(12f) A Deep Learning-Based Model Reduction and Control of an Ammonia Synthesis Process

The large-scale production of ammonia from nitrogen and hydrogen through the Haber-Bosch process is one of the most significant achievements in the chemical process industry. Ammonia is the primary precursor of fertilizers, a refrigerant gas, and is used to produce plastics, explosives, textiles, pesticides, and dyes [1, 2]. Furthermore, ammonia is considered an advantageous chemical-energy currency in integrated and sustainable polygeneration systems due to its easy storage, transportation, and conversion to other value-added chemicals [3,4]. In ammonia synthesis, the nitrogen and hydrogen are combined on the surface of a metal-based catalyst in a high-pressure-temperature reactor [5-8]. Managing such a reactor operation hinges on accurate descriptions of the elementary steps involved in the reaction and control approaches to direct reactions toward lower energy and more selective pathways.

Due to the much larger gradients in fluid flow, temperature, and concentration fields at the macroscopic reactor scale compared to the spatial heterogeneity of the properties at the microscale of catalysts, the reaction rates computed at the smaller length scales are only applicable over a specific discretization size. Therefore, the rates need to be re-evaluated at different regions of the macroscopic domain to account for changes in temperature and species concentration. The macroscopic behavior is then determined by integrating the transport-reaction spatiotemporal dynamics over the processes domain. Consequently, only an actual multiscale model can predict such intimate phenomena by linking microscopic properties to macroscopic process variables that can be measured and readily manipulated. Furthermore, a multiscale interpretation of transport-catalytic ammonia synthesis entails developing high-fidelity and computationally tractable fundamental models that can accurately predict the spatiotemporal dynamics of the system, enabling mutual information exchange through different scales. While the multiscale model improves the understanding of the ammonia synthesis process and the accuracy of the predictions, it poses various challenges, from complexity to computational efficiency, which significantly affects its applicability for real-time simulation and decision-making purposes.

In this work, we develop a multiscale model for catalytic synthesis of ammonia in an industry-scale packed bed reactor and utilize it for a model predictive control (MPC) design to regulate the temperature and ammonia concentration dynamics at optimal spatiotemporal profiles. Our proposed approach integrates ammonia synthesis rate data [5-8] with a detailed macroscale reactor model (momentum, mass, and energy balances closed through appropriate constitutive expressions over the three-dimensional reactor and one-dimensional catalyst particle geometry) to describe the spatiotemporal dynamics of the reaction mixture’s pressure, temperature, and concentrations. The resulting model comprises integrated nonlinear partial differential equations that must be solved simultaneously to obtain distributions of the system properties in time. However, the MPC solution hinges on the real-time solvability of its underlying dynamic optimization problem, which is challenging for the proposed high-fidelity model. Therefore, a deep-learning-based reduced-order model representation of the system is used to approximate its spatiotemporal dynamics to address this problem. The high-fidelity multiscale simulations are performed offline over various operating conditions to collect sufficient simulation data required for deep learning model training. Finally, the resulting reduced-order model is analyzed and used as the basis for the MPC design. The controller performance is evaluated on temperature and concentration regulation problems in different operating conditions.

References:

[1] V. Babu, R. Angira. Optimal design of an auto-thermal ammonia synthesis reactor. Computers and Chemical Engineering, 29, 1041-1045, 2005.

[2] J. Humphreys, R. Lan, S. Tao. Development and Recent Progress on Ammonia Synthesis Catalysts for Haber–Bosch Process. Adv. Energy Sustainability Res., 2, 2000043, 2021.

[3] D. Mukelabai, J.M. Gillard, K. Patchigolla. A novel integration of a green power-to-ammonia to power system: Reversible solid oxide fuel cell for hydrogen and power production coupled with an ammonia synthesis unit. International Journal of Hydrogen Energy, 46, 18546-18556, 2021.

[4] H.M. Al-Hamed, I. Dincer. A novel ammonia solid oxide fuel cell-based powering system with on-board hydrogen production for clean locomotives. Energy, 220, 119771, 2021.

[5] M. Aparicio, J.A. Dumesic. Ammonia synthesis kinetics: surface chemistry, rate expressions and kinetic analysis. Topic in Catalysis, 1, 233-252, 1994.

[6] M. Malmali, Y. Wei, A. McCormick, E.L. Cussler. Ammonia Synthesis at Reduced Pressure via Reactive Separation. Ind. Eng. Chem., Res. 55, 8922-8932, 2016.

[7] J. Qian, Q. An, A. Fortunelli, R.J. Nielsen, W.A. Goddard III. Reaction Mechanism and Kinetics for Ammonia Synthesis on the Fe(111) Surface. J. Am. Chem. Soc., 140, 6288-6297, 2018.

[8] A. Dumesic, A.A. Trevino. Kinetic simulation of ammonia synthesis catalysis. Journal of Catalysis, 1, 119-129, 1989.