http://www.paper.edu.

cn

Finite Elements in Analysis and Design 43 (2007) 1068 1081

www.elsevier.com/locate/nel

Identication of crack in a rotor system based on wavelet nite

element method
Jiawei Xiang a, , Xuefeng Chen b , Qiuyun Mo a , Zhengjia He b
a School of Mechantronic Engineering, Guilin University of Electronic Technology, Guilin 541004, PR China
b School of Mechanical Engineering, Xian Jiaotong University, The State Key Laboratory for Manufacturing Systems Engineering, Xian 710049, PR China

Received 10 April 2006; received in revised form 24 March 2007; accepted 6 July 2007
Available online 17 August 2007

Abstract
The dynamics and diagnosis of cracked rotor have been gaining importance in recent years. In the present study a model-based crack
identication method is proposed for estimating crack location and size in shafts. The rotor system has been modeled using nite element
method of B-spline wavelet on the interval (FEM BSWI), while the crack is considered through local stiffness change. Based on Rayleigh
beam theory, the inuences of rotatory inertia on the exural vibrations of the rotor system are examined to construct BSWI Rayleigh beam
element. The slender shaft and stiffness disc are modeled by BSWI RayleighEuler beam element and BSWI RayleighTimoshenko beam
element, respectively. Then the crack identication forward and inverse problems are solved by using surface-tting technique and contour-
plotting method. The experimental examples are given to verify the validity of the BSWI beam element for crack identication in a rotor
system. From experimental results, the new method can be applied to prognosis and quantitative diagnosis of crack in a rotor system.
2007 Elsevier B.V. All rights reserved.

Keywords: B-spline wavelet on the interval; Rayleigh beam element; Rotor system; Model-based method; Crack identication

1. Introduction changes in the spatial variation of the vibration response to de-

The interest in the ability to monitor a structure and detect
crack at the stage of the structure still has sufcient integrity to
allow repair, is pervasive throughout the civil, mechanical and
aerospace engineering communities. The vibration-based crack
identication methods include non-model-based and model-
based methods have been rapidly expanding over the last few
years [1]. The wavelets transform method for damage detec-
tion is one of the non-model-based crack identication meth-
ods [2]. The main advantage of the wavelet transform is its
ability to provide information simultaneously in time and scale
with adaptive windows. Wavelets methods combine the struc-
tural dynamic parameters such as modal frequencies, modal
shape and modal damp, etc. with the determined Lipschitz ex-
ponent to detect the damage location and depth. Loutridis and
Douka [3,4] extend wavelet transform to identication of cracks
in cantilever beams and plate structures by using the sudden
Corresponding author.
E-mail address: wxw8627@163.com (J. Xiang).
0168-874X/$ - see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nel.2007.07.001
termine the crack location. The intensity factor is also estab-
lished which relates the size of the cracks to the coefcients of
the wavelet expansion. Han and Ren [5] present wavelet packet
energy rate index (WPERI) method to detect damage in beam-
like structures. Zhu [6] propose a new method for crack iden-
tication of bridge beam structures under a moving load based
on wavelet analysis. However, the applications of the above-
mentioned wavelet methods are mostly in the elds of civil
engineering, such as beam and beam-like structures.
Vibration behavior of cracked structures, in particular
cracked shafts has received considerable attention in the last
three decades [79]. The model-based method for damage and
crack identication in structural components has acquired an
important role in recent years. The methods assume that the
monitored structure responds in some predetermined manner
that can be accurately discretized by nite element analysis,
such as the response described by EulerBernoulli beam the-
ory. Kisa [10] integrated the nite element method (FEM) and
component mode synthesis for cracked Timoshenko beam.
Lee [11] using the lowest four natural frequencies and cracked


