Development of Scale-Resolving Simulation Capabilities for Turbomachine Blade Design
Please login to view abstract download link
The ambitious environmental objectives assigned to the aeronautical industry in the upcoming years require radical innovation in aircraft propulsion, including groundbreaking engine architectures and usage of alternative fuels. As a result, the upcoming generation of aircraft engines will face more varied and demanding flow conditions, that challenge the capabilities of current RANS-based CFD tools. In this context, aerodynamic and aerothermal design methodologies integrating Scale-Resolving Simulation (SRS) methods, i.e. Large-Eddy Simulation (LES) and Direct Numerical Simulation (DNS), can significantly contribute to overcome the current barriers. In this talk, we will present continued efforts at Cenaero to develop an SRS capability applicable to realistic turbomachine flows. The developments are driven by two main aspects. The first one is the need for acceptable turnaround time, that is adressed by high-order Discontinuous Galerkin spatial discretizations efficiently running on large-scale GPU-based supercomputers. Time integration methods, appropriate mesh generation and methodological improvements are also crucial in reducing the computational time required to reach statistically converged SRS solutions. The second aspect is the ability to correlate SRS results with experimental data for (cross-)validation, involving both the physical representativeness of the simulation (in particular inflow turbulence) and the representation of real-world secondary effects not directly modeled by the simulation setup (e.g. streamtube contraction effects in cascade flows). Examples of applications highlighting the benefits of SRS over RANS will be given. Nevertheless, we will emphasize that SRS should not be considered as a drop-in replacement for RANS in engine design processes because of its computational cost. Consequently, a significant part of the talk will be dedicated to multi-fidelity strategies for the integration of SRS in shape optimization methodologies, including the use of LES and DNS results to benchmark, calibrate and improve turbulence models on a case-specific basis with the help of Machine Learning (ML) techniques.
