Proceedings of IGTI’04

2004 ASME International Gas Turbine Institute

Vienna, Austria, June 14-17, 2004

GT2004-53096

WHIRL FREQUENCY DEPENDENT ALFORD FORCES IN AXIAL COMPRESSORS

Claus M. Myllerup Graeme Keith

Ødegaard & Danneskiold-Samsøe A/S Ødegaard & Danneskiold-Samsøe A/S

Titangade 15 Titangade 15
DK-2200 Copenhagen, Denmark DK-2200 Copenhagen, Denmark
Email: [email protected] Email: [email protected]

ABSTRACT s Dimensionless eigenvalue.

Explicit closed form expressions are derived for the whirl vz Axial flow velocity.
frequency dependence of the Alford force in an axial compres- A Amplitude of flow disturbance.
sor operating in steady-state away from the onset of rotating AC Projected compressor
area, 2πRh.
stall. The analysis includes the compressor flow dynamics us- B Greitzer’s B-factor, aUs ACV
Peff .
ing the Moore-Greitzer approximation. By asymptotic expansion
Fτ , F p , Fu Reaction force to torque, pressure and unsteady flow.
in terms of the whirl orbit amplitude, expressions for the direct
H,W Pressure rise characteristic parameters.
and cross-coupling impedance are obtained analytically. Sev-
LC Compressor length.
eral components in the cross-coupling impedance are shown to
Leff Effective compressor length, LI + Rλ + LE .
change phase as the whirl frequency transits the rotating stall fre-
LI , LE , LT Length of inlet duct, exit duct and throttle.
quency. This implies that for a given compressor design the Al-
LZ Blade chord and blade axial chord.
ford force can be stabilizing or destabilizing depending on flow
N Number of compressor stages.
rate and whirl frequency. Further, the analysis shows that the
R Compressor mean radius.
coupling between mean flow quantities and rotor whirl is of sec-
R Gas constant.
ond order. Thereby, the present asymptotic expansion results in
TP Absolute gas temperature in plenum.
a fluid-structure coupling model, which is consistent in accuracy
U Compressor tip speed, RΩ.
with usual linearized rotordynamic analysis.
β Alford’s cross-coupling parameter.
βr1 , βr2 Relative flow angle at inlet and outlet of rotor stage.
γ Ratio of specific heats.
NOMENCLATURE γT Throttle coefficient.
√
as Isentropic speed of sound, γRTP . ε Dimensionless whirl amplitude, e/C.
c,C Local and mean compressor tip clearance. η Phase angle for flow disturbance.
e Circular whirl amplitude. λ Greitzer’s dimensionless time lag, NτU/R.
h Blade radial span (height). ρ Density.
h0 Enthalpy. τC Compressor torque.
k, K Cross-coupling and direct stiffness. τ Greitzer time lag.
m Exit duct diffusion parameter. φC Compressor axial flow coefficient.
ṁC , ṁT Compressor and throttle mass flow. φC1 , ψC1 Flow and head in steady-state test point.
p Pressure. ϕ Potential function.

1 Copyright 
c 2004 by ASME
ψC Total to static compressor head characteristic. Storace et al. [5] and Ehrich et al. [6] made a significant con-
ψC0 , ψCM Head at zero flow and maximum head at flow 2W . tribution to state-of-the-art by comparing experiment with three
ψε Sensitivity of ψC to tip clearance. different theories to predict β, namely the Two-Sector Parallel
ωε Whirl frequency. Compressor model of Ehrich [7], the Infinite-Segment-Parallel-
ΦC , ΦT Spatial mean compressor and throttle flow coefficient. Compressor model of Spakovszky [8] and the Two-Coupled Ac-
Ψ Plenum pressure coefficient. tuator Disc model of Song and Cho [9]. A fundamental finding
Ω Spin frequency. was that the sign and amplitude of β depends on the whirl fre-

(·) Dimensionless length, L(·) /R. quency and the axial flow coefficients (i.e., the proximity of onset
δ(·) Circumferentially varying disturbance. of rotating stall).
(·)(0) , (·)(1) Leading and first order expansion terms. By postulating a similarity between the flow non-uniformity
(·) Dimensionless parameter. induced by a rotating stall cell and a rotor eccentricity, Al-Nahwi
 Time averaged.
(·) et al. [10] – [11] adapted the four state “lag” model of Greitzer
(·) Space averaged over cross-section. [12] and Moore and Greitzer [13] to rotor whirl. This highly
inspired use of the Moore-Greitzer post-stall theory was intended
to “present a first principles-based model of the fluid-induced
INTRODUCTION forces acting on the rotor of an axial compressor.” The resulting
Rotordynamic stability of axial compressors is paramount model will be referred to as the Four-State Whirl (FSW) model.
for safe and reliable operation, and must be ensured by proper Al-Nahwi et al. [10] – [11] gave a thorough qualita-
control or mitigation of the destabilizing forces. The need for tive discussion of the FSW model as well as some validation
fundamental understanding and quantification of all significant against experimental data, and good agreement with the work of
destabilizing forces in axial compressors is therefore obvious, Spakovszky [8], Song and Cho [9], Storace et al. [5] and Ehrich
whether for gas compression in process plants or gas generation et al. [6] was found. However, the rotordynamic application was
in gas turbines. restricted to the stability analysis of the coupled fluid-structure
Ehrich and Childs [1] gives a review of potential rotordy- problem for an ideal Jeffcot rotor. While the analysis was ade-
namic instability mechanisms. For rotor vibration consisting of a quate to identify the governing parameters, it is not readily ap-
circular concentric whirl orbit of amplitude e and frequency ωε , plied to the rotordynamic analysis of real turbomachinery with
the reaction forces from a flow field about a rotor can in a linear axial compressors.
approximation be written as By asymptotic expansion of the FSW model in terms of the
rotor eccentricity, a linearized expression for the fluid-structure
   
