5

Unsteady Flow and Aeroelasticity

L He
School of Engineering, University of Durham, Durham, England

INTRODUCTION
An unsteady ow may vary in time either randomly (turbulence) or
periodically. It is the latter type that we are dealing with here. Regarding
turbomachinery performances, there are two main aspects associated with
unsteady ow effects: (1) aerothermal performance due to blade-row
interaction and ow instability (stall/surge); (2) blade mechanical integrity
due to ow-induced vibrations (forced response and utter).
For conventional blading at design conditions, blade-row interaction
effects (and unsteady losses) are normally small, and blades do not normally
experience serious vibration problems. This should be self-evident considering current blading designs with high aerothermal efciency have been
achieved so far largely by using steady-ow methodologies. Consequently,
the main design issues have been to nd sufcient aerodynamic and
aeroelastic stability margins. However, the drive in the future developments
toward higher loads and more compact congurations will result in
intensied unsteadiness even at design conditions, and its effects on timeaveraged performances need to be assessed during a design process from a
purely aerothermal point of view. In addition, the blade mechanical

Copyright 2003 Marcel Dekker, Inc.

integrity (stress levels) will have to be more adequately examined if aero

loading is to be increased, which challenges widely used empirical design
rules on blade utter and forced responses. All these requirements would
point to closer coupling and more simultaneous (concurrent) interaction
between aerodynamic design and structural mechanical design, since a
lengthy aeromechanical iterative path can be very costly.
The competitive situation demanding better understanding and
prediction of unsteady effects is further compounded by nonstandard
problems arising from increasingly important environmental concerns and
legislation. For example, a situation may arise where a required low NOx
combustion has to generate a large-scale circumferential temperature
distortion which can pass through a 5-stage low-pressure turbine with little
decay (Manwaring et al. 1997) [1]. The analysis of the blade aeromechanical
responses in this kind of nonstandard environments is becoming a necessary
part of standard procedures. An aero design that has a high efciency in a
normal undistorted ow condition but produces an excessive stress level
under the distortion condition is obviously not to be accepted. A similar
example is the increasing priority to reduce aerodynamics-related noise
generation and propagation in future aero-engine developments due to
increasingly important environmental concerns and legislation. Again, an
aerodynamically efcient blading design will not be acceptable if its
associated noise level (especially that of fans or LP turbines at off-design
conditions) is high.
Today computation of unsteady turbomachinery ows has become a
very active research and development area. Carefully executed and analyzed
unsteady CFD computations have provided useful information and insights
to help our understanding of complex unsteady ow problems. However, it
needs to be recognized that with computer resources currently available,
full-scale time-dependent CFD calculations for unsteady turbomachinery
ows are still very expensive and are not suitable for daily design purposes.
In order to make effective use of CFD methods, we need to have a clear
appreciation of relevant ow physics and length scales and be aware of the
capabilities as well as the limitations of different ow models and numerical
methods.
In this chapter, various basic sources of periodic unsteady disturbances and their aerodynamic and aeroelastic impacts are described rst.
Relevant parameters controlling the basic unsteady ow characteristics are
then described. Particular attention is paid to the role of circumferential
wavelength. Finally several modeling issues on computational efciency and
accuracy are discussed.

Copyright 2003 Marcel Dekker, Inc.

SOURCES AND EFFECTS OF UNSTEADINESS

Propagation of Wake and Pressure Disturbances
Unsteady disturbances travel in the form of ow characteristics. Each
characteristic has a distinct traveling velocity (magnitude and direction).
Although the basic characteristics are obtained from the linear Euler
equations (hence inviscid without any dissipation mechanisms), they are
extremely useful as the basis for developing computational models,
understanding ow physics, and interpreting computational (as well as
experimental) results. For a 2D inviscid compressible ow with V being
the local ow velocity and A the local speed of sound, there are four ow
characteristics: (1) entropy disturbance, convected at a local ow velocity V,
running downstream; (2) vorticity disturbance, convected at local velocity
V, also running downstream; (3) an acoustic (pressure wave) disturbance,
propagated at a speed of V A, always running downstream; and (4)
another acoustic (pressure wave) disturbance, propagated at a speed of
V
A, which runs either upstream for a subsonic ow, or downstream for a
supersonic ow.
In a turbomachinery context, the spatial nonuniform velocity and
pressure distributions in a blade row are sensed as unsteady disturbances in
adjacent blade rows. More specically, wakes shed from upstream blades
contain both entropy and vorticity components and normally have little
static pressure nonuniformity (until they impinge on downstream blades).
Therefore, a wake disturbance is convected downstream at a local ow
velocity. An example is shown in Fig. 1, where incoming unsteady wakes are
convected through a turbine passage and distorted in the nonuniform
velocity eld. For both compressor and turbine situation, a wake
disturbance mainly affects downstream rows, though unsteady pressures
generated when it hits the downstream blade also propagate upstream.
On the other hand, nonuniform steady static pressure necessarily
associated with blade loading in one blade row can be seen as unsteadiness
by both upstream and downstream relatively moving blade rows, assuming
that the axial-ow velocity is subsonic. The propagation of acoustic
(pressure) waves is largely an inviscid phenomenon, which is probably why
an interference with unsteady pressure disturbances is often called
potential interaction. Wakes have measurable velocity gradients and
therefore viscous dissipation always coexists with wake convection,
although the kinemics of wake convective transportation itself can be
described in an inviscid manner. An example of the potential interaction can
be seen in Fig. 2 for a transonic turbine stage, where pitchwise nonuniform
pressures of the upstream stator interact with the rotor in a complex
manner, largely due to reection of pressure disturbances. It is noted from

Copyright 2003 Marcel Dekker, Inc.

Figure 1
wakes.

Instant entropy contours of turbine blade passage subject to incoming

these snapshot contours that the downstream stator passage is more affected
by the potential interaction. This is typical because of higher loading
and ow speed in the rear part of turbine blading. Conversely for a
compressor case, an upstream blade row will typically be more affected by a
potential interaction, because of a typical frontal loading of the downstream
row.

Blade-Row Interference Effects

Having had some basic ideas about how unsteady disturbances propagate,
now we look at the aerodynamic and aeroelastic impacts due to the inherent
blade-row relative motion.

Copyright 2003 Marcel Dekker, Inc.

Figure 2 Instant static pressure contours for a transonic turbine stage.

Copyright 2003 Marcel Dekker, Inc.

Aerodynamic Interaction (Unsteady Losses)

Regarding aerothermal inuences of a wake shed from an upstream blade
on a downstream relatively moving blade row, there are several different but
related issues:
1.
2.
3.

How do we quantify ow loss in an unsteady environment?

How is the mixing loss of wake velocity decit inuenced by the
presence of a downstream blade row?
How does a wake disturbance affect the loss generation of
downstream blade rows?

Entropy/stagnation Pressure Decoupling. This is a basic issue concerning

transportation of ow losses in an unsteady environment, relevant to
presentation and interpretation of unsteady losses computationally and
experimentally. For a thermodynamic process, we have
T0 dS dh0

1
dP0
r0

Hence, in a steady adiabatic ow situation, a higher entropy is always

associated with a lower stagnation pressure, because the stagnation enthalpy
h0 remains constant along a streamline. Essentially, stagnation pressure
decit is a proper measure of entropy rise (loss) in steady ows. For an
unsteady ow, however, this is not the case. Neglecting the viscous and heat
conduction terms, the unsteady energy equation can be written as
dh0 1 qP

r qt
dt

Therefore, h0 T0 will not be constant along a stream line if static pressure is

varying in time, as it would be once an unsteady wake impinges on blade
surface.
An inviscid unsteady ow calculation by the author [2] for a lowpressure turbine cascade subject to periodic unsteady incoming wakes is
used to illustrate this point. Wakes shed from the upstream blade row are
modeled by specifying inlet wake proles, traveling relatively in the
circumferential direction at the rotor rotating speed. As can be seen from
the instantaneous entropy contours (Fig. 1), for the region upstream of the
blade passage, wakes are convected at an essentially uniform velocity, and
the time-averaged stagnation pressure and entropy would be more or less
uniformly distributed along the pitch upstream of the blade row. By looking
at the velocity triangles, we can see that within the wake there would be a
relative velocity component. The entropy within a moving wake is thus
convected relatively along the wake. Each wake rst hits the pressure surface

Copyright 2003 Marcel Dekker, Inc.

and acts like a negative jet [3]. Kinematically, these negative jets
effectively convect the high entropy within the wake toward the suction
surface (recall that entropy disturbances are convected at a local ow
velocity). If we look at the pitchwise distribution of time-averaged ow
parameters downstream of the blade row, we should have a higher loss
(entropy) region near the suction surface. This can be shown in Fig. 3(a),
which is the pitchwise distribution of time-averaged entropy rise at the exit
plane (about 50% chord downstream of the trailing edge). However, due to
unsteadiness, stagnation pressure decit is decoupled from entropy rise in
this unsteady ow case, as shown in Fig. 3(b). Note that at the position
around 60% pitch, we have a peak of stagnation pressure decit, while the
entropy rise is almost zero at the same position! Therefore, in an unsteady
situation, the time-averaged stagnation pressure variation might not
correspond to the true ow loss (time-averaged entropy rise) at all.
Wake/blade-row Interaction. First of all, we note that the velocity decit of
a wake shed from an upstream blade row will be eventually mixed out,
resulting in mixing loss. What is the effect of a downstream relatively
moving blade row on this mixing loss generation process? There are some
theories in a 2D context that the wake mixing loss can be reduced
(recovered) when passing through a compressor blade row, but it would
be increased when passing through a turbine row [4, 5]. The argument is
based on a simple kinematic consideration as shown in Fig. 4. One can see
that a wake passing through a compressor passage [Fig. 4(a)] is stretched
(elongated) by the velocity difference between the suction surface and
pressure surface, noting again a wake is convected at a local ow velocity.
We can simplify the argument by rst considering how the velocity decit
would be changed in a purely inviscid ow situation, and then thinking
about the consequent mixing loss due to the viscous dissipation of the
resultant velocity decit. An inviscid argument based on the conservation of
total vorticity (Kelvin theorem) would give a reduced velocity decit in this
stretched wake. It follows then that the corresponding mixing loss of the
wake passing through a downstream compressor blade row will be smaller
in comparison to that when the wake were mixed out in absence of the
downstream blade row. Following the same reasoning, one can reach the
opposite conclusion for an unsteady wake passing through a turbine blade
row [Fig. 4(b)] where the velocity decit is enlarged (hence larger viscous
dissipation and mixing loss) since the wake is compressed due to the
corresponding velocity difference between the suction and pressure surfaces.
A corollary is that a smaller intrarow gap should reduce wake-mixing losses
for compressors, but not for turbines.

