(375z) Optimizing CO2 Utilization Efficiency through Physics-Informed Modeling of Membrane Reactors

The alarming rise in atmospheric CO₂ concentrations, from under 300 ppm in the pre-industrial age to 421.85 ppm by the end of Q1 2024, highlights the urgent need to mitigate greenhouse gas emissions [1,2]. Carbon capture and utilization emerges as a key strategy for transforming CO₂ into valuable chemicals like methane, methanol, dimethyl ether, gasoline, and synthesis gas (syngas). In this context, the reverse water-gas shift (RWGS) reaction, which converts CO₂ and hydrogen into carbon monoxide and water vapor, has received growing interest in recent years due to advancements in hydrogen production from non-fossil fuel resources. The integration of water vapor-selective membranes in RWGS reactors enables in-situ water removal which can lead to process intensification by shifting the equilibrium reaction forward, resulting in increased CO₂ conversion and CO yield, as well as reduced operating temperatures, in accordance with Le Chatelier's principle [3]. Yet, accurately modeling the RWGS process in membrane reactors presents significant challenges due to the inherent complexities of a reversible reaction system with in-situ water removal. Mechanistic modeling of this intensified dynamic process is mathematically challenging due to high nonlinearity, high dimensionality, and ill-conditioning arising from the coupled reaction and membrane transport phenomena.

Motivated by this, we employ physics-informed neural networks (PINNs) as a hybrid modeling approach, to combine the predictive accuracy of machine learning with the robust understanding of first-principles models to simulate this intensified process. This approach allows us to avoid the complex calculations typically involved in solving highly nonlinear ordinary differential equations (ODEs), lowering the computational cost while preserving the model accuracy and validity [4]. By leveraging a custom loss function, PINNs not only fit the observed data but also adhere to the underlying physics of the problem, demonstrating exceptional predictive performance in blind testing (R² > 0.95). The validated PINN model serves as a membrane reactor simulator in the data-driven optimization process, where various optimizers are employed to refine operational parameters and maximize the overall reaction yield [5-7]. Through this approach, we identify the optimal operating conditions to enhance CO₂ utilization efficiency and sustain the model interpretability for further applications.

References

1. Tans, P., Dlugokencky, E. and Miller, B., 2020. The power of greenhouse gases.

2. NOAA Global Monitoring Laboratory. “Trends in atmospheric carbon dioxide”, https://gml.noaa.gov/ccgg/trends/gl_trend.html (accessed 04/01/2024).

3. Samimi, F., Hamedi, N. and Rahimpour, M.R., 2019. Green methanol production process from indirect CO2 conversion: RWGS reactor versus RWGS membrane reactor. Journal of environmental chemical engineering, 7(1), p.102813.

4. Raissi, M., Perdikaris, P. and Karniadakis, G.E., 2019. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics, 378, pp.686-707.

5. Beykal, B. and Pistikopoulos, E.N., 2024. Data-driven optimization algorithms. In Artificial Intelligence in Manufacturing (pp. 135-180). Academic Press.

6. Beykal, B., Aghayev, Z., Onel, O., Onel, M. and Pistikopoulos, E.N., 2022. Data-driven Stochastic Optimization of Numerically Infeasible Differential Algebraic Equations: An Application to the Steam Cracking Process. In Computer Aided Chemical Engineering (Vol. 49, pp. 1579-1584). Elsevier.

7. Boukouvala, F. and Floudas, C.A., 2017. ARGONAUT: AlgoRithms for Global Optimization of coNstrAined grey-box compUTational problems. Optimization Letters, 11, pp.895-913.