K k cos (ωεt) interaction forces of the form in Eq. (1) can be obtained, which
F= ·e (1)
−k K sin (ωεt) is readily implemented in commercial rotordynamic programs.

where K represents a centering stiffness and k a cross-coupling

stiffness. The latter will act stabilizing or destabilizing on a given AXIAL COMPRESSOR MODELLING
mode depending on its whirl direction. For modes with forward The axial compressor model is shown in Figure 1 and con-
whirl (i.e., whirl vector and rotor spin vectors have the same ori- sists of an inlet duct, a compressor stage of mean radius R, blade
entation) the cross-coupling will be destabilizing when k > 0. radial span h and mean tip clearance C, an exit duct, a plenum and
Among the flow induced rotor reaction forces is the well- a throttle. The compressor operates in steady-state at a constant
known tip clearance induced cross-coupling in axial turbines first spin velocity Ω. Figure 1 also illustrates the applied coordinate
proposed by Thomas [2] and later extended to axial compressors system.
by Alford [3], which predicts k = βτL /(Rh) and K = 0 in Eq. (1).
Paradoxically, this cross-coupling mechanism is known by most
turbomachinery engineers as the Alford force, yet for decades the Greitzer Axial Compressor Lag Model
experimental evidence from axial compressors was inconsistent Rotating stall, just as rotor whirl, induces a non-
even as to the sign of β. axisymmetric flow disturbance. At the compressor inlet the axial
Vance and Laudadio [4], the first published experimental flow velocity therefore has spatial mean value vz , and a super-
measurement of the Alford force, was restricted to static eccen- imposed circumferentially varying disturbance δvz (θ). This im-
tricity (i.e., ωε = 0). The existence of a positive cross-coupling plies that the axial flow coefficients can be written as
stiffness k was demonstrated, but it was reported that its magni-
tude was speed dependent and that it did not appear to be propor-      
tional to stage torque below a certain torque level. φC θ,t  = ΦC t  + δφC θ,t  (2)

2 Copyright 
c 2004 by ASME
 dynamics would be to write for the stator

∆p   
= F φC θ,t  − τ
2 ρU stator
1 2

      ∂φC   
≈ F φC θ,t  − τ F  φC θ,t  θ,t (6)
∂t 

and for the rotor

∆p        
= F φC θ t  − τ ,t  − τ ≈ F φC θ,t 
2 ρU rotor
Figure 1. Axial compressor section, adapted from Al-Nahwi et al. [10]. 1 2
 
 
  
 ∂φC    ∂φC   
− τ F φC θ,t θ,t + θ,t (7)
∂t  ∂θ
where UΦC (t  )
= vz |z=0 and UδφC (θ,t  )
= δvz |z=0 using the
nondimensional time scale, t  = tΩ, and the velocity scale, U =
showing that τU/R = τ F  (φC ). Thus, the Moore and Greitzer
RΩ. Neglecting density variations, the total inlet mass flow rate
“time lag” model formally neglects a flow dependence, but since
is therefore ṁC = ρACUΦC .
the quantification of τ is empirical, the consequence should be
Greitzer [12] observed that for axial compressors the clas-
limited. It was beyond the present scope to investigate any fur-
sical linearized surge model had a significant shortcoming in its
ther the effect of F  (φC ) in the FSW model.
assumption of the flow dynamics in the compressor. As a remedy
For N stages, each consisting of a rotor and stator pair, the
it was proposed to include the lags in the compressor by means
total pressure rise across the compressor therefore becomes
of the following expression for a single blade row
 
∆p ∂φC ∂φC
∆p = F (φC ) − λ 2 + (8)
= F (φC ) − τ
dφC
(3) 2 ρU
1 2 ∂t  ∂θ
2 ρU
1 2 dt
row
where
where F (φC ) is the axisymmetric steady-state head rise charac-
teristic and the “time lag” τ is a measure of the fluid inertia in NτU
λ= (9)
the passage. The underlying assumption is that the hub-to-tip ra- R
dius ratio is sufficiently high to warrant a two-dimensional flow
description. Compressor Precession
Moore [14] pointed out that the unsteadiness due to the rotor The flow in an axial compressor stage will be affected by
blades moving through a circumferentially nonuniform flow had rotor whirl (e.g., lateral shaft vibrations).
to be accounted for, such that the lag term in Eq. (3) would differ For a rotor whirl consisting of a precession about the casing
for rotor and stator blade rows centerline in a circular whirl orbit of radius e = Cε and whirl fre-
quency ωε = ωε Ω, the instantaneous blade tip clearance c (θ,t  )
dφC U ∂φC is
= (4)
dt stator R ∂t      
c θ,t  = C 1 − ε cos θ − ωεt  (10)

  It was early recognized by Thomas [2] and Alford [3] that

dφC U ∂φC ∂φC the local variation in tip clearance implies a variation in the local
= + (5)
dt rotor R ∂t  ∂θ load coefficient. This is best represented by restating the steady-
state static-to-static compressor head rise F (φC ) in terms of the
We have retained Greitzer’s and Moore’s notation to main- total-to-static compressor head characteristic ψC and the suction
tain consistency with the previous literature. However, it should flow coefficient φC
be noted that the “time lag” τ is not an independent system prop-
erty. A more consistent way to model the time lag in the flow F (φC ) = ψC (φC , ε) + φC2 (11)

3 Copyright 
c 2004 by ASME
 Integrating the radially symmetric Laplace equation (14)
over the cross section of the inlet gives the following equation
for ϕ (z ,t  )

∂2     
ϕ z ,t = 0 (16)
∂z 2

which, when Eq. (15) is substituted into equation (14), gives

directly that δϕ also satisfies Laplace’s equation,

 
∇2 δϕ θ, z ,t  = 0 (17)

Figure 2. Lateral motion of rotor disk.

