(391d) Nonlinear Model Predictive Control of Ammonia Synthesis and Separation Process Using Integrated Surrogate Modeling

Ammonia is the second most produced chemical globally, playing a critical role in agriculture and emerging as a carbon-free hydrogen carrier and energy storage vector. Conventional production of ammonia—primarily the high-pressure Haber-Bosch process followed by absorption-based separation—remains highly energy-intensive [1-4]. As industrial sectors transition toward energy efficiency and integration, the demand for real-time optimization and control strategies continues to grow [5]. This work presents a nonlinear model predictive control (NMPC) framework for integrated ammonia synthesis and adsorption. Although high-fidelity multiscale and multiphysics models accurately describe the thermochemical behavior of the system, their computational complexity renders them impractical for real-time control applications. An integrated surrogate modeling approach is developed using time-series deep learning to enable tractable and data-efficient optimization.

The framework comprises two interconnected surrogate models: one for the ammonia synthesis reactor and another for the adsorption-based separation unit. Each model is built using long short-term memory networks, trained on data generated through high-resolution multiphysics simulations using COMSOL [5,6]. The reactor model predicts key dynamics such as temperature, pressure, and ammonia composition, while the absorption model captures separation performance under varying inlet conditions. A critical component of this approach is the coupling mechanism between the two subsystems. Outputs from the synthesis reactor model serve as inputs to the absorption model. This integration maintains the interdependence between subsystems and enables the absorption model to reflect the real-time influence of upstream variations, preserving multiscale and cross-unit interactions. The integrated surrogate system is the foundation for the NMPC strategy, which dynamically regulates operating conditions across both synthesis and absorption stages. By leveraging data-driven models trained on physics-based simulations, the proposed control framework enables the regulation of the ammonia synthesis and separation dynamics with significantly reduced computational overhead.

References:

[1] M. Elbaz, S. Wang, T. F. Guiberti, and W. L. Roberts. Review on the recent advances on ammonia combustion from the fundamentals to the applications. Fuel Communications, 10:100053, 2022.

[2] R. MacFarlane, P. V. Cherepanov, J. Choi, B. H. R. Suryanto, R. Y. Hodgetts, J. M. Bakker, F. M. Ferrero Vallana, and A. N. Simonov. A roadmap to the ammonia economy. Joule, 4(6):1186-1205, 2020.

[3] Humphreys, R. Lan, and S. Tao. Development and recent progress on ammonia synthesis catalysts for Haber–Bosch process. Advanced Energy and Sustainability Research, 2:2000043, 2021.

[4] Aparicio and J. A. Dumesic, Ammonia synthesis kinetics: Surface chemistry, rate expressions and kinetic analysis. Topics in Catalysis, 1:233–252,1994.

[5] O. Cabral, A. Bagheri, and D. B. Pourkargar. Learning-based model reduction and predictive control of an ammonia synthesis process. Industrial and Engineering Chemistry Research, 63(23):10325-10342, 2024.

[6] Bagheri, T. O. Cabral, and D. B. Pourkargar. Integrated learning‐based estimation and nonlinear predictive control of an ammonia synthesis reactor. AIChE Journal, 71(5): e18732, 2024.