Combustion is a kind of violent chemical reaction with luminescence and heating,and the reaction process is extremely complex,which requires the basic measurement data to carry out the research.The traditional contact measurement methods have the problems of short sensor life,low resolution and serious interference to combustion.Based on the high frequency multi-source laser spectral diagnostic technique,the multi-dimensional flame dynamic information of the combustion field can be obtained by high-speed imaging the spectral information of the combustion intermediate components(such as CH radical and OH radical),which can be used for in-depth analysis of combustion characteristics.Lean blowout(LBO)is one of the most typical combustion instability phenomena in the aeroengine combustion chamber with high efficiency and low emission.Mastering the flameout critical mechanism,influencing factors and prediction methods are extremely important to realize the efficient,reliable,safe and intelligent operation of the aeroengine.The existing research methods mainly rely on the wall pressure data,and can only describe the combustion process qualitatively in one dimension,which has the problems of single data,low resolution and poor correlation.Therefore,there is an urgent need to carry out research on high-frequency multi-source spectral image sensing and recognition methods to obtain multidimensional feature information such as flame structure,heat-release,frequency spectrum to distinguish the combustion condition and analyze the combustion characteristics by feature layer fusion.High frequency multi-source laser spectral diagnostic techniques mainly include high frequency multi-source planar laser-induced fluorescence(CH/OH-PLIF)and high frequency CH*/OH* spontaneous emission imaging technology.By obtanining the image information of combustion intermediate components,the combustion characteristics of different reaction time-scales(nanoseconds and microseconds)and spatial-scales(reaction zone and burnout zone)are investigated.However,so far,the high-frequency multi-source laser spectral diagnostic technique only qualitatively displays the combustion phenomenon,and does not realize the organic fusion from the data and feature level.Therefore,through the research on analysis method of highfrequency multi-source spectral image sensing and fusion,the multi-source spectral image feature data fusion model is established to reveal the correlation law between image features and combustion state,and then establish the feature screening and evaluation criteria,finally forms a composite diagnosis strategy of high-frequency OH-PLIF and high-frequency CH* spontaneous emission for the flameout critical state.Firstly,aiming at the thermo-acoustic oscillation and lean blowout in swirl combustion chamber,this paper theoretically analyzes the feasibility of combustion instability process using high-frequency multisource spectral image.Conventional physical quantity features such as flame temperature and pressure have the relatively complex spectral inversion process,and it is easy to lose combustion details.Therefore,different from the quantitative measurement of absolute physical quantities,this paper creatively puts forward a new idea of directly studying the combustion state based on multi-source spectral image features.The method of extracting moment features and combustion mode features is developed,and the spectral image features set closely related to the critical state of flameout is established.Meanwhile,the idea of multivariate statistical analysis is introduced.The redundant information is omitted,and the difference features are highlighted.The construction scheme of multi-source spectral image feature data fusion model is proposed,and provides an important theoretical support for the efficient and in-depth analysis of combustion flow field data.Then,this paper carries out the research on the composite diagnosis method of high-frequency CH/OH-PLIF and high-frequency CH*/OH* spontaneous emission,and obtains the correlation law between the moment features and swirl combustion condition,which puts forward a new recognition method of flameout critical state.The results show that there are obvious differences between the local combustion characteristics characterized by high-frequency CH/OH-PLIF image and the global combustion characteristics characterized by high-frequency CH*/OH* spontaneous emission,which are reflected in flame structure,time oscillation spectrum and spatial pulsation mode.The recognition of stable flame and LBO flame are realized based on the image moment feature for the first time,and it is found that the oscillation frequency will change from tens of Hertz low-frequency oscillation to 200 Hz selfexcited oscillation from the stable to LBO flame,which is expected to be used as a novel feature parameter to predict the flameout limit.Finally,this paper introduces the multivariate statistical analysis idea to establish the spectral image feature difference analysis and data fusion method.The simplified fusion models of low-order moment feature and geometric feature are constructed,and significantly reduces the number of image features at the similar recognition accuracy level,which provides a new idea for the lightweight of combustion condition recognition model.The results show that the multi-source data recognition model based on image moment features has a higher recognition accuracy for the recognition of transitional combustion condition.Compared with the combustion condition recognition model based on only geometric image features such as flame area and perimeter et al,the recognition accuracy of transitional condition has been improved by more than 24%.Hypothesis testing and fold change can select the most directly related features to the swirl critical flame state from 32 kinds of multisource data fusion features.At the similar recognition accuracy level,the number of OH-PLIF image features is reduced to 9 and the number of CH* spontaneous emission image features is reduced to 4.The new fusion analytical method of high-speed multi-source spectral images established in this paper is of great significance to the analysis of engine combustion characteristics and the accurate prediction of combustion critical state,which expands the application of laser spectroscopy in the field of combustion science. |