Toward high-fidelity subsonic jet noise prediction using petascale supercomputers

The field of jet noise has become one of most active areas of research due to increasingly stringent aircraft noise regulations. A petascalable noise prediction tool-set based on the large eddy simulation (LES) technique is designed and implemented to improve the fidelity of subsonic jet noise predictions. Such tools are needed to help drive the design of quieter jets. The focus is to target computational performance and improved noise prediction fidelity through better matching experimental jet conditions and/or inclusion of the nozzle as part of the simulation. A communication-efficient SPIKE solver is used for spatial operations in conjunction with a non-overlapping multi-block topology based on a new concept of superblocks. These two choices have resulted in efficient scalability tested on up to 91,125 processors (or a theoretical speed of ∼1 petaflop/s). Other important optimizations include parallel file I/O and data buffering while gathering the acoustics. The noise from a Mach-0.9, isothermal jet is studied without and with a round nozzle. Production runs with up to first-ever one-billion-point simple-block topology grids without the nozzle and 125-million-point multi-block topology grids with the nozzle are performed. A vortex ring is used to excite the shear layers in the cases without the nozzle. The fine grid simulations with thinner shear layers have predicted higher sideline noise levels caused by the vortex ring and hence, established the need for nozzle inclusion. The problems of the centerline singularity and smaller time step size due to cylindrical grids have been addressed. A new, faster method based on a sinc filter is discussed for the time step issue in cylindrical grids. Two approaches are considered for nozzle inclusion by: 1) fully resolving the boundary layers at a lower Reynolds number; and 2) using a wall model to model the inner layer at the experimental Reynolds number. The wall-modeled cases exhibited numerical instabilities behind the nozzle lip which contaminated the far-field noise data, whereas the wall-resolved cases showed no such problems. The latter cases predicted noise and spectra that are in better agreement with the experiments. Overall, the inclusion of the nozzle as part of the LES is found to improve the noise predictions. Various innovative noise analysis tools have been used to understand the jet noise to a better extent. Lastly, specific guidelines have been suggested to improve jet noise predictions. It is hoped that the predicted noise levels with improved fidelity will help drive the design of quieter nozzles