(666e) Modeling and Multi-Objective Optimization of an Industrial Ammonia Synthesis Process | AIChE

(666e) Modeling and Multi-Objective Optimization of an Industrial Ammonia Synthesis Process


Ivanov, S., University of Western Ontario
Ammonia synthesis is a major important process in chemical industry. Ammonia application domain is very wide with major applications in manufacturing of agricultural fertilizers. Hence, it is particularly important for Canada where agricultural sector constitutes a significant part of national GDP. To keep up with continuously increasing demand, it is highly important to improve performance of existing facilities through process modeling, simulation and optimization. One of the way to achieve the goal is through mathematical modeling and optimization of operation and/or design parameters of a plant. Due to large scale of production, even minor improvements can result in significant profits or savings.

The conventional synthesis is based on Haber process - reaction of gaseous nitrogen and hydrogen over a solid catalyst to produce ammonia. The reaction is carried out at elevated temperature (400-500 °C) and pressure (80-230 atm.). The feedstock comes from a steam reforming unit at the same plant and is made of stoichiometric mixture of nitrogen and hydrogen with a minor portion of inert. The synthesis reaction occurs in one or few (usually, 1-4 beds) fixed catalytic beds. Since the process is exothermic, the vast amount of heat is generated and re-used in the process. In this work, a collaborative research was conducted with a large ammonia producing plant in Canada aiming to investigate possibilities of further improving ammonia production and heat duty in the ammonia converter. First-principle mathematical modeling of gas-solid catalytic reaction and heat transfer equipment was performed. A simple enough yet accurate model was developed to adequately simulate the ammonia converter. Main features of the model are one dimensional plug flow with Fick-type diffusion of reactant inside solid catalyst particle.

Importantly, kinetic parameters of catalyst are being the part of proprietary information, and hence, were not available. The model was first tuned with industrial data to provide estimate for catalyst kinetic parameters. For the proper estimation of kinetic parameters, a large set of industrial data was used covering time span over 12 years. Since data set is large and is full of information irrelevant to normal steady state operation, which included not only scheduled maintenance, but also short term disruptions, cold and hot start up and shutdown, sensor failures, etc., and hence, it was not evidently clear which data to use for the model validation. Hence, a data pre-treatment procedure based on data clustering around steady-states was performed to filter out relevant industrial data points. Thereafter, the model was fitted over several different time-dependent groups of data. It was discovered that even though catalyst life time for ammonia catalyst is large (> 10 years), the catalyst still experiences deactivation and it should be accounted at different stages of operation. Incorporating all the above, the model was validated over industrial data and showed good consistency with observed values. Finally, the validated model was used to perform multi-objective optimization with modified Non-Dominated Sorting Genetic Algorithm (NSGA-II). The primary goals were to maximize production of ammonia, minimize heat requirement for the reaction and minimize total catalyst loading in converter. Three optimization case studies were performed, where objectives were treated pair-wise to provide a Pareto optimal solution for converter operation.

In the first case, optimal values for three process parameters - namely, converter pressure, temperature of feed stream and ratio of quench stream to significantly reduce heat duty while keeping ammonia production above industrially acceptable level. In the second case, a different catalyst distribution among beds, was proposed to allow for higher ammonia production. Three cases with different distribution were obtained where total catalyst loading in the converter is lower, equal and higher than current. All three cases showed higher ammonia production compared to the base case. Additionally, a change in optimal conditions with respect to time-on-stream for the catalyst has been addressed and guidelines for the converter operation were provided like the first case.