SURROGATE-BASED OPTIMISATION FOR A MIXED-VARIABLE DESIGN SPACE: PROOF OF CONCEPT AND OPPORTUNITIES FOR TURBOMACHINERY APPLICATIONS

摘要

State-of-the-art turbomachinery design processes rely more and more on the extensive use of numerical simulations. To deal with expensive high-fidelity computations, surrogate-based optimisation (SBO) has become a very interesting approach. In order to cope with industrial cases, the capability to handle variables of a mixed nature appears key. Innovative and auto-adaptive surrogates have been implemented within MlNAMO that are capable of natively handling the different natures of the design parameters. The present work discusses this mixed-variable SBO framework applied to a multi-profile combinational problem inside the bypass duct. A proof of concept is given followed by a more advanced application. It is demonstrated that the proposed mixed-variable SBO efficiently delivers reliable results and that it offers many opportunities during a conceptual design phase.