Machine learning-based ensemble flame regime classification for mesoscale combustors based on insights from linear and nonlinear dynamic analysis

Gaining insights into flame behavior at small scales can lead to improvements in the efficiency of micro-reactors, compact power generation systems, and other micro/mesoscale combustion applications. In this study, we systematically examine the various flame regimes observed in mesoscale combustors, namely Stable Flame, Flames with Repetitive Extinction and Ignition, and Propagating Flame from both dynamical and statistical standpoints. Our experimental methodology involves stabilizing a flame inside a quartz tube with an inner diameter of 5 mm, using a premixed methane–air mixture as fuel, with equivalence ratio and Reynolds number as input parameters. Instantaneous OH* chemiluminescence and acoustic pressure signals, as well as high-speed flame imaging, were acquired for combustion dynamics characterization. The distinct dynamical signatures associated with the observed flame regimes were analyzed using Recurrence Quantification Analysis, followed by Statistical-Spectral Analysis based on the experimentally acquired signals. Subsequently, a stacking ensemble-based machine learning framework is implemented for mesoscale flame regime classification, using features extracted from the above analyses. To visually elucidate the evolution of system dynamics and the complex interaction of competing timescales in these flame regimes, we graph Continuous Wavelet Transform scalograms and phase plots. Our predictions demonstrate a clear distinction in dynamical characteristics observed in these flame regimes and indicate that the computationally efficient statistical-spectral measures are effective for classification. Overall, this study aims to provide an insightful understanding and establish a framework for the characterization and classification of the flame regimes observed in mesoscale combustors, which could potentially be scaled to monitor practical mesoscale systems.