Anda di halaman 1dari 13

Probabilistic Engineering Mechanics 21 (2006) 384–396

www.elsevier.com/locate/probengmech

Statistics of responses of a mistuned and frictionally damped bladed disk


assembly subjected to white noise and narrow band excitations
D. Cha, A. Sinha ∗
Department of Mechanical Engineering, The Pennsylvania State University, University Park, PA 16802, USA

Available online 23 March 2006

Abstract

In this paper, the effects of friction dampers on the vibration of a mistuned bladed disk assembly are examined for the following types of
random excitation: white noise and narrow band. To compute the statistics of nonlinear response, an analytical technique is developed based on
the equivalent linearization method. The validity of the analytical technique has been established by comparison with the results from numerical
simulations. The relation between the vibration mode of the system and the performance of friction dampers is investigated.
c 2006 Published by Elsevier Ltd

1. Introduction loads [3–5]. Recently, Cha and Sinha [6] have studied the
random response of a single-degree-of-freedom model of a
Friction dampers are widely used to reduce the resonant frictionally damped turbine blade.
vibration amplitude of a turbine blade, because excessive Since the numerical simulation of the random response of
resonant stress causes the fatigue failure of a turbine blade. In a nonlinear system is computationally very expensive, it is
practice, two types of friction damper, blade-to-ground (B–G) practically impossible to investigate the effect of mistuning on
and blade-to-blade (B–B), have been used and each friction the vibration of a bladed disk assembly with friction dampers by
damper can be simplified as a composition of a linear spring and the Monte Carlo method. Therefore, analytical techniques are
a Coulomb friction element [1]. The B–G damper provides slip required to compute the statistics of the variance of a blade’s
between points on the blade and a relatively rigid structure such displacement, and to obtain optimal normal loads of friction
as a coverplate, while the B–B damper provides a link between dampers that minimize the probability that a blade’s amplitude
adjacent blades. The amount of energy dissipation by a B–G will exceed a critical value. It should be noted that the system is
damper mainly depends on the blade’s vibration amplitude, nonlinear due to friction damping, and has random inputs and
whereas that by a B–B damper depends on the amplitude of random parameters as well. Using the equivalent linearization
the relative motion between adjacent blades. method (ELM) [7–9], statistics of responses of the nonlinear
In a mistuned bladed disk assembly, the vibration amplitude system with specified parameters and random excitations are
of a blade can be significantly greater than that of a perfectly computed. Then, statistics of these statistics of responses
tuned system. This mistuning phenomenon in the bladed due to random parameters (mistuning) are computed via a
disk assembly with friction dampers has been studied by linearization approach for which the required partial derivatives
numerical simulations [2] and the analytical technique [1]. with respect to random parameters have been obtained in an
These approaches are valid only for deterministic sinusoidal elegant manner. Therefore, even though standard linearization
excitation and the response of a frictionally damped bladed techniques have been used, the final computational algorithm is
disk assembly to random excitation has not been investigated. quite unique. In this paper, two types of random excitation have
It has been reported that turbine blades can experience random been considered: white noise and narrow band.
The model of the bladed disk assembly is chosen to be as
shown in Fig. 1 [1]. This model represents the fundamental
∗ Corresponding author. Tel.: +1 814 863 3079; fax: +1 814 863 4848. mode of vibration for each blade. The modal mass of each
E-mail address: axs22@psu.edu (A. Sinha). blade is considered to be identical and represented by m t . Also,

c 2006 Published by Elsevier Ltd


0266-8920/$ - see front matter
doi:10.1016/j.probengmech.2006.01.001
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 385

Nomenclature
A system matrix in the state space analysis
B input matrix in the state space analysis
BR input matrix of Lyapunov equation
C damping matrix
c damping coefficient of each blade
y y
c xB , k Bx , c B , k B equivalent linear coefficients of a B–B
damper
cGx , k x , c y , k y equivalent linear coefficients of a B–G
G G G
damper
dB displacement of a blade-to-blade damper
dG displacement of a blade-to-ground damper Fig. 1. Model of a bladed disk assembly with friction dampers.
E[ ] expected value
F(t) external force vector simulated by considering the perturbations in modal stiffnesses,
FN normal load of a friction damper ki , alone.
fB hysteretic force of a blade-to-blade damper First, analytical techniques are presented to compute the
fG hysteretic force of a blade-to-ground damper statistics of the variance of a blade’s displacement. Then, the
g(t) narrow band excitation results from these analytical techniques are compared with
h Gi , h Bi defined by Eqs. (5a) and (5b) those from numerical simulations.
In identity matrix
K stiffness matrix
Kc coupling stiffness between adjacent blades 2. Formulation
KB stiffness of a blade-to-blade damper
KG stiffness of a blade-to-ground damper The governing differential equations of motion for the
k0 modal stiffness of each blade for the tuned system system shown in Fig. 1 can be written as follows:
ki modal stiffness of ith blade
M mass matrix m t ẍi + c ẋi + (ki + 2K c )xi − K c xi+1 − K c xi−1 + f Gi
mt modal mass of each blade + f Bi − f Bi−1 = Fi (t), (i = 1, 2, . . . , n) (1)
n number of blades
P correlation matrix of state variables where f Gi and f Bi are hysteretic forces of B–G and B–B
Q0 intensity of white noise excitation friction dampers on the ith blade, respectively. Note that (i + 1)
Rw (τ ) correlation matrix of white noise excitation is taken as 1 if i = n and (i − 1) is taken as n if i = 1, where
w(t) white noise excitation n is the number of blades. The modal stiffness, ki , is a random
xB relative displacement, defined in Eq. (5e) variable with the mean value k0 ; i.e.,
x Bc slip distance of a blade-to-blade damper
x Gc slip distance of a blade-to-ground damper ki = k0 + δki (2)
xi displacement of ith blade
where δki is the statistical deviation due to mistuning. It
yB relative displacement of a blade-to-blade damper
is assumed that the distributions of δki are Gaussian and
yG relative displacement of a blade-to-ground damper
z state variables independent. In Fig. 1, dGi and d Bi represent displacements of
σxi standard deviation of ith blade displacement the B–G and B–B dampers, respectively. The friction forces,
σR defined in Eq. (20b) f Gi and f Bi , can be expressed as
µf coefficient of friction (i) when a friction damper is stuck:
µR defined in Eq. (20a)
K G |xi − dGi | ≤ µ f FN , K B |xi − d Bi | ≤ µ f FN (3a)
ξF damping ratio of the band-pass filter
ωF resonant frequency of the band-pass filter f Gi = K G (xi − dGi ), f Bi = K B (xi − d Bi ) (3b)