We can write the flow coefficient at the compressor inlet in
terms of the non-dimensional mean and spatially varying poten-
tial functions as follows

Introducing the sensitivity of the load coefficient to tip clearance   ∂ϕ ∂ϕ ∂δϕ
for a blade as φC θ,t  =  = + (18)
∂z I ∂z z =0 ∂z z =0

∂ψC ΦC (t  ) δφC (θ,t  )

ψε = −C (12)
∂c ε=0
This equation furnishes a boundary condition for ϕ at the com-
Fully in the spirit of Thomas’ and Alford’s intuitive approach, pressor inlet, which together with the condition that ϕ = 0 at the
Al-Nahwi et al. [10] proposed a linear approximation of the load start point A of the constant area inlet duct at z = −
I , allows us
coefficient in terms of the whirl amplitude to integrate Eq. (16) to find

   
ψC = ψC |ε=0 + εψε cos θ − ωεt  (13) ϕ = z +
I ΦC . (19)

where ψC |ε=0 corresponds to the compressor characteristic orig- Going back to dimensional notation we can write the un-
inally used by Moore and Greitzer [13]. steady, frictionless and incompressible momentum balance over
the inlet duct by means of Bernouilli’s equation between points
A and I
Compressor Inlet Duct
The flow in the inlet duct is incompressible, inviscid and ir- ∂ϕ 1  2 ∂ϕ
rotational. This implies that the velocity components can be ex- ptA + ρ = pI + ρ vz |z=0 + (20)
∂t 2 ∂t
pressed in terms of a potential function ϕ satisfying the Laplace A I
total inlet pressure
equation

Introducing the dimensionless notation, this translates into

∇2 ϕ(θ, z,t) = 0 (14)

ptA ∂ϕ pI ∂ϕ

A nondimensional axial coordinate z = z/R and a nondimen- + = + φ +
2
2 2 (21)
2 ρU
1 2 ∂t  2 ρU
1 2 C
∂t 
sional potential function ϕ = ϕ/RU are introduced and the po- A I

tential function is split into a spatial mean ϕ and a circumferen-

tially varying disturbance δϕ in accordance with the flow coef- From Eqs. (15) and (19) we find at z = 0
ficient.

      ∂ϕ dΦC ∂δϕ

         =
I + (22)
ϕ θ, z ,t = ϕ z ,t + δϕ θ, z ,t (15) ∂t  I dt  ∂t  I

4 Copyright 
c 2004 by ASME
The flow being steady at point A, we have in addition ∂ϕ /∂t  |A = −k in the inlet duct and k in the exit duct (i.e., the z derivative
0. Assuming finally that the compressor inlet guide vanes are of ϕ should give positive vz and ΦC on either side of the com-
lossless (i.e., ∂δφC /∂θ = 0), Eq. (21) becomes pressor despite a decreasing value of ϕ when moving away from
the compressor). Using therefore Eqs. (15) and (19) with re-
  versed sign for δϕ and integrating once with respect to z while
ptA − pI dΦC ∂δϕ
= φC + 2
I  +
2
(23) imposing pressure continuity at z =
E gives
2 ρU
1 2 dt ∂t  I

Actuator Disk Model for Compressor pP − p ∂      

   

= 2 
z −
E ΦC t − δϕ θ, z ,t + δϕ
The compressor model employed is based on Eq. (8), such 2 ρU
1 2 ∂t P

that the pressure rise across the compressor includes a steady- (28)
state, axisymmetric contribution as well as an unsteady contribu- Since the exit channel discharges into the plenum where the cir-
tion due to fluid inertia within the rotor and stator rows: cumferential flow nonuniformity vanishes, we impose δϕ |P = 0,
and applying Eq. (25) it therefore results at the exit of the com-
  pressor, z = 0
pE − pI ∂φC ∂φC
= ψC (φC , ε) + φC − λ 2  +
2

2 ρU
1 2 ∂t ∂θ
     
dΦC ∂ δϕ 2 ∂2 δϕ pP − pE dΦC ∂δϕ
= ψC (φC , ε) + φC2 −λ 2 + + = −2
E  + (29)
dt  ∂z ∂t  I ∂z ∂θ I 2 ρU
1 2 dt ∂t  I
steady − state
static to static
pressure rise Moore [14] and Moore and Greitzer [13] introduced the duct
(24) flow parameter m to describe the degree of diffusion in the exit
duct (i.e., m = 1 for sudden expansion, m = 2 for constant area
In the spirit of the actuator disk model for the compressor, duct), and accordingly the pressure balance over the exit duct
the flow nonumiformity is assumed to pass through the compres- becomes
sor unchanged  
pP − pE dΦC ∂δϕ
= −2
E  + (m − 1) (30)
δφC |E = δφC |I (25) 2 ρU
1 2 dt ∂t  I

Compressor Exit Duct Overall Pressure Balance

The deviation of duct pressure p from the plenum pressure As illustrated in Figure 1 the axial compressor is assumed to
pP is assumed to be sufficiently small to satisfy the Laplace equa- discharge through a diffuser into a plenum where, as discussed
tion (i.e., ∇2 (pP − p) = 0). in the previous section, any spatial disturbance δφ is eliminated.
The axial velocity component in the exit duct flow satisfies The plenum pressure pP is referred to total atmospheric pressure
the Euler equation (i.e., the axial momentum equation for invis- ptA and dimensionalized as follows:
cid flow when neglecting convection)

∂p ∂vz (pP − ptA )

− =ρ (26) Ψ=
2 ρU
1 2
∂z ∂t
(pP − pE ) (pE − pI ) (pI − ptA )
= 1 2 + 1 2 + 1 2 (31)
The plenum pressure being independent of z, this can be written 2 ρU 2 ρU 2 ρU
in nondimensional form as
exit duct actuator disk inlet duct
 
∂ pP − p ∂vz
=2 (27) Substituting Eqs. (23), (24), and (30) we find
∂z 2 ρU
1 2 ∂t 

dΦC ∂δϕ ∂2 δϕ ∂2 δϕ

The potential problem in the exit duct is therefore the same Ψ = ψC − 2
eff − 2m − 2λ − λ
as in the inlet duct, but with a potential function −ϕ . The neg- dt  ∂t  I ∂z ∂t  I ∂z ∂θ I
ative sign is necessary because the normal to the compressor is (32)

