Hybrid modeling of thermoacoustic oscillations: Integrating signal processing and machine learning for accurate detection

This study integrates machine learning models with traditional signal processing techniques to detect and analyze thermoacoustic instabilities in combustion systems. Time-domain analysis, Fourier transform, and phase space reconstruction methods are applied to pressure pulsation data to accurately identify the onset and evolution of thermoacoustic instabilities. Insights derived from these analyses are used to label datasets, which are subsequently employed as training data for two machine learning models: Extreme Gradient Boosting and Multilayer Perceptron. The trained models demonstrate high predictive accuracy in distinguishing between stable and unstable oscillatory states across multiple datasets. Moreover, quantitative comparisons with existing methods highlight the proposed approach's improvements in real-time monitoring capabilities, particularly in terms of prediction accuracy and computational efficiency. The study also evaluates two independent datasets and two public datasets to assess chaotic and irregular behaviors, further validating the models' robustness. The analysis reveals that the oscillatory states are characterized by distinct energy distributions across frequencies, as represented in the cumulative mass fraction diagram. These findings underscore the effectiveness of combining physical diagnostic methods with machine learning algorithms to enhance the detection, analysis, and real-time monitoring of thermoacoustic instabilities. This integrated methodology contributes significantly to the development of more efficient and reliable combustion systems.