(576h) Development of a Machine Learning-Driven Surrogate Model for Steam Methane Reforming-Kinetics Towards Sustainable Hydrogen Production

The role of numerical simulation in the field of engineering - especially in design and analysis, is very significant (Ganti & Khare, 2020). High-fidelity simulation has been useful in understanding different engineering frameworks at various scales in a quantitative manner (Ganti & Khare, 2020). Nonetheless, running computational and simulation codes remains a challenging task because of the complexity of simulation models, as well as the need for utmost accuracy of such models (Saves et al., 2024). For instance, in the chemical engineering field, simulation models are based on complex engineering concepts and equations including molecular dynamics, differential algebraic equations, density functional theory, computational fluid dynamics, Navier-Stoke equations, reaction kinetics, etc. Generating analytical solutions to these equations proves difficult, and input & output relationships could be stochastic (Na et al., 2021). Hence, expressing the output of a simulation specifically as a function of the input is particularly arduous. Again, considering the huge computational cost of these simulations, the concept of running different simulations to get outputs that are consistent with each input may be impracticable. As such, it isn’t uncommon for simulations to be modeled as a black box (Ganti & Khare, 2020; Na et al., 2021).

In recent times, surrogate modeling has been identified as a reliable approach to reduce the extent and costs of computational simulations (Schröder et al., 2020), and expensive simulations aimed at finding different numerical solutions are replaced with surrogate models (Saves et al., 2024). Because of this, surrogate models have been considered as a useful tool for different tasks in the field of engineering including the quantification of uncertain parameters, exploration of design space, together with optimization (Saves et al., 2024)

With the desire for clean and sustainable energy generation, hydrogen has been considered as one of the drivers. The ability to utilize hydrogen for energy generation with no release of toxic pollutants has resulted in it being generally regarded as the future fuel (Abbas et al., 2017), with growing research interest in hydrogen as a source of energy (Jo et al., 2018). The steam reforming of methane (SRM) is the most considered technique to produce hydrogen (Laosiripojana & Assabumrungrat, 2005). However, to understand the mechanism, intricacies, conversions, as well as the direction to which a reaction proceeds in a process like SRM, the need for a kinetic model is critically important (Lao et al., 2016). Several kinetic models have been developed for the SRM process (Quirino et al., 2020), however, Xu & Froment’s model is extensively considered as the most representative of methane reforming using steam and broadly utilized in research as well as industries (Lao et al., 2016; Quirino et al., 2020; Xu & Froment, 1989).

In this work, based on the different process parameters used for the development of Xu & Froment’s kinetic model for the SRM process, the simulation of this model will be carried out. We hypothesize that the high nonlinear nature of this model would result in the computational simulation being expensive, and owing to this, a framework is proposed for the development of a machine learning-driven surrogate model that will be developed based on the data obtained from high-fidelity simulation of the Xu & Froment’s kinetic model. A fully connected feed forward artificial neural network is anticipated for the development of the surrogate model in this work. The novelty of our work stems from the fact that presently, there is no existing study on the development of a machine learning surrogate framework for the kinetic model of steam methane reforming, specifically using Xu & Froment’s reaction kinetics. The output of this work would assist in understanding how surrogate models can be used to improve the predictions of kinetic simulations, specifically towards the optimization and sustainable production of hydrogen.

References

Abbas, S. Z., Dupont, V., & Mahmud, T. (2017). Kinetics study and modelling of steam methane reforming process over a NiO/Al2O3 catalyst in an adiabatic packed bed reactor. International Journal of Hydrogen Energy, 42(5), 2889–2903. https://doi.org/10.1016/j.ijhydene.2016.11.093

Ganti, H., & Khare, P. (2020). Data-driven surrogate modeling of multiphase flows using machine learning techniques. Computers and Fluids, 211. https://doi.org/10.1016/j.compfluid.2020.104626

Jo, T., Koo, B., Lee, Y., So, H., & Lee, D. (2018). Numerical and experimental study on the thermal characteristics of a steam reformer. Journal of Mechanical Science and Technology, 32(2), 679–687. https://doi.org/10.1007/s12206-018-0115-3

Lao, L., Aguirre, A., Tran, A., Wu, Z., Durand, H., & Christofides, P. D. (2016). CFD modeling and control of a steam methane reforming reactor. Chemical Engineering Science, 148, 78–92. https://doi.org/10.1016/j.ces.2016.03.038

Laosiripojana, N., & Assabumrungrat, S. (2005). Methane steam reforming over Ni/Ce-ZrO2 catalyst: Influences of Ce-ZrO2 support on reactivity, resistance toward carbon formation, and intrinsic reaction kinetics. Applied Catalysis A: General, 290(1–2), 200–211. https://doi.org/10.1016/j.apcata.2005.05.026

Na, J., Bak, J. H., & Sahinidis, N. V. (2021). Efficient Bayesian inference using adversarial machine learning and low-complexity surrogate models. Computers and Chemical Engineering, 151. https://doi.org/10.1016/j.compchemeng.2021.107322

Quirino, P. P. S., Amaral, A., Pontes, K. V., Rossi, F., & Manenti, F. (2020). Modeling and Simulation of an Industrial Top-Fired Methane Steam Reforming Unit. Industrial and Engineering Chemistry Research, 59(24), 11250–11264. https://doi.org/10.1021/acs.iecr.0c00456

Saves, P., Lafage, R., Bartoli, N., Diouane, Y., Bussemaker, J., Lefebvre, T., Hwang, J. T., Morlier, J., & Martins, J. R. R. A. (2024). SMT 2.0: A Surrogate Modeling Toolbox with a focus on hierarchical and mixed variables Gaussian processes. Advances in Engineering Software, 188. https://doi.org/10.1016/j.advengsoft.2023.103571

Schröder, L., Krasimirov Dimitrov, N., & Robert Verelst, D. (2020). A surrogate model approach for associating wind farm load variations with turbine failures. Wind Energy Science, 5(3), 1007–1022. https://doi.org/10.5194/wes-5-1007-2020

Xu, J., & Froment, G. F. (1989). Methane steam reforming, methanation and water‐gas shift: I. Intrinsic kinetics. AIChE Journal, 35(1), 88–96. https://doi.org/10.1002/aic.690350109