I Abm As 2014 Bridge Assessment

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Vibration Based Structural Model Updating and Damage Identification

of Bridges
Hua-Peng Chen1*
1
School of Engineering, University of Greenwich, Chatham Maritime, Kent, ME4 4TB, UK.

*Corresponding author: h.chen@gre.ac.uk

ABSTRACT: Effective structural model updating and damage identification methods are proposed by using
incomplete measured modal data. The exact relationship between the perturbation of structural parameters
and the modal properties of the tested dynamic structure is given on the base of dynamic perturbation method.
Structural perturbation parameters are properly selected to represent the differences in structural parameters
for structural model updating and damage identification at critical point level. Experimental studies for a la-
boratory tested steel structure model are undertaken to verify the effectiveness of the proposed model updat-
ing technique and to investigate the influence of structural damage on changes of modal parameters. Further-
more, a cable-stayed bridge is adopted for numerical investigations on inverse structural damage
identification. The results from experimental studies and numerical investigations show the applicability of
the proposed method to structural model updating and damage identification.

1 INTRODUCTION merical model by optimisation process which re-


quires the eigen-solutions and associated sensitivity
Condition assessment of existing bridge structures matrices of the numerical model to be calculated in
has become a critical issue due to the deterioration each iteration (Bakira et al. 2007; Modak et al.
of structural performance and the requirements of 2002). However, the sensitivity analysis used in the
structural safety. Structural health monitoring has iterative model updating methods may not perform
received significant attention in past decades in or- well, in particular in the cases when discrepancy be-
der to assess the safety and reliability of bridges dur- tween the initial FE numerical model and the actual
ing their service life (Brownjohn and Xia 2000; tested structure is significant and large modification
Chen and Alani 2013; De Roeck 2003). Structural of structural parameters is required.
damage identification techniques based on vibration The existing methods for damage identification
measurements show great promise because they al- using measured vibration modal data include model
low for quick and global damage detection at a rela- updating techniques (Mottershead and Friswell
tively low cost after a severe loading event such as 1993), sensitivity analysis methods (Kosmatka and
an earthquake or collision. These techniques rely on Ricles 1999) and dynamic perturbation methods
the fact that any change in stiffness caused by dam- (Chen 2008; Chen and Bicanic 2010). However,
age in a structure leads to a change in modal pa- most of these methods may be unable to provide ac-
rameters of the structure such as natural frequencies curate predictions for the severity of damage at spe-
and mode shapes (Chen 2010; Doebling et al. 1996). cific locations in complex structures such as bridges,
Although successful applications of these techniques because the chosen damage indicators are not capa-
have been developed to some extent, vibration based ble of reflecting the severity of the damage at local
local damage assessment of bridges remains a chal- level.
lenge in practice. In this paper, effective structural model updating
In vibration-based structural damage identifica- and damage identification methods are presented by
tion, a model updating procedure is often needed to using incomplete measured modal data, without re-
calibrate and update the initial finite element (FE) quiring modal expansion or model reduction tech-
model and to provide the baseline for damage identi- niques. A scale-down steel structure model was con-
fication. In general, iterative updating methods such structed in laboratory for experimental studies.
as sensitivity-based approaches are used for updat- Vibration testing for the structure model was per-
ing FE model of tested structures. The iterative formed to obtain the dynamic response of the struc-
methods update structural parameters of the FE nu- ture model under various dynamic excitations. Mod-
al parameters of the tested structure model, such as bration measurements is often limited in practice.
frequencies and incomplete mode shapes, are ex- The measured modes are paired to the numerical ei-
tracted from the measured accelerations by modal genvectors of dimension Ns (restricted to the same
analysis techniques. The measured incomplete mod- dimensions as ψ̂ i ), φ ak , by using Modal Assurance
al data are then utilised for updating the finite ele-
ment model of the tested structure model. The influ- Criterion (MAC) factors defined as
ence of structural damage at various locations and T 2

magnitudes on modal parameters is also investigated φ ak ψ


ˆi
by reducing stiffness of structural members. Fur- MAC (φ ak , ψ
ˆ i) = (4)
aT T a
thermore, a cable-stayed bridge is adopted for nu- φ φ ψˆ i ψˆ i
k k

