Journal of Fluids and Structures 85 (2019) 110–125

Contents lists available at ScienceDirect

Journal of Fluids and Structures

journal homepage: www.elsevier.com/locate/jfs

A comparative study of coupled and decoupled fan flutter

prediction methods under variation of mass ratio and blade
stiffness
∗
C. Chahine a,b , , T. Verstraete a , L. He b
a
von Karman Institute for Fluid Dynamics, Chaussée de Waterloo 72, 1640 Rhode-Saint-Genèse, Belgium
b
University of Oxford, Department of Engineering Science, Osney Lab, Oxford, OX2 0ES, United Kingdom

article info a b s t r a c t

Article history: Flutter is a long standing issue for fan blades of civil aero-engines and becomes of further
Received 22 May 2018 concern for modern light-weight designs with increasing fan diameters to reach ultra-
Received in revised form 11 December 2018 high bypass ratios. Accurate flutter prediction is therefore of prime consideration in the
Accepted 21 December 2018
design process in order to avoid catastrophic blade failure in operation or expensive
Available online 2 January 2019
redesign iterations if spotted in ground or flight tests. The traditional energy method,
Keywords: which is based on the assumption of negligible aeroelastic coupling, has been used to
Flutter great extend to predict flutter of turbomachinery components including aero-engine fan
Energy method blades, and is today by far the most widely applied technique. The underlying assumption
Decoupled of fluid–structure decoupling, however, has to be questioned for large fan blades that
Coupled are characterized by low mass ratios and low stiffness. Implications of the violation of
Transonic fan the system decoupling assumption on the prediction capabilities of the energy method
Composite fan
are important to understand for the fan designer in order to allow an informed decision
on the flutter prediction tool to use. In this work a comprehensive comparative study is
presented in which the energy method is contrasted to the predictions of a strongly coupled
fluid–structure interaction method for varying values of mass ratio and blade stiffness of
a transonic three-dimensional fan rotor. The strength of aeroelastic coupling is evaluated
in terms of the aeroelastic frequency shift and its impact on the prediction accuracy of the
energy method is investigated. The results show the capability of the energy method to
accurately predict flutter for a wide range of mass ratio and stiffness configurations, but
its prediction accuracy is reduced for combined low mass ratio and low stiffness blades.
Mechanisms governing the aeroelastic frequency shift are explained to allow a better
understanding of the effect and a method for its prediction based on results of a decoupled
analysis is shown.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction

Self-excited aeroelastic instabilities, known as flutter, are a dangerous phenomenon for fan blades of modern civil aero-
engines. Flutter occurs when small blade perturbations in certain structural vibrational modes become amplified by the
unsteady flow field surrounding the blades. If the introduced excess vibrational energy cannot be dissipated by some form
of structural damping, the vibration amplitude grows rapidly to reach harmful values leading to the destruction of the

∗ Corresponding author at: von Karman Institute for Fluid Dynamics, Chaussée de Waterloo 72, 1640 Rhode-Saint-Genèse, Belgium.
E-mail address: [email protected] (C. Chahine).

https://doi.org/10.1016/j.jfluidstructs.2018.12.009
0889-9746/© 2018 Elsevier Ltd. All rights reserved.
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 111

component. The severity of a flutter incident mainly stems from the fact that it can appear suddenly and its consequences
can be disastrous. In aero-engines, flutter mostly occurs in fan blades, which are characterized by highly three-dimensional,
large slender structures, but also low pressure turbine blades can suffer from this instability.
While flutter has long been known to be of importance for fan blades, the trend of modern engine design towards
increasing fan diameters to enable so-called ultra-high bypass ratios, as well as the simultaneously growing application of
low-density composite materials leads to an increased susceptibility of the blades to flutter. But not only does its likelihood
increase; a corresponding decrease of the structure-to-air mass ratio resulting from current design trends also influences in
which form flutter may occur and the prediction methods that can be used to accurately predict its onset.
To date, turbomachinery flutter has been treated mainly as a decoupled phenomenon, based on an early proposed method
by Carta (1967), most commonly known as the energy method. In this method, aeroelastic coupling effects are assumed to
be negligible, which implies that unsteady aerodynamic forces acting on the blade surface are small enough to not alter
the vibrational behaviour of the blade. This means, in the aeroelastic system with the blade subjected to an airflow, that
the blade is assumed to vibrate purely in one of its structural modes. The implications for the prediction of fan flutter are
that the feedback from the fluid to the structure does not need to be taken into account, and the flutter behaviour can be
predicted by imposing the structural free vibration in terms of bladed disk modeshape and eigenfrequency in an unsteady
flow computation, hence the term decoupled. To be more precise, the energy method is based on two major assumptions:
First, the aeroelastic vibration mode is identical to one of the structural blade modes (i.e. modal coupling does not occur).
And second, the aeroelastic response frequency is identical to the blade’s natural frequency.
While those assumptions are valid for turbomachinery blades with high mass ratios and stiffness, their validity for
large, low density and low stiffness fan blades is less clear. Despite the existence of a large body of research work on the
development and application of various flutter prediction methods, publications targeting the validity of the energy method
for lightweight blade designs are relatively scarce and are often limited to two-dimensional cases or subsonic flows. This is
likely due to the significant computational cost associated with fully coupled fluid–structure interaction analyses for three-
dimensional transonic cases, that are needed as a reference for the comparison with the decoupled approach. This cost is
further increased by the need to compute a large number of vibration cycles for an accurate determination of the aeroelastic
response frequency.
In an early comparative study conducted by He and Zhou (1985) a coupled eigenvalue method is proposed, based on
an unsteady potential flow model and a two degree of freedom structural model in two dimensions. In a comparison with
experimental flutter stability boundaries of two transonic test rotors, it was found that mode coupling played a significant
role for the aerodynamic damping characteristics. Consequently it was concluded that the single mode energy method cannot
correctly predict the flutter onset in the cases under investigation.
Srivastava and Reddy (1999) carried out a comparative study of three computational flutter prediction methods, including
a coupled, work-per-cycle and eigenvalue analysis method for a ducted propfan in subsonic flow. The unsteady flow was
modelled by solving the three-dimensional Euler equations in the time domain while the blade structure was modelled using
a normal modes method. No detailed account of the blade material or mass ratio is given in the paper, but based on the given
reference the blade is a composite blade with a mass ratio of 33. Coupled bending-torsion unstalled flutter was consistently
predicted by the coupled and eigenvalue approach. However, the work-per-cycle method, being intrinsically a single mode
analysis method, was naturally unsuited to predict this flutter mode.
Moyroud et al. (2000) investigated the effect of material changes on the predicted flutter stability using the energy method
and an eigenvalue analysis method for a fan configuration in subsonic flow. A quasi-three dimensional linearized potential
flow solver was applied based on a finite element discretization. A finite element blade structural model constituting a full
titanium, full graphite epoxy laminated composite and a sandwich configuration with titanium skins and honeycomb core
were investigated. The results indicate significant differences between the stability predicted by the energy method and the
eigenvalue analysis method. Mode shape approximation (i.e. one mode only) and frequency approximation (vibration at the
blade’s natural frequency) were found to introduce roughly similar error magnitudes in the case considered.
In a study on the flutter behaviour of a linear cascade of biconvex uncambered airfoils operating at zero degree incidence
in subsonic flow, Sadeghi and Liu (2005a) investigated the shift of the aeroelastic response frequency from the natural
vibration frequency for varying values of mass ratio. By identification of the frequency of neutral stability, i.e. the frequency
at which the net work exchanged between fluid and blade is zero, it was found that the potential frequency shift can be
as high as 5% for a typical titanium or steel material and significantly higher for lower density materials. It is subsequently
assumed, that at low mass ratios the assumption of blades oscillating at their natural frequency as applied in the energy
method would lead to significant prediction errors of flutter stability.
The goal of this paper is to build upon those earlier works by presenting a systematic comparative study contrasting the
results of the energy method with those of a strongly coupled fluid–structure interaction approach for a three-dimensional
transonic fan blade. We parametrically vary both mass-ratio and blade stiffness to assess their respective impact on the
strength of aeroelastic coupling in terms of the aeroelastic frequency shift, i.e. the difference between the aeroelastic
response frequency and the natural frequency of the blade. This strategy was chosen over the usage of a more specific
material system, such as for instance a laminated composite material with given stacking sequence and ply properties, in
order to avoid an over specification of the case with a consequent loss of generality. This work focuses on single mode fan
flutter, hence it neglects the modal coupling effects. For reasons of computational cost, the results shown are limited to
operation at an interblade phase angle of zero.
112 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

