Suresh Menon
Chris Stone
Balu Sekar
Joseph Zelina
Mohammed A. Mawid
|
|
Suresh Menon |
|
|
Computational Capabilities for High Performance Gas Turbine Combustion Modeling combustion phenomena has proven to be a rather daunting task requiring the synergy of several different research disciplines (e.g., chemistry, physics, computer science, etc.). The difficulty in modeling these systems lies in the complex, non-linear interaction of several physical processes (vortex/flame, flame/turbulence, acoustic/flame, etc.) with chemical reactions. For a modeling technique to be effective as an engineering design tool, it must be able to capture these complex processes (and many others) in a reasonable amount of time. Currently, the modeling techniques are based largely on the Reynolds Averaged Navier-Stokes (RANS) approach and to a lesser extent, on Large Eddy Simulation (LES) modeling. While RANS models all the scales in the flow, LES has the capability of capturing the highly convoluted, unsteady flow-field in a tractable amount of time. In the LES method, turbulent eddies larger than some pre-defined size, typically on the order of the computational grid, are directly simulated in both space and time while turbulence scaling laws are used to model those eddies below this cut-off. At AFRL and at Georgia Institute of Technology, much has been invested into the development of these state-of-the-art LES modeling techniques suitable for engineering-level design studies. These techniques have been used to model a wide range of combustion systems such as gas-turbines and ramjets/scramjets. While proven RANS software has existed for quite some time, LES technology is still evolving and appears attractive on modeling unstable combustion dynamics and pollutant emissions in gas-turbine combustors. The use of RANS methodology is applied where global features of the flow need to be captured and while detailed capturing of the smaller scales are not needed. With this approach, quick estimates of the flow can be obtained in a lesser time compared to LES. LES is well suited for studies in the field of combustion dynamics which is concerned with the unsteady instabilities often plaguing combustion systems. Combustion instabilities most often occur through a series of closed-loop feedback mechanisms in which flow perturbations interact with the flame or fuel-feed lines (or both). This, in turn, generates acoustic waves which are then reflected back causing additional perturbations, thus closing the loop. If the phase between these events are correct, they can amplify each other leading to high-amplitude oscillations and possible structural damage. In addition, the effect of swirl on the subsequent stability of the combustion system has been extensively studied by LES. In the prediction of pollutant emissions for gas-turbine combustors, LES is being used. By including models for the production/destruction rates of NO and CO (i.e., Nitrous Oxide and Carbon Monoxide) based on the local temperature, turbulence intensity and fuel content, the emissions of these pollutants can be predicted. By using empirical models for the pollutant production, finite-rate chemistry computation is avoided which greatly reduces the simulation cost. Another method for calculating pollutant predictions has been to directly compute the chemical species with finite-rate chemistry, within the LES framework. This method has been previously avoided due to the extreme computational expense associated with solving these very stiff chemical rate equations. To reduce the computational cost, artificial neural networks (ANN) are being developed which can approximate the chemical reaction rates, thus avoiding the stiff numerical integration. Typical examples of the RANS and LES methodologies, applied to combustor flow fields, are illustrated in the poster.
Suresh Menon and Chris Stone, Georgia Institute of Technology |