Research On The Detection Method Based On Image Processing For Furnace Flame

For large coal-fired boilers,it is very important to monitor the working state of suspension combustion for reliability, security and economic barometer of boilers. The burning process of pulverized coal boiler is a kind of very complex suspension combustion that its working condition is very unstable. And the security of boilers is mainly decided by the stability of flame. To prevent the explosive event of boilers, coal powder must be completely burnt. If the burning process is unstable or the temperature field is asymmetrical in boilers, it is very easy to get terrible results, and the efficiency of boilers will be synchronously reduced greatly. Therefore, the temperature field in furnace is the key to judge the burning state of a furnace, and regulate working parameters of boilers. Enhancing the capacity of judgment, prediction and fault diagnosis in burning process, and making operators to get clearer image, we research on the technology of image processing for furnaces flame emphatically.Although two-dimension temperature field cannot nicely reflect temperature changes in the furnace space along the altitude direction, it can reflect the distributive characteristics of the flame in two-dimension space. If CCD camera is installed appropriately, it can provide important information such as the flame burning state. Acquired temperature fields from the image provide temperature references to operators, and then they can supervise for further combustion operation, which has very important meaning for safety operation and combustion diagnosis in the plant. This paper proposed the principle of the colorimetric algorithm for temperature measure and error analysis, and also provides realization methods.In this paper, the noise source of image is analyzed according to the noise theory. There is a great deal of dynamic and stochastic noise in the burning process of the furnace flame. These noise signals must be eliminated. Median filtration method, a kind of non-linear technology of processing signal, is able to eliminate noise signal from image. However, this method has an obvious disadvantage that need sort by order of pixel value, which results in slow speed of calculation. To save time, we design fast median filtration method based on the characteristics of furnace flame. The new median filtration method can directly calculate the median value without sort. In order to obtain clear flame image, Gray transform and Pseudo-color are discussed by the technology of image processing. The traditional image segmentation technique is mostly based on the gradient of the gray, while the change of the overview gray of the furnace flame is unobvious. Therefore, adopting the traditional methods cannot obtain satisfactory results. In this paper, based on analyses of existing image segmentation technology, the genetic algorithm was used to solve this problem and applied in dividing and uniting which involves the problems of chromosome code and desiring genetic operation, individual adaptability evaluation. As a result, we finally can satisfactorily solve those problems. For the image of combustor, the method proposed in this paper is based on the grads operators. Through learning the samples of images, the parameters of the templets are trained by using the neural network. Then we can get the optimized grads operators. The result of the experiment shows that this method can effectively improve the edge detection of image processing.Basing on the analyses of existing diagnosing techniques of combustion, this paper presents an evaluation technique of the flame combustion state based on BP Neural Network. The input variables of neural network are the combustion parameters that symbolize flame image. In this paper, we give out the definitions for these parameters and their solution method. The combustion state exponent is a data that is between 0.0 and 1.0. The bigger exponent means the better combustion state, whereas the combustion state is worse. We also classify the output into three grades: flameout, combustion unnatu...