The test case is NASA’s Rotor 67 (Strazisar et al., 1989), which was been widely used in the framework of computational
aeroelasticity (e.g. Moffatt and He (2005), Sadeghi and Liu (2005b), Srivastava et al. (2002), Marshall and Imregun (1996);
Marshall (1996), Doi and Alonso (2002), He and Denton (1993), Vahdati and Imregun (1996), and Chuang and Verdon (1999)),
despite the lack of experimental flutter data in the open literature. The rotor has 22 blades with an aspect ratio of 1.56
(average span over root axial chord). Tip solidity (chord to pitch ratio) is 1.29 and the tip relative inlet Mach number at the
aerodynamic design point is 1.38.
In the next section we present the aeroelastic system and the derivation of the structural modal model that is used for
both flutter prediction methods applied in this work. The structural finite element model, which forms the basis for the
modal model is presented in the subsequent section. In this section also the parametric material changes are explained, that
are used to alter the blade mass ratio and stiffness. A short explanation of the applied CFD solver, mesh and an introduction
to the investigated operating points follows in Section 5. The decoupled and coupled flutter prediction methods applied are
detailed in Section 6. This is followed by the results and the conclusions of the paper.

2. Nomenclature

Symbols
Φ Modeshape matrix ζA Aerodynamic damping ratio
φ Single modeshape vector ζ Damping ratio
µ Mass ratio log dec Logarithmic decrement
ν Reduced frequency Fr Force ratio
x, ẋ, ẍ Vectors of physical displacements, ωn Angular natural frequency
velocities and accelerations
M Mass matrix fn Natural frequency
C Damping matrix ωd Damped angular natural frequency due
to viscous damping
K Stiffness matrix ωdyn Angular dynamic response frequency
due to aero added mass
Fd Vector of aerodynamic damping forces ρ∞ Undisturbed air density
Ff Vector of external forcing h Blade height
q, q̇, q̈ Modal displacement, velocity and ctip Tip chord length
acceleration
q̂ Amplitude of harmonic modal mb Blade mass
displacement
m Modal mass Vtip,inl,rel Tip relative inlet velocity
c Modal damping p Unsteady surface pressure
k Modal stiffness τ Wall shear stress
fd Modal aerodynamic damping force n Surface normal vector
fˆd Amplitude of harmonic modal u Local wall displacement velocity
aerodynamic damping force
WA Aerodynamic work T Vibration period
Abbrevia-
tions
bl Baseline
lm Low mass ratio
hm High mass ratio
ls Low stiffness
hs High stiffness

3. Aeroelastic system and modal equation

A discrete approximation of the continuous aeroelastic system comprising an elastic blade and surrounding air flow can
be given by the equation of motion
Mẍ + Cẋ + Kx = F d (+F f ) (1)
     
structure fluid

where M, C and K are respectively the mass, structural damping and stiffness matrices, while x represents the vector of
blade displacements and F d the vector of blade motion dependent aerodynamic damping forces on the blade surface. For
the purposes of this work, we omit externally forced excitations originating from blade movement independent forcing,
denoted above as F f , which can be due to, for example, inlet flow distortions and constitute the problem of forced response.
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 113

Table 1
Blade material properties used for the change of mass ratio and blade stiffness. Poisson’s ratio = 0.27.
Designation Mass density [kg/m3 ] Young’s modulus [GPa] Mass ratio
Baseline (bl) 4 110 125 178
Low mass ratio (lm) 797 125 34
High mass ratio (hm) 23 040 125 1000
Low stiffness (ls) 4 110 60 178
High Stiffness (hs) 4 110 205 178

The equation of motion, Eq. (1), represents a coupled system of (in the most general form non-linear) equations of size
equal to the number of degrees of freedom chosen for the system discretization. In the current state-of-the-art of CFD and
CSM modelling this number is typically of the order of several millions. A solution of the full system of equations is therefore
often prohibitively expensive for flutter analyses in an industrial setting.
However, it is well known that the structural response of turbomachinery bladings can be sufficiently well approximated
by only a few significant low order modes (e.g. Marshall (1996)). By making use of the natural orthogonality of the free
vibration modeshapes Φ = [φ1 , φ2 , . . . , φN ] with respect to the mass and stiffness matrices, the set of N coupled equations in
Eq. (1) in physical space can be transformed to a set of N decoupled equations in modal space, where each equation describes
the system state for one blade mode in separation. Defining a modal mass m = ΦT MΦ, modal damping c = ΦT CΦ and modal
stiffness k = ΦT KΦ, and using the coordinate transformation x = Φq with q being the vector of modal displacements, the
single degree of freedom modal equation for the first blade mode is given by

mq̈ + c q̇ + kq = fd (2)