(ii) when a friction damper is slipping:

the modal stiffness of the ith blade is represented by ki . The K G |xi − dGi | = µ f FN , K B |xi − d Bi | = µ f FN (3c)
mechanical coupling between adjacent blades due to the disk’s f Gi = µ f FN sign(ḋGi ), f Bi = µ f FN sign(ḋ Bi ) (3d)
flexibility has been represented by a spring with the stiffness
K c . The stiffnesses of friction dampers are represented by K G where µ f and FN represent the coefficient of friction and
and K B for B–G and B–B dampers, respectively. The damping the normal load of a friction damper, respectively. For such
coefficient c is used to represent the structural and aerodynamic systems, analytical calculation of the response depends on the
damping. It should be noted that the mistuning phenomenon is characteristic of the external force, Fi (t).
386 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396

2.1. White noise excitation and


∂ f Bi
 
If the external force, Fi (t), is assumed to be an ergodic and c xBi = E , (8a)
∂ ẋ Bi
Gaussian white noise with zero mean, wi (t), the ELM can be
∂ f Bi
 
used to obtain the statistics of response. x
k Bi = E , (8b)
The nonlinear forces of friction dampers can be described [7] ∂ y Bi
as ∂h Bi
 
y
c Bi = E , (8c)
f Gi (ẋi , yGi ; x Gc ) = K G yGi [U (yGi + x Gc ) ∂ ẋ Bi
− U (yGi − x Gc )] ∂h Bi
 
y
+ K G x Gc [U (yGi − x Gc )U (ẋi ) k Bi = E . (8d)
∂ y Bi
− U (−yGi − x Gc )U (−ẋi )] (4a)
Note that E[·] is the expected value. These coefficients are
f Bi (ẋ Bi , y Bi ; x Bc ) = K B y Bi [U (y Bi + x Bc ) evaluated by assuming a joint Gaussian probability distribution
− U (y Bi − x Bc )] of the multidimensional response process and are presented in
+ K B x Bc [U (y Bi − x Bc )U (ẋ Bi ) the Appendix A.
− U (−y Bi − x Bc )U (−ẋ Bi )] (4b) Now, Eq. (6) can be written in matrix form as
where M ẍ + (C + C Gx + C Bx )ẋ + K x + K Gx yG + K Bx y B = w(t) (9a)
y y
ẏGi = h Gi (ẋi , yGi ; x Gc ) ẏG = C G ẋ + K G yG (9b)
= ẋi [U (yGi + x Gc ) − U (yGi − x Gc ) y y
ẏ B = C B ẋ + K B yB (9c)
+ U (yGi − x Gc )U (−ẋi )
+ U (−yGi − x Gc )U (ẋi )] (5a) where M, K and C are mass, stiffness and damping matrices of
ẏ Bi = h Bi (ẋ Bi , y Bi ; x Bc ) the linear system without friction dampers, respectively. Also,
y
= (ẋ Bi )[U (y Bi + x Bc ) − U (y Bi − x Bc ) the matrices of linearized coefficients, C Gx , C Bx , K Gx , K Bx , C G ,
y y y
+ U (y Bi − x Bc )U (−ẋ Bi ) K G , C B , K B can easily be derived using Eq. (6).
+ U (−y Bi − x Bc )U (ẋ Bi )] (5b) Eq. (9) can be represented in the state space form; i.e.,

yGi = xi − dGi , (5c) ż = A z + Bw(t) (10)


µ f FN where
x Gc = (5d)
KG
z T = x T ẋ T T
y BT
 
yG (11a)
x Bi = xi − xi+1 , (5e)
 0 In 0 0

y Bi = xi − d Bi , (5f) −1 K x + Cx )
−M −1 (C + C G x
−M −1 K G x
−M −1 K B
−M B
µ f FN A= 0 (11b)

y y
. CG KG 0 
x Bc = (5g) y y
KB 0 CB 0 KB

Note that U (·) is the unit step function. Using the ELM,
T
M −1 .

B= 0 0 0 (11c)
equations of motion can be represented as
The correlation matrix of external forces is given as below:
m t ẍi + c ẋi + (ki + 2K c )xi − K c xi+1 − K c xi−1
+ (cGi
x x
ẋi + k Gi yGi ) + (c xBi ẋ Bi + k Bi
x
y Bi ) Rw (τ ) = E[w(t) w(t + τ )T ] = Q 0 δ(τ )In (12)

− (c xBi−1 ẋ Bi−1 + k Bi−1


x
y Bi−1 ) = wi (t) (6a) where Q o is the intensity of the white noise and δ(τ ) is the
y y dirac-delta function. Since E[w(t)] = 0, it has been shown
ẏGi = cGi ẋi + k Gi yGi (6b) that E[z(t)] = 0 in the statistical steady state. Let P be the
y y correlation matrix of state variables: i.e.,
ẏ Bi = c Bi ẋ Bi + k Bi y Bi (6c)
where P = E[z(t) z(t)T ]. (13)
∂ f Gi
 
Then the correlation matrix P satisfies the following algebraic
x
cGi =E , (7a)
∂ ẋi equations under steady state conditions:
∂ f Gi
 
x
k Gi =E , (7b) A P + P AT = B R (14)
∂ yGi
where
∂h Gi
 
y
cGi =E , (7c) B R = −B Q 0 B T . (15)
∂ ẋi
∂h Gi
 
y Because the linearized coefficients are functions of elements
k Gi =E (7d)
∂ yGi of the correlation matrix P, Eq. (14) can be solved numerically
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 387

to evaluate the correlation matrix P [8]. The variance of the 2.2. Narrow band excitation
system response, (σx2 )i , can be obtained from the correlation
matrix P: i.e., The external excitation, Fi (t), is assumed to be a narrow
band excitation, gi (t), which can be generated by passing the
(σx2 )i = E[xi2 ] = Pi,i . (16) white noise excitation, wi (t), through the band-pass filter: i.e.,
Because of mistuning, elements of the stiffness matrix are g̈i + 2ξ F ω F ġi + ω2F gi = ω2F wi (t) (21)
random variables. As a result, σx2 are random variables. Hence,
it is desirable to predict the mean and standard deviation of σx2 . where ξ F and ω F are the damping ratio and natural frequency
Using the Taylor’s series expansion, the solution of Eq. (14) for of the band-pass filter.
a mistuned system can be approximated as follows: Using ELM, governing differential equations are represented
n
by
X ∂P
P = P0 + δki . (17) m t ẍi + c ẋi + (ki + 2K c )xi − K c xi+1 − K c xi−1
i=1
∂ki
x x
yGi + c xBi ẋ Bi + k Bi
x
 
+ cGi ẋi + k Gi y Bi
Note that P0 is the correlation matrix of the perfectly tuned
− c xBi−1 ẋ Bi−1 + k Bi−1
x
y Bi−1 = gi (t)

system and the higher order terms of the Taylor’s series (22a)
∂P y y
expansion are neglected. ∂k i
is the solution of the following ẏGi = cGi ẋi + k Gi yGi , (22b)
linear algebraic equations, which are obtained by differentiating y y
Eq. (14) with respect to ki : ẏ Bi = c Bi ẋ Bi + k Bi y Bi . (22c)