5 Copyright 
c 2004 by ASME
Plenum and Throttle The presence of a flow nonuniformity implies that dτC varies
As in Greitzer [12] it is assumed that the compressor flow with θ. Each blade contributes with a lateral rotor reaction force
discharges into the plenum as a free jet. The balance between with cartesian components
the axial compressor flow ṁC into the plenum and the mean flow
ṁT through the throttle will induce an isentropic density fluctu-  
dτC − sin θ
ation in the plenum (i.e., d pP /pP = γdρP /ρP ). Using ideal gas dFτ = − (39)
R cos θ
properties, the continuity equation for the plenum is accordingly

and the net lateral rotor reaction to torque (i.e., the turning force)
VP d pP
ṁC − ṁT = (33) becomes
a2s dt
 2π  
1 − sin θ
The throttle flow coefficient is defined as ΦT = ṁT / (ρACU) and Fτ = − ρACU 2 N (tan βr1 − tan βr2 ) φC2 dθ
2π 0 cos θ
the Greitzers B-factor is
expressed using the effective length of (40)
the compressor, B = aUs ACV
Peff . This implies that Eq. (33) can The pressure being flow dependent, a pressure induced net
be written on nondimensional from as lateral rotor reaction will also arise from a flow nonuniformity

dΨ 2  2π  LC  
= 2 (ΦC − ΦT ) cos θ
B
eff
(34) Fp = − p Rdθdz
dt 0 0 sin θ
 2π  
1 cos θ
Since the pressure difference Ψ balances the throttle loss and ≈ −LC R (pI + pE ) dθ (41)
0 2 sin θ
any acceleration of the mass in the throttle duct

where it is convenient to write the mean compressor pressure as

dΦT
Ψ = FT (ΦT ) +
T (35) pI + pE = 2 (pI − ptA ) + (pE − pI ) + 2ptA . By using Equations
dt  (23) and (24), and making use of the fact that ptA is constant,
Equation (41) therefore implies
where Moore and Greitzer [13] propose a realistic throttle char-
acteristic as FT (ΦT ) = (ΦT /γT )2 . Assuming that the throttle   
duct is short,
T  1, we therefore obtain the throttle model as 1 LC 2π ∂2 δϕ
Fp ≈ − ρACU 2 N ψC − φC2 − 2λ
2π 4Nh 0 ∂z ∂t  I
√   
Φ T = γT Ψ (36) ∂2 δϕ ∂δϕ cos θ
−λ  −4 dθ
∂z ∂θ I ∂t  I sin θ
Rotor Reaction Forces (42)
The compressor torque per blade equals the change of angu-
lar momentum of the fluid over the rotor. Assuming that the axial Finally, a lateral reaction force arises from the unsteady mo-
velocity (UφC ) is constant (i.e., the blade height varies accord- mentum term
ing to the density variation), that the flow nonuniformity passes
 
unchanged through the compressor and that the N stages are sim- ∂ ∂
ilar, the blade enthalpy variation is Fu = − ρvabs dV ≈ − ρvrel dV (43)
∂t CV ∂t CV

UdτC 1 where the approximation relies on the fact that the difference be-
∆h0 (θ) = = U 2 φC N (tan βr1 − tan βr2 ) dθ (37)
Rd ṁC 2π tween absolute and relative unsteady fluid velocity is of O(
Z ).
Taking the average relative velocity in axial and radial directions
where the axial mass flow rate ”per blade” at the compressor over the rotor stages, the unsteady force can be expressed as
inlet, d ṁC , and the inlet and exit rotor flow angles, βr1 and βr1 ,
are functions of the blade flow φC . The total compressor torque  
1 LZ
is therefore Fu ≈ − ρACU 2 N ×
2π 2R
  
 2π ∂ 2π − sin θ
1 φ (tan β + tan β ) dθ (44)
τC = ρAC RU 2 N (tan βr1 − tan βr2 ) φC2 dθ (38) ∂t  0
C r1 r2
cos θ
2π 0

6 Copyright 
c 2004 by ASME
 Of the three rotor reaction forces it appears that the unsteady ψ
force Fu will be the smallest since the axial chord length is sig-
nificantly smaller than the compressor tip radius,
Z  1. The
axial length per rotor-stator pair LC /N being of similar size as ψ +2H
C0

the blade height h, it would a priori appear that the pressure force
F p and the turning force Fτ could be of equal significance.

APPROXIMATE SOLUTION
ψC0+H
Separation of Variables and Fourier Expansion
Eqs. (14) and (17) state that both the velocity potential in
the inlet duct z < 0 and the disturbance potential δϕ satisfy the
Laplace equation. We therefore write the solution as Fourier-
exponential series in θ and z respectively
ψC0

  φ
N
1        0 W 2W
δϕ θ, z ,t  = ∑ n enz An t  cos n θ − ηn t  (45)
n=1 Figure 3. Axial compressor performance map.

Applying this form of solution in Eq. (18) gives the compressor

flow disturbance as Another interpretation of the basic parameters W , H and ψC0
is achieved by assuming that compressor testing has been
  ∂δϕ N       possible in the point of maximum head rise (φCM , ψCM ) =
δφC θ,t  =
∂z
= ∑ An t cos n θ − ηn t  (46) (2W, ψC0 + 2H) as well as for a steady-state point (φC1 , ψC1 ) to
I n=1 the right φCM . With these known operating points it results

