(117b) Modeling of multi-component granule formation in a wet granulation process

One of the outstanding problems in pharmaceutical granulation is the issue of product composition non-uniformity. The composition of granules ? in particular the relative amount of the API (active pharmaceutical ingredient) ? often reveals substantial variation across granule size classes. This may lead to a variation of API content in tablets, an undesirable and potentially dangerous situation. The aim of our work is to understand the fundamental causes of the composition non-uniformity via a combination of experiments, multi-dimensional population balance modeling, and particle-level simulations. In this paper we will address the specific problem of finding the relationship between granule composition (porosity, binder/solids ratio, drug/excipient ratio), binder properties (spreading rate and equilibrium contact angle on primary particles), and the probability of coalescence between two granules of different composition or between a granule and a primary particle. Knowledge of this relationship is a pre-requisite for multi-dimensional population balance modeling of granulation, and the testing of various hypotheses that have been proposed as the cause of composition non-uniformity, such as difference in the wettability of the API and excipient particles, or difference in their initial particle size. Using a recently developed computational methodology for the construction of ?virtual granules? by random sequential deposition of primary particles of arbitrary shape coupled with spreading and solidification of binder droplets on the particle assembly (Chem. Eng. Sci. 60, 2005, 4019-4029), we have generated populations of granules by systematically varying the binder/solids ratio and drug/excipient ratio, and the liquid binder contact angle on the primary particles of each type. For each volumetric binder/solids ratio and particle wettability, the corresponding porosity of the virtual granule was found as well as the fraction of particle surface covered by the binder and the average binder layer thickness. This information was then used to determine whether two granules of given composition will coalesce upon collision. We have conducted computational experiments of agglomerate-agglomerate and agglomerate-particle collisions, and recorded the probability of wet contact over a statistically large number of simulated collisions as function of the composition and size of the colliding agglomerates. This probability was found to be a non-linear function of the fractional surface coverage, due to effects such as steric shielding of liquid forming capillary bridges (probability of collision smaller that expected from surface coverage), or imperfect spreading of droplets on the primary particles (larger probability than expected from surface coverage). The obtained dependence of coalescence probability on granule composition was supplied as input information to a multi-dimensional population balance model to predict the composition distribution across size-classes and compared with experimental data from a pilot-plant scale fluid bed granulator.