Copyright 2003 Marcel Dekker, Inc.

Figure 3 Pitchwise time-averaged entropy rise and stagnation pressure distributions (downstream of a turbine cascade subject to incoming wakes).

Copyright 2003 Marcel Dekker, Inc.

Figure 4 Schematic of wake distortion through a downstream blade row

(kinematics due to suction/pressure surface velocity difference).

Copyright 2003 Marcel Dekker, Inc.

The above considerations are all for pure 2D and highly simplied
situations. It should be noted that even in a 2D case with realistic blading
geometry, wakes may not be stretched all the time in a compressor passage
or compressed all the time in a turbine passage as depicted in Fig. 4. The 2D
computational result in Fig. 1 shows just how the wakes can be both
compressed in the frontal passage part and stretched in the rear in addition
to being distorted, giving a much more complex picture.
Furthermore, attention is drawn to the very 3D nature of wakeblade
row interaction in the near end-wall regions. An unsteady wake shed from
an upstream row is characterized by the cross-passage transportation by the
negative jet within the wake, which convects low-momentum uid from
the pressure surface to the suction surface for turbines and conversely for
compressors. On the other hand, end-wall ows in the blade row under
consideration are characterized by 3D secondary ow structures with
dominant streamiwse vorticity, e.g., passage vortex, tip-leakage vortex. A
basic feature of all these 3D ow structures is that the associated crosspassage uid movement near an end wall has an opposite sense to that close
to midspan. Here one can use a simple phasing argument to examine the
interaction between an incoming largely 2D unsteady wake with a 3D
steady passage vortex. Just imagine two separate activities that could
behave linearly and thus be superimposed. One is the 2D incoming wake,
and the other is a steady passage vortex. Note that the sense of the crosspassage uid movement due to the incoming wake is the same regardless of
spanwise positions, while the cross-passage movement due to the passage
vortex changes the directions depending on the spanwise position. In the
very near-wall regions, the low-energy uid is transported by the passage
vortex from the pressure to the suction surface and thus is in the same
direction as the wake transportation (for a turbine). On the other hand, in a
region away from the end wall, the cross-passage movement due to the
passage vortex will have an opposite sense to that due to the wake.
Consequently, the resultant ow structure when a 2D wake is interacting
with a passage vortex will have to be strongly 3D. The suppression or
enhancement of cross-passage movements at different spanwise sections has
been shown to generate a radial redistribution time-averaged entropy loss in
comparison with a steady ow without incoming wakes [6]. Given that a
major portion of aerodynamic losses is generated in highly 3D end-wall
regions for typical blading designs, one does need to be cautious in applying
the 2D ow arguments. More work is needed to identify 3D blade-row
interaction effects and understand the corresponding mechanisms.

Copyright 2003 Marcel Dekker, Inc.

Rotorrotor Interaction (Clocking and Aperiodic Effects). If a wake is not

completely mixed out after passing through the immediate downstream
blade row, the remaining disturbance will interfere with a further
downstream but relatively stationary blade row, causing a rotorrotor (or
statorstator) interaction. Although by intuition we might expect much
smaller rotorrotor (or statorstator) interactions than rotorstator
interactions, the picture is not entirely clear and may vary depending on
different designs and ow conditions. It can be argued that rotorrotor
(statorstator) interactions should be much smaller in compressors than in
turbines because of the wake recovery mechanism described earlier. Some
recent analysis using a multistage computational method [7], however,
indicates a marked difference between a rotorrotor interaction and a
statorstator interaction in a transonic compressor.
We rst note that the effects of rotorrotor (or statorstator)
interactions strongly depend on blade counts. This can be highlighted in
two extreme cases: (1) both rotors have the same number of blades; and
(2) two rotors have such different blade counts that a relative
circumferential positioning between two blades in the two rows can only
repeat after the whole annulus. Consider a two-stage compressor
consisting of rotor-1, stator-1, rotor-2, and stator-2. If both rotors have
the same blade count, every blade passage in one rotor row will see the
same pattern of spatial (time-averaged) disturbance from the other blade
row. Hence the time-averaged rotorrotor interaction pattern will be
spatially periodic with a wavelength of one blade pitch. In this case, we
can change the relative circumferential position (clocking or indexing)
between the two rotors by up to one blade pitch. Thus, for the case with
an equal rotor blade counts, we have the full range (one pitch) to identify
the maximum clocking effect.
However, the clocking effect can be signicantly reduced when the two
rotors have different blade counts. For example, if rotor-1 has 21 blades and
rotor-2 has 20 blades, then adjacent blades in rotor-2 will have slightly
different circumferential positions relative to their counterparts in rotor-1,
and the maximum clocking range would only be 1/20 blade pitch. In this
case, although the clocking effect is expected to be negligible, the timeaveraged ow patterns in the passages in each rotor row will not be spatially
periodic. This aperiodic effect needs to be assessed if a single-passage
domain/traverse area is adopted for either computation or experimental
measurement. It is worth pointing out that for rotor-2, although the
magnitude of velocity decit of wakes from rotor-1 might be smaller than
those from the immediate upstream stator row, the circumferential
wavelength of the disturbance due to the rotorrotor interference mode
might be much longer. In this case with 21 rotor-1 blades and 20 rotor-2

Copyright 2003 Marcel Dekker, Inc.

blades, the wavelength of the disturbance is the whole circumference. As will

be discussed later, disturbances with long circumferential length scales can
have marked effects on downstream blade rows.
So far, most of clocking studies have been carried out for stators, e.g.,
the experimental work for a low-speed compressor by Barankiewicz and
Hathaway [8]. Rotorrotor interactions (clocking and aperiodicity) might,
however, behave differently in particular when a downstream rotor row is
transonic. A relevant mechanism to be taken into consideration is that
associated with the passage shock wave of rotor-2 interacting with the wake
shed from rotor-1. The mixing loss of a wake would be higher when the
wake is subject to an adverse pressure gradient [9]. Thus, in this transonic
mean ow condition, the mixing loss of rotor-1 wake is certainly expected
to rise when passing through the rotor-2 passage shock wave, as shown by
the schematic in Fig. 5. Given that the strength of the rotor passage shock
varies pitchwise (typically stronger near the suction side), it would not be
surprising that the rotorrotor interaction loss is dependent on the clocking
position.
In contrast to typical ow conditions for transonic rotors, the mean
ow in which two stator rows interact is subsonic. Therefore, it seems
reasonable to expect more pronounced rotorrotor interference (aperidoic

Figure 5 Rotor-1 wakes crossing rotor-2 passage shock at two different clocking
positions.

Copyright 2003 Marcel Dekker, Inc.

and/or clocking) effects than their statorstator counterparts for transonic

compressors.
Aeroelastic Interaction (Blade Forced Response)
All the aforementioned interaction effects are of a pure aerodynamic nature
where the main concern is the corresponding time-averaged unsteady losses
resulting from the interactions, which are usually small. The blade
mechanical responses under those unsteady disturbances can have very
important consequences. Excessive stress levels at a blade resonant
condition, in which the frequency of an unsteady disturbance coincides
with that of a particular blade vibration mode, may occur due to lack of
damping (aerodynamic and/or mechanical). The frequencies of the blade
vibration modes at risk can be identied from the Campbell diagram (Fig.
6), when the frequency curves (e.g., those for the rst ap 1F and the rst
torsion 1T) cross the engine order excitation lines (1EO, 2EO . . . ). Even if
the designer can choose an appropriate condition to avoid resonance at
design, it is difcult or impossible to do so at off-design conditions. When
the machine is started off or shut down, blades will have to go through some
resonance conditions, and it is important to be able to assess the unsteady

Figure 6 Campbell diagram.

Copyright 2003 Marcel Dekker, Inc.

forcing and damping, and hence the blade stress levels at these crossing-over
points. Given that a component fatigue cycle life span is inversely related to
alternating stress levels, high cycle fatigue has to be an important blading
design aspect, if blades are to be designed to last for a required life span to
give an optimum economic benet (e.g., in terms of manufacture as well as
maintenance costs) and reliability.
The aerodynamic sources of excitation (forcing) are those nonuniformities due to blade wakes, pressure waves, and inlet/exit distortions. On the
aerodynamic side, there are two essential parts required for predicting a
blade stress level in a forced response problem:
1.

Unsteady aerodynamic forces on blades generated by relative

motions of adjacent blade rows and/or circumferential distortions
2. Aerodynamic damping, effectively the unsteady aerodynamic
forces induced by vibration of a blade once excited
Predictions of unsteady forcing/excitation as well as blade aerodynamic
damping with adequate accuracies are challenging tasks. In the past they
were largely avoided by adopting empirical design rules based on
experiences. But this situation with regard to blade forced response is
about to change. Currently, there are active developments in using advanced
unsteady ow methods to improve the capability for forcing and damping
prediction, as required for current and future high-load blading designs.
There are some basic aeromechanical considerations in designs. The
main objective here is, of course, to reduce or sustain unsteady forcing,
when steady aero loading is increased. It has been shown that the main
contributor to unsteady forces in compressors is the incidence variation,
while the velocity variation in wakes is mainly responsible for forcing in
turbines [10]. The 3D blading with radially leaned blades is widely used in
designs. Although the leaned blading may well be chosen to reduce
aerodynamic losses from a steady ow point of view, it does have a lower
unsteady loading compared to a radially straight one, since a wake shed
from an upstream blade with radial staking will not hit different radial
sections of a downstream blade at the same time (a simple example of 3D
relief of unsteady loading). Given that 3D blading designs are currently
exploited for benets in aerodynamic performances, it is equally (if not
more) important that such 3D aerodynamic designs are aero-mechanically
sound.