Substituting Eq. (46) into the overall pressure balance Eq. (32) 1
W = φCM
and making use of the orthogonality of the Fourier terms gives 2
the following set of ordinary differential equations ψC0 = ψCM − 2H
ψCM − ψC1 (49)
 H=    3
dΦC 1 2π 3 φC1 1 φC1
Ψ + 2
eff 
= ψC dθ 1− 2 −1 + 2 −1
dt 2π 0 2 φCM 2 φCM
m  dA 
n 1 2π
2 +λ = ψC cos (n (θ − ηn )) dθ (47)
n dt  π 0 A more well-founded approach would naturally be to deter-

dηn 1 2π mine the polynomial shape parameters ψC0 , H and W by least
2 (m + nλ) An  − nλAn = ψC sin (n (θ − ηn )) dθ
dt π 0 square fit to a larger set of empirical performance data. One
could even consider increasing the power of the polynomial, but
this requires carrying the additional terms through in the follow-
Polynomial Approximation
ing derivations.
A further fundamental approximation in the performance
model of Moore and Greitzer [13] was to assume a polynomial
expansion of ψC |ε=0 . Single Term Fourier Series
The steady-state compressor characteristic will typically As proposed by Moore and Greitzer [13], only the first term
have two local extremes, a minimum at (φC , ψC ) = (0, ψC0 ) and in the Fourier series is retained
a maximum at (φC , ψC ) = (2W, ψC0 + 2H). By restricting the
model to have only these two local extremes, a cubic polynomial δφC = A cos (θ − η) (50)
fit is obtained

    3 
and Eq. (18) reduces to
3 φC 1 φC
ψC |ε=0 = ψC0 + H 1 + −1 − −1 (48)
2 W 2 W φC = ΦC + A cos (θ − η) (51)

7 Copyright 
c 2004 by ASME
This approximation seems appropriate for capturing the effects and we can therefore conclude that the perturbation must satisfy
of the variation in head rise induced by rotor whirl, Eq. (12). the eigenvalue problem
Substituting Eq. (48) into Eq. (13) and making use of Eq.
(51), the ordinary differential in Eqs. (47) are readily expressed.    
J υ (0) − sI · δυ
υ=0 (57)
Together with plenum continuity and throttle characteristic, this
defines the FSW model for the compressor flow. The FSW model
has the state vector The four eigenvalues of this system are readily computed and the
sign of ℜ{s} will determine if the operating point υ (0) is stable.
υ = {Ψ, ΦC , A, η}T (52) In a steady-state operating point we have υ̇υ(0) = 0 in Eq.
(56). Considering first the uniform flow condition A(0) = 0 gives
and the non-linear governing equations of the FSW model can be a steady-state operating point on the cubic characteristic
stated as

(0)
ΦC = γT Ψ(0)
υ = N (υ
υ̇ υ) ⎛    (0) 3 ⎞
(0)
⎧ 2  √  3 ΦC 1 ΦC
⎪
⎪ Φ − γ Ψ Ψ(0) = ψC0 + H ⎝1 + −1 − −1 ⎠
⎪
⎪ B2
eff
C T 2 W 2 W
⎪
⎪     3 
⎪
⎪ 3 ΦC 1 ΦC
⎪
⎪
1
ψC0 − Ψ + H 1 + −1 − −1 (58)
⎪
⎪
⎪
⎪ 2
eff 2 W 2 W
⎪
⎪
⎪  2  
⎪
⎨ 3 A ΦC For the sake of argument, the stability of a steady-state operating
− H −1
= 4 W W point is inspected with A(0) = 0 (i.e., compression system surge
⎪
⎪       
⎪
⎪ 3 ΦC ΦC 3 A 2 A
is reached without significant rotating stall). At the threshold of
⎪
⎪
1
−H −2 +
⎪
⎪
surge stability the eigenvalues defined by Eq. (57) will be on the
⎪
⎪ 2 (m + λ) 2W W 4 W W complex axis, ℜ {s} = 0. When solving Eq. (57) it is found that
⎪
⎪   

⎪
⎪
⎪
⎪  +εψε cos ωεt − η  this condition imposes
⎪
⎪ 1 1  
⎩ λ + εψε sin ωεt  − η
2 (m + λ) A  4γT W
(53) Ψ(0) =  (0)
 (0)
(59)
Φ ΦC
3B2 H 2 − C
Since the whirl eccentricity represents a coupling to the axial W W
compressor rotordynamics, Eq. (53) does not represent a closed
form problem. which substitutes into Eq. (58) to give the conditions

 (0)
  (0) 2
SURGE AND ROTATING STALL 3 2 ΦC ΦC
γ2T,surge
= B H 2− (60)
To investigate the stability of the solution to Eq. (53) for 4 W W
ε = 0, υ is perturbed with a small harmonic term ⎛  (0)   (0) 3 ⎞
3 ΦC 1 ΦC
 ψC0 + H ⎝1 + −1 − −1 ⎠
υ = υ(0) + δυ
υest (54) 2 W 2 W
 (0)
−1
where s is non-dimensional with respect to the spin frequency Ω. 4W 2 Φ
= 2 2− C (61)
Substituting Eq. (54)into Eq. (53) gives directly 3B H W


 

υ̇ υ(0) + sδυ
υ = υ̇ υest = N υ (0) + δυυest thereby defining the flow and throttle coefficients at the onset
     of actual surge. At the threshold of surge stability the complex
= N υ (0) + J υ (0) · δυυest + O (δ)2 (55) eigenvalue, ℑ {s} = iωsurge , is the frequency of the flow distur-
bance
where J is the Jacobian of N. From Eq. (53) it follows
  (0)
 (0)
2
  1 3H Φ ΦC
(ωsurge )2 + = 2− C (62)
υ(0) = N υ (0)
υ̇ (56) B2
2eff 4W
eff W W

