Chemical Engineering Seminar Series: Prof. Konduri Aditya

Chemical kinetics for combustion simulations: mechanism development and dimensionality reduction

Direct numerical simulations of turbulent reacting flows resolve the detailed chemical kinetics that provide insights into the turbulence-chemistry interactions. While the kinetics models should accurately represent the chemistry, their sizes need to be small enough for the computations to be tractable. This talk first focuses on the mechanism development from MD simulations, followed by a methodology to identify low-dimensional chemical manifolds. Reactive molecular dynamics (MD) is an effective tool for unravelling the complex reaction processes in combustion systems. In this study, chemical kinetic modelling of hydrogen combustion was performed using reaxff-based MD to extract the complete reaction mechanism and statistically compute the rate constants for each elementary reaction at different temperatures from single-atomistic simulation data. The obtained reaction mechanism is in good agreement with the existing models for hydrogen combustion.

Dimensionality reduction aims to reduce the feature space of high-dimensional data while retaining the information and dynamics of the original system effectively. Widely used principal component analysis (PCA) achieves this for combustion data by transforming the original thermo-chemical state space into a low-dimensional manifold with eigenvectors of the covariance matrix of the input data. However, this may not effectively capture stiff chemical dynamics when the reaction zones are localised in space and time. Alternatively, a co-kurtosis PCA (CoK-PCA), wherein the principal components are obtained from the singular value decomposition (SVD) of the matricized co-kurtosis tensor, demonstrated greater accuracy in capturing stiff dynamics. In this study, we evaluated the efficacy of a co-kurtosis-based manifold. Our results show that while ANN outperforms linear reconstruction in general, the proposed CoK-PCA-ANN captures the stiff dynamics better than PCA-ANN.

Biography:

Konduri Aditya works as an Assistant Professor in the Department of Computational and Data Science at the Indian Institute of Science, Bengaluru. Prior to this, he was a postdoctoral researcher at the Combustion Research Facility, Sandia National Laboratories. He obtained his PhD from the Department of Aerospace Engineering at the Texas A&M University, College Station.