Copyright 2003 Marcel Dekker, Inc.

Self-Excited Unsteadiness
All the unsteady disturbances generated by blade-row interactions share one
common feature. Their frequencies are simply related to rotor speed and
blades counts in a form
f N n fr

Where
fr is the machine rotation speed (in Hz).
Nn is the number of blades in the exciting row (in general, Nn is the
number of nodes of a circumferential distortion pattern rotating
relatively at a frequency fr ).
On the other hand, when an unsteady ow pattern is triggered by a selfexcited aerodynamic or aeroelastic instability, there would be no such simple
frequency correlation.
Aerodynamic Instability
Vortex Shedding and Self-excited Shock Oscillation. Vortex shedding is a
signicant source of prole loss for turbine blading with a thick trailing edge
[9]. Prediction of the base pressure at the trailing edge is dictated by proper
modeling of unsteady periodic vortex shedding. But for most steady CFD
methods used in current blading designs, the trailing-edge vortex shedding
phenomenon and its associated effects on prole loss are completely missed
out. Even when a time-accurate method is adopted, vortex shedding may
still not be captured due to excessive numerical dissipation and/or
inadequate mesh resolution. An alternative approach is to introduce the
deterministic stress terms to the steady ow equations, and it has been
shown that a time-independent solution of the time-averaged ow eld due
to a vortex shedding can be obtained by using a steady ow solution method
[11]. For an unsteady ow eld inuenced by unsteadiness with a short
wavelength, the spatial gradient can be much steeper than its time-averaged
counterpart. Hence, a time-independent solution to the time-averaged ow
equations would not demand as high resolution as a time-domain unsteady
calculation. A practical question that remains to be answered is how to
model the deterministic stresses without resorting to full time-domain
unsteady calculations.
In an undisturbed ow, frequency characteristics of a vortex shedding
are largely dependent on the boundary-layer state and the geometry of the
trailing edge. However, its frequency can be easily locked into frequencies of
external disturbances. For example, in a turbine stage conguration, a
shedding from the upstream blade row can be locked into the blade passing

Copyright 2003 Marcel Dekker, Inc.

frequency and its higher harmonics [12]. Furthermore, because of its

sensitivity to ow conditions, vortex shedding may well couple with other
oscillatory acoustic or blade structural dynamic modes, and act as an
excitation source.
For passage shock oscillations, the ow physical mechanism is
typically associated with shock/boundary-layer interaction, resulting in a
thickened or separating boundary layer at the foot of passage shock wave.
The oscillating shock patterns (frequencies) are very often suspected to be
associated with acoustical modes within blade passages, similar to those
found in transonic ducts (e.g., [13]). An oscillating shock wave produces a
higher time-averaged loss than its steady counterpart. The existence of
shock oscillation can be easily identied from steady (time-averaged)
experimental data showing a smeared shock wave. The range of smearing
is normally a pretty good indication of the magnitude of the shock
oscillation.
Rotating Stall. Rotating stall is a circumferential ow instability, which is
normally regarded as a precursor of compressor and engine surge. Apart
from seriously detrimental effects on aerodynamic performances, it should
also be noted that stall onset generates severe blade vibration problems. The
vortical and reversal ow pattern produces transient aerodynamic loading
of a considerable magnitude, causing blades to be overstressed. There is a
need to estimate the maximum stress level in this fairly hostile environment,
especially if the stability boundary can only be identied by crossing the
boundary during experimental rig tests.
There has been a considerable amount of work recently resulting in
enhanced understanding of stall inception mechanisms based on experimental observations, e.g., [14]. It should also be mentioned that full-scale
CFD simulations of stall inception have started to emerge which can help to
understand complex physical mechanisms involved and identify relevant
inuencing parameters. For instance, computational studies using an
unsteady NavierStokes time-domain ow solver for a compressor stage
indicate that the initial stall inception pattern (number of cells, circumferential wavelength, and rotating speed) corresponds to that set by rotor
stator blade counts [15]. Figure 7 shows the results at an initial stage of stall
inception for the rotor and stator rows with blade numbers of 10 and 12,
respectively. With these blade counts, the rotor and stator disturbances beat
circumferentially twice per rotor revolution, resulting in an interference
disturbance with a wavelength of half an annulus. This should explain why a
two-cell structure is triggered as shown by a snapshot of entropy contours at
the inception [Fig. 7(b)]. Figure 7(a) shows the time traces of axial velocities
from four circumferential stationary positions upstream of the rotor row.

Copyright 2003 Marcel Dekker, Inc.

(May 2

Figure 7 NavierStokes solution of stall inception for a compressor stage

conguration (10 rotor blades, 12 stator blades).

Copyright 2003 Marcel Dekker, Inc.

The two-cell pattern with a relatively short circumferential wavelength

rotates in the absolute frame of reference at a faster speed (60% of rotation
speed) before merging to a more common single-cell pattern rotating at a
slower speed (48% of rotation speed). However, if the rotorstator blade
counts are such that the interference wavelength is the whole annulus, a
single-cell pattern will be triggered directly, as shown in Fig. 8, where the
rotor and stator blade numbers are 10 and 9, respectively. The overall
observation is that although a single-cell pattern rotating at about 50% rotor
speed is the most common one, a multiple-cell structure might occur at an
initial stage of stall inception if the stalling blade row is subject a distortion
disturbance of the same circumferential wave pattern. A numerical analysis
of 3D unsteady CFD results for a fan rotor at subsonic and relatively
supersonic ow conditions [16] suggests that the basic stall inception
mechanisms in high-speed transonic ows might be qualitatively different
from that at low speeds due to the difference between propagating
information upstream in subsonic ow and in supersonic ow.
So far, most of the research activities are largely aimed at providing
detailed (even qualitative) information to help understanding, motivated by
the prospect of active control/depression of rotating stall. We must
appreciate that true prediction of exact stall onset conditions for
determining suitable stall margins during designs is still extremely difcult,
depending strongly on numerical resolution, turbulence modeling, and
boundary condition treatment of truncated computational domains.
Aeroelastic Instability (Flutter)
Blade utter is a self-excited aeroelastic instability phenomenon. When
working blades are disturbed aerodynamically or mechanically, they tend to
vibrate in their natural modes with small amplitudes, which will in turn
induce unsteady aerodynamic forces. At certain conditions, the unsteady
forces resulting from the initial blade vibration will do a net work (energy
input) to the blade in each period of vibration. As a result, the initially small
vibration will be amplied and the instability (utter) will occur, often
leading to blade failure if there is not enough mechanical damping to
dissipate the energy.
Flutter can be a serious problem for frontal stages of compressors, in
particular fans of aero-engines, and needs to be addressed during a design
process. Figure 9 shows typical blade utter boundaries in an axial-ow
compressor/fan performance map. Corresponding aerodynamic conditions
of the instabilities are indicated by the names given, which clearly suggest
that steady and unsteady aerodynamics play an essential part in the
aeroelastic instability. At low off-design speeds, danger exists under high

Copyright 2003 Marcel Dekker, Inc.

Figure 8 NavierStokes solution of stall inception for a compressor stage

conguration (10 rotor blades, 9 stator blades).

Copyright 2003 Marcel Dekker, Inc.

Figure 9 Typical blade utter boundaries on compressor map.

incidence (high-pressure ratio) conditions, where ow separation on the

blade suction surface is likely to occur. The highly loaded blades (especially
with separating boundary layers) are very sensitive to small disturbances
and a separating ow pattern can easily couple with blade vibration, giving
a negative aerodynamic damping (i.e., a net energy input to blade vibration).
At higher operating speeds, a strong shock system is expected to play a role
in utter onset, either by intensive unsteady loading around an oscillating
shock (supersonic unstalled utter) or by inducing periodic boundarylayer separation (supersonic stalled utter). At medium speeds, throttle
opening (reducing incidence) might lead to choked blade passages, where a
supersonic region is ended by a weak shock wave. This kind of transonic
ow pattern is very stiff and can produce a large ow response to a small
disturbance and is likely to be responsible for choke utter.
It must be pointed out, however, that these names of the utter
boundaries need to be taken with cautionthe real mechanisms may be far
more complicated than they suggest. For instance, although it is expected
that subsonic stall utter should be associated with separated ows, there
have been some experimental results showing that this may not necessarily
be the case [17], and it is argued the instability may be better called high
incidence utter.
Flutter has been mainly a problem on the compressor/fan side, but
turbine utter (especially of low-pressure stages) has also been of concern.

Copyright 2003 Marcel Dekker, Inc.

Overall it is fair to say that because of the complex aerodynamic and

aeroelastic mechanisms involved in these instabilities, both current understanding and predictive capability are far from being satisfactory. The
situation will be more pressing as the trend of future designs heads toward
higher aero loading.

