Data-driven calibration of computational combustion models employing reduced chemical Kinetics

Abstract

In this thesis, a probabilistic embedded discrepancy approach to understanding the limits of the use of reduced chemical kinetics in computational combustion models and also to improve the ability of such models to predict key quantities of interest is adopted. Also, an embedded deep learning model discrepancy approach is proposed. More specifically, a deep neural network is embedded as an additive function to model the temporal evolution of chemical species concentrations that serves as a source to the flow field. A data-driven calibration is adopted with data set produced by numerical simulation of detailed mechanisms, in a model-to-model Bayesian calibration. These proposed strategies are evaluated in benchmark combustion scenarios widely used to evaluate the role played by chemical kinetics on main physicochemical properties characterizing combustion systems. The scenarios correspond to the application cases of homogeneous combustion during autoignition and flame propagation. The results demonstrate the ability of adopted approaches for model calibration in chemical kinetics. It is shown how the application range of the reduced chemical model can be extended to predict quantities of interest without increasing the number of reacting species in the combustion system and at a reduced computational cost.