(776a) Long-Term Coating Predictions for in a Wurster Fluidized-Bed Coater with Using a Combined CFD-DEM and Markov-Chain Monte Carlo Approach

Coating of particles, in particular, microspheres, using the Wurster fluid-bed coater is a common unit operation in the pharmaceutical industry. Such coatings are typically used to layer active drugs or other ingredients that control drug release, and also for taste/odor masking and enteric barrier coatings that prevent dissolution/disintegration. To ensure control over the drug-release rate and to achieve a consistent inter-particle drug loading (potency), a high degree of coating uniformity is essential. However, relating the coating uniformity of the beads to the process parameters (such as fluidization air-flow rate, atomizing air-flow rate, Wurster-gap height, and bead size) is not trivial.

Experimental methods to characterize bead-flow dynamics inside Wurster coaters are difficult for large systems; moreover, sampling techniques are insufficient to quantify the coating uniformity of millions of beads. In such situations, coupled CFD-DEM simulations (computational-fluid dynamics coupled with discrete-element method) can model the complex particle and airflow in a fluidized bed. However, due to the high computational costs (several days per process-second for an industrial system), simulating several hours of a coating process using CFD-DEM is infeasible. Therefore, a multi-scale model is required to predict the long-term coating uniformity, based on inputs from the CFD-DEM simulations. The Markov-chain Monte-Carlo-based compartment model is one such approach to develop a reduced-order model.

The compartment model subdivides the coater into non-overlapping regions, i.e., compartments. The exchange of particles between adjoining compartments is modeled as stochastic process described by a probability matrix P, where P_ij describes the probability of a particle moving from compartment i to compartment j during some small compartment-model time step Δt_CM. This probability matrix P is calculated from the detailed CFD-DEM particle trajectories; therefore, the exact positions of the particles are not tracked in the compartment model. Changes in the process parameters affect the flow behavior of the beads which is reflected as changes in the probability-matrix entries P_ij.

Certain compartments near the spray nozzle are designated as the spray zone; every particle that visits these spray compartments gains some coating mass, based on the spray rate, particle size, and the compartment-model time step Δt_CM. The probability matrix P is then used to update the positions of the particles in various compartments. By tracking the flow of a large, representative sample of beads, the evolution of the coating mass distribution can be tracked with time - these predictions can be generated in minutes for hours of real process time using the reduced-order Markov-chain Monte Carlo method.

Results from the compartment model for a given set of process conditions were successfully validated against experimental results of particle diameters obtained using Sympatec QicPic and coating thickness from optical-coherence tomography (OCT). After validation, a virtual design of experiments (DoE) study with varying process conditions has been performed. The results show that a larger Wurster-gap height reduces the variation of the coating mass and thickness but, owing to the initial size distribution of the uncoated beads, the variation in the coating potency increases. Depending on the application, it may be more desirable to optimize the coating potency (to maintain a constant coating-mass fraction for coatings that contain active drug), or optimize the coating thickness (e.g., for protective coatings where a predefined barrier thickness is needed). The present Markov-chain Monte Carlo approach, in combination with CFD-DEM simulations, can be used to design the optimal process parameters for both applications.