The modal aerodynamic damping force can be obtained by pre-multiplying the physical aerodynamic damping forces
with the transpose of the modeshape vector fd = φT F d . For simplicity, modeshapes are typically scaled to unity modal mass
m = 1, which results in the modal stiffness becoming the square of the natural frequency k = ωn2 . The structural damping
matrix can be similarly decoupled in case of proportional damping. However, in this work, structural damping is omitted in
all computations, as is a customary practice to yield a conservative estimate of flutter stability. In practice, fluid and structure
discretizations are normally not coincident, and hence the transfer of quantities between solid and fluid domains require an
interpolation in the non-matching mesh interface on the blade surface. A thin plate spline radial basis function is applied in
this work for that purpose. The modal equation, Eq. (2), is the basis for all flutter computations presented in this work.

4. Structural model

The free vibration modeshapes and eigenfrequencies of the blade required for the construction of the modal equation,
Eq. (2), were computed using a three-dimensional finite element code (Dhondt, 2012). The blade was discretized in space
using 17 512 quadratic tetrahedral elements. Centrifugal stiffening was taken into account by augmentation of the stiffness
matrix with the stiffness contribution from a preceding linear static analysis subject to centrifugal loading. The blade was
set to rotate at design rotational speed of 16 043 rpm. Nodes on the hub surface were constrained in all degrees of freedom
in order to model a clamped blade. The finite element mesh as well as the contour of the first bending mode, that is used for
all flutter computations in this work, is shown in Fig. 1.
The rotor 67 geometry used here corresponds to the running blade geometry reported in Strazisar et al. (1989). Hence
we did not perform a hot-to-cold transformation to recover the blade manufacturing geometry for the structural analyses.
Consequently, this geometry is subsequently assumed to represent the cold blade geometry.
Mass ratio and blade stiffness were altered by changing respectively the blade density and Young’s modulus of an
isotropic material model. A baseline set of properties was chosen to represent a typical titanium alloy. The blade density
and Young’s modulus were altered separately while keeping the other respective quantity constant. This resulted in five
blade configurations, which are listed in Table 1 along with the corresponding material properties used.
The mass ratio is a commonly used indicator for the amount of aeroelastic coupling in a system and is defined as the ratio
of blade mass mb over the mass of air of a blade enclosing cylinder
mb
µ= (3)
π (ctip /2)2 ρ∞ h
where ctip is the tip chord length, ρ∞ is the undisturbed air density and h is the blade height. High mass ratios typically
translate to low levels of aeroelastic coupling, hence representing a situation in which the aeroelastic response is mainly
influenced by the structural dynamics while the impact of the unsteady aerodynamic forces is small. The mass ratio range
shown in Table 1 was chosen to include a low mass ratio configuration to approximate values of modern lightweight aero-
engine fan blades (e.g. a composite sandwich construction), as well as one unrealistically high mass ratio to investigate the
agreement of both flutter prediction methods towards a high mass ratio limit.
114 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

Fig. 1. Solid mesh and mode shape contour of the first bending mode at design speed, calculated with the baseline titanium alloy.

Fig. 2. Overview of CFD mesh. Every second mesh point is shown.

5. CFD solver, mesh and operating points

For small perturbations, the unsteady flow is often assumed to behave linearly with respect to the vibration amplitude,
which has motivated the usage of time linearized methods for its solution. The time linearized flow equations can be solved
in the frequency domain, which is computationally much more efficient than a corresponding solution of the unsteady
flow equations in a time marching fashion (time domain). However, in transonic flows, even small harmonic oscillations
can lead to non-linear flow effects due to the presence of strong shocks and/or boundary layer separation (e.g. Huff et al.
(1991), Carstens and Belz (2001)). Non-linearities manifest themselves in the form of higher flow harmonics which alter
the time averaged unsteady flow from its steady counterpart. The non-linear harmonic method (He and Ning, 1998; Ning
and He, 1998), has been developed as an extension to linear frequency domain methods, allowing for the interaction of
higher harmonics to be taken into account. However, in order to avoid that any differences in flow modelling might affect
the comparison of decoupled and coupled flutter results, full non-linear unsteady time domain computations are used for
all flutter computations in this work. The solver employed is a commercial CFD code, which is a three-dimensional, density-
based, structured, multi-block Reynolds-Averaged Navier–Stokes (RANS) solver using a finite volume method. The unsteady
RANS equations are solved in time using a dual-time stepping approach. Turbulence closure is obtained using the one-
equation turbulence model of Spalart and Allmaras (1994).
The fluid domain is discretized with a multi-block structured mesh consisting of 0.3 million grid points with an average
non-dimensional height of the wall-adjacent cell of y+ = 3.6. The tip gap is discretized with five grid points in the radial
direction. Fig. 2 gives an overview of the mesh with details of the leading edge and blade tip discretizations. The computations
were performed for two operating points on the design speedline — one at peak efficiency and one at a higher loading,
off-design condition at 96% of the design point mass flow. The experimental measurements of the steady flow performance
reported in Strazisar et al. (1989) also include a near stall operating point at 93% of the design point mass flow. However,
this operating point was not considered in this work due to convergence reasons.
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 115

Table 2
Reduced frequencies for all blade configurations at peak efficiency and higher load condition.
Designation Peak efficiency Higher load condition
Baseline 0.708 0.711
Low mass ratio 1.262 1.268
High mass ratio 0.503 0.505
Low stiffness 0.595 0.597
High stiffness 0.821 0.825

Fig. 3. Comparison of numerical and experimental unsteady pressure amplitudes and phase angles for the three-dimensional turbine flutter test case
of Huang et al. (2009). cp1 is the pressure coefficient determined with the amplitude of the first harmonic pressure.

Table 2 summarizes the values of reduced frequencies for all blade configurations at both operating points. The reduced
frequency is given as
ωn ctip
ν= (4)
Vtip,inl,rel
where ctip is the tip chord length and Vtip,inl,rel denotes the relative tip inlet velocity. Note here the change of natural frequency
due to mass ratio changes, which is a result of the constant elasticity modulus. The impact of the operating point change on
the reduced frequencies can be seen to be negligible.

