Flame Furnace in Thermal Power Plant Condition Monitoring Using SVM

• Published in: 2009 Second International Conference on Intelligent Computation Technology and Automation Vol. 3; pp. 67 - 70
• Main Authors: Liu Dao-guang, Lv Li-xia, Liu Chang-liang, Cui Jing
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2009
• Subjects: Brightness; Combustion; Condition monitoring; Digital images; Fires; Furnaces; Image analysis; Monitoring; Pixel; Power generation; State; Support vector machines; SVM
• ISBN: 0769538045; 9780769538044
• DOI: 10.1109/ICICTA.2009.484

This paper presents a support vector machine algorithm to use to monitor the state of the furnace flame method. Through the comparison of several methods of monitoring found that the flame images, the basic brightness from the flame (through digital image processing methods will be broken down into its R, G, B RGB images), the flame area (camera video of the burning flame pixels area) in several aspects, such as extraction of data to analyze and eventually the state of the monitoring results of the flame. Support vector machine in the analysis algorithm based on the use of MATLAB language applications to achieve the classification of image analysis of raw data. Intensity in the furnace flame monitoring, the first application of digital image processing method to get the furnace burning flame of raw data (flame brightness R, G, B RGB data), and then use SVM classification algorithm to find the best possible image data plane (the largest interval hyper-plane), the original image data to identify the classification for flame intensity monitoring results. Flame of the furnace through the analysis of raw data, extract images combustion data points, and the total combustion flame image pixels are compared, can be more accurate analysis of the burning effect, that is, the results are divided into complete combustion, which are mostly full combustion and a small portion of the full combustion, and then classified by observing and analyzing the results to take appropriate measures. Simulation results show that the analysis of the use of this method flame images, can be effective, real-time to determine the brightness of the flame and combustion analysis of the status and accuracy of its combustion state.