화학공학소재연구정보센터
Combustion and Flame, Vol.223, 474-485, 2021
Data driven analysis and prediction of MILD combustion mode
Direct numerical simulation (DNS) data of moderate or intense low-oxygen dilution (MILD) combustion and a planar flame are analysed to identify quantities influencing the unique features of co-existing combustion modes and develop a model to identify them in MILD combustion. The results show the existence of direct relationship between the scalar dissipation and reaction rates in MILD combustion, whereas the correlation is weaker than the planar flame. Also, rotational turbulent motions identified by the enstrophy-strain balance also show a substantial influence on the MILD combustion field, via the principal component analysis, suggesting the mechanism by which the non-flamelet part of reaction zones is controlled. A neural network (NN)-based model was proposed to identify the filtered local combustion mode for MILD combustion fields in LES context. The NN is trained based on DNS data of MILD combustion at two different conditions for different filter sizes. The model assessment was performed using the third MILD condition having a higher Ka and dilution level than those of the two training data sets. Several filter sizes ranging from 0.5 to 2.0 times of a corresponding laminar flame thickness were considered, and also prediction performance of a "zeroth-order model" is compared with that of the NN-based prediction. The assessment shows very high correlation between the NN-based prediction and the mode directly obtained from DNS. The predicted local combustion mode could be used to exploit advantages of both flamelet and non-flamelet-type combustion models to predict co-existing MILD reaction zones. (C) 2020 The Authors. Published by Elsevier Inc. on behalf of The Combustion Institute.