RELEVANT PARAMETERS
Reduced Frequency
One of the most important parameters for unsteady ow problems is the
reduced frequency dened as
K

oL
V

Where
o 2pf is the angular frequency.
f(Hz) is the physical frequency of the unsteadiness.
L is a reference length scale, usually taken to be the chord length C.
V is a reference velocity, usually taken to be the inlet ow velocity.
The reduced frequency can be interpreted as
K

time scale for flow particles to be convected over 2pL

time scale of unsteadiness

5a

2pL
spatial length scale of a convected wave

5b

or

When modeling an unsteady ow problem, one must ensure that the

reduced frequency is the same as that in the real situation. The reduced
frequency is a measure of both temporal and spatial length scales. Thus at a
high reduced frequency, we need to use a scheme with high enough temporal
accuracy as well as ne enough spatial resolution since the high-frequency
unsteadiness would vary on a short spatial length scale.
As a simple measure of a true unsteady behavior against a steady one,
if the reduced frequency is much smaller than 1 (say K < 0.1), the behavior
can be regarded as quasi-steady, then the time-derivative terms in the
unsteady ow equations (and hence the temporal accuracy) can simply be
neglected. In other words, at each time instant, an effectively steady ow

Copyright 2003 Marcel Dekker, Inc.

eld is generated to satisfy the time-varying boundary conditions at that

instant.
For unsteady ows in turbomachinery, values of the reduced
frequency vary considerably. Resolving all different frequencies (length
scales) will be extremely difcult. Before doing unsteady ow calculations,
we should have some ideas about the relevant length scales of interest, so
that adequate and efcient discretization schemes and mesh arrangements
can be adopted.
For rotorstator interactions, unsteadiness experienced by a blade row
is caused by pitchwise blade-to-blade nonuniformities of adjacent blade
rows. Obviously, the dominant spatial length scale would be the blade pitch
(or chord) length, and the time scale would be blade passing period. A
typical value of the reduced frequency is about 510.
A high-frequency (short spatial wavelength) example is the blade
trailing-edge vortex shedding. By analogy to the Karman vortex street
behind a cylinder, the corresponding spatial length scale is of the blade
trailing-edge thickness d, and the corresponding time scale is mainly subject
to the convection time d=u, where u is the free-stream velocity. Both the
temporal and the spatial length scales may be by one order of magnitude
smaller than those due to rotorstator interactions. A low-frequency (long
wavelength) example is a rotating stall. A typical stall cell can cover several
blade passages and circumferentially travels at a speed about 50% of the
blade rotation speed. So both the temporal and spatial length scales can be
much longer than those for rotorstator interaction problems.
For blade utter problems, the time scale is given by the blade
vibration frequency. Usually the chord length is taken as the reference
length scale. A typical value of the reduced frequency for the rst bending/
ap mode is around 0.3, while that for the rst torsion is around 1. Simple
aeroelastic design guidance can be based on the observations that blades
tend to utter in low-order vibration modes and are less stable at lower
frequencies. Thus, blade-reduced frequencies need to be above certain
(empirical!) limiting values, e.g., K > 0:2 for the rst bending/ap and K >
0:6 for the rst torsion mode. These can be achieved by either increasing
blade natural frequencies (stiffness) or reducing ow velocity (loading).
Blade Count and Interblade Phase Angle (Inuence of
Circumferential Wavelength)
All practical turbomachinery stages have different numbers of blades in
neighboring rotor and stator rows to avoid resonance. And this difference in
blade counts denes the circumferential interference wavelength. Consider a
hypothetical compressor stage, for instance. If we have 10 rotor blades and 9

Copyright 2003 Marcel Dekker, Inc.

stator blades, the unsteady disturbances from the two rows can only beat
each other after the whole annulus, so the circumferential wavelength is the
whole annulus, as indicated by the instantaneous static pressure pattern
shown in Fig. 10. On the other hand, if the stator blade number is changed
to 12, the two rows beat twice for the whole annulus, so the wavelength is
half the annulus (Fig. 11). The fundamental interference wavelength is thus
determined by the difference of the blade numbers of the two blade rows.
On the aeroelastic side, for both utter and forced response problems
we have similar patterns. Vibratory patterns of blade and disk assemblies
are typically featured by a cyclic symmetry mode, rotating circumferentially
at a constant speed. For this kind of traveling wave modes, the radial lines
with zero displacements are called nodal diameters. So the circumferential
wavelength is dened by the number of nodal diameters.
In both blade-row aerodynamic interaction and aeroelastic problems,
each blade will be subject to unsteadiness of a circumferential traveling wave
pattern. Since the circumferential wavelength is not the same as the blade
pitch (usually much longer for those problems of interest), we no longer
have the direct periodicity between adjacent blade passages. Instead, there is

Figure 10 Instant static pressure contours (NBr 10, NBs 9, circumferential

interference wavelength whole annulus).

Copyright 2003 Marcel Dekker, Inc.

Figure 11 Instant static pressure contours (NBr 10, NBs 12, circumferential
interference wavelength half-annulus).

a constant phase lead or lag, called inter blade phase angle s (in radians):
s

2pn
N

Where
N is the number of blades of the blade row under consideration.
n the number of nodal diameters of the disturbance.
Hence, for a given number of blades, s also denes the circumferential
wavelength. Typically the utter instability occurs corresponding to an
interblade phase angle with a small number of nodal diameters.
For blade-row interaction, if a row of N blades is subject to unsteady
disturbances generated by an adjacent blade row with M blades, the most

Copyright 2003 Marcel Dekker, Inc.

apparent value of interblade phase angle would be

s 2p

M
N

7a

In general, there are multiple values of interblade phase angles corresponding to multiple possible interaction modes, as permitted by the circular
functions. For instance, the interference wavelengths shown in Figs. 10 and
11 will correspond to a form
s 2p

N
M
N

7b

The circumferential wavelength has a direct implication on computational

modeling, as will be seen later. It also has important physical inuences on
unsteady aerodynamic and aeroelastic characteristics. For blade utter
analysis, it has been well established that the aerodynamic damping at a
given blade vibration frequency and mode shape can change from a positive
value to a negative one depending on the interblade phase angle. For
aerodynamic interactions under a circumferential distortion, the spatial
wavelength is normally tied together with the temporal frequency, so it is
not easy to isolate the spatial wavelength effects. Nevertheless, the following
two examples should be able to illustrate the impacts of the circumferential
wavelength of disturbances on unsteady responses within blade passages as
well as on axial decay of the disturbances.
The rst example concerns the steam turbine operation at a practical
partial admission condition, in which one or several circumferential
segments of annular ow path are blocked to adjust the total mass ow
rate (and hence power output). Figure 12 shows computed instantaneous
entropy contours for two different blocking arrangements at the rst stator
leading-edge plane to achieve a 50% ow rate, 1650% and 2625%
admissions for a two-stage high-pressure turbine [18]. For both cases, the
entropy contours clearly show considerable boundary-layer separation on
blade pressure surfaces of the second stator row. Figure 13 shows the
pitchwise time-mean static pressure distribution downstream of the rst
stator row and that downstream of the second stator row. Clearly there is
very strong circumferential static pressure nonuniformity due to the partial
admission. Comparing Fig. 13(a) with Fig. 13(b), we note that the pitchwise
pressure nonuniformity at the 2625% admission has a signicantly faster
decay rate than that at the 1650% admission. The magnitudes of the
pitchwise pressure nonuniformity for both cases are roughly the same at the
position downstream of the rst stator row. However, downstream of the
second stator row, the magnitude of the nonuniformity for the 2625%

Copyright 2003 Marcel Dekker, Inc.

Figure 12 Calculated instantaneous entropy contours for two-stage turbine at

partial admission.

admission is less than half of that for the 1650% admission. Therefore, we
see that the nonuniformity with a longer circumferential wavelength decays
at a much slower rate in the axial direction.
The second example concerns responses of a blade row to inlet
distortions. In this case, stationary stagnation pressure distortions are
specied at the inlet to a transonic fan rotor row. And the results obtained at
the same distortion amplitude of 15% are compared for two different
circumferential wavelengths [19]. The distortion pattern in the rst case has
11 nodal diameters (wavelength 2 passages). In the second case, the
distortion has 1 nodal diameter (wavelength whole annulus, 22 passages).
The inuence of the circumferential distortion wavelength can probably be
best illustrated by time-space static pressure contours on the suction surface
at 90% span, as shown in Fig. 14. Note that at this spanwise section, the
passage shock wave in a steady solution is situated on the suction surface
around 8590% chord. The results demonstrate that the passage shock
oscillation is strongly dependent on the distortion wavelength. For the short
wavelength case [Fig. 14(a)] the shock oscillation is conned to 35% chord,

Copyright 2003 Marcel Dekker, Inc.

Figure 13 Time-averaged pitchwise static pressure distributions downstream of the

rst and second stator rows of a turbine at partial admission (Y pitchwise distance;
pitch stator blade pitch length).

Copyright 2003 Marcel Dekker, Inc.

Figure 14 Time-space static pressure contours on suction surface at 90% span

under different inlet distortion wavelengths.

while for the case with a long length scale [Fig. 14(b)] the moving shock
covers nearly 50% chord. Clearly, this long wavelength case with a large
shock oscillation is strongly nonlinear.
Given the importance of circumferential length scales, one should be
cautious about the practice of slightly changing blade counts to obtain a
much reduced multipassage computational domaina drastic change in the
corresponding circumferential wavelength by doing so may result in a very
different unsteady response!

Phasing of Unsteady Disturbances

It must be kept in mind that inuences of an unsteady disturbance depend
on its phase angle, not just the amplitude. The most obvious example is the
use of antisound where an acoustical noise disturbance can be completely
cancelled by articially introducing an acoustical disturbance with the same
amplitude but a 1808 phase shift.
Phasing Between Wake and Potential Disturbances
In the context of blade forced responses, there is a scope of utilizing the
phase difference between wake and potential disturbances to produce
positive results. For example, a higher unsteady loading might not

Copyright 2003 Marcel Dekker, Inc.

necessarily be generated at a higher steady loading condition if the wake and

potential disturbances are closer to being antiphase [10]. The phasing effect
is caused simply by the fact that a wake shed from an upstream row is
always convected downstream at the local ow velocity in the relative
streamwise direction, but the pressure disturbances from the upstream row
do not follow the same pattern at all! The resultant response is dictated by
the difference in propagation directions and speeds between the wake
(entropy and vorticity) and the pressure disturbances. A downstream
relatively rotating blade would see the pressure peaks originated from the
two disturbances at different times in one passing period, depending on its
axial location (the axial gap). Given that potential and wake disturbances
propagate in different paths, there might therefore be an optimum gap to
minimize the net unsteady forcing, where the two peaks due to the two
disturbances are 1808 apart circumferentially.