∂P ∂P T

∂A ∂ AT
 If the state vector is selected as
A0 + A =− P0 + P0 (18) T
∂ki ∂ki 0 ∂ki ∂ki z = g T ġ T x T ẋ T yG T
y BT
 
(23)
where then Eqs. (21) and (22) can be represented in the state space
∂A form as
=
∂ki ż = A z + B w(t) (24)
 
0 0 0 0
−M −1 ∂ K ∂C G
 x
∂C Bx

∂ K Gx ∂ K Bx where
−M −1 + −M −1 −M −1 
 ∂ki ∂ki ∂ki ∂ki ∂ki 

y y .
 A=
 ∂C G ∂ KG
0 0
   0 In 0 0 0 0

 ∂ki ∂ki 
 y
∂C B ∂KB
y
 −ω2F In −2ξ F ω F In 0 0 0 0 
0 0  
∂ki ∂ki  0 0 0

In

0 0 
 −1
−M −1 K x + Cx
−M −1 C + C G x
−M −1 K G x
−M −1 K B

(19) M 0 B 
 y y 
0 0 0 CG KG 0
The matrix P0 is known by solving Eq. (14) for a perfectly 0 0 0 CB
y
0 KB
y
tuned system. In Eq. (19), ∂∂kKi can be calculated analytically
y y y (25a)
∂C G
x ∂ K Gx ∂C G ∂ K G ∂C Bx ∂ K Bx ∂C B
and the elements of ∂ki , ∂ki , ∂ki , ∂ki , ∂ki , ∂ki , ∂ki
T
ω2F In .

y ∂c x ∂k x
B= 0 0 0 0 0 (25b)
∂K
and ∂kiB can be represented by combinations of ∂kGi j , ∂kGi j ,
y y y y Therefore, Eq. (14) can now be used to calculate the
∂cG j ∂k G j ∂c xB j ∂k Bx j ∂c B j ∂k B j
∂ki where ( j = 1, 2, . . . , n).
, correlation matrix P, and the statistics of response correlation
∂ki ∂ki , ∂ki , ∂ki , ∂ki and
The expression for each of these elements can be obtained can be obtained from Eq. (17). Compared with Eq. (19), the
∂σẋ2 ∂A
∂σ y2 ∂σ y2 expression for ∂k is changed as below:
j Gj Bj i
analytically and represented as a function of ∂ki , ∂ki , ∂ki ,
∂(E[ẋ j yG j ]) ∂(E[ẋ j ẋ j+1 ]) ∂(E[ẋ j y B j ]) ∂(E[ẋ j+1 y B j ]) ∂A
∂ki , ∂ki , ∂ki and ∂ki , which =
∂P ∂ki
are elements of the matrix ∂ki . Therefore, Eq. (18) can be 0 0 
0 0 0 0
∂P
solved in an iterative manner to evaluate the matrix ∂k . The
y y
i
y y
0 0 0 0 0 0 
∂cG
x
j ∂k G
x
j ∂cG j ∂k G j ∂c xB j ∂k Bx j ∂ cB j ∂k B j 0 0 0 0 0 0

expressions of , ∂ki, , ∂ki, ,∂ki , ∂ki and ∂ki ∂ki ∂ki ∂ki

x x x x

∂C G ∂C B ∂ KG ∂KB
!
∂K
 
are presented in the Appendix B. 0 0 −M −1 −M −1 + −M −1 −M −1
.

 ∂ki ∂ki ∂ki ∂ki ∂ki
From Eq. (17), the mean and standard deviation of the 
y y

 ∂C G ∂ KG 
response variance can easily be calculated as follows: 0 0 0 0 
 ∂ki ∂ki 
(µ R )i = E[Pi,i ],
 
(20a) ∂C B
y
∂KB
y
0 0 0 0
  ∂ki ∂ki
σ R2 = E[Pi,i2
] − (µ R )i2 (20b) (26)
i
∂A
where the expressions for (µ R )i and (σ R2 )i are provided in The expressions for the nonzero elements of ∂k i
are exactly
Appendix C. the same as those for the white noise excitation. Hence, each of
388 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396

Table 1
System parameters
m t = 0.0114 kg, c = 1.45 N s/m, k0 = 430,000 N/m, K c = 45,430 N/m
K G = 43,000 N/m, K B = 43,000 N/m

these elements can be represented as a function of the elements


∂P
of the matrix ∂k i
, and Eq. (18) can be solved iteratively to
∂P
evaluate the matrix ∂k i
.

3. Accuracy of analytical techniques


To verify the validity of analytical techniques for
nonlinear systems, analytical results are compared with those
from numerical simulations. The nominal values of modal
parameters are chosen from the paper by Griffin and Sinha [2],
and given in Table 1. The mean value of ki is equal to the Fig. 2a. µ R as a function of x Gc (B–G damper, white noise excitation).
modal stiffness k0 of a blade in the perfectly tuned system.
The standard deviation of the stiffness of each blade is chosen
to be the same. The number of blades is selected to be equal
to 10. The MATLAB routine ‘randn’ [10] is used to generate
the variations in modal stiffnesses. It yields random numbers
that have a Gaussian distribution with zero mean and specified
standard deviation. To obtain the statistics of each blade’s
response by Monte Carlo simulation, a large number (arbitrarily
chosen to be 1000) of bladed disk assemblies is considered.
For the numerical integration of differential equations,
the MATLAB routine ‘randn’ is used again to generate the
Gaussian white noise excitation, wi (t). The time response is
generated at 30,000 instants with the time interval equal to
5 × 10−5 s. The external excitation, wi (t), is assumed to be
constant during the time interval to facilitate the numerical
integration of differential equations.
In Section 3.1, the statistics of each blade’s response vari-
ance due to white noise excitation are calculated, whereas the Fig. 2b. σ R as a function of x Gc (B–G damper, white noise excitation).
results for the narrow band random excitation are obtained in
Section 3.2. The results are presented as functions of slip dis- for B–G dampers and x Bc = 0.12 × 10−3 , 0.45 × 10−3 ,
tances, x Gc and x Bc , which are proportional to normal loads. and 2.54 × 10−3 m for B–B dampers. For the tuned system,
The optimal slip distance (or normal load) is then sought to min- the response variances of blades are not identical (Table 2(a)
imize the mean and standard deviation of a blade’s amplitude. and 3(a)), because the generation of ideal white noise is
impossible in practice as excitation on one blade is not entirely
3.1. White noise excitation independent of those on other blades and the intensity of each
For white noise excitation, the intensity of external force, blade’s excitation is also not identical. In most of the cases,
Q 0 , is taken to be 1 N2 . The standard deviation of the blade’s the results from ELM are greater than those from the time
modal stiffness is chosen to be 17,200 N/m, which will corre- integration. For a given mistuned system (Table 2(b) and 3(b))
spond to about 2% mistuning in terms of the ratio of the stan- the differences in ELM and time integration results are of the
dard deviation and mean value of the modal frequency of each same order as variations in the statistics due to mistuning. From
blade. The mean and standard deviation of the response vari- the comparison with the results from time integration (Tables 2
ance, µ R and σ R , are calculated analytically from Eq. (20) and and 3), the ELM is shown to be well suited for the computation
plotted in Figs. 2 and 3 for B–G and B–B dampers, respectively. of the optimal normal load, because the trend of the analytical
The accuracy of ELM depends on system characteristics, results is fairly close to that of the simulation results.
because the ELM is an approximation to replace nonlinear To examine the accuracy of the analytical technique, the
terms by linear ones that minimize the error between them [11]. results obtained from Eq. (20) are compared with those from
Therefore, analytical results by ELM are compared with those the Monte Carlo simulation of Eq. (14). With values of slip
from the numerical integration of differential equations. The distances x Gc and x Bc same as those in Tables 2 and 3, results
results are presented in Tables 2 and 3 for B–G and B–B are presented in Tables 4 and 5 for B–G and B–B dampers,
dampers, respectively. Three different values of the slip distance respectively. The analytical results show good agreement with
are chosen: x Gc = 0.15×10−3 , 0.45×10−3 , and 1.95×10−3 m results from Monte Carlo simulation.
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 389

