(442g) Mechanistic Modeling of Exothermic Hydrogenation in a Packed-Bed Reactor for Model Predictive Control

Hydrogenation is a fundamental catalytic reaction in pharmaceutical synthesis [1]. Due to safety concerns related to temperature and pressure in batch systems, continuous flow systems using packed-bed reactors have emerged as a preferable alternative [2]. While flow synthesis improves yield and reduces leaching compared to batch synthesis, it poses risks of poor hydrogen transfer, potentially reducing yield [3]. Robust process control, grounded in mechanistic understanding, is therefore essential. However, modeling continuous hydrogenation is complex, involving gas, liquid, and solid phases alongside exothermic reactions. Despite extensive efforts (e.g., [3]-[5]), models that capture dynamic behavior with high predictability and reasonable computational cost are still needed.

This study presents a workflow for mechanistic modeling of hydrogenation in a packed-bed reactor, aimed at process control applications. The model structure was implemented in Python, following the framework of PharmaPy—an open-source library for pharmaceutical process analysis [6]. Two preliminary experiments were conducted to identify critical phenomena for model inclusion. First, step tracer experiments under varying process conditions were used to determine residence time distributions (RTDs). Second, hydrogenation experiments were performed while monitoring temperature profiles at three reactor locations, as well as conversion and yield over time. Based on these insights, the final model structure was established, followed by parameter estimation and validation.

The step tracer experiments revealed that liquid feed rate significantly influenced mean residence time, more so than hydrogen pressure, hydrogen feed rate, or temperature. Péclet numbers calculated from RTD data indicated notable axial dispersion across all conditions. In the hydrogenation experiments, temperature increased along the reactor length and varied with process settings. Catalyst activity decay was negligible, as conversion and yield remained stable over time under consistent conditions. Interestingly, yield increased at both low and high liquid flow rates due to the trade-off between mass transfer and residence time. The final model incorporates heat balance and mass transfer dynamics but neglects catalyst deactivation.

Moving forward, the model will be applied to process design and control. Nonlinear model predictive control (NMPC) will be implemented to enable advanced, mechanism-informed control strategies. All functionalities will be integrated into PharmaPy, enhancing its utility for pharmaceutical manufacturing applications.

[1] D. Wang and D. Astruc, Chemical Reviews, 115(13), 6621-6686 (2015)

[2] K. Chai et al., Chemical Engineering Science, 302(B), 120901 (2025)

[3] J. Kim et al., Computers & Chemical Engineering, 156, 107541 (2022)

[4] T. Kilpiö et al., Industrial & Engineering Chemistry Research, 51, 8858-8866 (2012)

[5] H. Xue et al., Industrial & Engineering Chemistry Research, 62, 6121-6129 (2023)

[6] D. Casas-Orozco et al., Computers & Chemical Engineering, 153, 107408 (2021)