Phasing Between Aerodynamic Force and Blade Vibration

(Energy Transfer in Blade Flutter)
As mentioned earlier, energy transfer is the fundamental consideration in
blade utter, and the role played by phasing in the instability is illustrated
here. Assume a blade is initiated (e.g., disturbed by mechanical or
aerodynamic noise) to vibrate in a sinusoidal form with a small amplitude.
Its vibration displacement XB is
XB AX sinot

Consider that this blade vibration movement induces an unsteady

aerodynamic force FA , in a sinusoidal form with a phase angle fFX relative
to the blade vibration displacement [although higher-order harmonics of
ow response may arise due to nonlinearity, as was seen earlier, it can be
shown that only the fundamental harmonic component of the ow response
will contribute to the net energy transfer between the blade oscillating in the
form Eq. (8) and the ow].
FA AF sinot fFX

The rate at which the aerodynamic force does work to the blade is
w FA VB

10

where VB is the blade vibratory velocity. The net work input to the blade in

Copyright 2003 Marcel Dekker, Inc.

one period Tp (neglecting change of the vibration amplitude) will be

Z Tp
FA VB dt
WA

11

pAF AX sinfFX
A utter instability would occur when the net work input to blade vibration
is positive (i.e., a negative aerodynamic damping when the induced unsteady
force leads the blade vibration displacement, 0 < fFX < p).

MODELING ISSUES
As for any other engineering problems, there are two basic considerations in
prediction of unsteady ow effects: accuracy versus efciency. Very often, a
blading designer or method developer may face a choice between a
seemingly more complete model and a truncated/reduced one. Usefulness of
a complete model is likely to be dictated by numerical resolution limited by
computing power, while the truncated model apparently would neglect some
phenomena deemed to be unimportant for the problems of interest. In the
unsteady aerodynamics and aeroelasticity area, a judgment can be helped by
appreciation of a common modeling issue, ow nonlinearity.
Unsteadiness/Time-Averaged Flow Interaction (Nonlinearity
and Deterministic Stresses)
In relation to blading designs, aerodynamic designers normally are only
interested in time-mean performances, rather than details of unsteady
disturbances. The basic question is, can we obtain the time-mean parameters
by solving the steady ow equations under the time-mean ow conditions?
Effectively, what we ask is the difference between the steady ow equations
and the time-averaged unsteady ow equations. The fundamental principles
can be illustrated in a very simple situation. Let us consider a 1D inviscid
ow governed by the following equations:
Continuity :
Momentum :
Energy :

qr qru

0
qt
qx
qru qruu
qP

qt
qx
qx
qre qruh0

0
qx
qt

12a
12b
12c

This set of equations is applicable to one-dimensional steady and unsteady

Copyright 2003 Marcel Dekker, Inc.

(random and periodic) ows. A purely steady ow will satisfy the equations
with the time derivatives being zero. To answer the question raised earlier,
we apply the standard time-averaging technique. An instantaneous ow
variable can be expressed in terms of a time-averaged part and a uctuation:
f x; t f x f 0 x; t

13

where
f x

1
T

Z
f x; tdt

and

f 0 x; t 0

If the ow is randomly unsteady (turbulent), the time scale T over which the
averaging is carried out should be much larger than the scales of turbulence
uctuations. For a periodic ow, T is the time length of the period. We
substitute the time-averaged and uctuation parts into the unsteady ow
equations [Eq. (12)]. Then the time-averaged equations would be as follows:
qru
0
qx
qruu qru0 u0
qP

qx
qx
qx
0 0
qruh0 qru h0

0
qx
qx

14a
14b
14c

The above time-averaged equations can be compared to their counterparts

for purely steady ows:
qru
0
qx
qruu
qP


qx
qx
qruh0
0
qx

15a
15b
15c

Comparing the time-averaged equations [Eq. (14)] with the steady equations
[Eq. (15)], we have the following observations:
1. The mass continuity equation remains unchanged, i.e., the timeaveraged mass ow at the inlet and outlet of a ow domain must
be conserved.
2. For the momentum and energy equations, the time-averaging
generates extra terms, due to nonlinearity of the equations. These
terms need extra relationships or equations to close.

Copyright 2003 Marcel Dekker, Inc.

It should be noted that the effects of the extra terms depend on the spatial
gradients of the unsteady disturbances (not just their absolute magnitudes).
Also note that the pressure term remains unchanged since it is linear. Thus,
an unsteady ow with large velocity uctuations (e.g., unsteady vortices) is
likely to be have more signicant effect on the time-averaged ow than one
with mainly pressure uctuation (e.g., pressure waves). Clearly, the timeaveraged results of an unsteady ow cannot be obtained by simply solving
the steady ow equations.
The most common and probably most complicated type of unsteady
ows is steady turbulent ows. It is somehow ironic that we hear much
more about how turbulence affects our steady ow results than we do
about a periodic unsteadiness. In a case of a turbulent ow, the extra terms
(e.g., ru0 v0 in a 2D case) are the turbulence (Reynolds) stress terms, which
need to be closed by turbulence modeling in one form or another. The basic
mechanism to generate turbulence stress terms is exactly the same as what is
illustrated above using the 1D ow equations, i.e., by the nonlinearity of the
inviscid part of the equations. The only difference is that the Reynolds
stresses are generated by essentially random unsteadiness, while for the cases
we consider here, the stresses terms are generated by unsteadiness with
distinctive frequencies, and for this reason they are called deterministic
stresses as cast in the framework for blade-row interactions by Adamczyk
[20].
Likewise, an appreciation of the effects of turbulence is helpful in
identifying those of a periodic unsteadiness on time-averaged ows. For
instance, we know that magnitudes of turbulence uctuations are generally
much smaller than an averaged ow velocity. It is therefore not surprising
that a periodic unsteadiness with a small magnitude could have a signicant
effect on a time-averaged ow. In general, as far as a time-averaged ow is
concerned, periodic unsteadiness might be regarded as regular turbulence.
The main difference is that the extra terms due to periodic disturbances can
be directly evaluated by unsteady ow calculations or modeling with less
uncertainties than conventional turbulence modeling for the Reynolds stress
terms.
Here are some simple examples concerning the loss associated with a
periodic unsteady ow. Firstly, consider a shock wave in a quasi 1D duct
ow (Fig. 15). In a steady ow situation with xed upstream conditions, the
solution of the shock wave (position, strength, and entropy rise, etc.) is
determined by the downstream static pressure P2 . Now the shock wave is
made to oscillate by the downstream static pressure changing in the
following form:
P P2 Am sinot f

Copyright 2003 Marcel Dekker, Inc.

16

Figure 15

Steady and time-averaged unsteady pressures around shock wave.

Assume the shock wave oscillates over the distance between location 1 and
location 2. We can then compare the time-averaged ow to a steady one at
the same steady (time-averaged) back pressure P2 . First it should be noticed
that the time-averaged pressure distribution is smeared between these two
extreme locations by the unsteadiness (not the articial viscosity!). For a
steady ow, conservation of mass, momentum, and energy will sufciently
determine the jump conditions across a shock (i.e., the RankineHugoniout
relations), so that the entropy loss across the shock will simply follow a
control volume analysis. Now we can take a control volume between
locations 1 and 2 that covers the region of shock motion. Then the
downstream momentum at the boundary surface of the control volume is
For the steady ow:
ruu2 ruu1
P2
P1

17a

For the time-averaged unsteady ow:

ruu2 ruu1
P2
P1
ru0 u0 2

17b

The unsteady stress term ru0 u0 2 can be roughly approximated by ru02 2 ,

which is a positive term. Therefore, for the unsteady ow case, there would
be less time-averaged downstream momentum, compared to the steady ow
case. Hence we would expect an extra aerodynamic loss in the case of an
oscillating shock.

Copyright 2003 Marcel Dekker, Inc.

There are some further points to note about the dependence of shock
oscillation characteristics on the mean shock wave strength. A quasisteady
analysis [2] shows that at the same relative amplitude of upstream Mach
number variation, a weak shock wave gives a higher relative increase of
time-averaged entropy rise than a strong shock wave. It is also observed
based on a time-domain CFD simulation (He, 1990) [21] that under the
same blade oscillation amplitude and frequency, a weak shock wave moves
over a larger blade surface distance with a more pronounced nonlinear
behavior than a strong one.
In addition to the difference between a time-averaged ow and a
steady ow, nonlinearity also manifests itself in the temporal form of
unsteady ow responses to incoming disturbances. For instance, for a given
blade sinusoidal oscillation:
Xb Ax sinot

18

where Ax is the amplitude of the blade vibration. Corresponding ow

velocity uctuating components will in general be in a Fourier series form:
u0 A1 sinot fu1 A2 sin2ot fu2   
v0 B1 sinot fv1 B2 sin2ot fv2   

19a
19b

If the ow responds to the blade oscillation (input) in a linear manner, then

any unsteady ow parameter should also vary in a pure sinusoidal form with
negligible higher harmonics, i.e.:
u0 A1 sinot fu1 ;

v0 B1 sinot fv1

20

Thus in the way, a linear response will be in the same harmonic form as that
of a boundary disturbance.
If, however, nonlinear effects are measurable, the ow response will be
inuenced by the nonlinear product terms of the ow equations. Then the
fundamental harmonic components can be seen to generate higher
harmonics:
u0 v0 A1 sinot fu1 B1 sinot fv1    
i
A1 B1 h

cosfu1
fv1
cos2ot fu1 fv1   

2

21

Hence, we can see that in addition to an extra steady (time-average) part,

there would also be a second harmonic component in the solution, which is
produced by the rst harmonics due to nonlinearity. In this way, a
nonlinear interaction can create other harmonics that are not present in the
forcing disturbances at domain boundaries. Thus, for a system subject to

Copyright 2003 Marcel Dekker, Inc.

only the rst harmonic disturbances at boundaries, an appearance of the

second- or higher-order harmonics is an indication of nonlinearity.
An example to illustrate a linear or nonlinear behavior is given for an
NACA-65 airfoil oscillating sinusoidally in a torsion mode. In this case,
surface unsteady pressure and shear stress were measured at different mean
incidence conditions [22]. Figure 16 shows ensemble-averaged unsteady
pressures for two periods measured at different chordwise surface locations
at 08 mean incidence. We can see that the pressure variations are almost
sinusoidal, indicating a linear response to the sinusoidal oscillatory
movement of the airfoil. However, at a high mean incidence of 108, there
is a separation bubble around the leading edge on the suction surface as
shown in the smoke ow visualisation [Fig. 17(a)]. The corresponding
ensemble-averaged unsteady pressure variations are nonsinusoidal [Fig.
17(b)], indicating a nonlinear pressure response due to the bubble-type
separation. In this case, subject to a sinusoidal unsteady disturbance of a
single frequency (blade vibration), the nonlinear effects manifest in the
higher harmonics.
In a more complex situation subject to two fundamental external
disturbances with frequencies o1 and o2 , it can be shown following the

Figure 16 Ensemble-averaged unsteady pressure data at various suction surface

locations of oscillating airfoil at a mean incidence of 08 (torsional oscillation at 28
amplitude).

