Early-stage chemical process screening through hybrid modeling: Introduction and case study of a reaction–crystallization process

Late-stage development of complex chemical processes presents significant challenges due to the high dimensionality and interactions of operating parameters. This complexity renders traditional factorial experimental designs impractical. Consequently, there is often a default reliance on suboptimal legacy technologies, which can lead to reduced overall performance and a larger environmental footprint. This work introduces a novel integrated methodology for combined process and product attribute screening specifically designed to overcome these limitations. The approach strategically integrates expert knowledge, high-fidelity first-principle modeling, and data mining techniques to accelerate the generation of critical process understanding. This supports the confident adoption of sustainable high-performance manufacturing routes. The sequential framework begins with expert knowledge to define promising technological pathways, which are then modeled using first-principle approaches, potentially enhanced by contemporary Artificial Intelligence (AI) techniques. Afterward, extensive parametric optimizations are performed, generating rich synthetic datasets. These data are then subjected to data mining algorithms for pattern recognition, identification of different clusters of the operational regime, and estimation of key product properties. The effectiveness of this methodology is demonstrated through a challenging case study that focuses on the crystallization of conglomerates, which combines deracemization and particle formation, steps traditionally performed sequentially with associated inefficiencies. Our analysis reveals that optimal operations form 12 distinct clusters within which the expected product properties can vary considerably. A key finding is that incorporating data from a strategically designed preliminary experiment enables the exclusion of difficult-to-measure material-specific parameters and enhances the cluster classification and product property estimation.