A new efficient algorithm for computational aeroacoustics on massively parallel computers

As transport aircraft engines get larger and larger, the dominant noise source component becomes the fan (forward and aft). A computational massively parallel aeroacoustics algorithm that has been developed to predict forward noise radiation from such engines is described in this paper. The numerical algorithm is a 4th order accurate (both in space and time), Euler/Navier-Stokes solver that is written in essentially High Performance Fortran (HPF). The scheme uses the classical four-stage Runge-Kutta time integration, central differencing for spatial derivatives, and a blend of second and sixth order, Jameson type artificial dissipation to sup press high frequency numerical oscillations. On the outer boundaries of the computational domain, non-reflecting boundary conditions are used. Specifically the algorithm has been optimized for the CM-5 of Thinking Machines Corporation. An overlap ping mixture of the fluxes of the interior points and the far-field boundary points is established by making use of the MERGE statement which the programming languages, CM Fortran and Fortran 90, facilitate. This procedure leads to a simultaneous discretization of the spatial derivatives for the three different sets of governing equations across the domain: Navier-Stokes/Euler equations, radiation boundary conditions and outflow boundary conditions. Hence, approximately 60% reduction in computing the residuals of the governing equations is obtained on the Connection Machine computers. Results for a steady engine inlet flow, oscillating flat plate in a viscous fluid, scattering from a sphere, sound radiation from a baffled oscillating piston, and sound propagation through an axisymmetric inlet are presented.