Abstract
Complex kinetic mechanisms involving thousands of reacting species and tens of thousands of reactions are currently required for the rational analysis of modern combustion systems. In order to represent, analyze and visualize effectively the ignition process, advanced computational techniques will be required. Recently, we introduced a novel approach that captured the principal elemental transformations in complex reaction mechanisms in the form of graphs. In this work, we propose new approaches in order to arrive at a compact representation of the information content of these graphs utilizing machine learning principles, such as feature selection in time series and hashing. These approaches allow the projection of the totality of the information contained in the graphs describing the chemical transformations onto a single scalar. The temporally evolving graphs are treated as streaming data and locality-preserving hashing allows the unique assignment of a scalar “motif” value to each such graph. Analysis of those motifs allows the quick identification of “clusters” of identical reaction graphs that correspond to regimes with similar kinetic characteristics. The approach is illustrated with highly complex kinetic mechanisms describing pentane autoignition. It is demonstrated how this novel representation allows the identification of regions where similar temporal history of the chemical transformations is experienced.