The proposed effort relies on machine learning strategies to develop a data-based framework to develop models for the low-temperature oxidation (LTO) of complex fuels. Machine-learning, a term associated with a range of data analysis and discovery methods, can provide enabling tools for effective data-based science in chemical kinetics as argued below. These tools can carry out a variety of automated tasks that include regression to develop models for the reaction rates, clustering to identify groupings of species that track a particular mechanistic phase of the oxidation, classification to identify the stages of ignition and dimensionality reduction. Dr. Echekki will team up with Drs. Aamir Farooq and S. Mani Sarathy (KAUST) and Matthias Ihme (Stanford) to develop the proposed data-based modeling framework starting with experimental data generated by the KAUST team. The principal milestones of the proposed effort are: 1) The development of a reduced chemistry description using a hybrid chemistry model for a set of simpler and mixtures of fuels based on the available measurements. 2) Testing of the modeling framework using existing chemical mechanisms for fuels and generating hybrid models for their chemistry. 3) Publication of the outcome of the research collaborations in archived journals at the end of each stage.

The objective of the proposed effort is to advance and implement the strategy of principal component (PC) transport as a model reduction strategy for combustion DNS. This approach has been implemented by the PI for combustion DNS on a much smaller scale than the inherent capability of exascale computing. The effort will 1) support the research activities at Sandia in the implementation of PC transport within the Regent programming framework to optimize data parallelism, 2) implement validation and benchmarking studies, and 3) develop strategies to evolve PCs within a simulation.

A novel framework for data-driven modeling in turbulent combustion is proposed and investigated. The framework relies on experimental and numerical data to develop comprehensive descriptions and characterizations of the composition space and scalar statistics relevant to problems of interest. The proposed effort attempts to address the main challenges to develop robust frameworks for data-driven modeling. The first challenge is related to the reconstruction of the composition space through an evaluation of missing data and the account for potential uncertainty in the measurements. The second challenge is related to the construction of reduced data-based models that reconstruct both the composition space and the scalars������������������ statistics. This is implemented using principal component analysis (PCA) and tabulation using artificial neural networks as well as the tabulation of joint statistics using the kernel density estimation (KDE) approach. Both detailed simulations from direct numerical simulations and experimental data are considered to validate the approaches designed to overcome the above challenges.

The Liquid Rocketry Lab (LRL) is a dedicated group of students at NC State University with a diverse background in rocket design, control system design and remote telemetry with the goal of constructing liquid rockets. The goal of the request is to participate at the HeroX Base 11 Space Challenge competition to launch a liquid-propelled, single stage rocket to an altitude of 100 km (the so-called Karman line). The goal of the proposed support the construction of the rocket engine that will be used for the competition.

Current designs for hydrocarbons-fueled scramjet propulsion systems rely on the concept of endothermic cooling, in which the fuel absorbs the heating load from the structure. This can induce a transition from a liquid to a supercritical-fluid state, and upon injection, the supercritical fuel can exhibit various phase transitions. This proposal focuses on the detailed numerical simulation of the injection event, including various phase transitions that might occur, followed by mixing and combustion. A hybrid large-eddy simulation / Reynolds-averaged Navier-Stokes (LES/RANS) turbulence-modeling framework will be extended to incorporate sub-models for phase separation, droplet vaporization, and turbulence effects on reactant mixing and combustion. The framework will be applied to various injection scenarios relevant to current scramjet designs.

The proposed effort seeks to develop a robust multiscale strategy for the simulation of turbulent reacting flows. The strategy comprises two important elements. The first element addresses the development of composition-space parameterization strategies that enable the optimum description of the thermo-chemical state of reacting system using principal component analysis (PCA) for parametrization and artificial neural networks for the recovery of the thermo-chemical variables through their tabulation in terms of the principal components. The second element consists of the development of a multiscale flame-embedding approach based on coupling large-eddy simulation (LES) with embedding flamelet solutions using the one-dimensional turbulence (ODT). The proposed strategy enables the development of high-fidelity simulations of complex reacting flows, but with a much reduced cost compared to directly computing with direction numerical simulations.

Combustion noise is an emerging issue associated with lean-burn systems, the potential way forward for environmentally friendly engines with reduced pollutants emission. This multi-physics problem remains very challenging that requires knowledge in fluid mechanics, acoustics and combustion. So far no precise predictive methods exist because high-fidelity numerical data of turbulent flames required for the modelling are not yet available. In this project it is proposed to bridge this gap by bringing together experts in acoustic modelling and combustion numerical simulation from the two institutions. The resulting prediction model that captures the spectral characteristics will eventually help mitigate noise emission from realistic lean-burn engines at design stages. The collaborative project will be based on the mutual visits and the placement of a PhD student. The outcome of the research will be published in a flagship international aeroacoustics conference and a high-impact journal. The project is also expected to establish the foundation of a long-term collaboration that aims to address a problem of global environmental importance.

The objective of the proposed research is to develop computational methods within a multiscale modeling framework in turbulent reacting flows. The framework is based on hybrid coarse-grained and fine-grained (resp.) simulations based on large-eddy simulations (LES) and the one-dimensional turbulence (ODT) model. The ODT model is a one-dimensional stochastic model that is designed to capture the coupling between turbulent and molecular transport and chemistry/heat release. The framework has been developed by the PI for combustion to directly model subgrid scale physics associated with turbulence-chemistry interactions; but, it can easily have broader applications for multiscale driven flows. The work proposed here extends the model formulation to address the development of multiscale computational tools to couple the two solution schemes and render the computations more efficient. Different physics-based and mathematically rigorous strategies will be formulated to address this coupling. The various formulations will be validated with direct numerical simulations (DNS).

The PI's propose collaborative computational-experimental efforts to understand the mechanisms that govern flame dynamics and structure at lean fuel conditions based on a canonical flame-flow configuration: the post-ignition turbulent premixed flame kernel. Turbulent flame kernels are encountered in spark-ignition engines as well as in other more complex configurations that involve the onset of autoignition in non-homogeneous mixtures or the break-up of turbulent flame structures to form flame pockets. Operation at fuel-lean mixture conditions may result in flame quenching, which may be followed by reignition. It also contributes to thermo-acoustic instabilities, which plague the operation of practical combustion devices operating at lean conditions, such as gas turbines. The combined computational-experimental efforts will address the mechanisms, which govern flame kernel dynamics (inherent instabilities) and structure (local quenching) under turbulent conditions.

The objective of the proposed research is to develop a multiscale modeling framework for turbulent and molecular transport, complex chemistry and radiative heat transport in turbulent reacting flows. The framework is based on hybrid coarse-grained and fine-grained (resp.) simulations based on large-eddy simulations (LES) and the one-dimensional turbulence (ODT) model. The ODT model is a one-dimensional stochastic model that is designed to capture the coupling between turbulent and molecular transport and chemistry/heat release. The LES-ODT framework can be contrasted to traditional models for turbulent combustion, which are based on the transport of moments of scalars; with emerging combustion technologies, which often exhibit strong non-equilibrium effects, these traditional models are often inadequate.