Optimization of an Ogv Cascade Using Multi-Fidelity Surrogate Modeling and Active Learning With Scale-Resolving Simulations

Achieving the European Union's 2050 climate-neutrality target requires substantial reductions in aviation emissions, motivating the use of high-fidelity scale-resolving simulations, and specifically LES, to improve engine efficiency, despite the prohibitive computational cost that limits their systematic deployment in design optimization. To address this challenge, we fuse LES and a low-fidelity RANS model in an automated Bayesian multi-objective optimization framework supported by a co-Kriging multi-fidelity surrogate. The framework autonomously explores the design space using Bayesian infill strategies and enforces both geometric and aerodynamic constraints: geometric constraints are imposed exactly, while aerodynamic constraints are handled probabilistically. We apply this framework to the optimization of an outlet guiding vane cascade at an inlet Reynolds number of 1.5 × 105, with the objective of minimizing mixed-out pressure losses at one design and two off-design incidences. To evaluate the effectiveness of the proposed MF approach, we compare the results with those from RANS-only and LES-only optimizations and examine the aerodynamic performance of the resulting optimal designs. The results show that the proposed framework is able to significantly improve the baseline geometry performance while providing infill candidates that satisfy geometric and aerodynamics constraints.