6. Description of flutter prediction methods

6.1. Decoupled method

The decoupled energy method introduced by Carta (1967), sometimes also referred to as work-per-cycle method, is based
on the assumption that the unsteady fluid forces are small compared to the blade’s inertia and stiffness forces and therefore
do not alter the vibrational mode in terms of modeshape and frequency. Under this assumption, the unsteady flow field
and in turn the flutter stability of a blade assembly can be computed by imposing a forced low-amplitude vibrational blade
116 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

motion in the unsteady fluid computation, which corresponds to one of the vibrational modeshapes and eigenfrequencies
of the structure either in vacuum, or at times with the steady state gas loads imposed. Flutter stability is inferred from the
work that the fluid performs on the blade over one vibration period. A positive net work input to the structure indicates
flutter. Structural damping is omitted, leading to a conservative estimate of the flutter stability.
Given a forced harmonic blade motion with period T and modal amplitude q̂, the aerodynamic work WA is computed as
the time integral of the total aerodynamic power on the blade suction side and pressure side surfaces over one vibration
cycle as
∫ ∫
WA = (pn + τ n) · u dAdt (5)
T Surface

where p is the unsteady pressure, τ is the unsteady wall shear stress, n is the unit surface normal and u is the local wall
vibratory velocity. While the viscous force contribution due to shear stress is typically small in comparison to the pressure
forces (and is therefore often neglected), its contribution is required for an accurate prediction of stall flutter (Vahdati et al.,
2001) and is therefore taken into consideration.
Based on the assumption of linear flow behaviour, aerodynamic damping can be cast in the form of an aerodynamic
damping ratio as
−WA
ζA = (6)
2π ωn2 q̂2
Flutter is consequently indicated by negative values of the damping ratio. The corresponding logarithmic decrement (log
dec) is given as
log dec = 2π ζA . (7)
While the energy method neglects any frequency shift resulting from aeroelastic coupling effects, it can be used to predict
the frequency shift. How this is achieved is shown later in the paper.
Validations of the CFD solver under forced oscillations were performed for various test cases, including the 11th
International Standard Configuration proposed by Fransson et al. (1999), as described in Debrabandere et al. (2013). In
addition, the authors of present paper performed in-house validations of the code with a three-dimensional turbine flutter
test case described in Huang et al. (2009). The computations were found to be in good agreement with the experimental
data, as shown for the comparisons of the unsteady pressure amplitudes and phase depicted in Fig. 3.

6.2. Coupled method

In contrast to the decoupled method introduced above, the coupled method enables the consideration of the two-
directional coupling between fluid and structure by allowing for the transfer of blade displacements as well as fluid forces
between the domains. Therefore, the blade response frequency is no longer predetermined, but results from the solution
of the aeroelastic system. The coupled fluid–structure interaction method applied in this work uses a strongly coupled
staggered partitioned approach (e.g. Farhat and Lesoinne (2000)). Strong coupling enables to achieve energy conservation
between fluid and structure domains. In this work, this is achieved via full convergence sub-iterations at every physical
timestep . A complementary function and particular integral method is used for the solution of the modal equation, which
was found to be more accurate than the commonly used Newmark algorithm (Debrabandere et al., 2012). Accurate energy
conservation of the scheme was verified as shown in Fig. 4 by comparison of total structural energy Etot and aerodynamic
work for rotor 67 based on the baseline full titanium structural model. A constant initial energy E0 in time indicates that the
energy transfer across the fluid–structure interface is accurately modelled in the computation.
Each vibration period is resolved with 50 physical timesteps. It was found that a higher number of physical timesteps
per period had no significant impact on the aeroelastic response in terms of stability and aeroelastic response frequency.
Aerodynamic damping is quantified in terms of the logarithmic decrement which is directly obtained from the resulting time
evolution of the generalized displacements. The logarithmic decrement can be computed from the amplitudes of successive
vibration cycles as
1 q̂(t0 )
log dec = log (8)
n q̂(tn )
where n denotes the number of cycles considered (here typically the last six).
The aeroelastic response frequency is computed by means of a Fast Fourier Transform of the overall generalized
displacement signal. Zero padding was applied in order to obtain a frequency resolution of approximately 0.025 Hz. The
aeroelastic frequency shift, i.e. the percentage difference of the aeroelastic response frequency f from the natural vibration
frequency fn then follows from
f − fn
freq. shift [%] = · 100 (9)
fn
The coupled computations at both operating points were initialized from the running rotor 67 geometry as “cold” blade
geometry, as noted in Section 4. To trigger the vibratory blade motion at the peak efficiency condition, the inlet total pressure
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 117

Fig. 4. Energy conservation test of coupled computation for baseline blade at peak efficiency.

has been initially lowered for the first 20 physical timesteps of the computations. A similar measure was not necessary to
initiate the blade vibration at the higher load condition.
Similarly to the decoupled computations, structural damping was neglected in all coupled computations.

6.3. Static aeroelastic blade untwist

Blade stagger changes due to static aeroelastic blade untwist were shown to significantly affect the aerodynamic
damping characteristics of transonic fans (e.g. Srivastava et al. (2002)). While the blade automatically assumes its aeroelastic
equilibrium position (if existing) in a coupled computation, it has to be imposed in the decoupled computations in order
to avoid that differences in blade untwist affect the comparison of the aerodynamic damping results. For a single mode
modal model this translates to the determination of the first mode’s generalized displacement at which static aeroelastic
equilibrium occurs. To correctly account for this effect, the average generalized displacement to be imposed in the decoupled
computations was predetermined for each blade configuration and operating point by a coupled untwist computation.

7. Comparative results and discussion

7.1. Global aero-damping comparisons and frequency shift results

7.1.1. Peak efficiency

The steady flow relative Mach number contours at 90% blade span for the peak efficiency operating point are shown in
Fig. 5, along with the corresponding experimental results from Strazisar et al. (1989). The CFD results are shown for the
running design point geometry of rotor 67 as obtained from the publication, hence no additional blade untwist was applied
here for the comparison to the experimental results. The flow field served as the initial solution for the coupled untwist
computations which were in turn used as starting solutions for the flutter computations.
It can be observed that a reasonably good agreement is obtained between CFD and experiment. The flow field is
characterized by a weak oblique shock wave attached at the leading edge of the blade which extends slightly downstream
into the covered blade passage where the shock tilts to become normal on the suction side surface.
In Fig. 6 the tip leading edge displacements (max blade displacements) resulting from the coupled computations for
all blade configurations are shown in terms of percentage tip chord. In total 30 vibration cycles were computed for each
configuration, with the exception of the low mass ratio case for which the computations were continued for a total of 52
vibration cycles to obtain an accurate frequency response. It can be seen that all blade configurations except for the high
mass ratio configuration are stable, indicated by a decaying vibration amplitude with time. Neutral stability is observed for
the high mass ratio case, indicated by a constant vibration amplitude.
The global comparison of the coupled and decoupled aerodynamic damping results for mass ratio and stiffness variations,
along with the corresponding coupled frequency shifts is shown in Fig. 7. The plots reveal straight away that coupled and
decoupled log dec predictions show an excellent agreement for all mass ratios and stiffness values under consideration.
In Fig. 7(b) a steep increase of the aeroelastic frequency shift with a reduction of mass ratio can be observed. As would be
expected, the frequency shift is zero at the highest mass ratio. However, even at a mass ratio of 34 the frequency shift remains
118 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

Fig. 5. Relative Mach number contours at 90 percent blade span for peak efficiency operation. Left: CFD, right: Experiments (Strazisar et al., 1989).