Table 2
Time simulation of the response variance (B–G damper, white noise excitation)
(a) Tuned system: E[x 2 ] (units: ×10−7 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Time simulation 3.85 4.02 3.96 4.28 3.72 3.70 3.92 3.91 3.99 3.95
x Gc = 0.00015 m
ELM 4.28
Time simulation 3.03 3.10 2.98 3.14 2.85 2.97 2.97 2.98 3.07 2.86
x Gc = 0.00045 m
ELM 3.36
Time simulation 5.74 5.67 5.60 5.72 5.54 5.77 5.58 5.66 5.74 5.24
x Gc = 0.00195 m
ELM 5.71
(b) A mistuned system: E[x 2 ] (units: ×10−7 m2 )
δk1 = 28,298.3 N/m, δk2 = 14,357.7 N/m, δk3 = −15, 845.5 N/m, δk4 = −47.3 N/m, δk5 = −22, 815.8 N/m, δk6 = −3762.5 N/m,
δk7 = 21,328.0 N/m, δk8 = −19, 096.3 N/m, δk9 = 3672.2 N/m, δk10 = −6088.8 N/m
Blade number 1 2 3 4 5 6 7 8 9 10
Time simulation 3.60 3.92 4.13 4.08 3.95 3.78 3.83 4.02 4.03 3.81
x Gc = 0.00015 m
ELM 4.01 4.10 4.32 4.21 4.38 4.23 4.05 4.35 4.18 4.25
Time simulation 2.88 3.05 3.14 3.10 3.03 2.93 2.86 3.12 3.07 2.81
x Gc = 0.00045 m
ELM 3.22 3.29 3.43 3.36 3.47 3.37 3.26 3.45 3.34 3.38
Time simulation 5.45 5.59 5.83 5.67 5.62 5.79 5.49 5.83 5.45 4.97
x Gc = 0.00195 m
ELM 5.46 5.58 5.86 5.71 5.93 5.75 5.51 5.89 5.68 5.77

Table 3
Time simulation of the response variance (B–B damper, white noise excitation)
(a) Tuned system: E[x 2 ] (units: ×10−7 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Time simulation 3.94 4.16 4.20 4.47 3.84 3.68 3.95 4.14 4.27 4.16
x Bc = 0.00012 m
ELM 4.52
Time simulation 2.95 3.03 3.10 3.19 2.95 2.80 2.82 3.01 3.05 3.04
x Bc = 0.00045 m
ELM 3.21
Time simulation 5.49 5.51 5.44 5.84 5.80 5.37 5.49 5.59 5.85 6.05
x Bc = 0.00254 m
ELM 5.45
(b) A mistuned system: E[x 2 ] (units: ×10−7 m2 )
δk1 = 28,298.3 N/m, δk2 = 14,357.7 N/m, δk3 = −15, 845.5 N/m, δk4 = −47.3 N/m, δk5 = −22, 815.8 N/m, δk6 = −3762.5 N/m,
δk7 = 21,328.0 N/m, δk8 = −19, 096.3 N/m, δk9 = 3672.2 N/m, δk10 = −6088.8 N/m
Blade number 1 2 3 4 5 6 7 8 9 10
Time simulation 3.29 3.81 4.47 4.19 4.41 3.90 3.63 4.57 4.17 4.05
x Bc = 0.00012 m
ELM 3.86 4.12 4.79 4.40 4.98 4.49 3.92 4.87 4.36 4.57
Time simulation 2.42 2.77 3.35 3.09 3.36 2.93 2.62 3.40 2.97 2.92
x Bc = 0.00045 m
ELM 2.71 2.91 3.45 3.21 3.65 3.24 2.81 3.51 3.13 3.28
Time simulation 5.22 5.53 5.98 5.92 5.88 5.57 5.56 6.23 5.99 5.80
x Bc = 0.00254 m
ELM 5.18 5.33 5.66 5.47 5.75 5.52 5.23 5.69 5.44 5.54

For B–G dampers, the slip distance x Gc for the minimum in normal loads and the σ R of the system with B–B dampers
value of µ R , which is the same as the variance of the tuned monotonically decreases with the increase in the slip distance
system response, is close to that for σ R (Fig. 2). It implies that x Bc until the B–B dampers become fully stuck. These results
the optimal normal load for the tuned system also minimizes show that the mistuning effect is relatively high in the system
σ R , which is a measure of the variations in response due to with B–B dampers. However, the system response is not very
mistuning. In contrast, this result is not true for B–B dampers sensitive to mistuning for the white noise excitation, as reflected
(Fig. 3). When normal loads of B–B dampers are increased by the values of σ R /µ R . This phenomenon is consistent with
from their optimal values for the tuned system, the energy the results of the linear system [12].
dissipated by friction dampers is decreased but the effective Using a Pentium P-400, the computation times required for
coupling stiffness is increased. Cha and Sinha [12] have shown the time integration, the Monte-Carlo simulation of ELM, and
that the sensitivity of the random response to mistuning is the analytical technique are 225 min, 40 min and 33 s for a given
reduced if the coupling stiffness increases. Hence, the effect slip distance, respectively. Hence, the analytical technique is
of the coupling stiffness becomes dominant with the increase highly efficient.
390 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396

