Quantitative Monitoring of Combustion Stability Based on Image Adversarial Convolutional Autoencoder

Accurate monitoring of combustion stability is of great significance in optimizing the combustion state. Traditional combustion stability monitoring methods are not only highly dependent on prior expert knowledge, but also difficult to achieve quantitative evaluation. To overcome these limitations, a novel quantitative assessment method for combustion stability is proposed in this study. In this method, an adversarial convolutional autoencoder (ACAE) is established to extract deep features of the flame images, and a quantitative assessment index is applied for feature analysis. Especially, the ACAE adopts a novel adversarial mechanism to improve the training efficiency and thereby enhance the feature learning ability. The numerical interval of the quantitative assessment index is [0, 1], and when the assessment index is lower than 0.5, the combustion state is stable. The feasibility of the stability monitoring method is verified by ethylene combustion experiments, and the testing results confirm that the deep image features extracted by the ACAE can be used to quantitatively estimate the combustion stability. Furthermore, the proposed monitoring method has a strong generalization performance that can accurately identify flame images beyond the scope of the training dataset. © 2024 Chin.Soc.for Elec.Eng.