(617f) Dynamic Embedded Gaussian Process Model Applied to Steam Methane Reforming

Uncertainties from reduced-order models derived from small-scale systems can impact the accuracy and reliability of predictions for large-scale systems [1]. Recent developments in machine learning (ML) techniques have opened avenues to propose new ways to build fast and accurate scalable models. For this purpose, a model selection framework for the generation of Gaussian Process (GPs) models with a Bayesian Smoothing Spline Analysis of Variance (BSS-ANOVA) using Karhunen-Loève (KL) decomposition has been proposed, known as FoKL-GPs [2]. This approach has been employed to model different scalable systems focusing on applications including kinetics, thermodynamics, and process control [1-5]. In this work, FoKL-GPs are employed to construct novel, accurate, and scalable dynamic embedded GPs, focusing on the case study of a Steam Methane Reformer (SMR). By combining a classical SMR reduced-order model and an ab initio (first-principles) model in a grey-box fashion with FoKL-GPs, a more accurate and scalable model can be achieved for online optimization and control.

The embedded GP model accounts for the mismatch between the classical kinetic SMR model proposed by Xu and Froment [6] using Langmuir equilibrium relations to eliminate adsorbed species and an ab initio reaction scheme derived from microkinetic modeling [7]. The ab initio microkinetic model implemented in this work is based on thermochemical data obtained from Density Functional Theory (DFT) and statistical thermodynamics for nickel-catalyzed steam methane reforming on Ni (111) under industrial reforming conditions (800 °C, 10 bar) [8]. This microkinetic model for the SMR reaction brings relevant information about the surface in the catalytic process, which allows the assessment of which adsorbed species have relevant or nonrelevant concentrations on the surface and how they influence concentration profiles of the gas-phase species leaving the system.

FoKL-GPs calibrate the mismatches between the ab initio and classical SMR models using a series of dynamic embedded Gaussian Process models. The calibrated models are then further incorporated into the rate of change equations of the classical model, and validation followed by upscaling are performed by comparing the concentration profiles of the embedded models in different scales against the classical and the first-principles models. These obtained models provide fast predictions for use in subsequent dynamic real-time optimization and advanced process control. Combining first-principles information within reduced-order models can accelerate the development of new technologies and algorithms in process systems engineering using accurate and reliable predictions.

References

[1] Bhat, K.S., Mebane, D.S., Mahapatra, P., Storlie, C.B., 2017. Upscaling Uncertainty with Dynamic Discrepancy for a Multi-Scale Carbon Capture System. Journal of the American Statistical Association 112, 1453–1467.

[2] Hayes, K., Fouts, M.W., Baheri, A., Mebane, D.S., 2022. Forward variable selection enables fast and accurate dynamic system identification with Karhunen-Loève decomposed Gaussian processes. e-print arXiv: 2205.13676.

[3] Ostace, A., Chen, Y.-Y., Parker, R., Mebane, D.S., Okoli, C.O., Lee, A., Tong, A., Fan, L.-S., Biegler, L.T., Burgard, A.P., Miller, D.C., Bhattacharyya, D., 2022. Kinetic model development and Bayesian uncertainty quantification for the complete reduction of Fe-based oxygen carriers with CH4, CO, and H2 for chemical looping combustion. Chemical Engineering Science 252, 117512.

[4] Mebane, D.S., Bhat, K.S., Kress, J.D., Fauth, D.J., Gray, M.L., Lee, A., Miller, D.C., 2013. Bayesian calibration of thermodynamic models for the uptake of CO2 in supported amine sorbents using ab initio priors. Phys. Chem. Chem. Phys. 15, 4355.

[5] Dinh, S., Mebane, D.S., Lima, F.V., 2023. A Framework for Implementation of Dynamic Discrepancy Reduced-Order Modeling in Advanced Process Control. Submitted for Publication.

[6] Xu, J., Froment, G.F., 1989. Methane steam reforming, methanation and water-gas shift: I. Intrinsic kinetics. AIChE J. 35, 88–96.

[7] Motagamwala, A.H., Dumesic, J.A., 2021. Microkinetic Modeling: A Tool for Rational Catalyst Design. Chem. Rev. 121, 1049–1076.

[8] Blaylock, D.W., Ogura, T., Green, W.H., Beran, G.J.O., 2009. Computational Investigation of Thermochemistry and Kinetics of Steam Methane Reforming on Ni(111) under Realistic Conditions. J. Phys. Chem. C 113, 4898–4908. https://doi.org/10.1021/jp806527q