Research activities at CTRFL broadly span high-fidelity numerical simulation of turbulent reacting flows. The ultimate goal of our research is to develop predictive computational tools for practical combustion devices. Such tools would shorten design cycles and allow for aggressive, high efficiency, low emissions combustion system designs to provide the energy for our future economy. To enable this goal, we must develop a true understanding of the chemistry and physics of interest, develop truly predictive models based this understanding that do not utilize arbitrary, ad-hoc constants, and develop computational tools that scale on current and future high performance computing platforms.
Our technical approach utilizes two fidelities: full-fidelity simulations to probe fundamental physics and chemistry and inform the development of predictive models and high-fidelity simulations appropriate for very accurate engineering calculations, both laboratory scale and device scale. Currently, we are interested in several specific areas and utilize both fidelities to explore these areas. In addition, we are actively developing techniques for assessing the uncertainties in our calculations. Additional details about each of these areas are provided on this page below.
Large Eddy Simulation of Turbulent Reacting Flows
Investigators: Jeffry Lew, Cody Nunno, Bruce Perry
Large Eddy Simulation (LES) is a high-fidelity approach for the simulation of turbulent reacting (or non-reacting) flows. In LES, the governing equations are spatially filtered, and only the large scales of the flow are resolved by the calculations. Therefore, the unresolved small scales, which include small scale turbulence and the chemical reactions, must be modeled. The majority of our work in LES is in modeling these unresolved phenomena. Computationally, our capabilities range from geometrically simple laboratory-scale flames to full-scale combustion devices such as gas turbine combustors. The figure on the left shows the temperature field from a LES of an ethylene turbulent nonpremixed bluff body flame, which consisted of approximately 5M gird points and ran for over two weeks on 96 processors.
Direct Numerical Simulation of Turbulent Reacting Flows
Investigators: Temistocle Grenga, Jon MacArt, Alex Novoselov, Sandra Sowah
Direct Numerical Simulation (DNS) is a full-fidelity approach for the simulation of turbulent reacting (or non-reacting) flows. In DNS, all of the scales of the flow are resolved, and no models are required. However, due to the large range of scales in turbulent combustion (microns to meters), such calculations are not feasible for full-scale device calculations. Nonetheless, DNS is a valuable tool to explore the fundamental physics and chemistry, albeit over limited parameter ranges, since no modeling is required. The major challenge with DNS is the sheer size of the calculations, which are performed on up to tens of thousands of processors on some of the largest supercomputers in the world. As a result, in addition to probing fundamental physics and chemistry, much of our work in this area also involves the development of numerical methods and software algorithms that can scale to these very large computers. The figure on the left is the temperature field from a DNS of a n-heptane turbulent nonpremixed temporally evolving planar jet flame, which consisted of approximately 500M grid points and ran for nearly one month on 32,768 processors.
Investigators: Jon MacArt, Sandra Sowah
The structure and mixing characteristics of turbulence are fundamentally important in developing predictive models for turbulent reacting flows. In LES, the small-scale turbulence is unresolved and must be model. These models are critically important, for this small-scale turbulence is responsible for mixing of fuel and air, wrinkling of flames, and transferring heat and mass toward or away from walls. We currently have two fundamental investigations of the structure of turbulent flows. In the first, we are using DNS to understand the transport of scalars (heat, etc.) in turbulent flows both unbounded and wall-bounded. This investigation will provide a better understanding of the transport of heat from hot combustion gases to cold walls and the mixing of fuel, air, and other chemical species, subsequently leading to improved models that can be incorporated in our LES framework. In the second, we are using DNS to understand the effects of chemical heat release on the small-scale structure of turbulence. Most of our understanding and models for turbulence come from the non-reacting flow community, so this investigation will critically assess the applicability of these models to reacting flows.
Turbulent Combustion Modeling
Investigators: Temistocle Grenga, Jeffry Lew, Cody Nunno, Bruce Perry
In LES, the small flame length scales are not resolved and must be modeled. Our approach utilizes the “flamelet” paradigm, in which a turbulent flame is conceptualized as a laminar flame wrinkled, stretched, and advected by a turbulent flow. Such a model allows for the one-dimensional laminar flame structure to be computed a priori and the results then tabulated and accessed during the LES calculation, leading to a substantial reduction in computational cost and simplified closure of the LES governing equations. However, a “flamelet” model is usually only valid in one asymptotic combustion regime: premixed combustion, nonpremixed (diffusion) combustion, and homogeneous ignition. Our efforts in flamelet modeling seek to develop more general flamelet models applicable to multi-mode and combined mode combustion in the presence of heat losses due to radiation or convection to cold walls as well as pressure changes due to acoustics or bulk compression/expansion.
Emissions: Soot and NOx
Investigators: Jeffry Lew, Cody Nunno
Soot and nitrogen oxides (NOx) are two undesirable by-products of combustion due to their detrimental effects on our environment and health. Unlike the oxidation of fuel to form (primarily) carbon dioxide and water, the formation of soot and NOx are kinetically-controlled processes (i.e., “slow”), and our usual turbulent combustion models do not account for kinetically controlled processes (i.e., chemistry is assumed to be sufficiently “fast”). Furthermore, the chemical kinetics governing the formation of these pollutants, the effects of fuel structure on their formation, and their evolution in turbulent reacting flows remain not well understand at a fundamental level. Our principal efforts in this area involve a combination of laminar flame calculations, LES, and DNS to understand these effects and subsequently development predictive models for LES. In addition, for soot, novel mathematical models are required to describe the evolution of the particle population, and we are currently developing new mathematical models that can predict not only gross amounts of soot but also its size distribution in practical combustion devices. The figure on the left shows the soot volume fraction (along with the location of fuel spray droplets and the iso-contour of a stoichiometric fuel-air mixture) from a recent LES calculation of a Pratt & Whitney aircraft combustor.
Investigators: Cody Nunno
In practical combustion systems at high pressure with large concentrations of soot, a substantial fraction of the chemical energy released from combustion can be lost as thermal radiation. In addition, thermal radiation is a significant heat redistribution mechanism in these systems. From a computational standpoint, radiation is a particularly challenging phenomenon to model due to its dramatic computational cost. We are currently developing new computational capabilities to integrate detailed, spectral radiation models into our LES framework at acceptable computational costs. With such a capability, we will then be able to fundamentally investigate the interactions between turbulence, radiation, and chemistry and, in particular, the role of thermal radiation in suppressing or enhancing the formation of pollutants such as soot and NOx.
Uncertainty Quantification in Turbulent Reacting Flows
While our approaches utilize state-of-the-art, high-fidelity models, we would be foolish to believe that our simulations are exact predictions of reality. All simulations are polluted by some degree of uncertainty, and quantification of this uncertainty is required to make informed decisions based on the simulation results. For simulations of turbulent reacting flows, this uncertainty may be parametric uncertainty in operating conditions, boundary conditions, etc. or parametric uncertainty in kinetic rates, thermodynamic properties, transport properties, etc. In addition, this uncertainty can be structural uncertainties due to improper model assumptions or model extrapolation. Quantification of this uncertainty is a daunting task requiring a characterization of these input uncertainties and the development of efficient methods for propagating these uncertainties through our expensive calculations. Our current efforts in Uncertainty Quantification (UQ) include the development of new, efficient methods for propagating high-dimensional parametric uncertainty through our LES calculations and the development of new methods for characterizing the impact of model assumptions on simulation predictions.