J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081
FEM model to detect structural crack. Saavedra and Cuitino
[12] presented an identication method combined FEM with
the additional exibility of crack for a cracked free-free beam
and a U-frame. Lele and Maiti [13], and Patil and Maiti [14], re-
spectively, investigated crack identication techniques based on
eight-node iso-parametric elements to make more efcient cal-
culation for single and multiple cracks identication in beams.
Dutta [15] using an adaptive h-version FEM for bridge damage
detection in order to control the discretization error because the
traditional nite elements are impotent to describe the singular
behavior of cracks and numerous elements are needed for nu-
merical computation. However, for the model-based crack iden-
tication method in a rotor system, fewer contributions were
given. Sekhar [16] proposed a model-based method which re-
placed the fault-induced change of the shaft by equivalent loads
in the nite element model. Dong [17] presented a parameter
identication method of a rotor with an open crack by using the
rst two natural frequencies and the rst mode shape, while the
shaft is modeled by an EulerBernoulli beam without regard to
the cross-section rotatory inertia.
A crack in a vibration structure results in stiffness decrease,
stress concentration, and local exibility, which are func-
tions of the crack location and size. Hence, crack change the
dynamic characteristics of structures, which can be applied to
estimate crack in structures. Because the rst three natural fre-
quencies can be easily and cheaply acquired in practice, they
are usually employed for crack identication. In the diagnosis
of structural crack fault, the model-based forward and inverse
problems based on FEM are studied for structural modal anal-
ysis. And the crack is replaced by a massless rotational spring
with a computable stiffness as mentioned in [18] and the rst
three frequencies response functions of normalized crack lo-
cation and size are approximated by means of surface-tting
techniques [19]. Hence, the rst three metrical frequencies
are employed as inputs of the frequencies response functions.
The intersection of the three frequencies contour lines [20]
predicted the normalized crack location and size.
The desirable advantages of WFEM are multi-resolution
properties and various basis functions for structural analysis
[21]. By means of two-scale relations of wavelets, the scale
adopted can be changed freely according to requirements to
improve analysis accuracy. Recently, 1D Daubechies wavelet
Euler beam element had been constructed by Ma [22] and the
2D Daubechies wavelet element for thin plate-bending prob-
lems had also been constructed by Chen [23]. However, for
discrete wavelets lacking of the explicit function expression,
traditional numerical integrals such as Gauss integrals cannot
provide desirable precision, the key problem is to calculate
connection coefcients, which is a complex process. Han [24]
extended the wavelet-based nite element method (WFEM)
to thick plate by using mixed variational principle. Xiang
[2527] construct some classes of 1D and 2D BSWI elements
for structural analysis. He [28] summarized the advanced in
theory study and engineering application of wavelet nite
element, especially for the model-based structural crack iden-
tication problems. Li [19] presented Daubechies WFEM for
crack identication with good performance. By using WFEM
http://www.paper.edu.cn
1069
model, Chen [29] gave the experimental identication results
in cantilever beams with small identication errors. Xiang [30]
proposed a BSWI Euler beam model for detecting crack in a
beam, the simulation and experimental results show the high
performance of the BSWI Euler beam element.
In the present work, BSWI RayleighEuler beam element
and BSWI RayleighTimoshenko beam element are con-
structed for the rst time to analysis and make a more effec-
tively identication in a rotor system. The experimental studies
show that the common used Euler beam element without con-
sider the cross-section rotatory inertia in some technical litera-
ture cannot provide reasonable frequencies in the present rotor
system. In addition, the model-based method by using wavelet
nite element is extent to identication cracks in shafts.
The outline of this paper is as follows. In Section 2, we
present a brief review of the basic theory of BSWI bases. Then
BSWI RayleighEuler and BSWI RayleighTimoshenko beam
elements are constructed in details. In Section 3, the forward
and inverse problems of crack identication method are given.
The effects of the crack location and depth are also analyzed.
Some experimental results are given in Section 4. The perfor-
mance of the proposed elements in dealing with the identica-
tion of crack in shaft is also veried. The inuence of bearing
stiffness is discussed in Section 5 for further researches.
2. BSWI Rayleigh beam element for structural modal
analysis
Classical approaches to wavelet construction deal with mul-
tiresolution analysis (MRA) on the entire real axis R or a
function space L2 (R). Sometimes the numerical oscillations
will be occurred when the boundary value problems are solved
by WFEM [31]. To overcome this limitation, Chui, Quak and
Goswami constructed BSWI functions, and presented a fast de-
composition and reconstruction algorithm [32,33]. All the scal-
j j
ing functions m,k () and wavelets m,k () for order m = 4 at
the scale j = 3 are shown in Fig. 1.
In the present work, a continuous model is presented for
vibration analysis and crack identication of a static (non-
rotating) shaft with an open crack, which is based on two
assumptions that the cracked shaft is a Rayleigh beam [34,35]
with circular cross-section and that the cracked region is
modeled as a local stiffness change with fracture mechanics
methods. In this section, two wavelet-based Rayleigh beam
elements are given for the model-based crack identication,
i.e. BSWI RayleighEuler beam element for slender shaft and
BSWI RayleighTimoshenko beam element for disc.
2.1. BSWI RayleighEuler beam element
The static slender shaft is modeled by a RayleighEuler beam
considering the effects of the cross-section inertia, the elemental
potential energy U e can be written as
  2
le EI z d2 w
U =
e
dx, (1)
0 2 dx 2
 http://www.paper.edu.cn
1070 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081

Fig. 1. m = 4, j = 3: (a) scaling functions and (b) wavelets on the interval [0, 1].

y
le y le
wr +1
w1
w2 w3 wr -1 wr 1 2 n n n +1
1 x
1
x 2 n+1
1 2 3 r-1 r r+1  z w1 w2 wn
r+1 wn+1
z

Fig. 2. The layout of elemental nodes and the corresponding DOF: (a) BSWI RayleighEuler beam element; (b) BSWI RayleighTimoshenko beam element.

j j j
where E is the Youngs modulus, Iz is the moment of inertia, where ={m,m+1 ()m,m+2 () . . . m,2j 1 ()} is the col-
w(x, t) is the transverse displacement, and le is the elemental umn vector combined by the BSWI scaling functions for order
length. m at the scale j, and the C1 type transformation matrix Teb is
The elemental kinetic energy T e of the RayleighEuler beam given by
allowing for the rotatory inertia effect can be expressed as 
 le    le   1 d(1 )
A jw 2 Iz j 2 Tb =
e
(1 ) (2 ) . . . (r )(r+1 )
Te = dx + dx, (2) le d
0 2 jt 0 2 jt
 1
where  is the density, A is the area of the cross-section, (x, t) 1 d(r+1 ) T
. (5)
is the rotation of the beam section due to bending and can be le d
given by  = dw/dx.
In order to satisfy the displacement and slope compatibility Substituting Eq. (4) into Eqs. (1) and (2), respectively, we obtain
among neighboring elements, the elemental edge nodes should 1 e T e e

U = 2 (w ) Kb (w ),
e
include the transverse displacements and slopes [2527]. There-
fore, elemental degree-of-freedom (DOF) on each edge node  e T  e     (6)
1 jwe T e jwe
in physical space should include one slope and one transverse T e = 1 jw