Table 4
Statistics of the response variance (B–G damper, white noise excitation)
(a) µ R : Mean value of response variance (units: ×10−7 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Simulation 4.21 4.21 4.21 4.21 4.21 4.21 4.21 4.20 4.21 4.22
x Gc = 0.00015 m
Analytical 4.28
Simulation 3.36 3.36 3.36 3.36 3.36 3.36 3.36 3.35 3.36 3.36
x Gc = 0.00045 m
Analytical 3.36
Simulation 5.71 5.71 5.71 5.71 5.72 5.72 5.71 5.70 5.71 5.72
x Gc = 0.00195 m
Analytical 5.71
(b) σ R : Standard deviation of response variance (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Simulation 1.31 1.28 1.30 1.21 1.25 1.29 1.32 1.25 1.26 1.33
x Gc = 0.00015 m
Analytical 1.28
Simulation 0.85 0.83 0.84 0.78 0.80 0.83 0.85 0.81 0.81 0.86
x Gc = 0.00045 m
Analytical 0.84
Simulation 1.62 1.58 1.60 1.50 1.54 1.60 1.63 1.54 1.56 1.54
x Gc = 0.00195 m
Analytical 1.55

Table 5
Statistics of the response variance (B–B damper, white noise excitation)
(a) µ R : Mean value of response variance (units: ×10−7 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Simulation 4.48 4.44 4.43 4.41 4.43 4.43 4.42 4.40 4.45 4.50
x Bc = 0.00012 m
Analytical 4.52
Simulation 3.23 3.18 3.18 3.17 3.18 3.19 3.17 3.15 3.19 3.25
x Bc = 0.00045 m
Analytical 3.21
Simulation 5.50 5.49 5.48 5.47 5.49 5.49 5.47 5.46 5.49 5.51
x Bc = 0.00254 m
Analytical 5.45
(b) σ R : Standard deviation of response variance (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
Simulation 3.83 3.78 3.84 3.58 3.73 3.82 3.94 3.69 3.68 3.87
x Bc = 0.00012 m
Analytical 3.90
Simulation 3.28 3.07 3.10 2.88 3.03 3.08 3.18 3.00 2.93 3.28
x Bc = 0.00045 m
Analytical 3.30
Simulation 2.02 1.95 1.99 1.86 1.91 1.97 2.03 1.92 1.91 2.05
x Bc = 0.00254 m
Analytical 1.96

3.2. Narrow band excitation perfectly tuned system. The peak with the lowest ω F
The band-pass filter is used to generate the narrow band corresponds to the resonance in the vibration mode with 0◦
excitation with ξ F equal to 0.01. The actual force on a blade, interblade phase angles. The accuracy of ELM is examined
gi (t), can be increased by as much as the quality factor of the again by comparison with the results from time integration.
band-pass filter, Q f , because the band-pass filter amplifies the Four different cases of slip distances are chosen, and the results
white noise excitation, w(t), near the resonance frequency of are presented in Tables 6 and 7. Note that the frequency ω F
the filter, ω F . The quality factor [13] can be defined as is chosen to correspond to the peak response for each slip
distance. The error due to ELM is increased compared to the
1 results for the white noise excitation. However, the accuracy
Qf = . (27)
2 ξF of ELM for the narrow band random excitation is sufficient to
Therefore, the intensity of the external white noise excitation, obtain the optimal normal load, because the trend of ELM is
Q 0 , is chosen to be 10−4 N2 . The standard deviation of each still fairly close to that of time integration.
blade’s modal stiffness is taken to be 8000 N/m. With the same values of slip distances and ω F as used in time
The analytical results obtained from Eq. (20) are plotted integration (Tables 6 and 7), the analytical results from Eq. (20)
in Figs. 4 and 5, for B–G and B–B dampers, respectively. are compared with those from the Monte Carlo simulations of
The multiple peaks in these figures are related to resonance Eq. (14). The results are presented in Tables 8 and 9 for B–G
conditions, and they occur near natural frequencies of the and B–B dampers, respectively. The analytical results compare
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 391

Table 6
Time simulation of the response variance (B–G damper, narrow band excitation)
(a) Tuned system: E[x 2 ] (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Gc = 0.00004 m Time simulation 1.11 1.14 1.37 1.52 1.34 1.00 0.91 1.07 1.07 1.09
ω F = 6300 rad/s ELM 1.26
x Gc = 0.00008 m Time simulation 0.61 0.69 0.88 0.78 0.68 0.58 0.56 0.60 0.57 0.53
ω F = 6240 rad/s ELM 0.81
x Gc = 0.00022 m Time simulation 1.43 1.57 1.56 1.35 1.50 1.60 1.65 1.59 1.47 1.34
ω F = 6480 rad/s ELM 1.60
x Gc = 0.0004 m Time simulation 2.08 2.29 2.49 2.27 2.39 2.36 2.45 2.67 2.55 2.09
ω F = 6540 rad/s ELM 2.37
(b) A mistuned system: E[x 2 ] (units: ×10−8 m2 )
δk1 = 13,162 N/m, δk2 = 6678 N/m, δk3 = −7370 N/m, δk4 = −22 N/m, δk5 = −10, 612 N/m, δk6 = −1750 N/m, δk7 = 9920 N/m,
δk8 = −8882 N/m, δk9 = 1708 N/m, δk10 = −2832 N/m
Blade number 1 2 3 4 5 6 7 8 9 10
x Gc = 0.00004 m Time simulation 0.85 1.06 1.55 1.42 1.50 0.96 0.77 1.30 1.11 1.11
ω F = 6300 rad/s ELM 1.02 1.14 1.41 1.17 1.40 1.23 1.02 1.45 1.26 1.34
x Gc = 0.00008 m Time simulation 0.42 0.61 1.05 0.88 0.90 0.59 0.47 0.73 0.58 0.52
ω F = 6240 rad/s ELM 0.57 0.66 0.94 0.89 1.05 0.84 0.67 0.96 0.84 0.81
x Gc = 0.00022 m Time simulation 1.07 1.39 1.86 1.30 1.64 1.52 1.33 1.84 1.43 1.31
ω F = 6480 rad/s ELM 1.29 1.46 1.80 1.42 1.77 1.56 1.27 1.82 1.57 1.70
x Gc = 0.0004 m Time simulation 1.67 2.09 2.72 2.01 2.46 2.16 1.95 3.00 2.45 2.13
ω F = 6540 rad/s ELM 1.94 2.18 2.65 2.10 2.58 2.29 1.90 2.68 2.36 2.53

