Fan design method improvement: an efficient use of aero-acoustic predictions in a multi-fidelity optimization framework

Abstract

This thesis proposes a new methodology for the formulation of a design strategy involving multi-fidelity optimization techniques. The latter offer additional design capabilities in the engineering toolbox by combining accurate predictions from sophisticated models, often time demanding, with models providing cheap and quick predictions, but less accurate. Their proper applications suffer, however, from the little literature and guidelines allowing the design community to refer to best practices or criterion on which the selection and combination of models may be based. To reproduce an aero-acoustic multi-fidelity environment, different strategies in model downgrading were considered. Based on the current state-of-the-art, models and methods of distinct accuracy and dedicated to the predictions of turbofan aero-acoustic performances were identified as candidates for optimization applications. The whole consists of RANS and HB methods to solve computational fluid dynamics (CFD) problems and whose solutions are, respectively, used to perform RANS-informed analytical and radial mode analysis acoustic methods. To extend the range of fidelity levels on which the models are performed, a set of degraded conditions, including mesh density reductions and blade geometry simplifications, are defined. The main part of the thesis consists into finding, by means of a statistical-based approach, the best combinations of aero-acoustic models. For this purpose, aerodynamic and acoustic quantities, predicted at different levels of fidelity, were intensively compared and the strength of their relationships assessed over a statistical population. Modelling parameters playing a role in the robustness of the predictions performed under degraded conditions are pinpointed. Subsequently, the most promising combination of models is implemented in an automated optimization method and its capabilities are demonstrated in the multidisciplinary optimization of a highly-constrained counter-rotating turbofan. To validate the design strategy derived from the statistical-based approach, the prediction quality of the model combination is compared with aerodynamic and acoustic experimental data, as well as with predictions from more accurate models. Finally, the effects of two different turbulence models used for the CFD simulations on the predictions are highlighted.