Fig. 6. Coupled time evolution of the tip leading edge displacements at peak efficiency. Teigen is the vibration period of the free vibration response
(=1/Eigenfrequency).

relatively small with a value of about −1.25%. Interestingly, aeroelastic coupling effects thus do not appear to be significant
at this condition.
Under stiffness variations, Fig. 7(c), the blade can be seen to be overall stable, with a reduction of stability towards
increasing blade stiffness. This result might appear to be somewhat counterintuitive, as stiffness and associated reduced
frequency increases are usually understood to increase blade stability (with the corresponding reduced frequency being a
frequently used design parameter). However, for a stable blade, frequency increases can lead to decreasing blade stability,
as was similarly observed in Vahdati et al. (2011). In terms of aeroelastic coupling, the effects of stiffness variations appear
to be negligible for the stiffness range considered, leading to a flat curve for the frequency shift as shown in Fig. 7(d).

7.1.2. Higher load condition

The steady relative Mach number contours for operation at the higher load condition are shown in Fig. 8. Experimental
measurements at this operating condition are not available. A comparison to the steady flow field at peak efficiency, Fig. 5,
shows that increased back pressure has now caused the passage shock to move upstream towards the leading edge resulting
in a strong and nearly normal shock.
The corresponding coupled time evolutions of the tip leading edge displacements for all blade configurations are depicted
in Fig. 9. A comparison of the plots with the peak efficiency results, Fig. 6, shows that the operating point change had an overall
destabilizing effect. Increasing vibration amplitude with time shows that the high mass ratio and low stiffness configurations
are now unstable, while the baseline configuration has become neutrally stable.
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 119

Fig. 7. Comparison of decoupled and coupled predictions of aeroelastic stability (left) and frequency shift (right) over mass ratio and stiffness changes at
peak efficiency.

Fig. 8. Relative Mach number contours at 90 percent blade span for operation at the higher load condition.
120 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

Fig. 9. Coupled time evolution of the tip leading edge displacements at the higher load condition. The black lines show the period-averaged displacement.
Teigen is the vibration period of the free vibration response (=1/Eigenfrequency).

The global comparison of decoupled and coupled values of aerodynamic damping and the associated frequency shifts for
mass ratio and blade stiffness variations are shown in Fig. 10. The agreement between both flutter prediction methods can be
seen to be of a similarly good level as previously observed for the peak efficiency operation. The overall system destabilization
with respect to the peak efficiency results is more clearly visible in these plots. However, unstable and stable behaviour, as
well as the neutrally stable baseline behaviour are accurately predicted by the energy method. The flutter mode observable
in Fig. 10(c) for the low stiffness design can be characterized as part speed stall flutter, which typically occurs for fan blades
in the first bending mode at a raised working line (increased loading) and part speed operation (i.e. reduced stiffness).
A comparison of Figs. 10(b) and 7(b) shows that the overall frequency shift curve has not changed drastically due to the
change of the operating point. The biggest effect can be seen to occur for the baseline configuration, where the frequency shift
roughly doubled. However, the impact on the low mass ratio case is minor, with an actual small frequency shift reduction
with respect to the corresponding peak efficiency result. A small frequency shift can now also be seen to occur for the high
mass ratio configuration. The sensitivity of the frequency shift towards stiffness variations (Fig. 10d) can be observed to have
slightly increased in comparison to the peak efficiency trend. However, it still remains small on a global level.

7.2. Discussion of frequency shift

It was seen in the results presented above that despite small values of mass ratio, the aeroelastic coupling strength in
the form of a frequency shift did not appear to be significant. As a result, a very good agreement was obtained between
flutter stability predictions by decoupled and coupled methods. In view of the frequency shift results reported elsewhere in
the literature for comparably low mass ratios (e.g. Sadeghi and Liu (2005a), Moyroud et al. (2000)) this result was slightly
surprising and motivated a more detailed investigation into the mechanisms governing the frequency shift, which will be
discussed next.
Frequency variations in aeroelastic systems can generally be attributed to two effects, from which the first is viscous
damping (or an equivalent damping force component), and the second is an alteration of the dynamic mass of the system
resulting from the presence of the fluid. The theoretical basis of the former effect is well known and is given by the relation
√
ωd = ωn 1 − ζ 2 (10)

representing the natural frequency of an underdamped system in dependency of the natural frequency and the damping
ratio. In the context of turbomachinery flutter, the aerodynamic damping ratio is typically small, and as a result the frequency
shift due to damping is normally negligible.
To explain the second contribution, consider the modal equation given in Eq. (2). If we assume a harmonic blade motion
i(ωt +φ )
q = q̂eiωt , and a harmonic modal aerodynamic damping force at the same frequency fd = fˆd , the force amplitudes
in the modal equation can be represented in the Argand diagram, as shown in Fig. 11. Note, that the blade displacement is
here considered to be purely real pointing in the positive direction. The force components are the modal force amplitudes
of inertia force, mω2 q̂, stiffness force, kq̂, (structural) damping force, c ωq̂, and aerodynamic damping force, fˆd . For the sake
of simplicity and clarity, we will drop the denotation ‘modal amplitude’ for the forces in the following discussion, and we
further neglect structural damping.
It is well known from vibration theory, that only a force component that is out-of-phase with the blade displacement,
is able to do work to the system, and therefore can contribute to the aerodynamic damping, be it positive or negative.
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 121

Fig. 10. Comparison of decoupled and coupled predictions of aeroelastic stability (left) and coupled frequency shift (right) over mass ratio and stiffness
changes at the higher load condition.

Fig. 11. Argand diagram of the forces in the aeroelastic system.

