A neural network (NN) aided model is proposed for the filtered reaction rate in moderate or intense low-oxygen dilution (MILD) combustion. The framework of the present model is based on the partially stirred reactor (PaSR) approach, and the fraction of the reactive structure appearing in the PaSR is predicted using different NN's, to consider both premixed and non-premixed conditions while allowing the use of imbalanced training data between premixed and non-premixed combustion direct numerical simulation (DNS) data. The key ingredient in the present model is the use of local combustion mode prediction performed by using another NN, which is developed in a previous study. The trained model was then assessed by using two unknown combustion DNS cases, which yields much higher dilution level (more intense MILD condition) and higher Karlovitz number than the DNS cases used as training data. The model performance assessment has been carried out by means of the Pearson's correlation coefficient and mean squared error. For both the present model and zeroth-order approximated reaction rate, the correlation coefficient with the target values shows relatively high values, suggesting that the trend of predicted field, by the present model and zeroth-order approximation, is well correlated with the actual reaction rate field. This suggests that the use of PaSR equation is promising if the fraction of the reactive structure is appropriately predicted, which is the objective in the present study. On the other hand, substantially lower mean squared error is observed for a range of filter sizes for the present model than that for the zeroth-order approximation. This suggests that the present filtered reaction rate model can account for the SGS contribution reasonably well.

Direct Numerical Simulations (DNS) data of Moderate or Intense Low-oxygen Dilution (MILD) combustion are analysed to identify the contributions of the autoignition and flame modes. This is performed using an extended Chemical Explosive Mode Analysis (CEMA) which accounts for diffusion effects allowing it to discriminate between deflagration and autoignition. This analysis indicates that in premixed MILD combustion conditions, the main combustion mode is ignition for all dilution and turbulence levels and for the two reactant temperature conditions considered. In non-premixed conditions, the preponderance of the ignition mode was observed to depend on the axial location and mixture fraction stratification. With a large mixture fraction lengthscale, ignition is more preponderant in the early part of the domain while the deflagrative mode increases further downstream. On the other hand, when the mixture fraction lengthscale is small, sequential autoignition is observed. Finally, the various combustion modes are observed to correlate strongly with mixture fraction where lean mixtures are more likely to autoignite while stoichiometric and rich mixtures are more likely to react as deflagrative structures.

Direct numerical simulations of Moderate or Intense Low-oxygen Dilution combustion inside a cubical domain are performed. The computational domain is specified with inflow and outflow boundary conditions in one direction and periodic conditions in the other two directions. The inflowing mixture is constructed carefully in a preprocessing step and has spatially varying mixture fraction and reaction progress variable fields. Thus, this mixture includes a range of thermo-chemical states for a given mixture fraction value. The combustion kinetics is modelled using a 58-step skeletal mechanism including a chemiluminescent species, OH*, for methane–air combustion. The study of reaction zone structures in the physical and mixture fraction spaces shows the presence of ignition fronts, lean and rich premixed flames and non-premixed combustion. These three modes of combustion are observed without the typical triple-flame structure and this results from the spatio-temporally varying mixture fraction field undergoing turbulent mixing and reaction. The flame index and its pdf are analysed to estimate the fractional contributions from these combustion modes to the total heat release rate. The lean premixed mode is observed to be quite dominant and contribution of non-premixed mode increased from about 11% to 20% when the mean oxygen mole fraction in the inflowing mixture is reduced from about 2.7% to 1.6%. Also, the non-premixed contribution increases if one decreases the integral length scale of the mixture fraction field. All of these results and observations are explained on physical basis.

Direct numerical simulation (DNS) data of freely propagating turbulent premixed flame of stoichiometric hydrogen air mixture inside a closed vessel is analysed to study a sub-grid combustion closure based on unstrained flamelet approach. This modeling framework needs closures for the sub-grid scale (SGS) reaction rate and scalar dissipation rate. The results show that the closure models for these two SGS quantities work quite well. The dissipation rate closure involves a scale dependent parameter, β_c, which is related to the flame curvature induced effects. The reactivity of reactant mixture increases with time in isochoric combustion because the mixture temperature and pressure increase with time. This also influences the parameter β_c and thus the dynamic evaluation of this parameter is investigated using the DNS data.