merical investigations on inverse structural damage


identification. Various structural damage scenarios From above definition, large MAC factors indi-
are simulated with different magnitudes at different cate a high degree of similarity between two mode
locations. The simulated damage scenarios are then shapes, while small MAC factors represent little or
employed to investigate the effectiveness of the pro- even no correlation between two vectors. In the ap-
posed inverse damage identification method using plication of governing equation Eq. (1), the meas-
only incomplete vibration modal measurements. ured incomplete mode shapes of the actual tested
structure need to be close to the corresponding part
of the original numerical eigenvectors φ ia . The
2 DYNAMIC PERTURBATION METHOD
measured DOF’s readings ψ̂ i are then scaled by a
The dynamic perturbation method (Chen 2005) pro- mode scale factor υ i , giving the scaled modal meas-
vides the exact relationship between the perturbation urement vector φ̂ ia of dimension Ns as
of structural parameters, such as difference in stiff-
T
ness ( ΔK ) and mass ( ΔM ) between the FE model φˆ ia = υ i ψˆ i , where υ i = φ ia ψˆ i / ψˆ i T ψ
ˆi (5)
and the tested structure, and modal properties, such
as the ith frequency ω i and the corresponding ei- From Eq. (2), the ith complete mode shape of the
genvector φ i of the FE model as well as the ith fre- tested structure, consisting of the measured compo-
nents and the remaining calculated components, is
quency ω̂ i and complete eigenvector φ̂ i of the tested then given by
structure, expressed here as N

N
φ k ( ΔK − ωˆ i ΔM )φˆ i
T 2 φˆ i = ϕ ia + C ik φ uk (6)
 2
ωi − ωk
ˆ 2
φ k − φˆ i = 0 (1) k =1, k ≠ i

k =1
where φ uk is original eigenvector corresponding to
The complete eigenvector of the tested structure the entries of the unmeasured components and ϕ ia is
with dimension of a total number of N degrees of
freedom (DOF) can be expressed as a linear combi- a known vector of dimension N, defined as
nation of the numerical eigenvectors, since the nu- ϕ ia = υ i ψˆ i + φ ui (7)
merical eigenvectors are linearly independent due to
In the proposed model updating and damage iden-
the symmetry of the stiffness and mass matrices of
tification method, structural system parameters to be
the FE model, namely
updated or identified, such as parameters for materi-
N
φˆ i =  C ik φ k (2) al and geometric properties, are employed to reflect
k =1 the perturbation of structural parameters, e.g. stiff-
ness matrix and/or mass matrix. The structural sys-
where C ik are mode participation factors, defined as
tem parameters can characterise the structural pa-
Cik = φ k Mφˆ i rameters at element level or at integration point
T
(3)
level. In the case where structural perturbation pa-
In structural dynamic testing, modal data about rameters are chosen at element level, the change in
the natural frequency ω̂i and mode shape readings element stiffness matrix ΔK e can be expressed by
ψ̂ i of the tested structure can be extracted from vi-
ˆ −K =α K
ΔK e = K (8)
bration measurements by modal analysis techniques. e e e e

The measured mode shapes ψ̂ i , with dimension of where α e is a stiffness perturbation parameter char-
Ns where Ns is the total number of effective sensors acterised at element level to be determined; K e and
installed, are usually incomplete with reference to
K̂ e are the eth element stiffness matrices for the FE
the FE model typically having large number of
DOFs, since the number of sensors installed for vi- model and the tested structure, respectively.
For structures with beam elements, it is difficult Nα N
~
to accurately model connections of these elements in [p ajii +
j =1
C
k =1, k ≠ i
ik p ujik ]α j
the FE model and the stiffness at the joints is often (13)
Nβ N
estimated with uncertainty. To effectively update the ~
[q +  Cik q umik ]β m − φ̂ ia = 0
2
− ωˆ i a
mii
bending stiffness at the ends of elements, the ele- m =1 k =1, k ≠ i
ment stiffness is now calculated from an integral
where the sensitivity coefficients associated with the
form. The change in element stiffness matrix ΔK be known eigenmodes and structural parameters are de-
between the element stiffness of tested structure K̂ be fined in a general form as
T
and the element stiffness of the FE model K be for φ k K j ϕ ia T u
, p ujil =  φk2 K j φl2 φak
N N
p ajii =  φ ak (14a)
the beam element is given by k =1 ωˆ i 2 − ω k 2 k =1 ωˆ i − ωk

