Audio-Based Classification of Swirl Combustion Regimes Using Deep Learning

Deep learning-based prediction of combustion reaction is of growing interest in combustion research community. Such predictions have potential to detect the onset of different combustion regimes and early predictions combustion instability during gas turbine operation. Advanced artificial intelligence-based combustion control system is conceptualized with audio feature-based recognition of swirl distributed combustion at 5.72 MW/m³-atm heat release intensity and at equivalence ratio (ϕ) 0.9. The main focus is to classify conventional swirl combustion from low oxygen diluted reaction zones leading to distributed combustion regime with methane fuel. Previous studies in this line of interest were performed at the Combustion Laboratory (University of Maryland) to classify swirl air combustion and distributed combustion using convolutional neural network (CNN) based on image features (flame shape, chemiluminescence intensity, and standoff height). However, images may not solely help to classify combustion regimes due to highly complex, non-linear dynamics, and geometrical dependence of combustion systems. Hence, an audio-based CNN is conceptualized in this research to examine capability of the deep-learning prediction framework. Model training was performed with acoustic signals at dilution levels of CO₂ ∼ 79, 81, 82, 84% while testing was performed with CO₂ ∼ 83 and 80%. Model optimization was considered with different epochs numbers of 20, 50, and 100. The trained model recognized new swirl flames at CO₂ levels 80 and 83% well using acoustic signatures based on shape of signal pattern and amplitude of signal.