Mb e jw
+ Mr ,
displacement DOF. The layout (here, we evenly divide the stan- 2 jt jt 2 jt jt
dard solving domain e into r = 2j + m 4 segments, the node where the bending stiffness matrix Kbe is
number = r + 1, the elemental whole DOF = r + 3) of ele-
mental nodes is shown in Fig. 2(a). The element is abbreviated EI z e T 2,2 e
Kbe = (Tb )  Tb (7)
to BSWImj RayleighEuler beam element. le3
Then elemental physical DOF can be represented by
the translational mass matrix Mbe is
w = {w1 1 w2 w3 wr wr+1 r+1 } ,
e T
(3)
Mbe = Al e (Teb )T 0,0 Teb (8)
where 1 = (1/ le )(dw1 /d) and 2 = (1/ le )(dwr+1 /d) denote
rotation on each elemental endpoint. and the rotatory inertia mass matrix is
The unknown eld function w() can be expressed as
Iz e T 1,1 e
Mre = (Tb )  Tb , (9)
w = Teb we , (4) le
 http://www.paper.edu.cn
J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081 1071

in which The elemental physical DOFs can be represented by

 1 d2 T d2  ue = {w1 1 w2 2 wn n wn+1 n+1 }T (18)
2,2 = 2 2
d, (10)
0 d d
w and  can be independently interpolated by the BSWImj
 1 dT d scaling functions as
1,1
= d, (11)
0 d d
w = Te we ,  = Te e , (19)
 1
0,0 = T d. (12) where we = {w1 w2 . . . wn+1 }T , e = {1 2 . . . n+1 }T , the col-
0
umn vector combined by the BSWImj scaling functions is
The elemental Lagrangian function L is j j j
={m,m+1 () m,m+2 () . . . m,2j 1 ()}, and the C0 type
    transformation matrix Te is given by
1 1 jwe T e jwe
L = U e T e = (we )T Kbe (we ) Mb
2 2 jt jt Te = ([T (1 ) T (2 ) . . . T (n+1 )]T )1 . (20)
 e T  e
1 jw jw
Mre . (13) Substituting Eq. (19) into Eqs. (16) and (17), respectively, we
2 jt jt
obtain
Upon applying Hamiltons principle to the elemental La- e 1 e T e,1 e 1 e T e,2 e 1 e T e,3 e
U = 2 (w ) K (w )+ 2 (w ) K ( )+ 2 ( ) K (w )
grangian function L, we can obtain the elemental free vibra-


tion equation + 21 (e )T Ke,4 (e ),

 e T  e  e T  e
 2 e
j w T e = 1 jw

M e,b jw 1 j
+2 M e,r j
,
(Mb + Mr )
e e
+ Kbe we = 0 (14) 2 jt jt jt jt
jt 2 (21)
and the corresponding elemental free vibration frequency equa- where the elemental stiffness matrices can by solved by
tions is
GA e T 1,1 e
|Kbe 2 (Mbe + Mre )| = 0, (15) Ke,1 = (T )  (T ), (22)
kl e
where  is the angular frequency (rad/s), and the natural fre- GA e T 1,0 e
quency is f = /2 (Hz). Ke,2 = (T )  (T ), (23)
k

2.2. BSWI RayleighTimoshenko beam element Ke,3 = (Ke,2 )T , (24)

EI z e T 1,1 e GAl e e T 0,0 e
The static disc is modeled by a RayleighTimoshenko beam Ke,4 = (T )  (T ) + (T )  (T ) (25)
le k
considering the effects of the cross-section inertia and shear
deformation, the elemental potential energy U e can be written the translational mass matrix Me,b is
as
 le    le  2 Me,b = Al e (Te )T 0,0 (Te ), (26)
EI z d 2 GA dw
U =
e
dx +  dx, (16)
0 2 dx 0 2k dx and the rotatory inertia mass matrix is
where E is the Youngs modulus, Iz is the moment of iner- Iz e T 0,0 e
tia, w(x, t) is the transverse displacement, le is the elemental Me,r = (T )  (T ), (27)
le
length, Gis the shear modulus, A is the area of the cross-section,
k is the shear deformation coefcient (in the present work, we in which
suppose k = 10 
9 ), and (x, t) is the rotation of the beam section 1 dT
due to bending. 1,0 =  d (28)
d
The elemental kinetic energy T e of the RayleighTimoshenko 0

beam allowing for the rotatory inertia effect can be expressed as and 2,2 , 1,1 and 0,0 are shown in Eqs. (10)(12).
 le    le   Likewise, applying Hamiltons principle to the elemental La-
A jw 2 Iz j 2
Te = dx + dx, (17) grangian function L = U e T e , we can obtain the elemental
0 2 jt 0 2 jt free vibration equation
where  is the density, I is the moment of inertia. j2 we
Fig. 2(b) shows the nodal layout of BSWI Rayleigh  e,b   e,1  e
M 0 jt 2 K Ke,2 w
Timoshenko beam element and the corresponding DOFs. + = 0. (29)
e,r j 
2 e e,3 e,4 e
Each node has two DOFs, i.e. wi , i (i = 1, 2, . . . , n + 1), 0 M K K
n = 2j + m 2. jt 2
 http://www.paper.edu.cn
1072 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081