In case of a purely real blade displacement as shown here, this force component is the imaginary component of the
aerodynamic damping force Im(fˆd ), c.f. Fig. 11, which contributes to the alteration of the vibration frequency via the same
mechanism as for viscous damping, as shown above in Eq. (10). On the other hand, the aerodynamic damping force
component that is in-phase with the blade displacement, Re(fˆd ), can be seen to be similarly in-phase with the inertia and
stiffness forces. It therefore effectively resembles the force contribution of an added mass in the aeroelastic system, which in
turn affects the system’s response frequency.
122 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

Table 3
Comparison of coupled and decoupled frequency shift values at the higher load condition. Decoupled values
are shown for both aero added mass (AAM) and viscous damping (VD).
bl hm lm hs ls
Coupled freq. shift [%] −0.426 −0.026 −1.217 −0.448 −0.375
Decoupled freq. shift (AAM) [%] −0.49 −0.022 −1.36 −0.46 −0.39
Decoupled freq. shift (VD) [%] −1.1e−7 −4.6e−5 −4.4e−3 −4.9e−5 −1.5e−4

Real and imaginary components of the aerodynamic damping force can be easily determined from its phase shift and
amplitude obtained either from the combined time evolution of the aerodynamic damping force and the generalized
displacement or from a frequency domain solution, from which the real and imaginary components of pressure and viscous
forces are directly available.
For a further explanation of this matter, consider the force equilibrium of the real components of the forces in Fig. 11

mω2 q̂ − kq̂ + Re(fˆd ) = 0. (11)

As can be seen here, the only mechanism in place to obtain a force equilibrium in the case of a non-zero real component of fˆd ,
is the adaptation of the inertia force (mω2 q̂) via a change of the response frequency ω, because the modal mass m and modal
stiffness k are constant and the generalized displacement amplitude q̂ linearly scales both the stiffness and inertia forces.
Such a change of frequency may alter the unsteady flow field and in turn the aerodynamic damping force fˆd . An iteration
in the form of a coupled computation would be required to find a converged solution of the frequency shift. However, in
a first approximation, it can be assumed that the aerodynamic damping force is invariant with ω, which enables to give a
prediction of the dynamic response frequency using a decoupled approach.
It follows directly from Eq. (11)
√
k Re(fˆd )
ωdyn = − (12)
m mq̂

where Re(fˆd ) and q̂ can be substituted for their corresponding pendants from a decoupled computation to yield a first order
approximation of the dynamic response frequency ωdyn . Note here again, that k = ωn2 , m = 1, so that the common relation
k/m = ωn2 holds. Very similar routes to the ones outlined here have been taken by Moffatt and He (2005) for the investigation
of aeroelastic coupling effects in the framework of turbomachinery forced response computations.
It was mentioned above that the apparent increase or decrease of the dynamic mass of the system due to the in-phase
component of the aerodynamic damping force can be suitably described as an added mass effect. However, it is important
to note that the effect discussed here should be clearly discerned from the more commonly debated added mass effect
that results from the acceleration of an attached portion of fluid in the framework of mainly incompressible hydroelastic
applications (e.g. De La Torre et al. (2013), Brennen (1982)). In contrast, the presently discussed apparent change of the
dynamic mass of the system is due to the phase shift between the unsteady aerodynamic forces and blade displacements
and results from the complex aeroelastic relations in a compressible flow. To more clearly differentiate the two physical
mechanisms that are both found in the scope of fluid–structure interaction, the authors of present paper believe that the
term aerodynamics added mass or aero added mass, in short, seems to be better suited to describe the phasing induced change
of the dynamic mass of the system that is the subject of the discussions presented here.
Before we use the force equilibrium in Eq. (11) to further investigate the factors influencing the magnitude of the
frequency shift, it is sensible to validate the assumptions made above about the physical effects which are governing it.
To this end, Eq. (12) can be used in combination with the decoupled computational results presented earlier, to yield a
prediction of the frequency shift. Those can be compared to the actual values of frequency shift as obtained from the coupled
computations. Such a comparison is shown in Table 3 for operation at the higher load condition. In addition to the decoupled
frequency shift originating from the aero added mass, the frequency shift due to the viscous (out-of-phase) force component
of the aerodynamic damping force is shown, resulting from Eq. (10). It can be seen that the decoupled frequency shift due to
the aero added mass closely matches the values of the coupled computations, therefore reassuring that the frequency shift
model, as well as the assumption of a frequency invariant aerodynamic damping force, are sensible for the cases considered.
In addition, it can be seen that the frequency shift due to the viscous aero damping force component is several orders of
magnitude smaller than the frequency shift due to the aero added mass effect, hence confirming that its contribution to the
overall frequency shift can be considered as negligible.
We can now investigate the force equilibrium in further detail to understand how the force interactions are governing
the actual frequency shift magnitude originating from the aero added mass effect. It is apparent at this point, that the
magnitude of the real component of the aerodynamic damping force plays a key role for the frequency shift, which depends
on the phase angle and the amplitude of the aerodynamic damping force. The larger the magnitude of the real component of
the aerodynamic damping force, the larger will be the induced frequency shift. However, the frequency shift magnitude is
established by the combined effect of all in-phase forces, and as we will show next, it can be attributed to the ratio between
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 123

Fig. 12. (a) Decoupled frequency shift prediction due to aero added mass as function of the force ratio. (b) the same plot with coupled results for the higher
load condition included.

