Examining the Potential of High-Order Scale-Resolving Simulation to Support RANS-Based Compressor Airfoil Optimization

Abstract Turbomachinery aerodynamic optimizations are predominantly carried out using Reynolds-averaged Navier–Stokes (RANS)-based computational fluid dynamics (CFD). The approach has reached a high level of maturity over the past years through extensive practical experience. With the ever-increasing demands on designs, the demands on simulation accuracy are also increasing, and efforts are being made to incorporate scale-resolving simulations (SRSs) into the design process of turbomachinery. Although SRS still remains too costly for primary use in industrial optimization, ongoing advancements favor its gradual integration. This is supported by design trends such as smaller core engines, resulting in locally reduced Reynolds numbers. Potential boundary-layer separation and a high level of unsteadiness in these low Reynolds number flows amplify the uncertainties of RANS. At the same time, the computational requirements of SRS are drastically reduced due to the reduced bandwidth of turbulent scales. To assess the potential of utilizing SRS in optimization frameworks, RANS-optimized airfoils are re-evaluated with large eddy simulations (LESs) based on a high-order discontinuous Galerkin solver. First, a RANS optimization is performed for a low Reynolds number airfoil with the aim of reducing the loss at the design point and increasing the operating range, while adhering to a constraint of nearly axial outflow angle. A subset of Pareto-front geometries is then re-simulated using LES to assess the impact of the chosen CFD methodology on the optimization result. Detailed flow analyses give insights on the deficiencies of RANS. The results demonstrate how optimizations can be driven into a sub-optimal direction when relying solely on RANS, underscoring the necessity of incorporating SRS into the process and providing initial insights into how this can be done. It is demonstrated how data obtained from only a few SRS can be fed back into the optimization process, leading to an improved optimization outcome.