Copyright 2003 Marcel Dekker, Inc.

Figure 17

Flow around an NACA-65 airfoil at a mean incidence of 108.

above reasoning that nonlinearity will result in secondary or induced

disturbances with frequencies: o1 o2 ; o1
o2 , and their higher harmonics. It is normally the long wavelength component at the low frequency
o1
o2 that has a more pronounced inuence. Generally speaking, in all
situations with measurable nonlinear effects, Fourier-transformed unsteady
pressure or velocities will reveal other harmonic components, which are not
present in the unsteady forcing disturbances at domain boundaries (inlet/
exit/blade surface).
Aerodynamic/Structural Dynamic Interaction
For utter and forced response problems, we must address how unsteady
aerodynamics interacts with blade structural dynamics. The situation

Copyright 2003 Marcel Dekker, Inc.

regarding the aeroelastic coupling effects for turbomachinery blades is

largely inuenced by the structural/ow mass ratio, measured by the mass
coefcient dened as:
Cmass

m
rpC=22

22

Where
m is the mass per unit span of blade or airfoil.
C is the chord length.
r is the air/gas density.
Typical values of the mass coefcient for conventional turbomachinery
blades are by an order of magnitude higher than those of aircraft wings (or
wind turbine blades). Thus, the changes of blade vibration mode shapes and
the frequencies due to aerodynamic damping and forcing are usually small
and may be neglected. This fundamental blade structural dynamic
characteristic has a signicant bearing in the ways blade utter and forced
response problems are dealt with.

Flutter Prediction
At a utter inception, blade vibratory movements (mode shape and
frequency) can be assumed to be the same as in the natural (i.e., without
unsteady aerodynamic forces) modes and frequencies. Then the stability of
the blade-ow aeroelastic system can be simply determined by the energy
transfer between the blade and its surrounding ow. The above energy
consideration forms the basis for the conventional blade aeroelastic
calculation procedure, the energy method [23].
In a utter prediction using the energy method, there are three main
steps. First, the blade natural vibration modes shapes and frequencies are
obtained by a nite-element (FE) analysis. Second, the unsteady aerodynamic forces induced by a specied blade vibration in a natural mode
(typically a lower-order one, e.g., rst ap or rst torsion) will be calculated.
Third, the blade aeroelastic stability will be determined purely based on the
total energy transfer (work sum) in one period between the blade vibration
and the ow. The mode shape only gives the relative displacement of
different points on a blade surface. An aerodynamic damping calculation
can in practice be carried out by taking a small maximum displacement (say
0.5% chord for the rst bending mode or 0.5 degree for rst torsion). If a
blade surface is covered by Nc mesh cells, then the total aerodynamic work
done to the blade in one period of vibration at a frequency o is [see Eq. (8)

Copyright 2003 Marcel Dekker, Inc.

(11) for a simple case]:

WA p

Nc
X

Af Ax sinffx i

23

i1

Where
Af is the amplitude of the induced aerodynamic force component on
mesh cell i in the direction of the vibratory motion.
Ax is the amplitude of the vibrating displacement of the cell.
ffx is the phase lead of the induced aerodynamic force relative to the
displacement for the cell.
Aerodynamic damping can be dened as
DA
WA

24

The total damping of the blade aeroelastic system is

D DM DA

25

A conservative prediction can be made by neglecting the mechanical

damping, DM , which is always positive. Aeroelastic stability of a blade is
directly indicated by a damping coefcient taking a form of
CA

WA
A2m

26

where Am is the maximum vibration amplitude taken in the unsteady

aerodynamic calculation. So, the somehow arbitrary vibrating amplitude
taken in the damping calculation does not matter provided the aerodynamic
response is of a linear nature.
Apparently, given natural vibration mode shapes and frequencies, a
blade utter analysis using the energy method will be solely determined by
unsteady aerodynamic calculations. This is why unsteady aerodynamics
around oscillating blades in a specied mode has attracted so much
attention and effort in utter analysis and predictions for turbomachinery
blades.
Forced Response Prediction
Forced response is more complex, because it involves two separate
aerodynamic elements, aerodynamic damping and aerodynamic forcing.
The analysis may be most effectively carried out in a Modal space, where
the original structural dynamic equations with N degrees-of-freedom are
decoupled into N single-degree-of-freedom equations. Each of these

Copyright 2003 Marcel Dekker, Inc.

decoupled equations corresponds to one vibration mode. For instance, if

mode k is identied in the Campbell diagram as the one with resonance
crossing, the natural mode at frequency ok is governed by
qk o2k qk 0

27

where qk is the modal displacement for this mode, relating to the physical
displacements through the mode shape matrix. Assuming that the mode
shape is not affected by the aerodynamics, the vibration displacement under
an external aerodynamic force should be a simple solution to an equivalent
equation:
qk ck q_ k o2k qk fk

28

where fk and ck are the modal aerodynamic force and damping coefcient,
respectively. The key link among different parts in the process is the mode
shape.
An alternative and somewhat simpler approach can be entirely based
on the energy consideration, similar to the energy method applied to utter
analysis. At an equilibrium (steady-state) vibration, the energy input from
the aerodynamic forcing to the blade must be balanced by the dissipation
work done by the aerodamping. So the question becomes one to nd a nite
blade vibration displacement to satisfy
WF D A

29

where WF is the work input to the blade by the aerodynamic forcing. This
work sum can be calculated similarly to the aerodynamic damping work in
Eq. (23), as long as the phase between the aerodynamic forcing and the
blade response vibratory movement is given. It is known that for a forced
oscillator in a single degree of freedom, the response displacement will lag
the forcing by exactly 908 if the frequency shift can be neglected.
The overall observation is that aerodynamic and structural interactions for conventional turbomachinery blades can be dealt with in a loosely
coupled manner. Caution should be taken about the linear assumptions
implied here about both aerodynamics and structural dynamics. On the
blade structural dynamic side, nonlinear effects are evident when underplatform dampers are used, resulting in measurable frequency shifts [24]. On
the unsteady aerodynamic side, the aerodamping characteristic for blade
utter is normally considered to be linear, but the situation for forced
responses is less clear. Nonlinear aerodynamic behavior has been revealed
for transonic fans under inuences of inlet distortions. But its impacts on
aeroelastic coupling are still to be identied.

Copyright 2003 Marcel Dekker, Inc.

Regardless whether loosely coupled approaches or fully time-domain

coupled approaches (e.g., Bakhle et al. [25], Vahdati and Imregun [26]) are
adopted, it is well recognized that the computing efforts for solving an
aeroelastic problem in turbomachinery is dominated by that required for
solving the unsteady aerodynamics.
Efcient Unsteady Flow Modeling Approaches
The most straightforward way to solve an unsteady nonlinear multipassage
multirow ow eld is to integrate the spatially discretized ow equations in
time (time-domain solutions). This kind of full-scale time-domain unsteady
CFD runs are very time-consuming. The following brief estimate can give
some indication of typical CPU requirements for a time-domain unsteady
solution compared to a steady ow solution:
1. Time accuracy: for the same number of mesh points, a time-accurate
solution would typically require about 1520 times more CPU than a steady
solution for blade-row interaction problems (e.g., Graf et al. [27]). The
difference can be much more for low-frequency long wavelength (e.g., inlet
distortion driven low engine order forced response) problems.
2. Multiple-passage domain: to model proper circumferential length
scales would normally lead to use of a whole annulus computational
domain, resulting in another factor between 20 and 100, depending on
numbers of blade passages.
Hence, a full-scale unsteady time-domain calculation would be by two
to three orders of magnitude more time-consuming than a steady
calculation. There certainly is a need for developing efcient numerical
modeling techniques.
Single-Passage Domain Methods
The large requirement for computing time can be reduced if we can truncate
the computational domain in the axial direction to a single blade row, or in
the circumferential direction to a single blade passage.
First some comments should be made on the axial truncation. Singlerow domains have been widely used in conjunction with nonreective inlet
and exit boundary conditions. The problem is that the presence of adjacent
blade rows will always reect! It has been shown recently [28] that the
potential interaction effects between adjacent blade rows can signicantly
change unsteady ow responses within the blade row concerned. Realistic
semireecting boundary conditions have not been available. It seems that
coupled (nontruncated) multiple-row solutions may become the only way to
include the complex interaction effects. Nevertheless, single-row solutions

Copyright 2003 Marcel Dekker, Inc.

with nonreecting boundary conditions are still useful when compared with
well-established analytic solutions (only available for single-row congurations) or specially designed single-row experimental tests for method
validation purposes.
Regarding the circumferential domain truncation, for both blade
utter and rotorstator interaction problems, adjacent blades in one blade
row would usually experience unsteadiness with a constant phase difference
(interblade phase angle). Therefore, if a single-blade passage computation
domain is taken (Fig. 18), a ow variable at the upper periodic boundary UU
and that at the lower periodic boundary UL will have to satisfy the phaseshifted (or time-lagged) periodic condition:
UL x; t UU x; t
Dt

30

Where
Dt s=o is the time lag of the ow variable at the lower boundary
compared to that at the upper boundary.
o is blade vibration (or blade passing) angular frequency.
s is the interblade phase angle [Eqs. (6) and (7)].

Figure 18 Single-passage computational domain with the phase-shifted periodicity

(s interblade phase angle).

Copyright 2003 Marcel Dekker, Inc.

