2025 AIChE Annual Meeting

(441b) Development of an End-to-End Digital Twin for the Scale-up of a Hydrothermal Synthesis Process

Authors

Olivier Gourdon, AstraZeneca
Haley Boyker, AstraZeneca
Anthony Samuel, AstraZeneca
Antonio Benedetti, GlaxoSmithKline
Managing the lifecycle of a commercial pharmaceutical product to meet increasing patient demand presents significant challenges. Traditional development approaches can be time-intensive, particularly when scaling up an active pharmaceutical ingredient (API) process from laboratory to commercial production. In this case, the scale-up of an API manufacturing process has been explored from 2 L to 20 L and ultimately to 2,000 L and 5,000 L commercial reactors, systematically investigating process parameters (PPs) to expand operational ranges. While the scale-up was largely successful, certain critical attributes exhibited variability between laboratory and commercial production. To bridge these inconsistencies, we developed an integrated, first-principles mechanistic model that unifies reactive crystallization (hydrothermal synthesis), filtration/rinsing, protonation, and drying into a single digital framework, enabling a more robust and predictive scale-up strategy.

This model captures critical phenomena in each unit operation, including colloidal silica (de)polymerization, gelling and crystal growth via Ostwald ripening and aggregation. It also predicts mass transfer during the solid–liquid protonation step under various sparging and recirculation conditions. In the final drying stage, the model combines mass and population balance equations to describe moisture evolution and particle size distribution.

By combining mechanistic descriptions with empirical data through parameter estimation, the model enables prediction of critical quality attributes (CQA) such as particle size distribution and cation exchange capacity (CEC)s. Model validation against experimental data shows predictions within acceptable error margins, confirming its accuracy and value for process development and scale-up.

Furthermore, we created a web application to democratize model usage, allowing scientists to leverage this “digital-twin” for designing more targeted experiments and reducing costly large-scale trials. The Python-based web application allows performing end-to-end “what-if-analysis” and Global Sensitivity Analysis (variance-based sensitivity analysis). Overall, this digital framework complements design-of-experiments approaches by providing a virtual DoE playground, minimizes trial-and-error experimentation, accelerates process optimization and scale-up, and advances Quality by Digital Design in pharmaceutical manufacturing.