摘要：

This PhD work deals with the development of a Computational Aeroacoustics (CAA) method for industrial applications. The constraints linked to this context impose the choice of a hybrid method based on the use of commercial computing codes adapted to turbulent low velocity flows. This approach is based on Lighthill’s Acoustic Analogy, and its application involves two steps. In the first step, the unsteady turbulent flow is computed to determine acoustic source terms, the latter being then propagated in a second step to produce the radiated acoustic field. The implementation is a variational formulation of Lighthill’s Acoustic Analogy with the coupling of Fluent CFD code and Actran/LA acoustic code. It is well adapted to the industry since complex geometries are easily handled in both finite volumes (Fluent) and finite elements (Actran/LA) methods. Two academic configurations are considered. The acoustic radiation produced by two corotating vortices with and without mean flow is first studied for validation. In particular, the goal is to show the necessity to take the local mean flow field into account when computing the source term. A Direct Numerical Simulation (DNS) is therefore performed within Fluent to yield a reference solution; this also reveals the rotating quadrupole nature of the acoustic source. The hybrid method is then applied with success: the source terms are computed from the velocity fields of the DNS, and then propagated to the far field in the spectral domain within Actran/LA. A second verification, in addition to the comparison with DNS results, consists in the analytical resolution of Lighthill’s equation using the Lighthill’s tensor obtained from the DNS. Another important conclusion of this study is that the presence of a mean flow field in both the propagation and source regions only acts on the acoustic waves refraction; however, it is not required to account for it in the source term determination. The second academic study concerns the handling of outgoing turbulent structures from the computing domain. These indeed produce a spurious dipolar acoustic radiation, of numeric nature purely, and with levels high enough to perturb the whole solution. This issue is modeled here with the convection of a perfect vortex through a virtual boundary. Several spatial filters are tested to smooth source terms down to zero at the boundary; the optimal filter tuning depends on the size and convection velocity of the structures to be dissipated. Finally, a real application is considered, the ducted diaphragm at low Mach number. A first Large Eddy Simulation (LES) is performed on a reduced geometry consisting of 10% of the total span. In spite of the model limitations, mainly due to the span reduction preventing a correct three-dimensional development of turbulence, the associated twodimensional acoustic computation yields consistent results. The full scale 3D flow field is then studied, with similarly a LES in which aerodynamic features conform well with the reference DNS. In order to reduce the acoustic model size, source terms are decimated through spatial interpolation. After propagation, the acoustic results suffer from this approximation that would require a more thorough validation.

展开