Several methods have been developed to deal with the phase-shifted

periodicity. The rst method has been proposed by Erdos et al. [29], known
as the direct store. In this method, ow variables at the periodic
boundaries are stored for one period of time. Then the stored parameters
and the current solution correct each other according to the phase-shifted
periodicity. An apparent disadvantage of the direct store method is that a
large amount of computer storage is required. The lower the frequency, the
more the storage. For a low-frequency case the storage required for the
periodic boundary condition can be 10 times more than that for the rest of
the computation. In a 3D ow situation, this restriction on computer
memory can be signicant.
Giles [30] has proposed a space-time transformation (time-inclination)
method to implement the phase-shifted periodic condition. In his method
the time plane in computational domain is inclined along the blade pitchwise
direction according to a given interblade phase angle. The phase-shifted
periodic condition can then be directly applied by equating ow variables at
the periodic boundaries on the transformed computational plane. Therefore,
no extra storage is required, and the convergence rate is much faster because
the solution procedure is less inuenced by the initial guess, compared to the
direct store method. However, the time-inclination angles (and therefore
allowed interblade phase angles) are very severely restricted by the
characteristics of the governing equations.
As described earlier, nonlinearity of a periodic ow manifests in terms
of the difference between the time-averaged ow and a pure steady one. It
also leads to higher harmonics generated by products of lower ones [Eq.
(21)]. In general, a nonlinear periodic ow variable for a known
fundamental frequency can be expressed by the Fourier series with its
zeroth order corresponding to a time-averaged value. Thus the Fourier
series can be used to implement the phase-shift periodicity at single-passage
domain boundaries, even when the ow unsteadiness is nonlinear. This is
the basis for a Fourier series-based method, called shape correction
[21, 31] for single-passage domain calculations. At the lower and upper
periodic boundaries (Fig. 18), we write an Nth-order timewise Fourier series
for the ow variables as
UL x; t Ux

N h
i
X
An x sinnot Bn x cosnot

31a

n1

UU x; t Ux

N h
i
X
An x sinnot s Bn x cosnot s

31b

n1

where Ux is the time-averaged ow part. Then the stored Fourier

components and the current solution can correct each other accordingly.

Copyright 2003 Marcel Dekker, Inc.

For most practical situations of interest, rst few harmonics (typically less
than 5) have been shown to be sufcient.
A major advantage of expressing unsteadiness in a Fourier series is its
ability to include multiple disturbances with unrelated frequencies. If the
total number of unsteady disturbances of interest is Nd , a ow variable at
periodic boundaries is expressed in the following general form:
Ux; t Ux

Nd
X

Ui x; t

32

i1

Each disturbance Ui is identied by its fundamental and higher harmonics

in a Fourier series as in Eq. (31). The capability of including multiple
disturbances has been demonstrated for a transonic fan rotor subject to inlet
distortion and blade oscillation [19].
Frequency-Domain Modeling
Frequency domain methods have been well established for unsteady ow
calculations, mainly for blade aeroelasticity applications, e.g., Verdon and
Caspar [32], Hall et al. [33]. The methodology of this kind starts with
decomposing a ow variable U into a steady part and an unsteady part:
Ux; t U0 x U 0 x; t

33

And the unsteady part is assumed to be very small so that nonlinear effects
(product terms of the unsteady perturbations) can be neglected. It follows
then that the time-averaged ow is the same as the steady one. On the basis
of a steady ow, a linear equation can be formed with respect to the
unsteady perturbation. The linear unsteady perturbation is normally cast in
a complex harmonic form, to remove the time dependence, resulting in the
perturbation equation with respect to the complex amplitude. The real and
imaginary parts of the complex amplitude determine the phase and
amplitude of unsteadiness. A solution to the complex harmonic amplitude
is obtained at a given frequency, thus is in frequency domain. Compared
to unsteady time-domain (time-marching) methods, frequency-domain
methods are much more efcient. A frequency-domain unsteady calculation
can be regarded as being equivalent to solving two steady ow problems
since the amplitude is not time-dependent, and can be conveniently solved in
a single-passage domain. The key limitation of the conventional frequencydomain methods is the linear assumption about unsteadiness.
To include nonlinear effects, a new methodology, called nonlinear
harmonic method, has been proposed [2] and developed for analysis of blade
aeroelasticity [34], [35] and blade-row interactions [6]. Here, a simple but

Copyright 2003 Marcel Dekker, Inc.

signicant difference is introduced to the ow decomposition:

Ux; t Ux U 0 x; t

34

where Ux is a time-averaged variable, instead of a steady one as dened in

Eq. (33). For a general case with Np disturbances, each disturbance at a
ek and U
e
k ,
frequency ok can be expressed by a pair of complex conjugates, U
so:
0

U x; t

Np h
X

e
k xe
iok t
ek xeiok t U
U

35

k1

The ow decomposition as dened in Eq. (34) leads to two sets of equations,

one for the time-averaged ow, the other for the unsteady perturbations.
The set of Np unsteady perturbation equations are obtained by balancing
harmonic terms for each frequency.
Following the earlier discussion, we know that the time-averaged
equations will contain extra deterministic stress terms, which depend on the
unsteady solution. On the other hand, the unsteady perturbation equations
cannot be solved without a known time-averaged ow. This interdependence between the two sets of equations reects the physical interaction
between the two parts of ow and is modeled by a simultaneous coupling in
the solution process of the nonlinear harmonic method, as shown in Fig. 19.
The nonlinear harmonic method offers a signicant gain in computational efciency, since neither a time-accurate integration, nor a multi-

Figure 19 Strongly coupled solution between two sets of decomposed ow

equations (n number of pseudotime iterations).

Copyright 2003 Marcel Dekker, Inc.

passage domain is required. The method is currently extended to a

multistage environment to analyze rotorstator and rotorrotor (stator
stator) interaction (aperiodic and clocking) effects on both aerothermal and
aeromechanical performances. The whole annulus unsteady ow eld is
reconstructed from the single-passage result according to the phase-shifted
periodicity. By introducing the nonrotating spatial harmonics, representing
upstream wakes, the aperiodic disturbances due to different rotorrotor
blade counts as well as the clocking effects are efciently included in a
single-passage solution.
Some further comments need to be made on this kind of unsteady ow
modeling based on ow decomposition, which is seemingly more complex
(undesirable!) than the straightforward time-domain unsteady CFD
methods. First, the unsteady disturbances we are dealing with are generally
small in magnitudes, though their effects may not be. To ensure small
unsteadiness to be adequately resolved, numerical errors (articial dissipation and dispersion) need to be much smaller compared to the physical
signals to give a reasonable signal-to-noise ratio. For instance, in the eld
of computational aeroacoustics, it has been evident that conventional
unsteady CFD algorithms with second-order temporal and spatial
discretizations are unable to adequately extract small but meaningful
physical acoustical disturbances from a background ow. In the context of
unsteady turbomachinery ows, it wont be difcult to nd situations where
physical unsteady disturbances are easily washed off by numerical
dissipations. Since magnitudes of numerical errors are normally in
proportion to those of ow variables, it would be more accurate to directly
solve the unsteady perturbations. Second, by decomposing the timeaveraged ow and the unsteady part, we can use the same numerical
algorithms and boundary condition treatments for both a time-averaged
ow and a pure steady ow. The past experiences suggest that usefulness of
comparing different unsteady CFD computations depends measurably on
the consistency of the ways these computations are carried out. Hence use of
the same numerics in both steady and time-averaged calculations should
help to identify and understand true unsteady effects.
Some Comments on Turbulence/Transition Modeling
It has been a standard practice that turbulence models tuned for steady
ows are used quasisteadily for unsteady ows. The uncertainties associated
with this must be recognized. At the same time, it is felt that this should not
always be used as an explanation (or excuse!) for poor agreements between
computational and experimental results. For instance, one may nd the
turbulence/transition modeling will strongly affect blade aerodynamic

Copyright 2003 Marcel Dekker, Inc.

losses, but it may affect to a minimal extent pressure distributions for

attached boundary layers. Figure 20 shows measured surface shear-stress
and corresponding unsteady pressures for an oscillating NACA-65 airfoil in
a low-speed wind tunnel [22]. The shear stress data clearly indicate that the
transition point periodically moves over a very large range on the airfoil
surface during one oscillating period [Fig. 20(a)], but the unsteady pressure
over the major portion of the surface appears to be almost unaffected [Fig.
20(b)]. So for this case, an accurate modeling of the large-scale movement of
the transition point (most likely to be nonlinear) should make a negligible
difference in an aerodynamic loading (aerodamping prediction).
A more complex situation where turbulence/transition modeling can
be important is with a bubble-type boundary-layer separation. Bubble-type
separations are common on both compressor and turbine blades at typical
Reynolds numbers. Depending on the bubble size, this type of ow
separation can affect local as well as global unsteady pressures. A typical
scenario is that a boundary layer separates at a laminar state, and a
transition happens in the separated shear layer. The reattachment point of
separation zone and its movement will strongly depend on the transition and
its modeling. For aerodynamic damping calculations involving blade
vibration at a small amplitude, a simple bubble separation/transition model
developed for steady low-speed ows [36] can be implemented, assuming
that an unsteady laminar separation bubble behaves quasisteadily [37]. In
transonic ows, the passage shock position at a highly loaded condition is
very sensitive to small blockage variations. In these cases, unsteady
transition would certainly inuence the unsteady shock boundary-layer
interaction and the shock oscillation. The associated aerodamping can be
thus dependent on the transition modeling [38].
For typical off-design conditions with much thickened and/or
separated boundary layers, it is probably fair to say that none of the
existing turbulence/transition models is shown to be reliable. Given that
viscous effects at off-design conditions are likely to be inuenced by large
turbulence eddy structures, which in turn interact with short-scale periodic
unsteadiness, large eddy simulation (LES) looks to be the way forward. The
need to pursue LES can also be viewed from a slightly different angle,
relating to mesh dependence of unsteady solutions. Generally it is more
difcult to get a mesh-independent solution for unsteady ows than for
steady ows. This is because that when we rene meshes (which would
normally be associated with using smaller time steps), we will pick up
disturbances at smaller scales. And this would be a neverending process
(imagine that at some stage down the line, we might start to pick up largescale random turbulence!). This fundamental uncertainty of modeling
turbulence while resolving it at the same time needs to be clearly appreciated

Copyright 2003 Marcel Dekker, Inc.

Figure 20 Unsteady shear stress and pressure on suction surface of NACA-65

airfoil (torsion oscillation at a mean incidence of 2.58).

