Evaluation of industrial practices for the prediction of blade flutter in a modern compressor

Abstract

Blade flutter may induce high cycle fatigue or sudden blade loss and must be avoided. Flutter prediction constitutes a rather recent activity at Safran Aero Boosters (SAB), and the current numerical modelling practices of the company require validation. The thesis presents an evaluation of these practices against experimental flutter results recorded during the test campaign of a low-pressure compressor designed at SAB. Using unsteady pressure signals, flutter of the first rotoric blade row is identified at 50% of the nominal rotation speed near the surge line. The instability is shown to occur on the first bending mode presenting ten nodal diameters. Extensive test data allow to establish a data-matched throughflow representation of the experimented flow, using complementary information from RANS calculations. The latter presenting considerable discrepancies with respect to the test data, matching requires the introduction of artificial corrections in the throughflow representation. This representation allows to establish representative boundary conditions for the subsequent URANS flutter calculations. A weakly coupled approach is adopted, whereby an harmonic blade motion is prescribed with the studied mode. Phase-shifted boundary conditions allow to reduce the domain to a single passage. Stability is inferred by computing the work done by aerodynamic forces over a vibration period. The present calculations do not agree with the experiments: the studied mode is predicted stable. SAB’s current practices appear thus unable to predict flutter in near stall conditions. They however enable to highlight the stabilising effect of the motion induced incidence. Blade-to-blade interactions through strong, unsteady tip vortices appear to be the main (if not unique) destabilising mechanism. The lacking quality of the present calculations is believed to originate in part in the mediocre accuracy of the k−" RANS model in separated flow regions and possibly in an inaccurate prediction of the unsteady tip vortices. The dependency of the results on numerical parameters may also be strong. A sensitivity analysis could constitute a basis for future work.