Impact of powder characteristics on a particle granulation model

• A granulation system is investigated with a combined experimental and modelling approach.
• Two characteristics of the initial powder are found to have a significant impact.
• Methods that use single and multivariate model results with experimental data are demonstrated.
• Redefining fixed parameter values as random variables is used to good effect with quasi-randomly generated sequences.

In this paper we present a combined experimental and modelling approach to understanding the wet granulation of lactose powder in a high-shear mixer and perform a sensitivity study of the model. Experimental data is produced by performing nine granulation runs using lactose monohydrate as the initial powder and deionised water as the binder. The granulation runs were performed with variations in impeller speed, massing time and binder addition rate. The granulation process is then simulated by a population balance model published by Braumann et al. (2007) (Chemical Engineering Science, 62, 4717–4728). The model contains five rate parameters requiring estimation: coagulation, compaction, attrition, penetration and chemical reaction. The rates are estimated by sampling with Sobol sequences over a pre-defined parameter space. A sensitivity study reveals two important properties. First, the model input value that quantifies the height of the asperities on the particles is found to limit the model's ability to simulate even simple characterisations of the particle ensemble. However, by allowing the parameter for the height of asperities to vary over a range while estimating the rates, the simulated particle size distribution demonstrates agreement with the experimental one when using a single value characterisation. Second, the input parameters which describe the initial particle size distribution are found to significantly affect the distribution of the end product. When the input parameters which define the initial powder are allowed to vary, the model demonstrates an ability to simulate the experimental empirical size distributions.