Copyright 2003 Marcel Dekker, Inc.

when we rene a mesh in an unsteady calculation. Again, the issue can be

consistently addressed in the framework of LES. It is anticipated that use of
ner and ner temporal and special resolutions for small deterministic
unsteadiness would lead to earlier rather than later realization of LES in
practical turbomachinery applications.

CONCLUDING REMARKS
On aerothermal performances, efforts have been (and continue to be) made
to enhance understanding and predictability of blade-row (wake and
potential) interactions. Generally speaking, unsteady effects tend to be more
signicant at off-design (or bad design!) conditions than at conventional
design conditions where unsteady losses tend to be relatively small.
However, future design conditions may be much more prone to
unsteadiness due to higher loading and/or other nonstandard requirements.
Aerodynamics-related vibration (utter and forced response) problems
directly affect reliability, cost, and life span of turbomachinery blades. This
is likely to be a major area where blading design procedures will further
evolve in the future with ever-closer coupling and more concurrent
integration between aerodynamics and structural dynamics.
Basic understanding of unsteady effects and appreciation of unsteady
ow modeling issues can be gained by examining how an unsteady
disturbance affects a time-averaged ow eld through nonlinearity in the
momentum and energy equations. The same principle applies to both
random turbulence uctuations and periodic unsteadiness.
With development of computer hardware, it is now possible to carry
out multipassage nonlinear time-domain calculations. While this kind of
full-scale unsteady CFD simulations will be increasingly applied, we should
ensure that relevant unsteady ow information is adequately and
consistently extracted from the results to enhance understanding and to
help design processes. It is expected that various truncated or reduced
computational models aimed at efciently including relevant unsteady ow
physics will continue to be developed, validated, and implemented in design
environments.

ACKNOWLEDGEMENTS
The author wishes to thank Stuart Moffatt (School of Engineering,
University of Durham) for his help to improve the presentation.

Copyright 2003 Marcel Dekker, Inc.

Figures 1, 2, and 3 are originally published in L. He, II. TimeMarching Calculations for Blade Row Interaction and Flutter, VKI
Lecture Series, Unsteady Flows in Turbomachines, by von Karman Institute
for Fluid Dynamics, March 1996, reprinted with permission.
Figures 7, 8, 10, and 11 were originally published in L. He,
Computational Study of Rotating Stall Inception in Axial Compressors
J. Power and Propulsion, 13(1), copyright# 1997 by American Institute of
Aeronautics and Astronautics, reprinted with permission.
Figures 12 and 13 are reproduced from L. He, Proc IMechE, Part A
Issue A3 211(1997) titled Computational Study of Unsteady Flows in
Steam Turbine Blade Rows at Partial Admission by permission of the
Council of the Institution of Mechanical Engineers.
Figures 16, 17, and 20 are reproduced from the ASME Paper 91-GT181 An Experiment on Unsteady Flow over Oscillating Airfoil by L. He
and J. D. Denton with permission from the American Society of Mechanical
Engineers.

REFERENCES
1. S. R. Manwaring and K. L. Kirkeng, Forced Response Vibrations of a Low
Pressure Turbine Due to Circumferential Temperature Distortions, Proc of
8th International Symposium on Unsteady Aerodynamics and Aeroelasticity of
Turbomachines, Stockholm, Sweden, Sept. (1997).
2. L. He, I. Modelling Issues for Computations of Unsteady Turbo-machinery
Flows; II Time-marching Calculations for Blade Row Interaction and Flutter,
VKI Lecture Series, Unsteady Flows in Turbomachines, von Karman Institute
for Fluid Dynamics, Belgium, (1996).
3. J. L. Kerrebrock and A. A. Mikolajczak, Intra-Stator Transport of Rotor
Wakes and Its Effect on Compressor Performance, ASME J. of Engineering
for Gas Turbine and Power, pp. 359370, Oct. (1970).
4. L. H. Smith, Wake Dispersion in Turbomachines, ASME J. of Basic
Engineering 88: 688690 (1966).
5. J. J. Adamczyk, Wake Mixing in Axial-Flow Compressors, ASME Paper 96GT-29 (1996).
6. L. He, Three-Dimensional Unsteady NavierStokes Analysis of Stator-Rotor
Interaction in Axial-Flow Turbines, IMechE J. of Power and Energy, 214: 13
22 (2000).
7. T. Chen, P. Vasanthakumar, and L. He, Analysis of Unsteady Blade Row
Interaction Using Nonlinear Harmonic Approach, AIAA J. Power and
Propulsion, 17(3), May (2001).

Copyright 2003 Marcel Dekker, Inc.

8. W. S. Barankiewicz and M. D. Hathaway. Effects of Stator Indexing on

Performance in a Low Speed Multi-Stage Axial Compressor, ASME Paper
97-GT-496 (1997).
9. J. D. Denton. Loss Mechanisms in Turbomachines, IGTI Scholar Lecture,
ASME Paper 93-GT-435 (1993).
10. R. Kielb, Forced Response, VKI Lecture Series, Aeroelasticity in Axial Flow
Turbomachines, von Karman Institute for Fluid Dynamics, May (1999).
11. W. Ning and L. He. Some Modelling Issues on Trailing-Edge Vortex
Shedding, AIAA J. 39(5) (2001).
12. D. L. Sondak and D. J. Dorney. Simulation of Vortex Shedding in a Turbine
Stage, ASME J. Turbomachinery, 121:(3) 428435 (1999).
13. T. J. Bogar, M. Sajben, and J. C. Kroutil, Characteristic Frequencies of
Transonic Diffuser Flow Oscillations, AIAA J. 21(8) (1983).
14. I. J. Day, Stall Inception in Axial Flow Compressors. ASME J.
Turbomachinery, 115: 19, Jan. (1993).
15. L. He, Computational Study of Rotating Stall Inception in Axial Compressors, AIAA J. Power and Propulsion, 13 (1) (1997).
16. L. He and J. O. Ismael, Computation of Blade Row Stall Inception in
Transonic Flows, Aeronautical J. 103 (1025): 317324 (1999).
17. F. O. Carta, Unsteady Aerodynamics and Gapwise Periodicty of Oscillating
Airfoils, ASME J. Engineering for Power, 105(3) (1983).
18. L. He, Computation of Unsteady Flows in Steam Turbine Blade Rows at
Partial Admission, IMech.E J. Power and Energy, 211: 197205 (1997).
19. H. D. Li and L. He, Prediction of Flutter and Inlet Distortion Driven
Response of a Transonic Fan Rotor Using Phase-Lagged Boundary Conditions, ASME Paper 2001-GT-0272 (2001).
20. J. J. Adamczyk, Model Equations for Simulating Flows in Multistage
Turbomachinery, ASME Paper 85-GT-226 (1985).
21. L. He, An Euler Solution for Unsteady Flows Around Oscillating Blades,
ASME J. Turbomachinery, 112(4): 714722 (1990).
22. L. He and J. D. Denton, An Experiment on Unsteady Flow over Oscillating
Airfoil, ASME Paper 91-GT-181 (1991).
23. F. O. Carta, Coupled Blade-Disk-Srooud Flutter Instabilities in Turbojet
Engine Rotors, ASME J. Engineering for Power, 89: 419427, July (1967).
24. J. S. Green and J. G. Marshall, Forced Response Prediction within the Design
Process, Proc. of 3rd European Conference on Turbomachinery, Paper C557/
076/99, London, March (1999).
25. M. A. Bakhle, R. Srivastava, G. L. Stefko, and J. M. Janus, Development of
an Aeroelastic Code Based on an Euler/Navier-Stokes Aerodynamic Solver,
ASME Paper 96-GT-311 (1996).
26. M. Vahdati and M. Imregun. Non-linear Aeroelastic Analysis using
Unstructured Dynamic Meshes, Proc. of the 7th International Symposium on
Unsteady Aerodynamics and Aeroelasticity in Turbomachines, Fukuoka, Japan,
Sept. (1994).

Copyright 2003 Marcel Dekker, Inc.

27. M. B. Graf, E. M. Greitzer, F. E. Marble, and O. P. Sharma, Effects of Stator

Pressure Field on Upstream Rotor Performance, ASME Paper 99-GT-99
(1999).
28. P. D. Silowski and K. C. Hall, A Coupled Mode Analysis of Unsteady MultiStage Flows in Turbomachinery, ASME J. Turbomachinery, 120(3) (1998).
29. J. I. Erdos, E. Alzner, and W. McNally, Numerical Solution of Periodic
Transonic Flow Through a Fan Stage, AIAA J. 15(11) (1977).
30. M. B. Giles, Calculation of Unsteady Wake Rotor Interaction, AIAA J.
Propulsion and Power, 4(4): 356362 (1988).
31. L. He, A Method of Simulating Unsteady Turbomachinery Flows with
Multiple Perturbations, AIAA J. 30(12) (1992).
32. J. M. Verdon and J. R. Caspar. A Linearized Unsteady Aerodynamic Analysis
for Transonic Cascades, J. Fluid Mechanics, 149: 403429 (1983).
33. K. C. Hall, W. S. Clark, and C. B. Lorence, A Linearized Euler Analysis of
Unsteady Transonic Flows in Turbomachinery, ASME J. Turbomachinery,
116(3) (1994).
34. W. Ning and L. He, Computation of Unsteady Flows around Oscillating
Blades using Linear and Nonlinear Harmonic Euler Methods, ASME J.
Turbomachinery, 120(3): 508514 (1998).
35. L. He and W. Ning, Efcient Approach for Analysis of Unsteady Viscous
Flows in Turbomachines, AIAA J. 36(11): 20052012 (1998).
36. H. P. Horton, A Semi-Empirical Theory for Growth and Bursting of Laminar
Separation Bubble, A.R.C. Report, CP-1073 (1969).
37. L. He, Unsteady Flows in Oscillating Turbine Cascade, Part 1. Linear
Cascade Experiment; Part 2. Computational Study, ASME J. Turbomachinery, 120(2): 262275 (1998).
38. K. Isomura and M. B. Giles, A Numiercal Study of Flutter in a Transonic
Fan, ASME J. Turbomachinery, 120(3) (1998).

Copyright 2003 Marcel Dekker, Inc.