velocity field and stress fields of granules in the granulator by DEM and PEPT to provide a better insight into the flow and stress profiles as a function of equipment scale.PEPT analysis shows that the macroscopic flow fields of 1 L and 5 L granulators are different, which could affect the final strength of the granules.PEPT results were used in DEM for modelling the microscopic interactions of granules within a shearing bed.The simulation results confirm that the flow conditions of the 1 L granulator make the granule more elongated and with a lower packing fraction, indicating that the granule in 1 L is weaker, a trend corroborated by experimental data.

Fig. 1
Fig. 1 Flow chart of the project.

Fig. 14 .
Fig.14.The granule within an assembly after shearing according to the conditions of the granulator, after Hassanpour et al.,39) .5 L 1 L BSc (Sharif University, Tehran) PhD (Leeds University) is currently a postdoctoral research fellow at the Institute of Particle Science and Engineering at the University of Leeds.His research activities include the use of the distinct element method (DEM) for the simulation of granular media, mechanical characterisation of particles, size reduction (breakage and attrition of the particles) and size enlargement (granulation of powders), scale-up of particulate processes and flowability and dispersion behaviour of cohesive powders.Yulong Ding Professor Yulong Ding BEng MSc PhD became a professor of Nanoparticle Engineering with the Institute of Particle Science and Engineering and Chemical Engineering Discipline at the University of Leeds after previous employment with the Imperial College London, University of Birmingham and University of Science and Technology (Beijing).He has been working on both the experimental aspects and mathematical modelling of particulate and multiphase reacting systems on both small (nano and micro) and large (meso and macro) scales over the past decade.Professor Ding's recent research has focused on two areas of nanofluids for thermal management and multiphase transport phenomena associated with particle classification and characterisation and separation enhanced reaction processes.His research has led to over 180 technical papers, 8 book chapters and 9 patents.

Table 1
Operating condition of different scales of Cyclomix

Table 2
The properties of particles and walls used in the simulation of the macroscopic flow field of a 1 L granulator

Table 3
Number and size of particles used in the simulations of different granulator scales

Table 4
The properties of particles and walls used in the simulation of microscopic interaction This means that the larger velocity gradient would create larger shear stress on the granule.The granule behaviour for both cases of shearing conditions is shown in Fig.14.It can be seen that the granules have two different shapes after shearing.

Table 5
Properties of granules after shearing according to the conditions of the granulators