TY - THES ID - 148639886 TI - Master thesis : Metamodeling for aerodynamic optimisation of a low-pressure compressor blade AU - Levaux, Nayan AU - Terrapon, Vincent AU - Ponthot, Jean-Philippe AU - Duysinx, Pierre AU - Nigro, Rémy PY - 2019 PB - Liège Université de Liège (ULiège) DB - UniCat KW - Metamodeling KW - Optimisation KW - Aerodynamics KW - Booster KW - Ingénierie, informatique & technologie > Ingénierie aérospatiale UR - https://www.unicat.be/uniCat?func=search&query=sysid:148639886 AB - The company Safran Aero Boosters for which this master thesis was conducted is the world leader in conception of low-pressure compressors, also called boosters, for commercial turbofan engines. In a search of continued improvement, the engineers of the company are regularly confronted to the aerodynamic design of blades, a time consuming iterative procedure between the aerodynamics and mechanical teams. Within the conception process, the design space exploration consists in carrying out small modifications to a reference blade and then reassess its performances with long running CFD computations until finding the optimal one. In order to spare computational cost and time, metamodels could be used instead. Constructed on a limited database obtained with CFD computations, the metamodels give predictions on blade performances in an instant and can be intensively employed with optimisation algorithms. 

The main objective of this master thesis is to gain experience in the optimal construction of metamodels. In this context, highly predictive metamodels have been constructed on a large database and the relationships between the design variables modifications and blade performances analysed. In an attempt to reduce the computational cost associated to the creation of metamodels, a comparison of their predictability was made when they are constructed on alternative databases of reduced size. This study concluded that a database of 20 elements obtained with the Latin Hypercube sampling method appeared to be the most adequate choice in this application. 

After that, an enlargement of the design space was considered as modifications on the initial blade induced changes in other design variables. This enlarged design space was once again explored with the metamodels constructed this time with a reduced size database, result of the previous comparison, and with CFD computations on a smaller domain. In fact, with the reduced domain, the blades too different from the initial one encountered issues in convergence of their CFD computations. Nevertheless, the surrogate models were sufficiently predictive for this work and optimised blades were determined. Among those ones, the blade for which was predicted an improvement of both efficiency and operability satisfied all expectations. To finish this thesis, an analysis of the flow around these optimised blades was conducted in order to better understand the reasons behind the performances improvement with the blade geometry modifications. ER -