High Frequency Plif Image Processing Based On Dynamic Mode Decomposition

In order to study the complex causes of combustion oscillation,high spatial and temporal resolution diagnostic methods,such as planar laser-induced fluorescence(PLIF),are often used to process the data obtained by dynamic mode decomposition(DMD).However,the algorithm has the defects of systematic error,weak noise robustness and high complexity.Although many optimization algorithms have been developed,there are few reports on the application of these optimization algorithms in oscillatory combustion data processing.Combined with the high frequency PLIF diagnostic data,the optimization algorithm suitable for combustion oscillation research is screened and improved,which has a certain significance for the quantitative research of combustion oscillation.High frequency PLIF technology is mature,with the advantages of large detection range,non-interference and high time-space resolution.It is an important channel to obtain combustion oscillation diagnosis data.Based on the principle,a scheme of hydroxyl PLIF was designed to diagnose the flame structure,which was applied to methane air swirling combustion experiment,and 12 groups of dynamic flame images under different conditions were obtained.An important result of using DMD method to process combustion oscillation diagnosis data is characteristic frequency.It comes from the eigenvalues extracted by the algorithm,so the accuracy of the extracted eigenvalues is taken as the criterion to evaluate the advantages and disadvantages of the optimization algorithm.For this reason,a scheme to evaluate the accuracy of extracted eigenvalues is developed.By analyzing the factors that affect the accuracy of the algorithm,the optimization direction is determined,that is,to truncate and improve the noise robustness.The former is mainly ratio threshold truncation and optimal hard threshold truncation.In the test of diagnostic data,the results are not ideal,because the relationship between truncation number and rank of transformation matrix is ignored.Combined with the singular value logarithmic curve,the judgment method of the rank of the transformation matrix is given.The comparison of noise robustness is divided into two modules: numerical simulation test and combustion oscillation diagnosis data test.The results of the former show that the additive noise robustness of hodmd-1 algorithm is the strongest,while the multiplicative noise robustness is the weakest;DMD algorithm is just the opposite;The performance of the least squares dynamic mode decomposition method is close to that of hodmd-1 algorithm.In the second module test,the HODMD-1 algorithm is not always the best,but the stability is better than the other two algorithms.Since the multiplicative noise can be partially eliminated by experimental improvement,hodmd-1 algorithm is more suitable for combustion oscillation diagnosis.The improvement of dynamic mode decomposition algorithm is not only truncation operation,but also snapshot reconstruction and selection of main characteristic frequency.The former combines sparse reconstruction,compares four reconstruction algorithms,and determines the optimal reconstruction algorithm by numerical simulation test.On the basis of others,the latter introduces modal amplitude factor to improve the evaluation standard of main characteristic frequency,which is helpful to the diagnosis data analysis of methane air swirling combustion.The analysis results show that the characteristic frequency distributions of cone-shaped and rootless flames are more in low frequency,less in middle frequency and less in high frequency,so they have strong stability;The upper and lower separated flame has more high frequency components.