(426d) Discovery of Pt Trimetallic Electrocatalysts for Ammonia Oxidation with Interpretable Deep Learning | AIChE

(426d) Discovery of Pt Trimetallic Electrocatalysts for Ammonia Oxidation with Interpretable Deep Learning


Pillai, H. - Presenter, Virginia Tech
Li, Y., Jiangsu University
Wang, S. H., Virginia Tech
Mu, Q., Virginia Tech
Pokrywka, C., Virginia Tech
Achenie, L. E. K., National Science Foundation
Abild-Pedersen, F., SLAC National Accelerator Laboratory
Wu, G., University At Buffalo
Xin, H., Virginia Tech
Electrooxidation of ammonia (AOR) to dinitrogen plays a crucial role within the nitrogen cycle, and has important applications in wastewater remediation and direct ammonia fuel cells. While platinum (Pt) based catalysts, specifically terminated with (100) facets, have shown promising activity, they suffer from a large overpotential and surface deactivation. These issues have been tackled via the use of bimetallic and trimetallic Pt-based alloys, however the lack of a clear mechanistic understanding of ammonia electrooxidation has hindered significant improvements. Moreover, traditional trial-and-error approaches struggle searching through the high-dimensional design space of trimetallic systems. Although modern machine learning (ML) approaches have shown great potential in screening such immense design spaces, their black-box nature makes it impossible to understand model predictions and draw physical insights.

We first present a brief analysis of the thermodynamics and kinetics for the NH3 oxidation to N2 on various transition metals from density functional theory (DFT) calculations. It is shown that due to the balanced *N bridge and hollow binding energies only Pt and Ir are able to oxidize NH3 to N2. Subsequently, linear energy-scaling relationships were integrated with a microkinetic model to develop an activity volcano map that can explain activity trends on metals and state-of-the-art Pt-based alloys. To screen for active and stable trimetallic electrocatalysts, we use the activity map and TinNet within an active learning scheme with different filters, e.g., reactivity, synthesizability, and stability metrics. From a vast design space consisting of Pt-based trimetallic alloys we find promising stable and active candidates, e.g. Pt3Ru doped with Co. This approach is verified via controlled synthesis and electrochemical testing of the catalyst. Lastly, SHapley Additive exPlanations (SHAP) analysis is used to gain fundamental understanding of physical factors that govern the site reactivity and to identify key design rules to further aid catalyst discovery.