The study presented in this manuscript takes place in the context of flow control within tur - bomachinery. Following the demonstration of the effectiveness of the boundary layer suction technique in axial compressors for improving their performance, the present work aims at determining if this same technique may be relevant in radial compressors. Boundary layer suction is then applied on a centrifugal compressor stage, designed and built by Turbomeca, Safran group. The working of the compressor is first analyzed thanks to steady-state and unsteady numerical simulations, performed using the elsA solver developed by Onera, the French Aerospace Laboratory. The turbulence is modeled with the two equations k-l model of Smith. The validity of the numerical results is ensured by comparison with available experimental measurements results: one-dimensional performance coefficients and time-dependent pressure signals. This data were obtained by both Turbomeca, during the characterization of the compressor, and the fluid mechanics laboratory of ISAE, Université de Toulouse, were the compressor is mounted in an experimental test rig dedicated for studying the impeller-diffuser interactions. The detailed analysis of the numerical results reveals the growth of a corner separation within the diffuser between the hub endwall and the vane suction side when the operating point moves toward surge. The precise location and the internal structure of that boundary layer separation are then elucidated thanks to a topological study, which allows to identify the singular points and the separating lines of the skin-friction pattern. In steady-state numerical simulations, the development of that corner separation leads to the stall of the diffuser, which compromises the compressor stage performance and finally the stability of the numerical model. Unsteady numerical simulations results allow to specify the temporal behavior of the corner separation: the extent of the separated zone is modulated by the scrolling of pressure waves created by the impeller-diffuser interaction, but the trajectories of fluid particles within the separation match with the streamlines of the time-averaged flow field. The unsteady separation is then fixed, and its topology is in agreement with the prediction of the steady-state numerical simulations. In particular, the location of the separation is predicted similarly by the steady-state and the unsteady numerical models. Afterward, a control strategy using suction technique is developed thanks to the previous conclusions. The suction slot is set in the neighborhood of the main saddle of the separation, which corresponds to the location of the maximum of the time-averaged adverse pressure gradient. This strategy is implemented within both the steady-state and unsteady numerical models. The steady-state numerical model predicts the complete control of the separated zone with a removal of 1% of the total massflow through the compressor. This leads to a significant increase of the numerical stable range. With a removal of 0.3% of the compressor total massflow, the numerical model predicts only a partial control of the corner separation, but it also leads to a significant increase of the numerical stable range. With the unsteady numerical model, the initial hub corner separation is again controlled thanks to aspiration. But the results also highlight the major role played by the scrolling of pressure waves. They reinforce when crossing the diffuser throat, and generates a strong instantaneous adverse pressure gradient. This maximum provokes a new boundary layer separation, further downstream of the suction slot. The stable range of the unsteady numerical model is not increased. These results put into evidence the possibility to act on boundary layer separation that occur in radial diffusers thanks to the boundary layer suction technique. [...]