According to the layout of elemental physical DOFs Eq. (18), on beam elements are able to model the effects of an open
Eq. (29) can be rewritten as crack. Because the natural frequencies can be easily and cheaply
e acquired in practice and the linear rotational spring model can
M1,1 e
M1,2 M1,n+1
e
effectively describe open crack, the present work is based on
Me e M2,n+1
e
2,1 M2,2 2 e the open transverse crack in shafts and using the rst three
j u
natural frequencies to identify crack location and size.
jt 2
e Fig. 3 shows a simply supported rotor system with a crack in
Mn,1 e
Mn,2 Mn,n+1
e shaft (suppose the crack occurred on segment L2 ). A transverse
e
Mn+1,1 e
Mn+1,2 e
Mn+1,n+1 crack of depth  is considered on a shaft of diameter d1 (the
e corresponding radius is r1 ) as shown in Fig. 4. The shaft has
K1,1 e
K1,2 K1,n+1
e
Ke local exibility due to the crack, in many directions, depending
2,1
e
K2,2 K2,n+1
e
on the direction of the applied forces. Here, only bending de-
e
+ u = 0,
(30) formation will be considered. Axial forces which give coupling
e with transverse motions of the cracked shaft will not be con-
Kn,1 e
Kn,2 e
Kn,n+1
sidered. Therefore, the shaft is bent by a pure bending moment
e
Kn+1,1 e
Kn+1,2 e
Kn+1,n+1 and the additional angular deection of the shaft end relative
where to the other will be computed. The local stiffness kt due to the
  crack is [7]
e,1 e,2
kp,q kp,q
e
Kp,q = (p, q = 1, . . . , (n + 1)2 ), (31) Er 81
e,3
kp,q e,4
kp,q kt =
32(1 )
e,1 e,2 e,3 e,4
in which kp,q , kp,q , kp,q , kp,q are the entries of stiffness matri- 1
,
e,1 e,2 e,3
ces K , K , K , K , and e,4
 r1 
1(12 ) 2 (r12 2 )[ 0a() F 2 (/H )d]d
2

 e,b  r1 1(12 )
mp,q 0
e
Mp,q = (p, q = 1, . . . , (n + 1)2 ), (32) (34)
0 me,r
p,q
where E is the Youngs modulus,  is the Poissons ratio,
in which me,b e,r
p,q , mp,q are the corresponding entries of mass ma- = /2r1 denotes 
normalized crack size, a() = 2r1 (r1
trices Me,b and Me,r . r12 2 ), H = 2 r12 2 and the function F (/H ) can be
The corresponding elemental free vibration frequency equa- given by the experimental formula [37].
tion is
F (/H ) = 1.122 1.40(/H ) + 7.33(/H )2
|K  M | = 0,
e 2 e
(33)
13.08(/H )3 + 14.0(/H )4 . (35)
where  is the angular frequency (rad/s), and the natural fre- Eq. (34) is a function of normalized crack size only and
quency is f = /2 (Hz). can be computed by numerical integration.
= e/L2 denotes
The assembly procedures are similar to traditional FEM. To normalized crack location.
deal with the boundary conditions, the corresponding DOFs are The physical model and rotational spring model with stiffness
set to zero and eliminated from the equation. kt , are shown in Fig. 5.
The corresponding DOF of the left neighboring element
3. Crack identication and estimation method around crack are woleft ={ wj j }T and the corresponding DOF
of the right neighboring element are wright = {wj +1 j +1 }T .
3.1. Forward problem If we change the two entries of woleft to wleft = { j wj }T , the
corresponding row and column entries of the global stiffness
Friswell and Penny [36] compared the different approaches matrix K and mss matrix M should be exchanged their loca-
to crack modeling, and demonstrate that for structural health tions each other. The DOF of wleft and wright can be assembled
monitoring using low-frequency vibration, simple models based to { j wj +1 j +1 , . . . }T .

e crack

= e / L2

L1 L2
L1
d1

L3
L
d2

Fig. 3. Simply supported rotor system with a crack in shaft.

 http://www.paper.edu.cn
J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081 1073

Hence, we can assemble cracked stiffness submatrix KS into by using BSWI beam element. The solution of the eigenvalue
the global stiffness matrix easily. The global mass matrix of problem can then proceed as usual.
cracked rotor system is equal to the uncracked one. From now For the determination of the natural frequencies  for a
on, the cracked rotor system nite element model is constructed given crack location (determine the location of cracked stiff-
ness submatrix in global stiffness) and size (determine Kt ),

the normalized crack location
and size are given as input.
The relationship between the natural frequencies and the crack
parameters is
 d
a() fj = Fj ( ,
) (j = 1, 2, 3). (36)

Because the functions Fj (j = 1, 2, 3, . . .) are unknown and
o H the discrete values can be obtained by solving WFEM model,
the surface-tting techniques are needed for the 3D plots of
r1
Eq. (36). Then the crack identication database of forward
problem for cracked rotor system has been built up.

3.2. Vibration analysis

Fig. 4. Geometry of a cracked section in shaft.
In the simulation, the rotor geometries and the material prop-
erties are as follows: L = 300 mm, L1 = 8 mm, L2 = 133 mm,
j
kt L3 = 18 mm, d1 = 9.5 mm, d2 = 76 mm, E = 2.06 1011 N/m2 ,
j-1 j+1 j+2
 = 7860 kg/m3 ,  = 0.3.
13 BSWI43 RayleighEuler beam and 1 BSWI43 Rayleigh
Fig. 5. The layout of the corresponding nodes around crack with stiffness Timoshenko beam elements are employed to the vibration
coefcient kt . analysis of rotor system. Fig. 6 shows the relationship between

Fig. 6. The relationship between i (i = 1, 2, 3) and

, ( ,
[0.1, 0.9]).
 http://www.paper.edu.cn
1074 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081

Fig. 7. The rst three natural frequencies as a function of normalized crack location
for some of the normalized crack size .

i (i = 1, 2, 3) and all possible normalized crack location Unlike to the effects of the normalized crack location of sim-

