Recognition Of Turbulent Combustion Regimes Based On Unsupervised Learning

In the research of turbulent combustion,whether based on experimental or numerical simulation,it is often necessary to define the specific flame surface and identify the reaction zone.However,in previous studies,the methods are generally problematic: the variables that can be measured in the experiment are very limited,and it is difficult to obtain the complete flow field data,and the measurement accuracy is limited by the experimental conditions;although the numerical simulation can provides full flow field data of a variety of variables,but usually only select a specific isosurface of a certain variable to define the flame front,which is highly subjective.In this paper,we combine machine learning and large eddy simulation(LES)to identify flame surface in a turbulent combustion field.This method can not only take advantage of LES’s ability to provide multi-dimensional physical variable data in the three-dimensional flow field,but also solve the shortcomings in traditional method of defining flame surface.This paper implements nonlinear large eddy simulations for three different jet flames(Sandia Flame D/F and MILD combustion HM1 case),and then selects the mass fractions and temperatures of multiple components as the initial data set.Divide the flow field area and identify the flame surface from the initial data set.Specifically,it uses principal component analysis to reduce the dimensionality of the original data,and then uses two clustering algorithms(K-means algorithm and self-organizing map neural network)to perform clustering respectively.The results show that the two clustering algorithms have high consistency in the classification results of these three flame cases,especially for Sandia Flame D and HM1 case.This indicates that the combination of unsupervised learning and LES can more uniformly classify complex turbulent combustion fields.This new method provides support for further analysis of the characteristics of each area(especially the combustion regime)in the turbulent combustion field.