Koji Maeta, Takenao Ohkawa

Abstract: In this paper, a simple combustion device consisting of a premixed burner, a rectangular cylinder, and a visualization window was used to measure the pressure fluctuation level and flame images while varying the flame position and operating conditions. A Convolutional Autoencoder (CAE) was applied to the acquired images to extract the features of the images. The images reconstructed from the extracted features and the original acquired images were then used to define the Combustion Instability Index (ΔCAETI_err), which can be used to quantify the flame conditions. By organizing the correlation between the proposed Combustion Instability Index and the combustion oscillation levels, we evaluated the possibility of detecting signs of increase in the combustion oscillation. The results showed that the proposed Combustion Instability Index and the combustion oscillation level were highly correlated. Using Grad-CAM data analysis, which enables visualization of the Combustion Instability Index on a two-dimensional plane, the mechanism that causes the increase in the combustion oscillation level was discussed by evaluating the effects of operating conditions on the flame distribution.

Keywords: Premixed combustion, Combustion instability, Convolutional Autoencoder, Grad-CAM.

Date Published: July 15, 2024
DOI: 10.11159/jffhmt.2024.017

View Article