Table 7
Time simulation of the response variance (B–B) damper, narrow band excitation
(a) Tuned system: E[x 2 ] (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Bc = 0.00002 m Time simulation 1.77 1.87 2.31 2.29 1.79 1.42 1.84 2.34 2.38 2.13
ω F = 6220 rad/s ELM 2.15
x Bc = 0.00008 m Time simulation 1.50 1.47 1.79 1.97 1.70 1.25 1.23 1.59 1.76 1.70
ω F = 6300 rad/s ELM 1.63
x Bc = 0.00016 m Time simulation 2.09 1.76 1.90 2.36 2.29 1.93 1.58 1.78 2.25 2.40
ω F = 6380 rad/s ELM 1.78
x Bc = 0.00023 m Time simulation 2.50 2.02 2.18 2.83 2.84 2.34 1.85 2.07 2.69 2.92
ω F = 6380 rad/s ELM 2.06
(b) A mistuned system: E[x 2 ] (units: ×10−8 m2 )
δk1 = 13,162 N/m, δk2 = 6678 N/m, δk3 = −7370 N/m, δk4 = −22 N/m, δk5 = −10, 612 N/m, δk6 = −1750 N/m, δk7 = −9920 N/m,
δk8 = −8882 N/m, δk9 = 1708 N/m, δk10 = −2832 N/m
Blade number 1 2 3 4 5 6 7 8 9 10
x Bc = 0.00002 m Time simulation 1.32 1.62 2.51 2.26 2.01 1.39 1.60 2.75 2.49 2.13
ω F = 6220 rad/s ELM 1.65 1.85 2.37 2.15 2.45 2.05 1.74 2.48 2.30 2.26
x Bc = 0.00008 m Time simulation 1.24 1.38 1.91 1.88 1.80 1.24 1.18 1.84 1.84 1.66
ω F = 6300 rad/s ELM 1.44 1.52 1.66 1.47 1.69 1.53 1.31 1.65 1.65 1.77
x Bc = 0.00016 m Time simulation 1.90 1.84 2.08 2.15 2.27 1.90 1.54 2.08 2.32 2.43
ω F = 6380 rad/s ELM 1.74 1.61 1.46 1.29 1.75 1.70 1.31 1.39 1.47 1.88
x Bc = 0.00023 m Time simulation 2.23 2.14 2.45 2.58 2.72 2.25 1.85 2.51 2.81 2.91
ω F = 6380 rad/s ELM 1.99 1.73 1.47 1.36 1.95 1.92 1.41 1.40 1.54 2.09

well with the results from Monte Carlo simulation. In general, dampers is not significant in the first mode of vibration (Figs. 4
the errors in the analytical results of B–G dampers are smaller and 5). If ω F is chosen to be close to the first natural frequency
than those of B–B dampers. of the system (6141.6 rad/s), the blades will oscillate primarily
While the performance of B–G dampers is not directly in the first mode of vibration with 0◦ interblade phase angles
related to the vibration mode of the system, the effect of B–B and the relative motion between adjacent blades will be small.
392 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396

Table 8
Statistics of the response variance (B–G damper, narrow band excitation)
(a) µ R : Mean value of response variance (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Gc = 0.00004 m Simulation 1.23 1.23 1.24 1.23 1.24 1.24 1.23 1.22 1.24 1.24
ω F = 6300 rad/s Analytical 1.26
x Gc = 0.00008 m Simulation 0.82 0.82 0.82 0.82 0.83 0.83 0.82 0.81 0.82 0.83
ω F = 6240 rad/s Analytical 0.81
x Gc = 0.00022 m Simulation 1.55 1.55 1.56 1.55 1.56 1.56 1.55 1.54 1.56 1.56
ω F = 6480 rad/s Analytical 1.60
x Gc = 0.0004 m Simulation 2.30 2.30 2.31 2.30 2.31 2.32 2.31 2.28 2.31 2.32
ω F = 6540 rad/s Analytical 2.37
(b) σ R : Standard deviation of response variance (units: ×10−9 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Gc = 0.00004 m Simulation 1.47 1.48 1.45 1.41 1.43 1.47 1.52 1.43 1.47 1.49
ω F = 6300 rad/s Analytical 1.48
x Gc = 0.00008 m Simulation 1.66 1.61 1.65 1.53 1.59 1.62 1.67 1.57 1.60 1.70
ω F = 6240 rad/s Analytical 1.58
x Gc = 0.00022 m Simulation 1.97 1.96 1.90 1.89 1.90 1.94 2.00 1.91 1.93 1.95
ω F = 6480 rad/s Analytical 1.77
x Gc = 0.0004 m Simulation 2.79 2.82 2.72 2.72 2.71 2.77 2.85 2.73 2.79 2.79
ω F = 6540 rad/s Analytical 2.60

Table 9
Statistics of the response variance (B–B damper, narrow band excitation)
(a) µ R : Mean value of response variance (units: ×10−8 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Bc = 0.00002 m Simulation 2.08 2.08 2.08 2.05 2.06 2.07 2.05 2.04 2.09 2.10
ω F = 6220 rad/s Analytical 2.15
x Bc = 0.00008 m Simulation 1.66 1.58 1.50 1.51 1.58 1.58 1.52 1.49 1.59 1.57
ω F = 6300 rad/s Analytical 1.63
x Bc = 0.00016 m Simulation 1.79 1.49 1.30 1.44 1.70 1.70 1.45 1.29 1.49 1.80
ω F = 6380 rad/s Analytical 1.78
x Bc = 0.00023 m Simulation 2.01 1.55 1.29 1.53 1.93 1.93 1.54 1.29 1.55 2.02
ω F = 6380 rad/s Analytical 2.06
(b) σ R : Standard deviation of response (units: ×10−9 m2 )
Blade number 1 2 3 4 5 6 7 8 9 10
x Bc = 0.00002 m Simulation 2.79 2.70 2.70 2.58 2.62 2.64 2.79 2.61 2.75 2.87
ω F = 6220 rad/s Analytical 2.92
x Bc = 0.00008 m Simulation 1.52 1.37 1.30 1.33 1.38 1.38 1.40 1.31 1.34 1.52
ω F = 6300 rad/s Analytical 1.43
x Bc = 0.00016 m Simulation 1.37 1.45 1.14 1.29 1.47 1.42 1.31 1.15 1.43 1.35
ω F = 6380 rad/s Analytical 1.44
x Bc = 0.00023 m Simulation 1.36 1.56 1.10 1.40 1.64 1.60 1.42 1.09 1.54 1.35
ω F = 6380 rad/s Analytical 1.53

In this situation, B–B dampers do not slip significantly and Note that there are six distinct natural frequencies of the tuned
the vibration energy is hardly dissipated. As mentioned before, system.
the number of peaks in plots for µ R is related to the natural Using a Pentium P-400, computation times required for the
frequencies of the tuned system. However, only four peaks are time integration, the Monte Carlo simulation of ELM, and the
shown in Figs. 4a and 5a, because the bandwidth of the random analytical technique are 245 min, 61.7 min and 53 s for a given
excitation is not very narrow with ξ F equal to 0.01. Therefore, slip distance and ω F , respectively. Therefore, the computational
the first and second peaks are merged, and it happens again for efficiency of the analytical technique is extremely high. In
peaks corresponding to the fifth and sixth natural frequencies. fact, without techniques developed in Section 2 for white noise
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 393

and narrow band excitations, it would have been cumbersome


(or almost infeasible) to generate Figs. 2–5, do parametric
studies, and draw conclusions about the selection of slip loads
of friction dampers.

