Thermoacoustic Instability Vortex Shedding Characteristics And SVM Time Series Prediction Model Based On Rijke Tube

With a wide application in lean premixed combustion technology of gas turbine, thermoacoustic instabilities problem get more and more researchers’attention. Thermoacoustic instability is a low frequency high amplitude pressure oscillation of heat release and acoustic field coupling in the combustion chamber. Its appearance will affect the performance of the combustion chamber and reduce the combustion efficiency. This work based on the Rijke type combustor test bench, conducted two researches on thermal acoustic vibration problems.First, vortex shedding driven thermoacoustic instability characteristics was studied. Build a Rijke-type thermal acoustic vibration system which can be observed by quartz glass tube. Using different width strip plate formed eddy shedding process. High-speed imaging acquisition system and dynamic pressure acquisition system were used to collected high frame rate of flame image and combustion chamber sound field vibration. Based on Matlab image processing system, using high-frequency flame image analyzed vortex shedding frequency. The results show that the sound field excited by vortex shedding consists of the dominant frequency in 160 Hz and harmonic components in 320 Hz. Flame image is superimposed by several different frequencies of peak intensities. Vortex shedding frequency is coupled with acoustic frequency at component of 160 Hz.Then in order to overcome the time delay problem of present thermoacoustic active control system, a method of fuzzily predict the time series of thermoacoustic instability using support vector machine was proposed.In self-designed Rijke thermoacoustic instability test bench, conduct experiments in speaker disturbance, capture the thermoacoustic dynamic pressure sequence. Use phase space reconstruction law to build the input and output datasets and build the prediction model by support vector regression. It was a satisfactory agreement between simulated and experimental data by checking with test datasets. In addition, the effects of embedding dimension and delay parameters on model’s performance were analyzed. Models at the higher delay parameter can still maintain a high degree of fitting 0.95, showed the feasibility of the time series prediction method.