= e/L2 and size = /d1 using surface-tting techniques ply supported shaft without mass disc, as mentioned by Dong
(here, ,
[0.1, 0.9]). [17], the changes in the rst three natural frequencies of a
cracked rotor system with a mass disc, would not monotoni-
3.2.1. Effects of the normalized crack location
cally decrease with the increment of the normalized crack lo-
Fig. 7 shows the rst three natural frequencies as a function cation. The reason is that only the crack of shaft segment L2
of normalized crack location
for some of the normalized crack is considered in the present study, and this is not affecting the
size . It can be seen from Fig. 7(a)(c) that the change in the crack identication results.
rst three natural frequencies were affected when the crack was
located at every normalized location
. For a certain normalized 3.2.2. Effects of the normalized crack size
crack size , Fig. 7(a) show the fundamental frequency would The rst three natural frequencies as a function of normal-
be decreased signicantly with respect to the larger normalized ized crack size for some of the normalized crack location

crack location
. While a correlation between normalized crack are shown in Fig. 8. It is noticed that the change in the rst three
location and size was given, as shown in Fig. 7(b) and (c). natural frequencies were affected when the crack was occurred
Taking = 0.8, for example, the second natural frequency was at every normalized size . For a certain normalized crack lo-
mostly affected when the crack was located at the center of the cation
, Fig. 8(a)(c) show the rst three frequencies would
shaft segment L2 , and the third natural frequency was mostly monotonically decrease if the larger normalized crack size
affected when the crack was located at
= 0.25. Moreover, was given.
Fig. 7(c) shows that the third natural frequency was almost From the above observations, it could be stated that the
unaffected for a crack located at
= 0.75; the reason for this change in frequencies is not only a function of the normal-
zero inuence was that the nodal point for the third mode was ized crack location
but also the normalized crack size . In
located at that position. the procedures of constructing crack identication database of
 http://www.paper.edu.cn
J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081 1075

Fig. 8. The rst three natural frequencies as a function of normalized crack size for some of the normalized crack location
(1 :
= 0.1; 2 :
= 0.2;
3 :
= 0.3; 4 :
= 0.4; 5 :
= 0.5; 6 :
= 0.6; 7 :
= 0.7; 8 :
= 0.8; 9 :
= 0.9).

forward problem for cracked rotor system, the inuence of the and size . Because the rst three frequencies can be measured
normalized crack size and location
was considered through easily and accurately; they are usually served as inputs to solve
the computation of rotational spring stiffness kt and the loca- the inverse problem in structural damage identication.
tion of inserting cracked stiffness submatrix into global stiff- Supposing the rst three frequencies are known, the fre-
ness, respectively. quency contour plots of the crack identication database of for-
ward problem for cracked rotor system, can be acquired and
3.3. Inverse problem plotted on the same axes. The common intersection of all the
three contour lines indicates the normalized crack location and
The crack identication inverse problem can be described by size. This intersection becomes unique due to the fact that any
cracked structural natural frequency can be represented by a
( ,
) = Fj1 (fj ) (j = 1, 2, 3). (37) frequency equation (see Eq. (36)) that is dependent on normal-
The measurements of any two natural frequencies enable us ized crack parameters [20].
to dene the normalized location and size of a crack if Eq. (36)
has been determined, i.e. the crack identication database of 4. Experimental verication
forward problem for cracked rotor system has been constructed.
However, when we use the method of frequencies contour plots 4.1. Experimental setup
for crack identication of a rotor system, two natural frequen-
cies contour plots may intersect at more than one point. There- Fig. 9 shows the experimental setup used for measuring the
fore a minimum of three frequencies is required to identify the rst three frequencies of the cracked rotor system with a single
two unknown parameters, i.e. the normalized crack location
mass disc using the Doppler signal laser vibrometer. A Polytec
 http://www.paper.edu.cn
1076 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081

high-computational precision. And the relative errors of the rst

OFV-5000 PXI signal three frequencies are less than 0.3%. Moreover, the elemental
controller conditioner DOF of 5 BSWI43 Rayleigh beam elements (consider rota-
tory inertia effect) are 58, which are smaller than 402 DOF of
OFV-505 sensor head

BEAM3 elements. The results of this example indicate the accu-

racy and efciency of the BSWI Rayleigh beam element, which
has been constructed herein. Meanwhile, the BSWI Rayleigh
Window XP
beam element can be employed to build up the crack identi-
1.7 MHz
cation database for cracked rotor system.

