Tracking the structure of metal catalysts under operando conditions remains a challenge due to the paucity of experimental techniques that can provide atomic-level information for active metal species. Here we report on the use of X-ray absorption spectroscopy (XANES and EXAFS) and supervised machine learning (SML) for determining the three-dimensional geometry of metal catalysts. Artificial neural network (NN) is used to unravel the hidden relationship between the XANES features and catalyst geometry. In the case of EXAFS, NN is used to obtained the partial radial distribution function (PRDF) directly from the spectra. In other words, we trained computer to learn how to ‘invert” the unknown spectrum and obtain the underlying structural descriptors. Our approach allows one to solve the structure of mono- or heterometallic nanoparticles or size-selected clusters from their experimental XANES or EXAFS spectra. Our new directions in the NN-assisted structural refinement include the studies of single atom catalysts and non-metallic catalysts, such as nano-oxides. These applications are demonstrated by reconstructing the average size, shape and morphology of nanoparticles with narrow size and composition distributions from the coordination numbers obtained by NN-XANES or from the PRDF obtained by NN-EXAFS. First applications of this method to the determination of nanomaterial structure in operando conditions, such as studies of synthesis, nucleation, growth and reactivity of metal catalysts will be demonstrated.

Neural network approach for operando characterization of catalysts: From ultra-small clusters to nanoparticles

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.