A study of sound generation from turbulent heated round jets using three-dimensional large eddy simulation

Improvements in computing speed over the past decade have made Large Eddy Simulations (LES) amenable to the study of jet noise. The study of turbulent hot jets is required jets since all jet engines fitted on aircraft operate at hot exhaust conditions. The primary goal of this research was to further advance the science of jet noise prediction with a specific emphasis on heated jets using 3-D LES. For the 3-D LES methodology, spatial filtering is used as an implicit subgrid scale (SGS) model in place of an explicit SGS model, such as the classical Smagorinsky or Dynamic Smagorinsky models. To study the far-field noise, the porous FfowcsWilliams-Hawkings (FWH) surface integral acoustic formulation is employed. Results obtained for the heated jets in terms of jet development are in good agreement with other LES results and experimental data. The predicted overall sound pressure level (OASPL) values for heated jets exhibited the same trend as experimental data. The levels were over-predicted by approximately 3 dB, which was deemed satisfactory. An investigation of noise sources for heated jets was also performed within the framework of Lighthill’s acoustic analogy. It is discovered that when a high-speed is jet heated, significant cancellations occur between shear and entropy noise sources compared to an unheated high speed jet. This could explain why a high speed heated jet is quieter than an unheated jet at the same ambient Mach number. High-order compact finite difference schemes along with high-order filters are used extensively in LES, especially for aeroacoustics problems, since these schemes have very high accuracy and spectral-like resolution as well as low-dispersion and diffusion errors. Due to the implicit nature of compact schemes, one technique of parallelization is based on the data transposition strategy. However, such transposition strategy is near impossible to apply to jets with complex geometries. Hence, an alternative parallelization methodology based on the Schur complement technique was proposed to address the decomposition deficiency of the transposition strategy. Good scalability with a nearly linear was obtained for the 3-D Schur complement up to 1,024 processors on a CRAY XT3 supercomputer. The 3-D Schur complement is slightly slower compared to the transposition scheme by about 10% on a CRAY XT3. On a cluster with ethernet connection between compute nodes, the Schur complement was faster than the transposition scheme by approximately 20%. The computational overhead associated with the Schur complement matrices may be significant, and offset the two fold reduction in the communication time in some instances when compared to the transposition strategy. Nonetheless, the Schur complement is robust and has been able to handle a massive grid size of 2 billion grid points which ran on 4,096 processors. Kinetic based methodologies such as the Lattice-Boltzmann Method (LBM) have been used extensively to model complex fluid flow phenomena. The LBM was used to study the far-field noise generated from a Mach 0.4 unheated turbulent axisymmetric jet. A commercial code based on the LBM kernel is used to simulate the turbulent flow exhausting from a pipe. Near-field flow results such as jet centerline velocity decay rates and turbulence intensities are in agreement with experimental results and results from comparable LES studies. The predicted far field sound pressure levels are within 2 dB of published experimental results. Weak unphysical tones are present at high frequency in the computed radiated sound pressure spectra. These tones are believed to be due to spurious sound wave reflections at boundaries between regions of varying voxel resolution and do not affect the overall levels significantly. The LBM appears to be a viable approach, comparable in accuracy to LES, for the problem considered. The main advantages of this approach over Navier-Stokes based finite difference schemes may be a reduced computational cost, ease of including the nozzle in the computational domain, and ease of investigating nozzles with complex shapes such as lobed mixers and chevrons