1. Home ›
  2. Research

Research

Turbulent Combustion Modeling and Control

Investigators: Priyesh Kakka, J. Jacobowitz, Nathan Ziems, Jonah Ikeda

Current Large–Eddy Simulation (LES) and Reynolds-Averaged Navier–Stokes (RANS) closure models are predicated on the forward transfer of kinetic energy from the resolved scales to the unresolved scales, which is overly limiting for turbulent combustion regimes in which energy transfer away from the small scales significantly affects the large-scale turbulence. Using high-fidelity Direct Numerical Simulation (DNS) and physics-constrained machine learning (ML), we investigate and develop new models for the energy dynamics for regimes in which correctly modeling flame–turbulence interactions is crucial to predictive accuracy. These predictive capabilities may be essential to the design of next-generation combustors for low-emission jet engines and high-efficiency stationary gas turbines.

At the same time, we are developing combustion modeling and control techniques using adjoint-based ML. These techniques have two overarching objectives: (i) to reduce the computational cost of simulations of highly compressible, shock-coupled, and igniting/extinguishing flames by developing and deploying reduced-order thermochemical representations, and (ii) to increase combustion stability through active control in highly unstable environments such as scramjet combustors and rotating detonation engines.

Publications:

Nonequilibrium Flow Modeling

Investigators: Ashish Nair, Mikolaj Kryger

Hypersonic flows in the transition-continuum regime (Knudsen numbers 0.1 to 10) are inaccurately predicted by the Navier–Stokes equations due to the failure of the continuum approximation and associated viscous boundary conditions. While direct simulation Monte Carlo (DSMC) solutions of the Boltzmann equation provide accurate predictions in this regime, their cost increases as the Knudsen number decreases (increasing molecular number density), hence Navier–Stokes solutions are desirable for the transition-continuum regime. We have developed several methods to extend the validity of the Navier–Stokes equations to subcontinuum flow by optimizing the parameters of first-order neural network-based transport models, second-order models for the viscous stress and heat flux, and slip-wall boundary conditions based on modeled nonequilibrium distribution functions. Extensions to geometrically more complex flows and reacting flows are ongoing.

Publications:

Aerodynamics, Turbulence, and Flow Control

Investigators: Xuemin Liu, Saif Elmaleh

We have developed adjoint-based ML methods to optimize over turbulent flows for LES and RANS closure modeling and flow control, with applications to nonreacting laminar and turbulent separated flows. These methods optimize neural network closure and control models over the governing Navier–Stokes PDEs, which provide a physics constraint that ensures consistency of the learned models with the flow dynamics. This aspect is crucial, as a successfully trained network is guaranteed to improve the predictive accuracy of LES/RANS or to control a flow according to its control objective. Ongoing work is focusing on online and data-free optimization methods, in particular methods that are stable for long-time and time-averaged optimization over chaotic flows.

Publications:

Numerical Methods and Machine Learning

Investigators: Mikolaj Kryger, J. Jacobowitz

It is no secret that the Age of Data has resulted in the availability of a wealth of tools for high-dimensional data fitting and decision-making, many of which are optimized for high-performance, heterogeneous compute architectures and which are being rapidly assimilated into the world of science. Machine learning methods have proven wildly successful in areas such as image recognition, but work remains in their application to physics-constrained problems in engineering and applied science. Using physics-constrained training methods, including an adjoint-based a posteriori training algorithm for LES, we apply tools such as the PyTorch library to turbulence model development for canonical turbulent flows. So far, we have used these methods to produce models of comparable accuracy to “traditional” turbulence models on coarser computational grids, which translates to significantly reduced computational cost. Alternatively, for fixed computational cost, the deep learning-based approach can produce models of much higher accuracy. Extensions to more complicated flows, including turbulent reacting flows, are underway.

Plasma-coupled Combustion

Investigators: J. Jacobowitz

Flame ignition and stabilization are critical operational concerns in the design of supersonic-combusting ramjet engines (scramjets) for high-speed flight. One possible route involves flame ignition by focused-laser-induced optical breakdown, in which charged-species kinetics and post-shock hydrodynamics evolve simultaneously. The combination of these phenomena has been observed to give rise to flame instabilities, including the well-known cellular instability, but the precise transition mechanism from micro-scale evolution to macro-scale instability is unknown. We explore these mechanisms using highly detailed one-, two-, and three-dimensional numerical simulations that include plasma kinetics, multi-component diffusion, and low-dissipation shock-capturing methods. Possible future directions include the use of adaptive-grid and wavelet-based discretizations and investigation of the transition mechanism from combustion instability to turbulent combustion.

Publications: