A rational, automated knowledge framework for reaction kinetic modeling and catalyst design
The continuing development of high throughput experiments (HTE) in the field of catalysis has dramatically increased the amount of data that can be collected in relatively short periods of time. The key questions in the current scenario are (1) Even when HTE can afford "Edisonian" discovery, how can the increasing amounts of data be converted to knowledge that will guide the next search in the vast design space that encompasses catalytic materials? and (2) How can HTE data lead to fundamental understanding? In order to address these questions, we propose a catalyst design architecture that involves (1) a forward model to predict the performance of a given material structure and (2) a genetic algorithm-based inverse problem that uses the forward model to search the descriptor space for a material that meets specific design objectives. We have developed a rational, automated knowledge extraction engine to aid the forward model building process. The Reaction Modeling Suite* (RMS) is a set of tools based on artificial intelligence and optimization techniques that enables the expert to initiate the kinetic modeling sequence in a simple reaction chemistry language. The software then interprets this information into a reaction sequence, writes the appropriate equations, optimizes the model parameters within physically and chemically allowed bounds, and statistically analyses the results. RMS is also capable of refining models based on the comparison of the features of the data and model predictions. These steps have been demonstrated for propane aromatization on HZSM-5. The iteration of model predictions as compared to HTE data and subsequent model refinement and formulation of experiments, leads to convergence to a predictive kinetic model which is the knowledge repository in this framework. Our hybrid modeling approach enables transformation of approximate initial models to increasingly sophisticated ones that strengthen our understanding of the fundamentals of catalytic behavior and, at the same time, guide HTE. A successful forward model has considerable value in its own right, but its power is dramatically leveraged by the inverse model, which forecasts successful specific catalyst formulations. This potential to truly design catalysts has been demonstrated as a proof-of-concept in this thesis. ^ *U.S. patent pending^
Major Professors: Venkat Venkatasubramanian, Purdue University, James M. Caruthers, Purdue University.