Fig. 4b. σ R as a function of x Gc and ω F (B–G damper, narrow band


excitation).

Fig. 3a. µ R as a function of x Bc (B–B damper, white noise excitation).

Fig. 5a. µ R as a function of x Bc and ω F (B–B damper, narrow band


excitation).

Fig. 5b. σ R as a function of x Bc and ω F (B–B damper, narrow band excitation).


Fig. 3b. σ R as a function of x Bc (B–B damper, white noise excitation).
4. Conclusions
This paper represents a computationally efficient solution of
a realistic problem in gas turbine design and analysis. Using
this technique, which is based on the equivalent linearization
method (ELM), it is now feasible to optimally select design
parameters such as slip loads etc. to minimize the vibratory
response of a frictionally damped gas turbine blade subjected
to random excitation.
For white noise excitation, it is found that the normal load of
each B–G damper for the minimum value of the mean (µ R ) of
the response variance is close to that for the minimum value of
the standard deviation (σ R ) of the response variance. However,
σ R of the system with B–B dampers monotonically decreases
Fig. 4a. µ R as a function of x Gc and ω F (B–G damper, narrow band with an increase in normal load, while the minimum value of
excitation). µ R does exist. Also, the impact of mistuning is not very high.
394 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396
!
For the narrow band random excitation, the performance of y −ρ Bi x Bc σẋ Bi 2
−x Bc
B–G dampers is not directly related to the vibratory mode of k Bi = √ exp
σ y2Bi 2π 2σ y2Bi
the system, which is related to the bandwidth of excitation.   
However, the performance of B–B dampers is influenced by the ρ Bi x Bc
bandwidth of excitation, because the effect of B–B dampers is × 1 + erf  q 
not significant in the first mode of vibration with 0◦ interblade σ y Bi 2(1 − ρ Bi2 )
q
phase angles. σẋ Bi 1 − ρ Bi 2 2
!
−x Bc
− exp (A8)
Appendix A. Expressions for equivalent linear coefficients σ y Bi π 2(1 − ρ Bi
2 )σ 2
y Bi
of friction dampers
where
Assuming a joint Gaussian probability distribution of
the multidimensional response process, the equivalent linear σẋ2i = E[ẋi2 ], (A9a)
coefficients are obtained as follows [7]:
σ y2Gi = 2
E[yGi ], (A9b)
 
K x x E[ẋi yGi ]
x
cGi =
G Gc
√ erf c 
Gc
(A1) ρGi = (A9c)
σẋi σ yGi

σẋi 2π
q
σ yGi 2(1 − ρGi ) 2

!
2
! σẋ2Bi = E[ẋi2 ] + E[ẋi+1
2
] − 2E[ẋi ẋi+1 ], (A10a)
x x Gc K G x Gc −x Gc
k Gi = K G erf √ − √ exp σ y2Bi = E[y Bi
2
], (A10b)
σ 2 σ yGi 2π 2σ y2Gi
 yGi 
E[ẋi y Bi ] − E[ẋi+1 y Bi ]
ρGi x Gc ρ Bi = . (A10c)
× erf c  q  (A2) σẋ Bi σ y Bi
σ yGi 2(1 − ρGi 2 )
y y
" !# Coefficients cGi and c Bi require numerical integrations.
y 1 x Gc However, they can be approximated for small values of ρGi and
cGi = 1 + erf √
2 σ yGi 2 ρ Bi as,
 
ρ ν
Z ∞
1 2 Gi
e−ν erf  q  dν " !#
−√ (A3) 1 x Gc
π xc /σ yGi √2 1 − ρGi 2 y
cGi = 1 + erf √
2 σ yGi 2
! !
y −ρGi x Gc σẋi −x Gc2
1 ρGi −x Gc2
k Gi = √ exp −√ q exp (A11)
σ y2Gi 2π 2σ y2Gi π 1 − ρ2 2
2σ yGi
   Gi
ρ Gi x Gc
" !#
× 1 + erf  q  y 1 x Bc
σ yGi 2(1 − ρGi ) 2 c Bi = 1 + erf √
q
2 σ y Bi 2
!
σẋi 1 − ρGi 2 ρ Bi 2
!
2
−x Gc 1 −x Bc
− exp (A4) −√ q exp . (A12)
σ yGi π 2(1 − ρGi 2 )σ 2 π 1 − ρ2 2σ y2Bi
yGi Bi
and
∂A
 
K B x Bc x Bc Appendix B. Expressions for elements of ∂ ki
c xBi = √ erf c   (A5)
σẋ Bi 2π
q
σ y Bi 2(1 − ρ Bi 2 )
∂A
! ! In the case of B–G dampers, the elements of ∂ ki in Eq. (18)
2
−x Bc
x x Bc K B x Bc can be represented as
k Bi = K B erf √ − √ exp
σ 2 σ y Bi 2π 2σ y2Bi
 y Bi    !
ρ Bi x Bc ∂cGi
x
K G x Gc x Gc −1 ∂σ ẋ i
× erf c  (A6) = √ erf c  
∂k j σẋ2i ∂k j
 q
2π σ yGi 2(1 − ρGi 2 )
q
σ y Bi 2(1 − ρ Bi 2 )
!
" !# K G x Gc2 −x Gc 2
y 1 x Bc +√ exp
c Bi = 1 + erf √ 2π σẋi σ y2Gi (1 − ρGi2 ) 2σ y2Gi (1 − ρGi 2 )
2 σ y Bi 2
   
ρ ν ∂σ ρ ∂ρ
Z ∞
1 2σ
q
2 Bi y y Gi Gi
−√ e−ν erf  q  dν (A7) ×  2(1 − ρGi 2 ) Gi
− q Gi  (B1)
π xc /σ y Bi √2 1 − ρ Bi
2 ∂k j 2(1 − ρ ) 2 ∂k j
Gi
D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396 395
√ !
∂k Gi
x
− 2K G x Gc −x Gc2
∂σ yGi and K G :
= √ 2 exp
∂k j π σ yGi 2σ y2Gi ∂k j
∂σẋ2i
!
  ∂σẋi 1
K G x Gc ρ x = (B5a)
−√ erf c 
Gi Gc  ∂k j 2σẋi ∂k j
2π σ y2Gi
q
σ yGi 2(1 − ρGi 2 )
∂σ y2Gi
!
∂σ yGi 1
= (B5b)
! !
2
x Gc −x Gc2
∂σ yGi ∂k j 2σ yGi ∂k j
× − 1 exp
σ yGi
2 2
2σ yGi ∂k j
∂ρG i 1 ∂ (E [ẋi yGi ])
! =
2
K G x Gc −x Gc2 ρGi2 x2
Gc ∂k j σẋi σ yGi ∂k j
+ exp −
π σ y3Gi (1 − ρGi
2 ) 2σ y2Gi 2σ y2Gi (1 − ρGi
2 )
ρG i ∂σẋi ρG i ∂σ yGi
− − (B5c)
σẋi ∂k j σ yGi ∂k j
 
