Investigation of a fast, robust method for flutter prediction of a transonic compressor at off-design The von Karman Institute for Fluid Dynamics

Abstract

In recent years, the demand for higher compressor performance has led to structures that are increasingly susceptible to flutter instability. Flutter is an aeroelastic phenomenon driven by the exchange of energy between the structure and the surrounding fluid. Although the growth in computational resources now enables precise flutter modelling through simulations, the associated cost remains high, with a typical flutter simulation requiring approximately ten days to complete. Faster methods for flutter assessment are therefore worth investigating.

This thesis evaluates the performance of a coarse mesh using only one harmonic within the Nonlinear Harmonic framework to model flutter behaviour in the rotor blade of an intermediate compressor. The compressor is part of the RISE engine developed by Safran Aero Booster for aviation applications. The assessment focuses solely on the computational fluid dynamics component, which is carried out using the FineTurbo solver.

The method produced an error below 10% regarding the approximated damping ratio across most of the compressor map, except near the stall line at low rotational speeds and in the high speed regime. The first source of error lies in inadequate leading edge flow description and poor resolution of the separation zone under high incidence. The second source of error comes from inaccurate transonic and supersonic shock position capture and differences in modelling shock interactions. The discrepancy between the conventional and low-fidelity methods increases with flutter sensitivity, which exposes the main limitation of the low-fidelity approach. The method proved suitable for a first approximation of aerodynamic damping outside high incidence and shock interaction regimes, but it cannot provide a precise evaluation of the flutter boundary. Moreover, when varying the nodal diameter, the method failed to capture neighbouring effects on both shock position and leading edge flow.

The results also showed large variability depending on the turbulence model, with a 14% difference near stall at low regime and a 37% difference near choke at high regime. The error at low speed arises from spurious activation of the limiter around the leading edge in the k-omega SST model. For the shock interaction regime, the difference comes from variations in turbulent kinetic energy production across the shock. However, because the turbulence model depends on grid size, these differences could not be formally attributed to the turbulence model alone.

The method offers a first approximation of flutter stability outside high incidence and shock interaction regimes, reducing computational cost by a factor of 32 to 62. However, it cannot formally assess compressor stability. Further simulations with mesh refinement or additional harmonics could identify an intermediate optimum between low-fidelity and conventional methods. Alternatively, leading edge refinement may improve accuracy while preserving cost savings.