Large scale reactive molecular dynamics using ReaxFF force field (ReaxFF MD) is a promising method to obtain detailed reactions in fuel pyrolysis. The skeleton reactions classification of typical fuel can help to screen out important skeleton reaction classes that is required in building kinetic model for numerical simulation of industrial reactor. However, automatic classification method is not available for extracting skeleton reactions from the massive and detailed reactions obtained in large-scale ReaxFF MD pyrolysis simulations. This talk presents the extraction and description of core reaction information with cheminformatics approach in an attempt for automated reaction classification for ReaxFF MD simulations with machine learning method.<br/>The skeleton mechanism can be constructed by merging detailed reactions on basis of structural and physicochemical similarities of reactions. Based on the detailed reactions generated by VARxMD from ReaxFF MD simulation trajectory files, the atom-mapped SMARTS strings of a reaction containing the global atomic information of unique Atom_ID were calculated as the basic reaction representations for traversing and extracting substructure of reaction center and its surroundings. The extended reaction centers (ERCs) were employed for a reaction description other than a full reaction to reduce description redundancy of less relevant structural parts in a reaction. The ERCs are extracted by extending reaction center to its closely relevant structures, using both the neighbor atoms and surrounding functional groups. The 4-level description of the ERCs include features of bonds, atoms, relevant neighbor structures, and integral ERCs. The ERCs properties were calculated using RDKit for aromaticity, number of free radicals, bond-type or bond order, bond length, net charge etc. taken from ReaxFF MD trajectory. The 32 descriptors ultimately aggregated into a set of feature vectors as input to machine learning scheme for building the reaction classification model.

Development of reaction description for fuel pyrolysis reaction classification of large-scale ReaxFF MD simulations

ACS Spring 2022

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