Towards Autonomous Crystallization Systems through Advanced Process Modelling and Optimization

Promovendus/a: Gustavo Lunardon Quilló

Promotor(en): Prof. dr. ir. Jan Van Impe, De heer Satyajeet Sheetal Bhonsale, Dr. Christos Xiouras

Crystallization is a key step in making many medicines. It is the process where a substance changes from dissolved in a liquid or a disorganized solid into an ordered crystal. This transformation affects product features like particle size, crystal form, and purity, which impact how easily the material can be processed in later steps like washing, drying, and tablet-making. Despite its importance, developing crystallization processes is difficult due to complex behavior and large volumes of experimental data that are hard to interpret.

This thesis addresses these challenges by developing new methods that combine traditional scientific models, machine learning, and smart experimental design. It first improves solubility prediction by blending physics-based equations with machine learning into a hybrid model. The approach is demonstrated using ketoconazole in 2-propanol and water.

Next, the work examines how crystals grow, focusing on how different growth mechanisms occur and coexist depending on the concentration of the dissolved substance. It shows that using more accurate methods to calculate supersaturation improves understanding of crystal growth rates.

The thesis also emphasizes the importance of good experimental data. As crystallization experiments become more complicated, with varying temperature profiles and other factors, calibration and measurement techniques must keep up to ensure data quality. This helps create reliable models that reflect what is happening in the process.

Finally, it presents and solves a powerful computational method called population balance modelling. Within an optimization framework, it identifies optimal temperature and dosing strategies to produce crystals with desired properties while respecting process limits. This moves beyond trial-and-error, data-driven decisions in crystallization development.

Together, these contributions provide a flexible, efficient way to model and optimize crystallization. Though focused on pharmaceuticals, the methods also apply to other sectors and enables faster development of crystallization processes, which is an important step toward industry digitalization.