σy ∂ρGi 2 ∂σ yGi 
q
×  q Gi − ρGi 1 − ρGi (B2)
∂k ∂k j 1 ∂σẋi 1 ∂σẋi+1
2 2
1 − ρGi2 j ∂σẋ Bi
= +
y
! ∂k j 2σẋ Bi ∂k j  2σẋ Bi ∂k j
∂cGi −x Gc −x Gc2
∂σ yGi 1 ∂ E ẋi ẋi+1
= √ exp
∂k j 2π σ yGi
2 2
2σ yGi ∂k j −
σẋ Bi ∂k j
(B6a)
!
2
−x Gc
1 ∂σ y2Bi
!
∂σ y Bi 1
−q  exp 2σ 2 = (B6b)
π 1 − ρ2 Gi
yGi ∂k j 2σ y Bi ∂k j
 !
∂ρ Bi ∂ (E [ẋi y Bi ]) ∂ E ẋi+1 y Bi
! 
1 ∂ρGi ρGi x Gc2
∂σ yGi 1
× + (B3) = −
1 − ρGi 2 ∂k
j σ y3Gi ∂k j ∂k j σẋ Bi σ y Bi ∂k j ∂k j
   ρ Bi ∂σẋ Bi ρ Bi ∂σ y Bi
y
∂k Gi −x Gc  ρ x − − . (B6c)
= √ 1 + erf 
Gi Gc  σẋ Bi ∂k j σ y Bi ∂k j
∂k j 2π σ y3Gi
q
σ yGi 2(1 − ρGi 2 )
!
2
−x Gc ∂ρGi ∂ σẋi
× exp σẋi σ yGi + ρGi σ yGi Appendix C. Expressions for µ R and σ R2
2
2σ yGi ∂k j ∂k j
! !
2
x Gc ∂ σ yGi 1. The mean of the response correlation (µ R )i
+ ρGi σẋi −2
σ yGi
2 ∂k j
! " #
ρGi σẋi x Gc
2 −x Gc 2 ρGi
2 x2 n
X ∂ Pi,i
− exp − Gc
(µ R )i = E P0i,i + δkl = P0i,i (C1)
π σ y4Gi (1 − ρGi2 ) 2σ y2Gi 2σ y2Gi (1 − ρGi 2 )
l=1
∂kl
 
σy ∂ ρGi 2 ∂σ yGi 
because
q
×  q Gi − ρGi 1 − ρGi
1 − ρGi 2 ∂k j ∂k j
E[δkl ] = 0. (C2)
!
−x Gc 2
1 2. The standard deviation of the response correlation (σ R )i
− exp
π σ y2Gi 2(1 − ρGi 2 )σ 2
yGi
 is:
2 ∂σẋi − σq ẋi σ yGi ρGi
q
× σ yGi 1 − ρGi (σ R2 )i = E[Pi,i
2
] − (µ R )i2 (C3)
∂k j 1 − ρGi 2
! where
2
x Gc ∂ρGi
× 1+
∂k j
 !2 
(1 − ρGi 2 )σ 2
yGi n
X ∂ Pi,i
   2
E[Pi,i ] = E  P0i,i + δkl 
q x 2
∂ σ l=1
∂kl
y
+ σẋi − 1 − ρGi 2 + q Gc  Gi 
.
∂k
 !2 
σ 2 1−ρ
yGi
2
Gi
j n
X ∂ Pi,i
= E[(P0 )i,i
2
]+ E δkl 
(B4)
l=1
∂kl
y y
∂c x ∂k x ∂c ∂k
" !#
n
For B–B dampers, ∂kBij , ∂kBij , ∂kBij and ∂kBij can be obtained by X ∂ Pi,i
+ 2E P0i,i δkl (C4)
substituting σẋ Bi , σ y Bi , ρ Bi , x Bc and K B into σẋi , σ yGi , ρGi , x Gc l=1
∂kl
396 D. Cha, A. Sinha / Probabilistic Engineering Mechanics 21 (2006) 384–396

and [4] Sogliero G, Srinivasan AV. Fatigue life estimates of mistuned blades via
a stochastic approach. AIAA Journal 1980;18(83):318–23.
E[(P0 )i,i
2
] = (P0 )i,i
2
(C5) [5] Minkiewicz G, Russler P. Dynamic response of low aspect ratio blades in a
 two stage transonic compressor. In: 33rd AIAA/ASME/SAE/ASEE/SME
!2  joint propulsion conference and exhibit. AIAA paper 97-3284; 1997.
n
∂ n 
∂ Pi,i 2

X Pi,i
X
E  δkl  2
= E[δk ] (C6) [6] Cha D, Sinha A. Computation of the optimal normal load of a friction
l=1
∂kl l=1
∂kl damper under different types of excitation. ASME 99-GT-413. ASME
Journal of Engineering for Gas Turbines and Power 2003;125(4):1042–9.
"
n
!# [7] Asano K, Iwan WD. An alternative approach to the random response
X ∂ Pi,i of bilinear hysteretic systems. Journal of Earthquake Engineering and
E P0i,i δkl = 0. (C7)
l=1
∂kl Structural Dynamics 1984;12:229–36.
[8] Sinha A. Friction damping of random vibration in gas turbine engine
airfoils. International Journal of Turbo and Jet Engines 1990;7:95–102.
References [9] Roberts JB, Spanos PD. Random vibration and statistical linearization.
Chichester: John Wiley & Sons; 1990.
[1] Chen S, Sinha A. Probabilistic method to compute the optimal slip load [10] MATLAB Manual. Massachusetts: The MathWorks, Inc.; 1995.
for a mistuned bladed disk assembly with friction dampers. ASME Journal [11] Socha L, Soong TT. Linearization in analysis of nonlinear stochastic
of Vibration and Acoustics 1990;112:214–21. systems. Applied Mechanics Reviews 1991;44:399–422.
[2] Griffin JH, Sinha A. The interaction between mistuning and friction in the [12] Cha D, Sinha A. Statistics of response of a mistuned bladed disk assembly
forced response of bladed disk assemblies. ASME Journal of Engineering subjected to white noise and narrow band excitation. ASME 98-GT-379.
for Gas Turbines and Power 1985;107:205–11. ASME Journal of Engineering for Gas Turbines and Power 1999;121(4):
[3] Whitehead DS. The analysis of blade vibration due to random excitation. 710–7.
Reports and Memoranda No. 3253, Cambridge Univ.; 1960. [13] Rao SS. Mechanical vibrations. Addison-Wesley Co.; 1995.

Anda mungkin juga menyukai