(493d) The Effect of Particle Shape in Drum Mixers.
AIChE Annual Meeting
2017
2017 Annual Meeting
North American Mixing Forum
Mixing Scale-Up/Scale-Down Issues in Pharmaceutical and Biopharmaceuticals Processes
Wednesday, November 1, 2017 - 9:06am to 9:28am
Investigating the effect of operational parameters and variations in the properties of input particulate materials on the homogeneity of a product is a tedious and expensive task to perform experimentally. In addition, certain changes may inadvertently not be monitored and their corresponding changes mistakenly attributed to other changes. Here, computational simulation offers a cheap and attractive investigative tool to interrogate changes and consider scenarios that would be impracticable or even impossible to perform experimentally. Early investigative tools relied on Computational Fluid Dynamics (CFD) to provide insight into mixing. However, these simulations lack the resolution to consider changes at particle level on the macroscopic behaviour of the system. Modern approaches include the Discrete Element Method (DEM) that accounts for the particulate nature of materials by explicitly modeling the interaction between individual particles and thus providing insight into the effects of particle properties on the macroscopic mixing. However, the number of particles has been limited and its application to relevant industrial scale simulations hindered by the large computational demand to resolve the particle interactions millions of times per second. As a consequence, the majority of the numerical mixing studies that utilized DEM employed a number of restrictive approximations to reduce the computational burden. These include, the assumption of mono-disperse size distributions, scaling up of particle size and spherical particle shapes.
While recent studies on the GPU have been able to significantly increase the number of particles in a simulation, this has been done at the expense of sacrificing physics by simplifying force models, omitting secondary results such as attrition rates, stress and energy and assuming spherical particle shapes. The omitted secondary information is vital in understanding how operational parameters affect the final product and whether hot spots arise. A major simplification remains the simplification of particle shape as spheres. This assumptions could be overly restrictive for the pharmaceutical industry where feed powders are made from crystalline solids where the shape of the individual particles are distinctly polyhedral with significant angularity. This is significant in that the underlining dynamics of polyhedral particles is vastly different to that of spherical particles, resulting in tighter packing fractions and different segregation rates. The former is critically important in that it is the packing fraction that determines the mass of product in a given volume. In this paper we use the GPU based DEM codes Blaze-DEM and XPS to study the effect of particle shape and particle dispersion in drum mixers using multiple GPUs to simulate millions of polyhedral and spherical particles in a comparative study. We also investigate the use of machine learning algorithms in conjunction with Blaze-DEM to optimize operational parameters for given specifications.