A Novel Computational Approach Coupled with Machine Learning to Predict the Extent of Agglomeration in Particulate Processes

Solid particle agglomeration is a prevalent phenomenon in various processes across the chemical, food, and pharmaceutical industries. In pharmaceutical manufacturing, agglomeration is both desired in unit operations like wet granulation and undesired in unit operations such as agitated filter drying of highly potent active pharmaceutical ingredients (API). Agglomeration needs to be controlled for optimal physical properties of the API powder. Even after decades of work in the field, there is still very limited understanding of how to quantify, predict, and control the extent of agglomeration, owing to the complex interaction between the solvent and the solid particles and stochasticity imparted by mixing. Furthermore, a large size of industrial scale particulate process systems makes it computationally intractable. To overcome these challenges, we present a novel theory and computational methodology to predict the agglomeration extent by coupling the experimental measurements of agglomeration risk zone or "sticky zone" with discrete element method. The proposed model shows good agreement with experiments. Further, a machine learning model was built to predict agglomeration extent as a function of input variables, such as material properties and processing conditions, in order to build a digital twin of the unit operation. While the focus of the present study is the agglomeration of particles during industrial drying processes, the proposed methodology can be readily applied to numerous other particulate processes where agglomeration is either desired or undesired.