 r r
T T
ˆ b − Kb = α Kb
ΔK be = K (9)
N
φ k M m ϕ ia a , N
φ k M m φ ul a
e e q amii =  φ k q u
mil =  φk (14b)
r k =1 ωˆ i 2 − ωk 2 k =1 ωˆ i 2 − ωk 2
where α r is stiffness perturbation parameter charac- On the basis of the governing equations Eq. (13),
b
terised at critical point level to be determined, K is
r
a direct solution procedure is sufficient to solve for a
stiffness contribution at critical point level to the total number of Nx (= Nα + Nβ ) structural perturba-
global stiffness matrix. tion parameters α j and β m . In the case when a total
Consequently, the change of global stiffness ma- number of Nm experimental modes are measured
trix between the tested structure and FE model is from a total number of Ns sensors installed on the
written as tested structure, the governing equation Eq. (13)

should be used for the total number of Nm measured
ΔK =  α j K j (10)
j =1
modes to generate a total number of Np ( = Nm × Ns )
equations available in order to find a solution for a
where Nα represents the total number of stiffness
total number of Nx structural perturbation parame-
perturbation parameters adopted in calculations; α j ters to be determined. A regularisation method such
is stiffness perturbation parameter characterised at as the Tikhonov regularisation (Tikhonov and
element level or integration point level; K j is the Arsenin 1993) is applied to find a solution for what
contribution of the jth element or integration point to is in general an ill-conditioned system.
the global stiffness matrix. For structural model updating, the structural up-
Similarly, the the change of global mass matrix dating parameters α j and β m are directly estimated
between the tested structure and FE model is ex- from these linear governing equations in Eq. (13).
pressed as For structural damage identification, the globe
Nβ mass matrix of the structure M is considered un-
ΔM =  β m M m (11) changed before and after damage because the mass
m =1
is assumed here not affected by structural damage,
where Nβ represents the total number of mass per- i.e. β m = 0 for m = 1, Nβ . The structural perturbation
turbation parameters β m ; M m is the mth contribu-
parameters α j represent here as structural damage
tion to the global mass matrix.
indicators for inverse damage identification.
In order to avoid iterative procedures in structural
model updating and damage identification, the mode
participation factors C ik for the ith measured mode 3 LABORATORY VIBRATION TESTING
are simply estimated by using Eq. (3), where the ith
complete mode shapes of the tested structure φ̂ i is A small scale steel model structure, as shown in
Figure 1, was utilised for experimental studies to
approximately replaced by the known vector ϕ ia de- demonstrate the accuracy and effectiveness of the
fined in Eq. (7), giving proposed FE model updating method and to investi-
~ Τ gate the influence of structural damage on modal pa-
Cik = φ k Mϕ ia (12)
~ rameters. The structure has four stories comprising
By using the estimated Cik and the perturbations of 10 structural members at each storey, i.e. four
of stiffness and mass matrices, the governing equa- columns, four beams and two diagonal braces. The
tions in Eq. (1) for the ith measured mode are now beams and columns are modelled as conventional
expressed in a set of linear equations as beams and the braces as axial bar elements. The
connection joints of beams and columns are mod- The first six measured frequencies and the associ-
elled as rigid joints to maintain bending moments. ated incomplete mode shapes for the laboratory test-
ed model structure are shown in Fig. 3. These meas-
ured incomplete modes are adopted for updating the
FE model.

1st mode, 2nd mode, 3rd mode,


10.569Hz 26.386Hz 32.258Hz

Figure 1. Steel structure model used in laboratory structural


dynamic testing.