cracked rotor system with a single mass disc

4.2. Experimental results

We tested six cracked shafts each having an open crack at

shaft with four crack cases as shown in Table 2. The material of
Fig. 9. The experimental setup. workpiece for experiment was 40Cr steel, and the rotor geome-
tries and the material properties are as follows: L = 300 mm,
L1 = 8 mm, L2 = 188 mm, L3 = 18 mm, d1 = 9.5 mm, d2 =
Doppler laser vibrometer OFV-505/5000 was used to measure 76 mm, Youngs modulus E = 2.06 1011 N/m2 , material den-
the velocities of one point in the shaft. We point out here that
to measure the rst three frequencies only requires one mea-
surement point. The reason is that for the simple structure,
single-input and single-output (SISO) modal analysis by using
= /d1 are 0.2
a hammer as excitation, which is a usually used method. To
reduce the reection of the laser beam and spectral noise, retro-
reective tapes were put on the measurement point in the shaft.
= e/L2 are 0.4, 0.6 and 0.8, respectively.
The laser vibrometer OFV-5000 uses the principle of the
heterodyne interferometer to acquire the characteristics of me-
chanical vibrations. For each cracked shaft, the high-metrical
frequencies can be obtained by using the standard FFT program
of the software Matlab.
Taking the uncracked rotor system with the same geome-
tries and the material properties, as given by Section 3.2, and
the corresponding impulse response signal and power spectrum
are shown in Fig. 10. In the experimental study, the sampling
frequency fs is 5000.
Different elements are employed to estimate the rst three
frequencies, respectively, i.e.:
1. 5 BSWI43 beam elements (not consider rotatory inertia
effect);
2. 200 traditional beam elements (not consider rotatory inertia
effect);
3. 5 BSWI43 Rayleigh beam elements (consider rotatory in-
ertia effect);
4. 200 BEAM3 (Software Ansys) elements (consider rotatory
inertia effect).
The results are shown in Table 1. Compared with the exper-
imental frequencies, both 5 BSWI43 beam element (not con-
sider rotatory inertia effect) and 200 traditional beam elements
(not consider rotatory inertia effect) have the same computa-
tional precision. The relative errors of the second frequencies
and size
. In the experimental studies, when the three lines
arrive at 61.5%, which can not be used to the model-based
crack identication method. However, if we consider the ro-
tatory inertia effect, both 5 BSWI43 Rayleigh beam elements
(consider rotatory inertia effect) and 200 BEAM3 (Software
and
and the predicted crack parameters
and
Ansys) elements (consider rotatory inertia effect) can achieve
. For the given cases, the relative errors of
are not more
sity =7860 kg/m3 , Poissons ratio =0.3. The cracks actually
are slots cut by a wire electrical discharge machine, the crack
width is 0.2 mm, and the crack depths  are 3.8 and 7.2 mm.
In other words, the normalized crack sizes
and 0.4. The crack locations e are 75.2, 112.8 and 150.4 mm,
respectively, and the corresponding normalized crack locations
In most cases, however, the three lines do not accurately
intersect at one point because of inaccuracies in the modeling
as compared to measured results. For this purpose, the zero-
setting procedure that described by Adams [38] is used. In
this procedure, Youngs modulus of the structure is changed
by using the undamaged natural frequencies of the structure
to determine an effective value, and given by the following
iterative approach:
 
 K 
 2
i M Em  = 0, (38)
 E
where Em is the corrected value of Youngs modulus E, which
can be acquired through solving Eq. (38) for each frequency. It
should be noted that the physical signication for the correction
of Youngs modulus is not to change the value of E but to make
the metrically uncracked natural frequencies match the compu-
tational ones. This procedure can greatly reduce the error be-
tween theoretical analysis and the experimental studies, which
are caused by boundary conditions and material parameters.
In this section, the rst three metrical frequencies are em-
ployed as the inputs of inverse problem for crack quantitative
identication. Fig. 11 shows the crack identication results in
a shaft by using the frequency contour plots. The intersection
point A of three lines indicates the normalized crack location
do not meet exactly, the centroid of the three pairs of intersec-
tions is taken as the normalized crack location and size [13].
Table 2 shows the comparison of actual normalized crack pa-
rameters
 http://www.paper.edu.cn
J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081 1077

Impulse response signal

1

0.5

Amplitude 0

-0.5

-1
0 0.1 0.2 0.3 0.4 0.5 0.6
t/s

Power spectrum
0.2
86.11
0.15
Amplitude

0.1
557.82 1727.54
0.05

0
0 500 1000 1500 2000 2500
f /Hz

Fig. 10. The impulse response signal and power spectrum.

Table 1
The experimental and differently elemental results of the rst three natural frequencies

Method Computational frequencies/Hz Metrical frequencies/Hz Computational errors/%

f1 f2 f3 f1 f2 f3 f1 f2 f3

1 86.33 900.95 1732.07 0.26 61.51 0.26

2 86.34 901.00 1732.25 0.27 61.52 0.27
86.11 557.82 1727.54
3 86.33 558.17 1728.25 0.26 0.06 0.04
4 86.33 558.17 1728.20 0.26 0.06 0.04

Table 2
Crack cases of shaft and identication results

Case
Metrical frequencies/Hz Predicted
(Error/%) Predicted (Error%)

f1 /Hz f2 /Hz f3 /Hz

1 0.4 0.2 98.2 582.04 1049.88 0.415 (3.75) 0.196 (2)

2 0.4 0.4 96.62 568.43 1016.25 0.41 (2.5) 0.42 (5)
3 0.6 0.2 98.12 583.13 1055.80 0.585 (2.5) 0.17 (14)
4 0.6 0.4 95.76 572.14 1053.12 0.605 (0.83) 0.385 (3.75)
5 0.8 0.2 97.34 583.84 1051.64 0.785 (1.88) 0.225 (12.5)
6 0.8 0.4 93.04 580.87 1032.62 0.794 (0.75) 0.425 (6.25)

than 4% while the relative errors of arrive at 15%. Hence, point out here that the identication precisions of the crack sizes
the proposed model-based crack identication method by using are much larger than those of the crack location. The probable
BSWI Rayleigh beam element is considered to be valid for reason is that the effects of the normalized crack size are much
actual application to detect open cracks in rotor systems. We complicated than those of the normalized crack location
.
 http://www.paper.edu.cn
1078 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081

Fig. 11. Frequency contour plots (- - -: 1st frequency; -.-.-: 2nd frequency; - - -: 3rd frequency): (a) case 1; (b) case 2; (c) case 3; (d) case 4; (e) case 5 and
(f) case 6.

Therefore, for the identication of
, the tiny metrical errors structural crack detection is challenging because actual rotor
would be signicantly enlarged.
5. Discussion and further researches
Except the open crack, there are many other types of open
and closed cracks and defects in rotor systems, and different
types of defects have different controller equations and because
damage estimation is essentially an inverse problem, it is dif-
cult to obtain a unique or accurate estimation of cracks, espe-
cially for on-line crack identication of rotor systems. On-line
system dynamic responses are always contaminated by un-
known external loads as well as other harmless conditions, such
as material composition uncertainly, geometric variation, sen-
sor noise, and so on. Therefore, after the possible existence of
damage is detected by an on-line damage detection method as
mentioned by Green [9], a reliable inspection of the stopped
system to nd crack location and size is necessary in order
to assure the existence of damage, and the model-based crack
identication method by using BSWI Rayleigh beam element
is intended for such use.
 http://www.paper.edu.cn
J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081 1079

e crack

= e / L2

L2

d1
L1 L3 L1

d2
L

Fig. 12. Typical rotor-bearing system considered.

