Pattern recognition for measuring the flame stability of gas-fired combustion based on the image processing technology

• Published in: Fuel (Guildford) Vol. 270; p. 117486
• Main Authors: Wang, Yu, Yu, Yuefeng, Zhu, Xiaolei, Zhang, Zhongxiao
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 15.06.2020; Elsevier BV
• Subjects: Algorithms; Combustion; Combustion diagnosis; Decision making; Diagnostic systems; Entropy; Feature extraction; Flame stability; Fuzzy pattern recognition; Gas-fired flame; Image processing; Image segmentation; Logarithmic entropy multi-threshold segmentation; Neural networks; Parameter identification; Pattern recognition; Q and T2; Self organizing maps; Support vector machines; Fuzzy pattern recognition; Logarithmic entropy multi-threshold segmentation; Combustion diagnosis; Q and T2; Gas-fired flame
• ISSN: 0016-2361; 1873-7153
• DOI: 10.1016/j.fuel.2020.117486

This study proposes a diagnostic method for gas-fired combustion based on the image processing technology, for identifying an abnormal combustion situation in a gaseous flame. The proposed algorithm is divided into four aspects: (1) a logarithmic entropy multi-threshold segmentation method for segmenting the flame region utilized to extract image features; (2) 12 typical characteristic parameters representing gas-fired flame images, with five of them extracted for identification; (3) a fuzzy pattern recognition algorithm using an S-type membership function and a maximum-minimum distance function to distinguish between variable flame states; and (4) two statistics, Q and T2, used to evaluate the decision-making results of the fuzzy pattern recognition. The results are also compared to those from several other algorithms, including the self-organizing map, neural network, and support vector machine methods. The experimental results indicate that the proposed method has better performance in identifying different combustion situations in a gaseous flame and is superior to the other algorithms. Through a two-parameter adjustment, normal gas-fired combustion state can be accurately identified; for abnormal combustion, the prediction accuracy can become more than 90%. There can be a slight misjudgment; this may be owing to the relatively less training data for abnormal flame states.