the real component of the aerodynamic damping force and the vibrational inertia force of the system, which we express as
the force ratio
Re(fˆd )
Fr = (13)
mωn2 q̂
Note that the inertia force in the denominator equals the inertia force at the blade’s natural frequency. Inserting Eq. (13)
into Eq. (11) and solving for the frequency yields again an expression for the dynamic response frequency, but this time as
a function of Fr
√
k √
ωdyn = − Fr · ωn2 = ωn (1 − Fr) (14)
m
Eq. (14) can now be expressed in terms of a frequency shift in percent as
ωdyn √
∆ ω [ %] = ( − 1) · 100 = ( (1 − Fr) − 1) · 100 (15)
ωn
The variation of frequency shift with the force ratio is plotted in Fig. 12(a). In this figure, we can see that, in the theoretical
case of the force ratio increase towards a value of one, i.e. the real component of the aerodynamic damping force would reach
the same magnitude as the inertia force, the frequency shift reaches -100%. However, due to the quadratic dependency of
the inertia force from the vibration frequency, only an approximately 41% increase of the frequency would be needed to
compensate for the case of Fr = −1. Again, those are theoretical limits.
Values of force ratio and frequency shift obtained from the coupled rotor 67 computations can now be included in the
plot as shown in Fig. 12(b) in which the higher load condition results are depicted. As can be seen, the predicted dependency
of the frequency shift on the force ratio is indeed very well matched by the coupled results. Results for the low stiffness and
high stiffness configurations are clustered closely around the baseline design, while the mass ratio changes have a higher
impact on the force ratio and consequently the frequency shift. However, the overall values of the force ratio remain rather
small for all blade configurations considered thus far in this paper, where either mass ratio or blade stiffness were altered in
isolation. This strategy was chosen in order to allow an investigation of the respective impact of each individual parameter.
However, in reality, blade mass and stiffness are typically not decoupled from one another. The notable exception might be a
composite material in which the directional stiffness of the material can be tailored without necessarily having a significant
effect on the blade mass. Nevertheless, a low mass ratio blade, e.g. a blade made from carbon fibre reinforced plastic, would
normally have a lower stiffness than a corresponding integral titanium blade, which should be taken into consideration in
the assessment of potential aeroelastic coupling effects. Based on the derivations presented above, c.f. Eq. (13) to Eq. (15), it
can be expected that a corresponding low mass ratio and low stiffness blade would potentially show an increased frequency
shift magnitude. An additional case was therefore added to the investigations, as will be shown in the next subsection.

7.3. Low mass ratio, low stiffness case

An additional blade configuration with the same mass ratio as the previously discussed low mass ratio configuration, but
with a decreased stiffness to match the eigenfrequency of the baseline design is presented in this subsection. The material
properties for this low mass ratio, low stiffness configuration are listed in Table 4.
124 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125

Table 4
Material properties used for the low mass ratio, low stiffness blade configuration.
Mass density [kg/m3 ] Young’s modulus [GPa] Poisson’s ratio Mass ratio
797.15 24.2 0.27 34

Table 5
Comparison of decoupled and coupled logarithmic decrement predictions and frequency shift for the low
mass ratio, low stiffness configuration as well as the low mass ratio configuration at the higher load condition.
The decoupled frequency shift shown is due to the aero added mass effect.
log dec freq. shift [%]
Coupled Decoupled Difference [%] Coupled Decoupled
Low mass ratio, low stiffness 0.022 0.028 27.3 −2.078 −2.1
Low mass ratio 0.059 0.058 −1.7 −1.217 −1.36

The coupled results obtained for this blade were added to the plot shown in Fig. 12(b). Also this new case matches the
predicted curve well, however, it can be seen that the force ratio and the corresponding magnitude of the frequency shift
have increased considerably in comparison to the case where the mass ratio was altered in isolation. The frequency shift
magnitude originating from the coupled computations has increased to about 2% for this new case. It can also be seen that
the point is deviating slightly from the decoupled prediction, which might be an indicator that the frequency shift begins to
affect the aerodynamic damping force.
The corresponding logarithmic decrement values resulting from the decoupled and coupled computations for the low
mass ratio, low stiffness configuration are shown in Table 5 and to allow for a straightforward comparison, the values for
the low mass ratio case (mass ratio change in isolation) were added to the table. It can be seen that the reduction of the
blade stiffness leads to a destabilization with respect to the isolated mass ratio change. In addition, the stability prediction
mismatch between the coupled and decoupled methods can now be seen to be of significance for the low mass ratio
and low stiffness blade. The decoupled method is seen to overpredict the stability by 27.3%. A comparison of the coupled
and decoupled frequency shift values included in Table 5 for the two cases shows here more clearly the capability of the
decoupled method to predict the frequency shift with good accuracy, therefore providing a useful means to estimate the
applicability of the decoupled approach for stability predictions.
This result illustrates that, while the mass ratio can be an indicator for the potential strength of aeroelastic coupling
effects, the manifestation of aeroelastic coupling in the form of a frequency shift is mainly resulting from the ratio of
the in-phase component of the aerodynamic damping force and the vibrational inertia force of the blade. While the
former is dependent on the operating conditions and the vibration mode (blade and disk modeshape and frequency), the
vibrational inertia force depends solely on the blade mass and the blade stiffness, which are purely structural parameters.
The consequence is that the frequency shift experienced by low mass ratio blades with high stiffness might be negligible,
while blades with comparable mass ratios and low stiffness (large, lightweight fan blades) can be expected to show a higher
susceptibility to aeroelastic coupling effects. As a result, stability prediction differences between decoupled and coupled
methods may become of further significance as well, which should be of consideration in the analysis and design process.

8. Conclusions

In this paper comparisons of decoupled and coupled flutter analyses of NASA’s transonic fan rotor 67 subject to parametric
variations of mass ratio and blade stiffness are presented. The aeroelastic coupling strength is evaluated in terms of the shift
of the response frequency from the natural frequency of the blade and its effect on the agreement between the stability
predictions of the two methods is shown.
From the results obtained, the following main conclusions can be drawn:

1. Overall, decoupled and coupled stability predictions show a very good agreement for all cases in which isolated
variations of mass ratio and stiffness from the baseline, full titanium blade are performed. The corresponding
frequency shift values reach maximum values of about 1.2 percent for the lowest mass ratio considered (µ = 34)
without negatively affecting the prediction quality of the energy method. For a combined low mass ratio and low
stiffness blade configuration, which more closely resembles a realistic lightweight fan blade design, the stability
prediction accuracy of the energy method is seen to decrease with a corresponding frequency shift increase to about
2 percent. For this case the energy method overpredicts the stability by about 27 percent.
2. It was shown that the aeroelastic frequency shift mainly originates from an aerodynamics added mass effect
which results from the component of the aerodynamic damping force that is in-phase with the vibrational blade
displacement. Consequently, the magnitude of the frequency shift is shown to depend on the ratio between this
in-phase damping force component and the blade’s vibrational inertia force, which is affected by blade mass and
blade stiffness. The mass ratio in isolation can therefore only be an indicative measure for the expected aeroelastic
coupling strength in transonic fans and other turbomachinery blades operating in high-speed compressible flows. Fan
 C. Chahine, T. Verstraete and L. He / Journal of Fluids and Structures 85 (2019) 110–125 125

designers should evaluate carefully whether good experience with the energy method with previous blade designs
can be transferred to new designs with comparable mass ratios but modified stiffness. This aspect might become of
increasing importance if advanced composite blade design and manufacturing techniques are applied, which enable
the targeted tailoring of blade stiffness, without having a major impact on the blade mass.
3. The decoupled method can be used to predict the aeroelastic frequency shift rapidly and with acceptable accuracy,
using the technique shown in this paper. While the prediction accuracy can be expected to decline for larger values
(e.g. in excess of 2%), it provides a good indication of whether the aeroelastic coupling strength is negligible, and
therefore presents a useful means to check the potential applicability of a decoupled (energy method) approach
without resorting to a costly coupled method.