8 Copyright 
c 2004 by ASME
By setting A(0) = 0, the surge stability analysis becomes equiva- operating point is excited by a harmonic, but the flow non-
lent to the classical linearized surge model of Emmons et al. [15]. uniformity A is not trivial and care has to be taken to treat term
The strength of the Moore-Greitzer performance model is to il- A−1 in the η̇-equation appropriately.
lustrate the coupling between rotating stall and surge dynam- For a steady-state operating point the solution will consist
ics, so the above surge discussion may therefore seem at cross- of a homogenous part (i.e., for ε = 0) and driven part. A ho-
purposes. However, Eq. (62) could be used to quantify
eff within mogenous solution A(0) = 0 was seen to exist, but only in the
engineering accuracy for an axial compressor prototype based on (0)
flow range 0 ≤ ΦC ≤ 2W , and it is beyond the present scope
low pressure performance and surge testing. to determine the whirl amplitude past the aerodynamic stability
Another steady-state solution υ̇ υ(0) = 0 with A(0) = 0 repre- limit. On this basis we can assume A = O (ε), or in other words
sents a conditions with rotating stall in equilibrium A(0) = 0. This condition creates a priori a conflict with the η̇-
term A−1 in Eq. (53). Assuming therefore the following regular

(0) expansions of the state variables
ΦC = γT Ψ(0)
⎛  3 ⎞
(0)
ΦC ⎧ ⎫ ⎧ (0) ⎫ ⎧ (1) ⎫
Ψ(0) = ψC0 + H ⎝1 + −1 ⎠ ⎪
⎪ Ψ⎪ ⎪ ⎪
⎪ Ψ ⎪ ⎪ ⎪
⎪
Ψ ⎪
W ⎨ ⎬ ⎨ (0) ⎬ ⎨ (1) ⎪ ⎬
ΦC Φ ΦC
 υ= = C +ε + O (ε)2 (65)
 (0)   (63) ⎪
⎪ A ⎪
⎪ ⎪ ⎪ ⎪ (1) ⎪
⎭ ⎪ ⎩ (0) ⎪ ⎪
⎩ (1) ⎪
(0) 0 A
A(0)  ΦC ΦC ⎩ ⎭ ⎭

= 2 2− η η η
W W W
λ
η(0) = t  + η0 we find that Eq. (66) decomposes as follows:
2 (m + λ)
O (1) :
The rate of variation in η(0) is the nondimensional rotating stall
frequency dΨ(0) 2  (0) 
(0)

= Φ − γ T Ψ
dt  B2
eff C
λ (0)
   (0) 
ωstall = (64) dΦC 1 3 Φ
2 (m + λ) = ψC0 − Ψ(0) + H 1 + C
−1
dt  2
eff 2 W
 (0) 3 ⎞⎞
and η0 sets the initial time. 1 ΦC (66)
A steady-state solution with A(0) = 0 is seen to require − − 1 ⎠⎠
(0) 2 W
0 ≤ ΦC ≤ 2W . Furthermore, the presence of the rotating stall
(0)
cell forces the operating point (ΦC , Ψ) away from the cubic A(0) = 0
(0)   
characteristic. Indeed, for given ΦC we find that Ψ(0) is less dη(0) 1 1   (0)
(0) = λ + ψε sin ωε t − η
than predicted by the cubic characteristic for W < ΦC < 2W dt  2 (m + λ) A(1)
(0)
and vice-versa for 0 < ΦC < W . This property of the solution
agrees well with the empirically proven drop in head at the onset O (ε) :
(0)
of rotating stall. It is also observed that A(0) exceeds ΦC when
(0)  
ΦC < 4W /3, which indicates that the rotating stall amplitude is
dΨ(1) 2 (1) γT Ψ(1)
so large that flow reversal occurs locally. = 2 ΦC − √
It should also be observed that the rotating stall frequency dt  B
eff 2 Ψ(0)
predicted by Eq. (64) will be less than 50% for positive dimen- (1)
 (0)
 (0)  (1) 
dΦC 1 (1) 3 ΦC ΦC ΦC
sionless “time” lag, λ > 0, and all degrees of diffusion in the exit 
= −Ψ − H −2
duct, 1 ≤ m ≤ 2. dt 2
eff 2 W W W
 (0)
 (0)
 (67)
dA(1) 1 3 ΦC ΦC A(1)
= − H −2
ROTOR WHIRL dt  2 (m + λ) 2 W W W
 
Asymptotic Expansion +ψε cos ωεt  − η(0)
The FSW model for lateral vibrations is Eq. (53) together
with a rotordynamic model for the compressor. An undisturbed η̇(1) = irrelevant

9 Copyright 
c 2004 by ASME
 Note that A(1) appears in the O (1) phase equation, and that Before discussing the impact of this result in terms of cross-
η(1) is irrelevant to an O (ε) accuracy because it should be re- coupled stiffness we observe that the phase ηε tends towards
called that with A(0) = 0 the flow coefficient becomes ±π/2 when approaching the maximum head point ΦC = 2W .
(0)

   The sign of ηε is determined by the whirl frequency, being pos-

(0) (1)
φC = ΦC + ε ΦC + A(1) cos θ − η(0) + O (ε)2 (68) (0)
itive when ωε ≥ ω stall and ΦC > 2W . The amplitude of the
flow non-uniformity increases as the flow is reduced towards the
(0)
It is concluded that whirl induced effects are of O (ε) for the op- maximum head point ΦC = 2W and the vibration frequency ap-
(0)
erating point (ΦC , Ψ(0) ). Furthermore, it is seen that the cor- proaches ωε = ω stall . This influence of the rotating stall fre-
(1) quency exists despite the fact that the operating point was as-
rection (ΦC , Ψ(1) ) is uncoupled from the flow non-uniformity
sumed to be free of rotating stall.
A(1) , implying that the mean operating point is uncoupled from
the rotordynamics.
The two equations governing A(1) and η(0) are restated
Rotor Reaction Forces

dη(0) (1)   The steady-state excitation from a single harmonic flow

(1)   (0)
2 (m + λ) A − λA = ψε sin ωε t − η nonuniformity consist of the torque and pressure reaction forces.
dt    The flow angles βr1 and βr2 depend on φC , but for small ampli-
(0) (0)
dA(1) 3 H ΦC ΦC tude A  ΦC of the flow non-uniformity we can linearize this
2 (m + λ) + − 2 A(1) (69)
dt  2W W W dependence
 
