Generalized simulation and processing of TAP reactor data

Abstract

An understanding of intrinsic kinetics is necessary for the rational design of new materials for catalytic processes. One way to obtain this information is by running and interpreting Temporal Analysis of Products (TAP) reactor experiments. Though these experiments provide rich, transient kinetic information, converting the raw TAP data to knowledge of the material is a major bottleneck. Steps have previously been taken to reduce this burden, but further developments, as well as refinement of current methods, are needed. One change that could improve processing is the application of automatic differentiation (AD), which offers a highly accurate calculation of the derivative. For this reason, a workflow to simulate and process TAP reactor pulses built around the FEniCS Python package has been developed. This package allows for the efficient evaluation of the necessary PDEs for TAP, and allows for efficient AD by taking advantage of the adjoint operators of the PDE. A method to convert elementary reactions directly into the FEniCS PDE format was developed. The first steps to generate reaction mechanisms, as well as sets of rate-limiting reaction expressions, based on the gaseous reactants and products observed during TAP experiments were also taken. The new, general method of simulating pulses around FEniCS was validated and benchmarked. The time required for each simulation was not found to limit the workflows ability to quickly handle TAP data, but improvements can be made. The Degree of Flux Control was also introduced as an alternate form to the commonly used Degree of Rate Control. This is the first example of a transient sensitivity analysis performed on TAP pulses, and one of few implementations of the Degree of Rate Control analysis to transient processes. A method to fit parameters was also implemented, and objective functions were constructed with a reduced number of points to improve efficiency. The parameter fitting method was applied to several examples, including pure diffusion, a linear reaction mechanism, and multiple carbon monoxide oxidation reaction mechanisms, and was found to accurately determine diffusive and kinetic parameters. The methods developed in this Thesis show the utility of AD and should lead to a more efficient processing of TAP data. This workflow will act as a foundation on which more advanced methods can be developed, including forward and reverse uncertainty quantification, the generation of initial parameter estimates for fitting, and the application to increasingly complex reaction networks. Ultimately, it is envisioned that these methods can work in concert with experiments, providing a route to adaptive TAP experiments that automatically interrogate the intrinsic kinetics of real catalytic materials.