In laboratory vibration testing, a total number of 16


accelerometers are placed at beam-column joints to
measure only translational displacement readings in
horizontal directions. The modal data such as fre-
quencies and mode shape readings at measured
4th mode, 5th mode, 6th mode,
DOFs are extracted from the acceleration measure-
46.018Hz 51.847Hz 65.522Hz
ments by using modal analysis technique, as shown
in Fig. 2. Figure 3. First six measured modes of the steel structure model
from laboratory vibration testing.

The FE numerical model used for structural mod-


el updating has 20 nodes and 40 elements with a to-
tal number of 96 DOFs, as shown in Fig. 4.

a) Acceleration measurements

b) Identified frequencies
Figure 4. Finite element model for the steel structure model
Figure 2. Vibration test results for the steel structure model. adopted for model updating.
Damage Pattern 2: No stiffness in any of the brac- posed approach for inverse structural damage identi-
es in the levels 1 and 3 fication from limited incomplete modal data by nu-
Damage Pattern 3: No stiffness in any of the brac- merical investigations. The elevation and dimension
es in the levels 1, 2 and 3 of cable-stayed bridge are shown in Fig. 5. The
Damage Pattern 4: No stiffness in any of the brac- cross-sectional properties of each component are
es in all stories. listed in Table 5. The numbers of structural members
shown in Table 5 correspond to those indicated in
Laboratory vibration testing was carried out to Fig. 5. The modulus of elasticity is taken as E=2.1×
measure natural frequencies for each of damage pat- 1011N/m2 for steel of the structure and E=3.2×
tern by reducing stiffness of braces in various levels. 1010N/m2 for concrete.
The experimental results for various damage pat-
terns are summarised below in Table 4. 10

4x3.5=14.0
9
8
Table 4. Experimental natural frequencies of the steel structure

28.8144
7

14.8144
model for various damage patterns.

29.86
Intact Damage Damage Damage Damage 11
8.0 4x9=36.0 4.867
Mode
(Hz) pattern 1 pattern 2 pattern 3 pattern 4 5.4 9.733

1 2 3 4 3 2 5 6 128/2=64.0m 56.0m
10.6 10.5 10.5 10.4 10.4 Bending
26.4 18.8 18.0 14.9 14.6 Torsion
32.3 32.2 32.3 32.7 32.8 Bending Figure 5. Elevation and dimension of fan-system cable-stayed
46.0 45.9 45.0 44.2 43.0 Torsion
bridge.
51.8 52.2 52.8 52.9 52.7 Bending
Table 5. cross-sectional and material properties of the ca-
65.5 63.5 63.4 64.5 65.7 Mixed
ble-stayed bridge
The results in Table 4 show that the largest Number of Area Moment of Mass
change of frequency is the second mode (first tor- Member Members (m2) Inertia (103kg/m)
sion), reducing from 26.4Hz for the intact structure (10-4m4)
to 18.8Hz for damage pattern 1 and further to Girder 1 4.976 2.730 16.213
14.6Hz for damage pattern 4. The forth mode (sec- Girder 2 4.976 2.730 17.194
ond torsion) is also affected by structural damage, Girder 3 5.420 3.462 18.415
reducing from 46.0Hz for the intact structure to Girder 4 6.012 4.662 18.522
43.0Hz for damage pattern 4. This is because the Girder 5 4.560 2.814 16.971
brace members of the steel structure model make Girder 6 3.444 2.125 11.094
significant contributions to the torsional stiffness. Girder 7 0.007032 / 0.120
Damage in brace members by reducing their stiff- Cable 8, 12 0.009897 / 0.168
ness causes decrease in the torsional stiffness, con- Cable 9, 13 0.012722 / 0.218
sequently leading to decrease in torsional frequen- Tower 10 4.800 1.600 12.274
cies. Pier 11 30.000 19.980 79.894
The frequencies corresponding to bending modes
appear no significant change due to damage in the
brace members of the steel structure model. The fre- Four damage scenarios are generated with dam-
quencies should be reduced after structural damage age at different locations and magnitudes in the
in theory. However, due to uncertainties in laborato- structure, as summarized in Table 6.
ry vibration testing such as disturbance of structure,
Table 6. Hypothetical damage scenarios by reducing stiff-
different vibration excitation methods and noise in ness at various locations with different magnitudes
vibration measurements, the measured frequencies
will have some uncertainties with small increase af- Scenario Element No Gauss Point No Damage Amount
ter damage occurs in the structure. 1 35, 36, 37 105, 106, 108, 109 -30%
2 35, 36, 37 105, 106, 169 -30%
3 18, 19, 25, 30 54, 55, 75, 90 -30%
4 NUMERICAL INVESTIGATIONS 4 14, 15, 30 51, 57, 62 -20%
42, 43, 88 151, 172, 186 -30%
A model of the actually constructed fan-system ca-
ble-stayed bridge (Wang and Huang 1992) is now The simulated structural damage scenarios are al-
adopted to demonstrate the effectiveness of the pro- so illustrated in Fig. 6, where structural damage oc-
curs at various locations such as in towers, cables