= ψε cos ωεt  − η(0)
∂βri
βri = βri + A cos (θ − η) + O(A2 ) (73)
∂φC
We seek a steady-state solution. Setting dA(1) /dt 
= 0 in the φC =ΦC

latter equation we see that time dependence is maintained unless

(ωεt  − η(0) ) is constant, a condition that can be stated as
where βri = βri |φC =ΦC , i = 1, 2, are the inlet and exit relative
rotor flow angle averaged in both time and space at the consid-
η(0) = ωεt  − ηε (70) ered steady-state operating point. Neglecting the dependence of
the flow angles on the flow non-uniformity in the same way as
where ηε is a constant phase between the minimum whirl clear- Nahwi [10] did, it follows that whirl does not induce a fluctua-
ance and the maximum flow non-uniformity. Substituting Eq. tion in the total compressor torque at an O(ε2 ). Indeed, Eq. (68)
(70) into Eq. (69) with dA(1) /dt  = 0 and making use of Eq. and the approximation βri ≈ βri imply that Eq. (38) becomes
(64), the steady-state problem reduces to
 
ωε (0)
τC ≈ ρACU 2 NR(ΦC )2 (tan βr1 − tan βr2 )
λ − 1 A(1) = ψε sin ηε (74)
ω stall
(0)
 (0)  (71)
3 H ΦC ΦC
− 2 A(1) = ψε cos ηε The lateral rotor reaction force to whirl induced torque non-
2W W W
uniformity is readily expressed in terms of the asymptotic expan-
sion by substituting Eqs. (68) and (70) into Eq. (42)
which has the solution

ψε
A(1) =   2
   2
 3 H Φ(0) Φ(0) ωε
(0)
Fτ ≈ −εA(1) ρACU 2 NΦC (tan βr1 − tan βr2 )×
 C C
−2 +λ 2
−1  
2W W W ω stall − sin(ωεt  − ηε )
   (75)
ωε cos(ωεt  − ηε )
λ −1
ω stall
tan ηε = (0)
 (0)

3 H ΦC ΦC where A(1) and ηε are given by Eq. (71).
−2
2W W W Proceeding in the way for the contribution to the rotor reac-
(72) tion force from whirl induced pressure non-uniformity and un-

10 Copyright 
c 2004 by ASME
steady momentum, Eqs. (42) give The employed asymptotic expansion was only valid for op-
eration away from the onset of rotating stall (i.e., ΦC > 2W in
 
LC Figure 3), and Eq. (71) therefore gives cos ηε > 0. Accordingly,
F p ≈ −εA(1) ρACU 2 N × within the framework of this paper torque contributes to making
4Nh
   
   Alford’s β parameter more positive (i.e., destabilizing for for-
(0) cos(ωε t − ηε )  − sin(ωεt  − ηε )
−ΦC − (2 − m) ω ward whirl).
sin(ωεt  − ηε ) stall cos(ωεt  − ηε )
If the whirl frequency is in the range 0 < ωε < ωstall , Eq.
(76) (71) gives sin ηε < 0 and both pressure and unsteady momen-
tum effects contribute to reduce β. Should these contributions be
and large enough to make β negative, the cross-coupling force will
  be destabilizing for backward whirl.

Z In the case of ωε > ωstall Eq. (71) gives sin ηε > 0 and the
Fu ≈ −εA(1) ρACU 2 N (tan βr1 + tan βr2 )×
4R only contribution that does not tend to make β more positive is
  
cos(ωεt  − ηε ) the pressure term containing (2 − m)ωε cos ηε , which disappears
−ωε (77)
sin(ωεt  − ηε ) in an axial compressor with exit duct of near constant area.
Finally, for backward whirl ωε < 0, the contribution from
because the analysis is restricted to steady-state conditions. torque, the pressure term containing (2 − m)ωε cos ηε and the un-
For the purpose of rotordynamic analysis the lateral ro- steady momentum force will act stabilizing by making β more
tor reaction forces are more conveniently expressed in terms positive, while the pressure term containing sin ηε acts destabi-
of impedance coefficients. Using the nomenclature from Eq. lizing.
(1), but allowing frequency dependence of the direct K(ωε ) and The relative importance of the four contributions to the lat-
cross-coupled k(ωε ) “stiffness” coefficients, it results eral rotor whirl reaction force (i.e., one torque term, two pressure
terms and one unsteady momentum term) depends on the com-
⎛ pressor design, i.e., the blade height to stage length ( L h/N ), the
C
h⎜ rotor stage length to mean compressor radius ( LRZ ), the exit duct
K= ⎝ − sin ηε
C diffusion parameter (1 ≤ m ≤ 2) and stage loading expressed in
torque contribution terms of the flow angles (βr1 and βr2 ).
  
LC (2 − m)ωε The amplitude of the flow non-uniformity A(1) is highest
+ cos ηε + (0)
sin ηε when ωε = ωstall , see Eq. (72). Regardless of other design fea-
4Nh(tan βr1 − tan βr2 ) ΦC tures, fluid-structure interaction in an axial compressor with near
pressure contribution constant area exit duct will therefore be most detrimental to the
    rotordynamic stability margin if the first forward bending mode
LZ tan βr1 + tan βr2 ωε A(1) τC
+ (0)
cos ηε (0)
(78) coincides with the rotating stall frequency. The amplitude of
4R tan βr1 − tan βr2 ΦC Φ Rh
C A(1) grows unboundedly as the operating point approaches the
unsteady contribution
onset of rotating stall. This represents a singularity in the solu-
tion, which will need separate treatment. Until then, it is rec-
ommended not to extend the use of Eq. (78) and (79) beyond
(0)
⎛ conditions where εA(1) exceeds 10% of ΦC .
τC h⎜ These qualitative features of the asymptotic solutions appear
k=β = ⎝ cos ηε to agree with the results published by Spakovszky [8], Song and
Rh C
torque contribution Cho [9], Storace et al. [5], Ehrich et al. [6] and Al-Nahwi et
   al. [10] – [11].
LC (2 − m)ωε
+ sin ηε − (0)
cos ηε
4Nh(tan βr1 − tan βr2 ) ΦC
pressure contribution
CONCLUSION
⎞ The FSW model allows to establish explicit expressions for
   ⎟ the direct and cross-coupling impedance from fluid-structure in-
LZ tan βr1 + tan βr2 ωε ⎟ A(1) τC teraction in an axial compressor. The validity of the expressions
+ (0)
sin ηε ⎟
⎟ (0) Rh (79)
4R tan βr1 − tan βr2 ΦC ⎠ ΦC derived here is limited to operating condition away from rotating
stall, but their accuracy is consistent with usual linearized rotor-
unsteady contribution dynamic analysis.

