Investigation On Fuel Recognition Using Flame Digital Image Processing

In thermal power plants, the type of fuel is related to the safety of boiler and combustion efficiency. Due to the irreproducibility of coal resources, multiple kinds of coal are used in coal-fired power plants commonly and the biomass fuel as green energy has the same diversity. Flame, which is the external form of fuel combustion, is the immediate reaction of the combustion stability and economy. Flame radicals is the intermediate products of combustion which contains lots of information.In this paper, based on the digital images of combustion flame radicals, the model of fuel identification has been established by extracting image features and analyzing the inner link between radicals and fuel types. It has important engineering significance for the efficiency and optimization of combustion, reducing of pollution emissions, saving energy and reducing consumption. In this paper, the main work is as follows:Firstly, aiming at coal and biomass fuel, it is discussed that the similarities and differences of the industrial analysis and elemental analysis of two fules. The structure and combustion mechanism of fuel are also analyzed. The produce mechanism and application of combustion flame radicals is introduced and the feasibility of fuel type recognition by the flame radicals is analyzed.Secondly, the acquisition system and acquisition process about the combustion flame radical images are profiled in detail. Based on the digital images of four flame radicals (OH*, CN*, CH* and C2*), adaptive wiener filtering and homogeneity measurement method are used severally for the image filter and threshold segmentation. In addition, three groups of features are extracted:the intensity on the contours of flame radical images; the intensity ratios of different flame radicals (CH*/OH*, CH*/C2*, CH*/CN*, OH*/C2*, OH*/CN* and C2*/CN*); the average intensity, area and eccentricity of flame radical images.Next, based on the three groups of features extracted from combustion flame radical images, three kinds of fuel identification models are established respectively with the BP neural network, RBF neural network and probabilistic neural network (PNN) based on the radial basis function. The experimental data of flame radical images, including three kinds of biomass fuel, are used for the three kinds of fuel recognition models. The experiment results show that the three kinds of neural networks combined with three groups of flame radical image features all have good recognition effects. Among them, using the average intensity, area and eccentricity of flame radical images as feature is slightly better than separate application of the intensity on the contours of flame radical images and the intensity ratios of different flame radicals for fuel identification; PNN is slightly better than the other two kinds of neural networks on the fuel recognition effect and speed.Finally, the main content is summaried and future work is presented.