Table 3
The rst three natural frequencies under different bearing stiffness

Frequencies Simply supported Bearing stiffness kb (N/m)

1 105 5 105 1 106 1 107 1 108 1 109 5 109

f1 /Hz 96.97 60.25 84.77 90.31 96.25 96.90 96.96 96.97

f2 /Hz 569.13 225.03 398.18 464.15 556.38 567.83 569.002 569.11
f3 /Hz 1019.25 452.90 693.05 827.37 1001.13 1017.48 1019.08 1019.22

Fig. 13. The rst three natural frequencies inuenced by bearing stiffness.

1080 J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081
The effect of the bearing on the fault diagnosis of the rotor
system is very obvious for the rotating shafts. In this section,
the bearing stiffness is considered, as shown in Fig. 12. The
stiffness matrix considers the stiffness of the shaft elements
including cracked shaft element and the bearing stiffness kb .
The vibration analysis has been carried out using BSWI
Rayleigh beam elements for the rotor system with the same ge-
ometrical and physical parameters as shown in Section 4.2. The
bearing stiffness kb is varying from 1105 to 5109 N/m, sup-
posing the normalized crack location
= 0.4 and crack depth References
= 0.4. The rst three natural frequencies under different bear-
ing stiffness are shown in Table 3. The rst three natural fre-
quencies inuenced by bearing stiffness are shown in Fig. 13.
To describe clearly, we adopt the logarithm coordinates. From
Table 3 and Fig. 13, the inuence of bearing stiffness can be
neglected when kb > 1 108 . Both simply supported rotor sys-
tem as shown in Fig. 3 and the bearing stiffness is considered
as shown in Fig. 12 have the same computational results.
In the present experimental studies of this paper, only static
(non-rotating) rotor system is studied, the bearing stiffness
nearly to 5 109 N/m. Hence the simply supported model of
rotor system can give reasonable results. The bearing damping
and other factors are not considered. It obviously introduces
tiny errors of the wavelet-based nite element model. So the
zero-setting procedure is introduced to eliminate those inu-
ences. However, it is worth to point out here that the present
method is only a reliable inspection of the stopped system to
nd crack location and size for the sake that the bearing effect
is not considered. The further research is that a more compli-
cated rotating rotor-bearing model which includes bearing stiff-
ness, bearing damping, gyroscopic effect, etc., will be studied
for on-line crack identication use.
6. Conclusions
A methodology based on BSWI element to detect crack loca-
tion and size is presented. Because the good character of BSWI
scaling functions, the BSWI element presented in this paper
is a useful tool to deal with high-performance computation in
structural crack identication of rotor system. In the construc-
tion of BSWI Rayleigh beam element, the inuence of rota-
tory inertia on the natural frequencies for rotor systems have
been examined in detail. For a rotor system, the slender shaft
and stiffness disc are modeled by BSWI RayleighEuler beam
element, and BSWI RayleighTimoshenko beam element, re-
spectively. Hence, the rst three natural frequencies are in good
agree with the experimental ones. Then, experimental investi-
gations verify that the proposed method can be utilized to detect
crack location as well as crack size with high performance. The
procedure for detection of crack is simple and need little effort
for various boundary conditions and different rotor systems. It
is also believed that this procedure can be easily extended to
complex rotor systems with multiple cracks if more natural fre-
quencies are measured as inputs of inverse problem. Therefore,
it may provide new elements and technique for quantitatively
diagnosing cracks on engineering rotor systems.
http://www.paper.edu.cn
Acknowledgments
This work is supported by the key project of National Natu-
ral Science Foundation of China (Nos. 50335030, 50505033),
Natural Science Foundation of Guilin University of Electronic
Technology (No. Z20701). Besides, we express our sincere
thanks to the anonymous reviewer for his suggestions for the
improvement of the paper.
[1] S.W. Doebling, C.R. Farrar, M.B. Prime, A summary review of vibration-
based damage identication, Shock Vibr. Dig 30 (2) (1998) 91105.
[2] W.J. Staszecski, Structural and mechanical damage detection using
wavelets, Shock Vibr. Dig. 30 (6) (1998) 457472.
[3] S. Loutridis, E. Douka, A. Trochidis, Crack identication in double-
cracked beams using wavelet analysis, J. Sound Vibr. 277 (2004)
10251039.
[4] S. Loutridis, E. Douka, L.J. Hadjileontiadis, et al., A two-dimensional
wavelet transform for detection of cracks in plates, Eng. Struct. 27
(2005) 13271338.
[5] J.G. Han, W.X. Ren, Z.S. Sun, Wavelet packet based damage
identication of beam structures, Int. J. Solids Struct. 42 (2005)
66106627.
[6] X.Q. Zhu, S.S. Law, Wavelet-based crack identication of bridge beam
from operational deection time history, Int. J. Solids Struct. 43 (2006)
22992317.
[7] A.D. Dimarogonas, C.A. Papadopoulos, Vibration of cracked shafts in
bending, J. Sound Vib. 91 (4) (1983) 583893.
[8] A.S. Sekhar, B.N. Srinivas, Vibration characteristics of slotted shafts,
J. Sound Vibr. 251 (4) (2002) 621630.
[9] I. Green, C. Casey, Crack detection in a rotor dynamic system by
vibration monitoringPart I: analysis, ASME J. Eng. Gas Turbines
Power 127 (2005) 425436.
[10] M. Kisa, J. Brandon, M. Topcu, Free vibration analysis of cracked beams
by a combination of nite elements and component mode synthesis
methods, Comput. Struct. 67 (1998) 215223.
[11] Y.-S. Lee, M.-T. Chung, A study on crack detection using eigenfrequency
test data, Comput. Struct. 77 (2000) 327342.
[12] P.N. Saavedra, L.A. Cuitino, Crack detection and vibration behavior of
cracked beams, Comput. Struct. 79 (2001) 14511459.
[13] S.P. Lele, S.K. Maiti, Modeling of transverse vibration of short beams
for crack detection and measurement of crack extension, J. Sound Vibr.
257 (3) (2002) 559583.
[14] D.P. Patil, S.K. Maiti, Detection of multiple cracks using frequency
measurements, Eng. Fract. Mech. 70 (2003) 15531572.
[15] D. Anjan, S. Talukdar, Damage detection in bridges using accurate modal
parameters, Finite Element Analy. Design 40 (2004) 287304.
[16] A.S. Sekhar, Crack identication in a rotor system: a model-based
approach, J. Sound Vibr. 270 (2004) 887920.
[17] G.M. Dong, J. Chen, J. Zou, Parameter identication of a rotor with an
open crack, Eur. J. Mech. A/Solids 12 (2004) 325333.
[18] P.F. Rizos, N. Aspragathos, A.D. Dimarogonas, Identication of crack
location and magnitude in a cantilever beam from the vibration modes,
J. Sound Vibr. 138 (3) (1990) 381388.
[19] B. Li, X.F. Chen, Z.J. He, et al., Detection of crack location and size
in structures using wavelet nite element methods, J. Sound Vib. 285
(45) (2005) 767782.
[20] G.M. Owolabi, A.S.J. Swamidas, R. Seshadri, Crack detection in beams
using changes in frequencies and amplitudes of frequency response
functions, J. Sound Vibr. 2003 (265) (2003) 122.
[21] X.F. Chen, Z.J. He, J.W. Xiang, et al., A dynamic multiscale lifting
computation method using Daubechies wavelet, J. Comput. Appl. Math.
188 (2) (2006) 228245.
[22] J.X. Ma, J.J. Xue, Z.J. He, et al., A study of the construction and
application of a Daubechies wavelet-based beam element, Finite Element
Anal. Design 39 (2003) 965975.