11 Copyright 
c 2004 by ASME
 In the design phase of an axial compressor prototype, the [5] Storace, A., Wisler, D., Shin, H.-W., Beacher, B., Ehrich,
rotordynamic stability margin in the normal operating range can F., Spakovsky, Z., Martinez-Sanchez, M., and Song, S.,
therefore be assessed by the following simple steps: 2001. “Unsteady flow and whirl-inducing forces in axial-
flow compressors. part i - experiment”. ASME Journal of
1. Fit the analytical or empirically scaled compressor perfor- Turbomachinery , pp. 433–445.
mance map (i.e., total to static head ψC versus flow φC for [6] Ehrich, F., Spakovsky, Z., Martinez-Sanchez, M., Song,
concentric compressor shaft ε = 0) to the cubic polynomial S., Wisler, D., Storace, A., Shin, H.-W., and Beacher, B.,
from Eq. (48). 2001. “Unsteady flow and whirl-inducing forces in axial-
2. Evaluate the sensitivity of the compressor head characteris- flow compressors. part ii - analysis”. ASME Journal of Tur-
tic to tip clearance, ψε , using cascade test data. bomachinery , pp. 446–452.
3. Determine if the exit duct represents a sudden expansion [7] Ehrich, F., 1993. “Rotor whirl forces induced by tip clear-
(m = 1) or tends towards a constant area duct (m = 2). ance effect in axial flow compressors”. Journal of Vibration
4. Quantify Greitzer’s dimensionless time “lag” λ. Depend- and Acoustics , pp. 509–515.
ing on the novelty of the design, empirical rotating stall fre- [8] Spakovszky, Z., 2000. “Analyzis of aerodynamically in-
quency data may assist by means of Eq. (64). duced whirling forces in axial flow compressors”. ASME
(0) Journal of Turbomachinery , pp. 761–768.
5. Starting with a flow rate ΦC well away from the aerody-
namic stability limit, the rotordynamic stability analysis is [9] Song, S., and Cho, S., 2000. “Non-uniform flow in a com-
performed by setting ωε in Eqs. (78) and (79) equal the fre- pressor due to turbine tip clearance”. ASME Journal of Tur-
quency of the most critical mode in bearings and seals only, bomachinery , pp. 751–760.
normally the first forward bending mode. The resulting fre- [10] Al-Nahwi, A., Paduano, J., and Nayfeh, S., 2003.
quency for the considered mode is used to update K and k, “Aerodynamic-rotordynamic interaction in axial compres-
and the rotordynamic stability analysis is repeated. A mod- sion systems part i: Modeling and analysis of fluid-induced
erate number of iterations should suffice. forces”. ASME Journal of Turbomachinery , pp. 405–415.
6. Repeat the procedure for decreasing flow rate until the vicin- [11] Al-Nahwi, A., Paduano, J., and Nayfeh, S., 2003.
(0) “Aerodynamic-rotordynamic interaction in axial compres-
ity of the singularity at ΦC = 2W .
7. Repeat the procedure for the first backward mode in case the sion systems part ii: Impact of interaction on overall system
axial compressor design has significant negative β. stability”. ASME Journal of Turbomachinery , pp. 416–424.
[12] Greitzer, E., 1976. “Surge and rotating stall in axial flow
compressors parts i, ii”. ASME Journal of Engineering for
Explicit formulation as in Eqs. (78) and (79) have the ad-
Power, 98 , pp. 190–217.
vantage of permitting simultaneous treatment of all known con-
[13] Moore, F., and Greitzer, E., 1986. “A theory of post-stall
tributions to rotordynamic instability. The behavior of the the
(0) transients in axial compression systems part i: Develop-
FWT model at the limit ΦC → 2W , as well as into the post-stall ment of equations”. ASME Journal of Engineering for Gas
region, was included in Al-Nahwi et al. [11] for a Jefcott rotor. Turbines and Power, 108 , pp. 68–76.
However, for the same reasons, it would also appear that asymp- [14] Moore, F., 1984. “A theory of rotating stall of multistage
totic solutions near the onset of rotating stall could be of use for compressors part i, ii, iii”. ASME Journal of Engineering
rotordynamic analysis of real axial compressors. for Power, 106 , pp. 313–336.
[15] Emmons, H., Pearson, C., and Grant, H., 1955. “Compres-
sor surge and stall propagation”. Transaction of the ASME,
REFERENCES 77 , pp. 455–469.
[1] Ehrich, F., and Childs, D., 1984. “Self-excited vibration in
high performance turbomachinery”. Mechanical Engineer-
ing , pp. 66–79.
[2] Thomas, H., 1958. “Unstable natural vibration of turbine
rotors induced by the clearance flow in glands and blading”.
Bulletin De l’A.I.M., 71 (11/12) , pp. 1039–1063.
[3] Alford, J., 1965. “Protecting turbomachinery from self-
excited rotor whirl”. ASME Journal of Engineering for
Power , pp. 333–334.
[4] Vance, J., and Laudadio, F., 1984. “Experimental measure-
ment of alford’s force in axial flow turomachinery”. ASME
Journal of Engineering for Power, 106 , pp. 585–590.

12 Copyright 
c 2004 by ASME