スパース動的モード分解によるエチレン超音速燃焼の不安定性解析

抄録

Applications of machine learning methods to combustion science and technology have been carried out in recent years. Combustion instabilities in a scramjet engine are significant problems to be solved, leading to devastating damage to the propulsion systems. In the previous study, although the dynamic mode decomposition (DMD) was applied to time-resolved CH^* chemiluminescence observed in a laboratory-scale scramjet combustor, there were many peaks including noise components, making it difficult to extract the features. Therefore, in the present study, the sparsity-promoting DMD (SP-DMD) with the L1 regularization term was applied to remove the noise and extract the dominant modes. The dynamic pressure and shadowgraph images were also measured with a pressure sensor and a high-speed video camera. The fast Fourier transform was used for the pressure data. The peaks around 1600 Hz were observed for the power spectral density of pressure. This instability was also explained by the SP-DMD modes and a low-rank expression of 8 modes with an observation of time-resolved images of a shadowgraph. The data-driven approach of time-resolved CH^* chemiluminescence and the shadowgraph with SP-DMD clarified the combustion instability mechanisms in detail.