Use this URL to cite or link to this record in EThOS:
Title: GPU based aeroacoustic computation with prefactored compact schemes
Author: Miao, Shuming
ISNI:       0000 0004 6347 3948
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
In this work a computational aeroacoustic (CAA) solver, based on finite difference method, used for sound propagation in engineering practice, is accelerated on graphics processing units (GPUs) by using CUDA FORTRAN. The high-fidelity CAA solver is governed by linearized Euler equations (LEE), which features high-order, optimized prefactored compact schemes with low dissipation and dispersion. Solving prefactored compact schemes gives rise to bidiagonal matrices and it is the dominant computational cost in the CAA solver. Multiple methods for solving the bidiagonal matrix are investigated on GPUs. The numerical methods achieve different performance in the x, y and z directions due to anisotropic memory access pattern. The anisotropic memory access pattern refers to coalesced memory access in a direction, which increases the computational performance, and redundant memory access in the other directions, which increases the computational cost. A new hybrid method is proposed in terms of the anisotropic memory access and a strategy is formulated for solving the bidiagonal matrix in 3D computations. In addition, multiple-GPU implementation is added to the solver. Different parallel strategies are applied to different subroutines in accordance with different memory access patterns. The data transfer between multiple GPUs is also optimized by a direct data transfer between GPUs. Based on a comparison of the wall-clock time on the same amount of CPU cores and GPUs, speed-ups of 40-80 are achieved in double precision. The new solver is used to investigate the scattering of propeller noise off a cylinder and the refraction effect of boundary layer. It is found that the propagation of thickness noise concentrates on the ring plane whereas that of loading noise concentrates on the ring plane and inclines upstream. In addition, the refraction effect of the boundary layer is weak and negligible at M = 0.205 whereas it is significant at M = 0.75. An extension of computation shows that the refraction effect of the boundary layer becomes important when M ≥ 0.3. Finally, the new solver is employed to predict the scattering of propeller noise off an ATR-72 like wing-body at a full scale. Results show that the current solver can be used to predict the large-scale engineering cases and an acceptable wall-clock time and speed-up is achieved.
Supervisor: Angland, David Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available