Identification of governing physical processes of irregular combustion through machine learning

This study is concerned with the data-driven analysis of predictive parameters associated with regime transition for ignition and detonation. To this end, we introduce machine learning as a technique to analyze the predictive capability of these parameters. Machine learning enables the critical evaluation of combustion regimes from disparate sources and allows for a generalized comparison of the parameters. The parameter set is composed of features that have been found to be effective in classifying ignition regimes and are supposed to have a universality in predicting the regularity of the ignition process. Three different configurations are examined: weak ignition shock tubes, mild ignition in rapid compression machines, and the regularity of cellular detonations. All configurations are demonstrated to show a strong sensitivity to the heat deposition time of the chemical reaction. Additionally, detonations are found to be primarily sensitive to the heat release and show that an increased chemical sensitivity has a regularizing effect when plotted against the heat release rate. The data are found to be well classified with only one or two parameters, indicating that a universal, governing parameter is plausible.