Research Of A Flame Detection Real Time System Based On ARM And Wavelet Analysis

The energy and power is necessary to people by the coal firing.The flame detection system is the sticking point of making the furnaces safely work. There are two kind of flame in fired furnace.One is furnace-body flame,the other is flame from furnace nozzles.In order to judge the operation status of furnace,workers need to know whether the furnace nozzle is burning or not under the background flame of furnace.Traditional flame detection method finds it much difficult to extract the signal feature and recognize the status realiably as the signal-to-noise ratio is very low.we proposed a new method, based on combined the Wavelet transform and Back Propagation neural network,which has been used after the acquisition and analysis of experimental data from the power plant's coal firing furnace. Firstly,a set of flame intensity data contains of lots of noise,which acquired via a sensor,are decomposed based on multi-scale Wavelet. Next,the feature information is distilled from original signal and then against noise with soft threshold value wavelet transform. Last,the feature information,which distilled by power analysis,are used as the neural network's input signals for training purpose.This thesis,which is based on the aspect of signal precessing, introduces the advantages of wavelet time-frequecy localization and the particular method of denoising,also concretely analyzes the experimental data using the theory of wavelet transform.The experimental results demonstrate that the time-frequency localization method improved the SNR of the flame detection signal.It overcome the disadvantages of traditional flame detection method that easy to incorrectly judge.This method not only recognizes combustion status more efficiently,but also has its extensive utility.