Acknowledgements

The research leading to these results has received funding from the Engineering and Physical Sciences Research Council
of the UK and the European Union Seventh Framework Programme under grant agreement no 316394. The financial support
is greatly acknowledged.

References

Brennen, C.E., 1982. A Review of Added Mass and Fluid Inertial Forces. Tech. rep.. Naval Civill Engineering Laboratory, Port Hueneme, CA, USA.
Carstens, V., Belz, J., 2001. Numerical investigation of nonlinear fluid-structure interaction in vibrating compressor blades. J Turbomach. Vol. 123 (April
2001), 402–408.
Carta, F.O., 1967. Coupled blade-disk-shroud flutter instabilities in turbojet engine rotors. J. Eng. Power 89 (3), 419–426.
Chuang, H.A., Verdon, J.M., 1999. A nonlinear numerical simulator for three-dimensional flows through vibrating blade rows. J. Turbomach. 121 (2), 348–357.
De La Torre, O., Escaler, X., Egusquiza, E., Farhat, M., 2013. Experimental investigation of added mass effects on a hydrofoil under cavitation conditions. J.
Fluids Struct. 39, 173–187.
Debrabandere, F., Tartinville, B., Hirsch, C., Coussement, G., 2012. Fluid-structure interaction using a modal approach. J. Turbomach. 134 (5).
Debrabandere, F., Tartinville, B., Hirsch, C., Coussement, G., 2013. Extension of the non-linear harmonic method to flow computations around moving and
deforming structures. In: International Forum on Aeroelasticity & Structural Dynamics 2013. Bristol, UK.
Dhondt, G., 2012. CalculiX CrunchiX USER’ S MANUAL version 2.5.
Doi, H., Alonso, J.J., 2002. Fluid/structure coupled aeroelastic computations for transonic flows in turbomachinery. In: ASME Turbo Expo 2002: Power for
Land, Sea and Air.
Farhat, C., Lesoinne, M., 2000. Two efficient staggered algorithms for the serial and parallel solution of three-dimensional nonlinear transient aeroelastic
problems. Comput. Methods Appl. Mech. Engrg. 182 (3–4), 499–515.
Fransson, T.H., Jocker, M., Bolcs, A., Ott, P., 1999. Viscous and inviscid linear/nonlinear calculations versus quasi-three-dimensional experimental cascade
data for a new aeroelastic turbine standard configuration. J. Turbomach. 121 (4).
He, L., Denton, J.D., 1993. Three dimensional time-marching inviscid and viscous solutions for unsteady flows around vibrating blades. J. Turbomachinery
116 (3), 469–476.
He, L., Ning, W., 1998. Efficient approach for analysis of unsteady viscous flows in turbomachines. AIAA J. 36 (11), 2005–2012.
He, L., Zhou, S., 1985. A check on the energy method of predicting blade transonic stall flutter. In: Beijing International Gas Turbine Symposium and
Exposition. The American Society of Mechanical Engineers, Beijing, China.
Huang, X.Q., He, L., Bell, D.L., 2009. Experimental and computational study of oscillating turbine cascade and influence of part-span shrouds. J. Fluids Eng.
131 (5).
Huff, D.L., Swafford, T.W., Reddy, T.S.R., 1991. Euler flow predictions for an oscillating cascade using a high resolution wave-split scheme. In: ASME Turbo
Expo, paper 91-GT-198.
Marshall, J., 1996. Prediction of Turbomachinery Aeroelasticity Effects using a 3D Non-Linear Integrated Method (Doctoral dissertation), Imperial College
of Science, Technology and Medicine London.
Marshall, J.G., Imregun, M., 1996. An analysis of the aeroelastic behaviour of a typical fan-blade with emphasis on the flutter mechanism. In: Proceedings
of the International Gas Turbine and Aeroengine Congress and Exhibition, ASME paper 96-GT-78. ASME, Birmingham, United Kingdom.
Moffatt, S., He, L., 2005. On decoupled and fully-coupled methods for blade forced response prediction. J. Fluids Struct. 20 (2), 217–234.
Moyroud, F., Jacquet-Richardet, G., Fransson, T.H., 2000. Aeroelasticity in turbomachines: Some aspects of the effect of coupling modeling and blade material
changes. Int. J. Rotating Mach. 6 (4), 265–273.
Ning, W., He, L., 1998. Computation of unsteady flows around oscillating blades using linear and nonlinear harmonic Euler methods. ASME J. Turbomach.
120 (3), 508–514.
Sadeghi, M., Liu, F., 2005a. Computation of cascade flutter by uncoupled and coupled methods. Int. J. Comput. Fluid Dyn. 19 (8), 559–569.
Sadeghi, M., Liu, F., 2005b. Coupled fluid-structure simulation for turbomachinery blade rows. In: 43rd AIAA Aerospace Sciences Meeting and Exhibit.
American Institute of Aeronautics and Astronautics, Reston, Virigina, pp. 10–13.
Spalart, P.R., Allmaras, S.R., 1994. A one-equation turbulence model for aerodynamic flows. Rech. Aerosp. (1), 5–21.
Srivastava, R., Bakhle, M.A., Keith Jr, T.G., Stefko, G.L., 2002. Flutter Analysis of a Transonic Fan, Nasa Technical Memorandum, (211818).
Srivastava, R., Reddy, T., 1999. Comparative study of coupled-mode flutter-analysis methods for fan configurations. J. Propul. Power 15 (3), 447–453.
Strazisar, A.J., Wood, R., Hathaway, M.D., Suder, K.L., 1989. NASA Technical Paper 2879.
Vahdati, M., Imregun, M., 1996. A non-linear aeroelasticity analysis of a fan blade using unstructured dynamic meshes. Proc. Inst. Mech. Eng. C 210 (6),
549–564.
Vahdati, M., Sayma, A.I., Marshall, J.G., Imregun, M., 2001. Mechanisms and prediction methods for fan blade stall flutter. J. Propul. Power 17 (5), 1100–1108.
Vahdati, M., Simpson, G., Imregun, M., 2011. Mechanisms for wide-chord fan blade flutter. J. Turbomach. 133 (4).