(434g) Beyond R2: Cautionary Tales Using Reduced-Order Kinetic Models for Reactor Optimization

In this talk, we share some possible pitfalls for using reduced-order kinetic models in reactor design optimization.

The shale gas revolution [1] has led to so-called stranded production creating new opportunities for novel distributed chemical conversion processes. For example, Tan and Barton [2] show modular scale shale gas processing units, if optimally redeployed, can outperform conventional scale processing units in the Bakken shale play. Similarly, techno-economic analysis by Yang and You [3] shows modular methanol manufacturing and shale gas processing units have larger net present values (NPV) compared to conventional scale plants based on data from the Marcellus, Eagle Ford, and Bakken shale plays. These economic advantages may further improve by incorporating new catalysis [4] and separations science innovations into modular chemical conversion platforms. As part of the NSF-funded Center for Innovative and Strategic Transformation of Alkane Resources (CISTAR), we are creating a multi-scale optimization framework for detailed oligomerization [5] reactor design to (i) tractably integrate current molecular-level research insights from microkinetic models [4] into process design [6] and (ii) determine conversion and selectivity targets under process constraints to help guide catalysis research.

Undoubtedly, microkinetic models of catalytic processes help elucidate reaction phenomena [7, 8] such as catalyst-reactant interactions, catalyst lifespan, and final product distributions ultimately accelerating catalysis research and development. Unfortunately, microkinetic models are seldom utilized in design, optimization, and scale-up of industrial reactors due to their computational complexity [9 - 11]. Typical multi-product microkinetic models track O(500) to O(10,000) chemical species, whereas process and reactor optimization methods are best suited for O(10) to O(100) pseudo/apparent chemical species. Recently published microkinetic model reduction strategies [11, 12] achieve up to 50% reduction in size for reaction networks with O(10) reaction rates. However, in CISTAR, we want to consider multi-component oligomerization reaction networks with O(1,000) reaction rates.

In this talk, we propose and evaluate six families of reduced-order kinetic models for oligomerization reactor optimization based on the varying functional form of the kinetic parameters [4, 13, 14]. We seek to (i) quantify the variability in model fit quality resulting from the functional form of reduced-order kinetic models chosen to emulate microkinetic model behavior and (ii) analyze how this variability affects reactor performance and output at the process-scale. We calibrated the reduced-order kinetic models using over 70 microkinetic model [4] simulations and the reduced-order kinetic models, statistically, showed decent agreement with the training data. However, some of these reduced-order kinetic models became numerically infeasible when simulated at out-of-sample operating conditions. Next, we embed the reduced-order kinetic models in a staged reactor design framework to maximize the production of heavier olefins by varying the number of reactors in series and optimizing the reactor temperatures. We find that although the optimal number of reactors and reactor temperature profiles are independent of model choice, the predicted olefin distribution varies significantly across model forms; and, for some model forms, the predicted olefin distribution at low propylene conversion regime is rich in butene which is fundamentally different from the microkinetic model simulations where dimers and trimers like hexene and nonene, respectively, are the major products in the low propylene conversion regime. These results emphasize the need to look beyond statistical quality of fit as the sole metric of model goodness and quantify the ability of the model to capture microkinetic model physics when working with reduced-order kinetic models.

As ongoing work, we are extending the reduced-order kinetic models to distinguish between branched and linear olefins. We are also exploring model-based design of experiments [15] to determine experimental and simulation conditions to best discriminate between alternative reduced-order kinetic models. We are especially interested in designing experimental (simulation) campaigns that minimize variability in the process designs (e.g., G-optimality).

References

1. EIA. Growing U.S. HGL production spurs petrochemical industry investment - Today in Energy - U.S. Energy Information Administration (EIA), 2016.
2. Tan, S. H., & Barton, P. I. (2015). Optimal dynamic allocation of mobile plants to monetize associated or stranded natural gas, part I: Bakken shale play case study. Energy, 93, 1581-1594.
3. Yang, M., & You, F. (2018). Modular methanol manufacturing from shale gas: Techno‐economic and environmental analyses of conventional large‐scale production versus small‐scale distributed, modular processing. AIChE Journal, 64(2), 495-510.
4. Vernuccio, S., Bickel, E. E., Gounder, R., & Broadbelt, L. J. (2019). Microkinetic model of propylene oligomerization on Brønsted acidic zeolites at low conversion. ACS Catalysis, 9(10), 8996-9008.
5. O'connor, C. T., & Kojima, M. (1990). Alkene oligomerization. Catalysis today, 6(3), 329-349
6. Ridha, T., Li, Y., Gençer, E., Siirola, J., Miller, J., Ribeiro, F., & Agrawal, R. (2018). Valorization of Shale Gas Condensate to Liquid Hydrocarbons through Catalytic Dehydrogenation and Oligomerization. Processes, 6(9), 139.
7. Salciccioli, M., Stamatakis, M., Caratzoulas, S., & Vlachos, D. G. (2011). A review of multiscale modeling of metal-catalyzed reactions: Mechanism development for complexity and emergent behavior. Chemical Engineering Science, 66(19), 4319-4355.
8. Stamatakis, M., & Vlachos, D. G. (2012). Unraveling the complexity of catalytic reactions via kinetic Monte Carlo simulation: current status and frontiers. ACS Catalysis, 2(12), 2648-2663.
9. Mhadeshwar, A. B., & Vlachos, D. G. (2005). Is the water–gas shift reaction on Pt simple?: Computer-aided microkinetic model reduction, lumped rate expression, and rate-determining step. Catalysis Today, 105(1), 162-172.
10. Salciccioli, M., Chen, Y., & Vlachos, D. G. (2011). Microkinetic modeling and reduced rate expressions of ethylene hydrogenation and ethane hydrogenolysis on platinum. Industrial & engineering chemistry research, 50(1), 28-40.
11. Karst, F., Maestri, M., Freund, H., & Sundmacher, K. (2015). Reduction of microkinetic reaction models for reactor optimization exemplified for hydrogen production from methane. Chemical Engineering Journal, 281, 981-994.
12. AvÅŸar, E. (2017). Dimensionality reduction for predicting CO conversion in water gas shift reaction over Pt-based catalysts using support vector regression models. International Journal of Hydrogen Energy, 42(36), 23326-23333.
13. Oliveira, P., Borges, P., Pinto, R. R., Lemos, M. A. N. D. A., Lemos, F., Vedrine, J. C., & Ribeiro, F. R. (2010). Light olefin transformation over ZSM-5 zeolites with different acid strengths–a kinetic model. Applied Catalysis A: General, 384(1-2), 177-185.
14. Nguyen, C. M., De Moor, B. A., Reyniers, M. F., & Marin, G. B. (2011). Physisorption and chemisorption of linear alkenes in zeolites: a combined QM-Pot (MP2//B3LYP: GULP)–statistical thermodynamics study. The Journal of Physical Chemistry C, 115(48), 23831-23847.
15. Galvanin, F., Cao, E., Al-Rifai, N., Gavriilidis, A., & Dua, V. (2016). A joint model-based experimental design approach for the identification of kinetic models in continuous flow laboratory reactors. Computers & Chemical Engineering, 95, 202-215.