Combustion Condition Identification though Flame Imaging and Convolutional Autoencoder

Identifying the combustion conditions in power-plant furnaces is crucial for optimizing combustion efficiency and reducing pollutant emissions. Traditional image-processing methods heavily rely on prior empirical knowledge, limiting their ability to comprehensively extract features from flame images. To address these deficiencies, this study proposed a novel approach for combustion condition identification through flame imaging and a convolutional autoencoder (CAE). In this approach, the flame images are first preprocessed, then the CAE is established to extract the deep features of the flame image, and finally the Softmax classifier is employed to determine the combustion conditions. Experimental research is carried out on a 600MW opposed wall boiler, and the effectiveness of the proposed method is evaluated using captured flame images. Results demonstrate that the proposed CAE-Softmax model achieves an identification accuracy of 98.2% under the investigated combustion conditions, significantly outperforming traditional models. These findings reveal the method feasibility, offering an intelligent and efficient solution for enhancing the operational performance of power-plant boilers.