Our framework follows a three-step approach: (1) establishing foundational process knowledge through lab-scale experiments to build scale-independent models, (2) model validation at scale for digital process design, and (3) system model deployment for process monitoring and control. We demonstrate this methodology through a flow chemistry case study featuring a continuous process with a plug flow reactor (PFR), a mixer-settler, and a liquid-liquid extraction column.
The process is characterized by using knowledge-driven mathematical models calibrated with both laboratory and pilot-scale data. We highlight the system model value by performing Global Sensitivity Analysis (GSA) to identify critical process parameters, define robust design spaces, and maximize yield. The in-silico experiments enable thorough exploration of a design space comprising nearly 20 process parameters and raw material properties, saving a minimum of 300 experiments required from a classical development approach. Additionally, dynamic models based on Residence Time Distribution (RTD) enable optimization of divert-to-waste protocols and start-up/shut-down procedures. Finally, we showcase model-based soft-sensing capabilities within a digital twin framework that facilitates real-time monitoring, control, and continual learning throughout the product lifecycle.