Session: 22-05 Flutter models

Paper Number: 101894

101894 - Development of a Reduced Order Model Based on Aerodynamic Influence Coefficients to Simulate Aeroelastic Phenomena in Axial Compressor Blades 

Recent advancements in aerodynamics and materials of turbomachinery are leading the development of

blades towards lightweight and slender shapes to fulfil the requirements of increased aerodynamic

efficiency and reduced engine weight and fuel burn. This in turn could make the blades more prone to

aerodynamic and aeroelastic instabilities. For this reason, prediction and avoidance of flutter are becoming

increasingly important in designing new blade shapes. Aeroelastic computations in high-speed

turbomachinery are challenging due to the inherent nonlinear nature of the fluid and structural

phenomena and the state of the art still relies on uncoupled simulations where the blade is forced to

oscillate according to a predefined motion, with the exchanged energy used as a measure to assess

stability. On the other hand, time-marching coupled models require large amounts of computational

resources to proper describe aeroelastic interactions. In this work a reduced order model (ROM) is

proposed to simulate the aeroelasticity of compressor blades in both coupled and uncoupled mode at low

computational cost. The ROM describes the structural dynamics of blades using the mode superposition

technique, centrifugal stiffening can also be accounted for, while the aerodynamic forces are modelled as

lumped in the centre of pressure using various system identification techniques. The ROM takes advantage

of the aerodynamic influence coefficients (AIC), a technique to model the aerodynamic forces of vibrating

turbomachinery blades based on the effect superposition hypothesis. In this approach, a training

computation with 2N+1 blades is run, where only the central blade is forced to oscillate according to a

predefined law of motion. The forces on all blades are related to the input motion signal in order to build

I/O models, the overall force on each blade is then computed in the ROM by summing up the contribution

of the 2N neighbouring blades. Polynomial models as the nonlinear exogenous input moving average

(NARX) and recurrent neural networks (RNN) are adopted in the system identification of aerodynamic

forces, using blade displacement and velocity as inputs. The ROM is first run in uncoupled mode, adopting a

travelling wave vibration pattern, and the computed aerodynamic damping is compared to the Fourier

method and harmonic balance simulations of reference, showing a very good agreement at a fraction of

the computational cost. Then a comparison is performed between the coupled ROM and the full order fully

coupled model of reference, which links a transient structural finite element model to a transient

computational fluid dynamics model of the blades. The two models are compared in both a single-passage

case, adopting periodic boundary conditions, and in a full-anulus case of the rotor. Again, an overall good

matching is achieved in terms of vibration amplitude, frequency, and aerodynamic damping, with a

computational cost two orders of magnitude lower, showing the ability of the ROM to accurately describe

aeroelastic phenomena in compressors in a preliminary design phase.

Presenting Author: Marco Casoni Università  Di Padova

Presenting Author Biography: M.S. in aerospace engineering at the University of Padova in 2019, currently PhD candidate at the same institution.
His fields of research are aeroelastic computations in turbomachinery and optimization methods for engineering

Authors:

Marco Casoni Università  Di Padova
Ernesto Benini Università di Padova
Andrea Magrini Università di Padova