Président du jury Degrez, Gérard
Promoteur Anthoine, Jérôme P.L.R.
;Leduc, Bernard Publication Non publié, 2011-09-14
| Résumé : | This work proposes to address the computation of flow-generated noise, including the high frequency components in an acceptable computational time, relative to industrial applications. In this framework, hybrid methods based on aeroacoustic analogies are relevant to predict the corresponding sound, in term of computational time and easy implementation, and are separating the sources of sound computation from the sound propagation itself. Both parts of the hybrid method are then investigated. Firstly, the sources of sound should be computed accurately, reproducing most of the energy content of the turbulent flow. The present study is mainly based on LES computations, that are a good compromise between the accuracy and the time required to obtain the sources, especially for low Mach number application, where incompressible flow computations can be used. The influence of the computational parameters is then studied through the investigation of the influence of the boundary conditions, flow solver, mesh refinement, LES subgrid-scale model and computational domain definition. Secondly, the sound sources propagation is computed using classical analogies based most of the time on free field Green's functions. They are limited to the low frequency range of the emitted sound spectrum, where the source is compact, if incompressible acoustic sources are used. They are then proposed to be combined with Amiet's theory for airfoil noise to cover the remaining high frequency part of the sound spectrum. The limitations of Amiet's theory are analysed through the geometrical and acoustical far-field assumptions. Furthermore, an inverse strip method is proposed to extend Amiet's theory to spanwise varying flow conditions and taking correctly into account the spanwise wave number effects. The leading-edge noise mechanism is studied through the interaction of the turbulent region of a jet with a NACA0012 airfoil. The hybrid method, consisting in an incompressible LES computation combined with Curle's analogy, showed its limitation to the low frequency range of the sound spectrum when low-order CFD are used. The Amiet's theory, based on the modelisation of the upstream velocity spectrum, provides a good sound prediction at high frequency, as far as the airfoil thickness is accounted for. The Trailing-edge noise mechanism is studied through the flow around a mid-span cut of a blade (CD airfoil) of an automotive cooling fan. Three different acoustic methods are used : Curle's analogy is using wall-pressure fluctuations over the airfoil surface, Ffowcs-Williams and Hall's analogy is using volumetric velocity informations around the trailing-edge and Amiet's theory for trailing-edge noise is using the wall-pressure spectrum around the trailing-edge and the corresponding spanwise correlation. The methods using wall-pressure informations are shown to directly transfer the informations of the wall-pressure spectrum to the sound spectrum. Differences are appearing in the higher frequency range where the airfoil starts to be non compact, invalidating the use of Curle's analogy at such frequencies compared to the other methods, taking scattering effects into account implicitly in their formulations. Finally, in order to reduce the computational cost, Amiet's theory for trailing-edge noise is proposed to be driven from steady RANS computations. Two methods to compute the wall-pressure spectrum from boundary-layer informations are studied. Their respective robustness and reliability are analysed in an uncertainty quantification framework, in case of varying velocity profiles upstream the airfoil. Both models showed similar results and tendency compared to the wall-pressure LES spectrum, as far as the flow topology remains similar. |