Damage Amount (in %)


-40

and girders. In this study, it is assumed that only -30

translation displacement readings at the nodes -20


-10
marked with "●" on the girder shown in Fig. 6 are 0

available, generating incomplete mode shapes with 10


1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191

limited DOFs readings. Gauss Point No.

Damage Area a) Predicted damage for damage scenario 1


x x x Damage Area

Damage Amount (in %)


1 2 3 18 Damage Area -40
51
Gauss Points 24 19 29 57 52 62
-30
25 58 63
30
-20
66
-10
1 2 12 13 14 15 16 17 34 35 36 37 38 50
0
10
Damage Area
1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191

Gauss Point No.

Figure 6. Details of simulated damage scenarios for the cable-


b) Predicted damage for damage scenario 2
stayed bridge.

Damage Amount (in %)


-40
-30
The girder is supported vertically at the towers -20

but is independent of the towers. Therefore, the -10

bending moments are not transferred between the 0


10
girder and the tower. The cables have initial tensile 1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191

forces due to the dead loads, so that they are capable Gauss Point No.

of resisting compressive forces during vibration of c) Predicted damage for damage scenario 3
the structure.
Damage Amount (in %)

-40
Different element stiffness matrices are used in -30

the cable-stayed bridge to generate the global stiff- -20


-10
ness matrix, and nonlinearity in the inclined cable 0

stays is considered in element stiffness for cables. 10


1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181 191
The bridge structure has a total number of 66 struc- Gauss Point No.

tural elements with a total number of 136 DOFs.


