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

Area of Science:

• Chemical Engineering
• Materials Science
• Pharmaceutical Manufacturing

Background:

• Solid particle agglomeration is a critical phenomenon in chemical, food, and pharmaceutical industries, impacting product quality and process efficiency.
• Controlling agglomeration is essential for optimizing the physical properties of active pharmaceutical ingredients (APIs), particularly in processes like wet granulation and agitated filter drying.
• Current understanding and predictive capabilities for agglomeration are limited due to complex particle-solvent interactions and computational challenges at industrial scales.

Purpose of the Study:

• To develop a novel computational methodology for predicting the extent of solid particle agglomeration.
• To integrate experimental measurements of agglomeration risk zones with discrete element method (DEM) for enhanced prediction.
• To build a machine learning model for predicting agglomeration extent and creating a digital twin of particulate processes.

Main Methods:

• Coupling experimental measurements of the agglomeration risk zone ('sticky zone') with discrete element method (DEM).
• Developing a machine learning model to predict agglomeration extent based on material properties and processing conditions.
• Validating the proposed computational model against experimental data.

Main Results:

• The developed computational model demonstrated good agreement with experimental results in predicting agglomeration extent.
• The machine learning model successfully predicted agglomeration based on input variables, enabling the creation of a digital twin.
• The methodology provides a pathway to better quantify, predict, and control particle agglomeration.

Conclusions:

• The novel theory and computational methodology offer a robust approach to predict and control particle agglomeration in industrial settings.
• This integrated approach, combining experimental data, DEM, and machine learning, overcomes limitations of previous methods.
• The proposed methodology is broadly applicable to various particulate processes where controlling agglomeration is crucial, including pharmaceutical manufacturing.