Precise Control of Tungsten Carbide in Joule Reactors via Natural Language Processing
PI: Dr. Marc D. Porosoff, University of Rochester (UR)
Transforming the chemical industry to meet United States energy independence goals requires major innovations in catalyst design, earth-abundant materials, and the electrification of high-temperature catalytic processes for compatibility with intermittent energy sources. This research project aims to develop a new approach for catalyst development by combining natural language processing (NLP) and pulsed Joule heating to create and control tungsten carbide (WC)-based catalysts for the reverse water-gas shift (RWGS) reaction, a key step in the production of value-added chemicals and fuels.
Our project introduces a novel framework that uses NLP to enable rapid discovery and understanding of WC-based catalysts from text-based catalyst synthesis procedures and reaction conditions. We hypothesize that pulsed Joule heating will enable us to uncover previously inaccessible catalyst phases by exploiting the rapid heating rates of our Joule reactor to control carburization timescales and synthesize entropically favorable WC phases. The hypothesized active phase for RWGS, orthorhombic WC (β-W2C), is less thermodynamically stable than hexagonal WC (δ-WC), but possesses electronic features similar to the highly active and thermodynamically stable orthorhombic molybdenum carbide (β-Mo2C), making WC an ideal model system for our study.
This project is organized into three primary research thrusts: (1) Synthesizing phase-controlled WC using pulsed Joule heating; (2) predicting RWGS performance using text-based descriptions of experimental procedures, improving rigor and reproducibility of catalysis science; and (3) exploiting transient catalysis with Joule heating to better understand RWGS kinetics by leveraging temperature oscillations to control mass transfer, surface coverages, and catalyst restructuring.
The broader significance of our work extends beyond WC-based catalysts, as the applications of NLP and Joule heating can be adapted to a wide range of catalysts, supporting discovery of next-generation catalysts and electrification of industrial processes. The fundamental insights gained from this work will contribute to a new methodology for representing catalysts with the text of synthesis procedures and reaction conditions, accelerating catalyst understanding and development by applying recent advancements in AI/ML methods.