d) Predicted damage for damage scenario 4
Structural damage is characterized at critical point
level for all structural members, e.g. cable, girder, Figure 7. Inverse damage predictions for different damage sce-
tower and pier members, with a total number of 198 narios at critical point level, information on incomplete dam-
damage parameters to be identified. aged modes 1, 3, 5 and 6 with the sensor array used.
A finite element analysis was performed for both
the intact and damaged structures to calculate natural From the results, it is found that structural dam-
frequencies and the corresponding mode shapes. In age can be located correctly for each of the simulat-
the following, the computed frequencies and the cor- ed damage scenarios, while the extent of structural
responding DOF’s readings of the damaged structure damage can also be estimated if the total amount of
are adopted in inverse damage predictions in place structural damage within a cable spreading within its
of the measured experimental modal data, which integration point locations is considered. The results
would normally be furnished from experiments. The presented here show that the inverse predictions for
first 6 noise free incomplete modes of the original structural damage may not be unique, as the same
and damaged structures are considered for the in- damaged stiffness can result from different structural
verse prediction of simulated structural damage sce- damage scenarios, e.g. the cases of element stiffness
narios in the bridge. for cables.
The results in Fig. 7 are for inverse predictions
for different damage scenarios simulated at critical
5 CONCLUSION
point level as shown in Fig. 6. Here, numerically in-
tegrated element stiffness matrices obtained from in- On the basis of the results from the experimental
tegrations for all structural members of the bridge studies for the steel structure model tested in labora-
are utilised, and three integration points are consid- tory and numerical investigations for a actually con-
ered for all structural elements. The information structed cable-stayed bridge, the following conclu-
about the incomplete damaged modes 1, 3, 5 and 6 is sions are noted:
used for inversely identifying four simulated damage 1) The proposed approach is capable of success-
scenarios. fully updating structural parameters and identifying
structural damage such as stiffness at critical point National Laboratory Los Alamos, New Mexico, Report No.
level, so that the stiffness parameters at the connec- LA-13070-MS.
Duan, Z., Yan, G., Ou, J.. Spencer, B.F. 2007. Damage detec-
tions of structural members such as beam-column tion in ambient vibration using proportional flexibility ma-
joints can be evaluated from information on meas- trix with incomplete measured DOFs. Structural Control
ured incomplete modes. and Health Monitoring, 14(2): 186-196.
2) The proposed method could be directly used Friswell M. I., Inman D. J., Pilkey D. F. 1998. Direct updating
for updating the finite element models for real engi- of damping and stiffness. AIAA Journal, 36(3): 491-493.
neering structures to avoid an iterative solution pro- Kosmatka, J. B., Ricles, J. M. 1999. Damage detection in struc-
cedure, since the dynamic perturbation method pro- tures by modal vibration characterization. Journal of Struc-
vides direct relationship between structural updating tural Engineering, 125(12): 1384-1392.
parameters and the measured modal data. Modak S. V., Kundra T. K., Nakra B. C. 2002. Comparative
3) The experimental results show that the fre- study of model updating methods using simulated experi-
mental data. Computers and Structures, 80: 437-447.
quencies associated with torsion modes may change
significantly due to damage in the brace members by Mottershead J. E., Friswell M. I. 1993. Model updating in
reducing their stiffness, the frequencies associated structural dynamics: a survey. Journal of Sound and Vibra-
tion, 167 (3): 347-375.
with bending modes may not have remarkable shifts
even severe damage exists in braces. Tikhonov A. N., Arsenin V. Y. 1993. Solutions of ill-posed
4) Only information about incomplete modal data problems. Wiley, New York.
Wang, T. L., Huang, D. Z. 1992. Cable-stayed bridge vibration
with a limited number of DOF's readings is suffi- due to road surface roughness. Journal of Structural Engi-
cient to correctly update finite element models and neering, ASCE, 118(5): 1354-1374.
to inversely identify damage in structures, making Zhao, J., DeWolf, J. T. 1999. Sensitivity study for vibrational
the proposed method promising in practices. parameters used in damage detection. Journal of Structural
5) The proposed approach is capable of not only Engineering, 125(4): 410-416.
identifying the multiple locations of damage in dif-
ferent members of a structure, but also can provide
information on the extent of the structural damage
from incomplete modal data.

REFERENCES

Bakira P. G., Reynders E., De Roeck G. 2007. Sensitivity-


based finite element model updating using constrained op-
timization with a trust region algorithm. Journal of Sound
and Vibration, 305: 211-25.
Brownjohn J. M. W., Xia P. Q. 2000. Dynamic assessment of
curved cable-stayed bridge by model updating. Journal of
Structural Engineering, 126 (2): 252-260.
Chen, H-P. 2005. Nonlinear perturbation theory for structural
dynamic systems. AIAA Journal, 43(11): 2412-2421.
Chen, H-P. 2008. Application of regularization method to
damage detection in plane frame structures from incomplete
noisy modal data. Engineering Structures, 30(11): 3219-
3227.
Chen, H-P. 2010. Application Mode shape expansion using
perturbed force approach. Journal of Sound and Vibration,
329(8): 1177-1190.
Chen, H-P., Alani, A. M. 2013. Optimized maintenance strate-
gy for concrete structures affected by cracking due to rein-
forcement corrosion. ACI Structural Journal, 110(2): 229-
238.
Chen, H-P., Bicanic, N. 2010. Identification of structural dam-
age in buildings using iterative procedure and regularisation
method. Engineering Computations, 27(8): 930-950.
De Roeck, G. 2003. The state-of-the-art of damage detection
by vibration monitoring: the SIMCES experience. Journal
of Structural Control, 10(2): 127-134.
Doebling, S. W., Farrar, C. R., Prime, M. B., Shevitz, D. W.
1996. Damage identification and health monitoring of
structural and mechanical systems from changes in their
vibration characteristics: a literature review. Los Alamos

You might also like