(38c) Robust Catalyst Characterization and Estimation of Mechanistic Pathways through Data Science

Robust Catalyst Characterization and Estimation of
Mechanistic Pathways through Data Science

Transient kinetic experiments provide a significant volume
of data as the material is evaluated as a function of time rather than at a
singular steady-state.   This allows more detailed microkinetic information to
be collected where variations in temperature, pressure or concentration on the
materials surface is reflected in the measured response.  However, the
resolution of this detail also requires the development of new analysis to
accurately inform the user of the underlying mechanism while maintaining the
temporal nature of the data. A novel approach presented here utilizes a
combination of modern machine learning techniques and physics-based assumptions
to analyze the full information available from transient kinetic experiments
without assumption of a kinetic model.  Furthermore, this approach applies
feature selection within a global reaction network to provide an optimal
trade-off between the selection of a complex estimated mechanism and the
minimization of the fit on the measured reaction rate.  As an example, a set of
catalysts are compared when performing ammonia decomposition in transient pulse
response experiments.   We were able to
quantitatively measure the ammonia gas interaction with different surface
species and the degree of impact on the transformation rate.  As such, the
global sparse covariance matrix describes the network path of individual
catalysts.  More specifically, we are able to distinguish differences in each
catalyst based on the degree, if at all, of the complete set of surface and gas
phase interaction combinations.  The figure below shows an example of the
obtained network where the surface species are indicated as z and the
gas species are indicated as c.  The coloration of each node indicates
the grouping of the interactions while the edges indicate a significant interaction.
These connections between the gas and surface species are remarkably distinct
for different materials and represent a mechanistic signature for
discriminating catalysts. The results of this novel method provide
interpretability through integration of physical assumptions, estimation of the
mechanistic reaction pathways and a unique experimental kinetic fingerprint for
each material whereby catalysts can be microkinetically discriminated.