The Flame Stability Diagnostic Methods For Study

The strict environment laws in our country demand the reduction of the pollutant from power plant boilers. It is the optimism of the combustion in furnaces that can effectively approach this goal. What's more, combustion optimism can not only save the fuel, but also avoid tube explosion in furnaces. In this way, the efficiency and safety of power plant boilers can be ensured. In these years, a lot of studies have been done on the alteration of combustor, the numerical simulation in air dynamical field in furnaces and so on, which have played the important roles in improving combustion. However, the research on the monitoring and control of combustion in furnaces is behindhand. The purpose of this thesis is to establish a new combustion monitoring and control system based on the image processing and artificial intelligence technology to carry out the combustion visualization and diagnosis, and then give the instruction information for power plant staff.In this paper, we use advanced computer image processing technology to research for a quantitative method of judging the stability of combustion flame.Firstly, researching at a lab and at a utility boiler respectively, the flame imagesof stable combustion, deflagration, flameout were acquired. Then pick up some image characteristic values form the flame image, which can reflect the state of combustion. These characteristic values are flame luminance, high temperature flame luminance, flame area, high temperature flame area, high temperature ratio, centroid offset distance and circularity, Which are universal and out of the effect of the flame ignition point excursion, so that simplify greatly the degree of difficulty and can pick up these seven features on line.Secondly, put forward a method to monitor and to diagnose the stability of burning flame base upon image processing technology and PCA (principal component analysis) method, using the two statistics of Hotelling T2 and SPE monitor the every-time image data vectors, to check them whether they exceed their own controllable limit. As long as one of them exceeds the limit, abnormity combustion can be concluded. The experimental research showes: the method can recognize andjudge the combustion state of burning flame real-time and availably, and visually shows the result with SPE fig, Hotelling T2 fig and PCA fig, demonstrating a promising potentiality for real-time monitoring combustion and farther diagnosis.Besides, in order to overcome the existing problems in linear PCA method, I present a nonlinear principal component analysis method based upon input training neural network and BP neural network. Through experiment, this method can effectively diagnose the state of the flame.At last, for the sake of perfecting the combustion diagnosis system, this paper has the further studies on the image processing technology and pattern identification. In the way of the image processing technology, this paper introduces the fractal theory into the study of flame stability, and presents a new method of image segmentation based on fractal feature and also fine a new feature fractal dimension which can reflect the combustion state. In the way of model identification, apply the EM-PCA algorithm to treat with the incomplete data space. Therefore, Appling the principal component analysis to set up real -time combustion diagnosis system with the information of flame image and power planet operational factor will come to be true.