J. Xiang et al. / Finite Elements in Analysis and Design 43 (2007) 1068 1081
[23] X.F. Chen, S.J. Yang, Z.J. He, et al., The construction of wavelet nite
element and its application, Finite Element Anal. Design 40 (2004)
541554.
[24] J.-G. Han, W.-X. Ren, Y. Huang, A multivariable wavelet-based nite
element method and its application to thick plates, Finite Element Anal.
Design 41 (2005) 821833.
[25] J.W. Xiang, X.F. Chen, Z.J. He, H.B. Dong, The construction of 1D
wavelet nite elements for structural analysis, Comput. Mech. 40 (2)
(2007) 325339.
[26] J.W. Xiang, X.F. Chen, Y.M. He, Z.J. He, The construction of plane
elastomechanics and Mindlin plate elements of B-spline wavelet on the
interval, Finite Element Anal. Design 42 (2006) 12691280.
[27] J.W. Xiang, Z.J. He, X.F. Chen, The construction of wavelet-based
truncated conical shell element using B-spline wavelet on the interval,
Acta Mech. Solida Sin. 19 (4) (2006) 316326.
[28] Z.J. He, X.F. Chen, Advanced in theory study and engineering application
of wavelet nite element, Chinese J. Mech. Eng. 41 (2005) 111 (in
Chinese).
[29] X.F. Chen, Z.J. He, J.W. Xiang, Experiments on crack identication in
cantilever beams, Exp. Mech. 45 (3) (2005) 295300.
[30] J.W. Xiang, X.F. Chen, Z.J. He, et al., Identication of crack in a beam
based on nite element method of B-spline wavelet on the interval,
J. Sound Vibr. 296 (45) (2006) 10461052.
http://www.paper.edu.cn
1081
[31] S. Bertoluzza, G. Naldi, J.C. Ravel, Wavelet methods for the numerical
solution of boundary value problems on the interval, in: C.K. Chui,
L. Montefusco, L. Puccio (Eds.), Wavelets: Theory, Algorithms, and
Applications, Academic Press, Inc., New York, 1994, pp. 425448.
[32] C.K. Chui, Q. Ewald, Wavelets on a bounded interval, Numer. Methods
Approx. Theory 1 (1992) 5357.
[33] J.C. Goswami, A.K. Chan, C.K. Chui, On solving rst-kind integral
equations using wavelets on a bounded interval, IEEE Trans. Antennas
Propag. 43 (1995) 614622.
[34] Y. Kang, A.-C. Lee, Y.-P. Shih, A modied transfer matrix method for
asymmetric rotor-bearing systems, ASME J. Vibr. Acoust. 116 (1994)
309317.
[35] H.D. Nelson, A nite rotating shaft element using Timoshenko beam
theory, ASME J. Mech. Design 102 (1980) 793802.
[36] M.I. Friswell, J.E.T. Penny, Crack modeling for structural health
monitoring, Struct. Health Monit. 1 (2) (2002) 139148.
[37] H. Tada, P.C. Paris, G.R. Irwin, The Stress Analysis of Cracks Handbook,
third ed., ASME Press, New York, 2000 pp. 487492.
[38] R.D. Adams, P. Cawley, C.J. Pye, B.J. Stone, A vibration technique for
non-destructively assessing the integrity of structures, J. Mech. Eng. Sci.
20 (2) (1978) 93100.