Upload
others
View
7
Download
0
Embed Size (px)
Citation preview
저 시-비 리- 경 지 2.0 한민
는 아래 조건 르는 경 에 한하여 게
l 저 물 복제, 포, 전송, 전시, 공연 송할 수 습니다.
다 과 같 조건 라야 합니다:
l 하는, 저 물 나 포 경 , 저 물에 적 된 허락조건 명확하게 나타내어야 합니다.
l 저 터 허가를 면 러한 조건들 적 되지 않습니다.
저 에 른 리는 내 에 하여 향 지 않습니다.
것 허락규약(Legal Code) 해하 쉽게 약한 것 니다.
Disclaimer
저 시. 하는 원저 를 시하여야 합니다.
비 리. 하는 저 물 리 목적 할 수 없습니다.
경 지. 하는 저 물 개 , 형 또는 가공할 수 없습니다.
Finite Element Model Updating
using Rotational Response
Myunghwan Bae
Department of Urban Environmental Engineering
(Urban Infrastructure Engineering)
Graduate school of UNIST
2015
Finite Element Model Updating
using Rotational Response
Myunghwan Bae
Department of Urban Environmental Engineering
(Urban Infrastructure Engineering)
Graduate school of UNIST
Finite Element Model Updating
using Rotational Response
A thesis
submitted to the Graduate School of UNIST
in partial fulfillment of the
requirements for the degree of
Master of Science
Myunghwan Bae
December 18th, 2014
Approved by
Advisor
Sung-Han Sim
Finite Element Model Updating using Rotational Response
Myunghwan Bae
This certifies that the thesis/dissertation of Myunghwan Bae is approved
December 18th, 2014
Sung-Han Sim
Assistant Professor of Urban and Environmental Engineering
Ulsan National Institute of Science and Technology
Advisor
Myoungsu Shin
Associate Professor of Urban and Environmental Engineering
Ulsan National Institute of Science and Technology
Marco Torbol
Assistant Professor of Urban and Environmental Engineering
Ulsan National Institute of Science and Technology
i
ABSTRACT
The Finite Element (FE) models have been used in civil, mechanical and aerospace engineering field
for system identification and response simulation under various unexperienced loadings. However, the
original FE models are difficult to use directly in simulation because the original FE model is different
with the present state of actual structure. Therefore the FE models are updated to minimize the
differences with real structure. In the existing FE model updating method, generally translational
responses measured by accelerometers have been employed to identify the structural properties and
update numerical model controlling parameter. This paper proposes FE model updating method only
uses rotational response such as angular velocity measurement because the rotational response is more
sensitive to damage than translational response in numerical analysis on simply supported beam. First
sensitivity of translational and rotational response is investigated by sensitivity analysis on a
numerical simply-supported beam. The FE model updating is carried out for a numerical simply-
supported beam using an optimization algorithm which minimizes the gap between responses from
the actual structure and the FE model reduce. The used responses in FE model updating are natural
frequencies and rotational mode shape obtained from angular velocities measured by gyroscope
sensor. The sensitivity analysis shows that rotational responses have higher sensitivity than
translational responses to the structural and boundary condition changes. Then, the updated model
using existing translational response from the experiment two FE models updated using translational
and rotational responses are compared to validate he improvement by proposed FE model updating
method. From the experiment, using rotational responses is a good enough in FE model updating
compared to existing method, using translational responses.
ii
iii
TABLE OF CONTESTS
ABSTRACT ...................................................................................................................................... i
TABLE OF CONTENTS ............................................................................................................... iii
LIST OF FIGURES ........................................................................................................................ iv
LIST OF TABLES.......................................................................................................................... vi
CHAPTER 1. INTRODUCTION ......................................................................................................................................... 1
CHAPTER 2. BACKGROUNDS ........................................................................................................................................ 3
CHAPTER 3. PRELIMINARIES ....................................................................................................................................... 7
3.1 Modeling for Numerical Analysis & System Identification of Model ........................................ 7
3.2 Sensitivity Analysis of Translational and Rotational Mode Shapes Using MAC value .... 10
3.3 FE Model Updating Process Approached by Using Rotational Response .............................. 16
CHAPTER 4. NUMERICAL SIMULATION ............................................................................................................... 17
4.1 Selecting Parameter ........................................................................................................................................ 17
4.2 Numerical Simulation using Exact Values ............................................................................................ 19
4.3 Numerical Simulation using Excitation ................................................................................................. 22
4.4 Validation of FE Model Updating using Rotational Response..................................................... 34
CHAPTER 5. EXPERIMENT ............................................................................................................................................. 35
5.1 Experimental Set up ....................................................................................................................................... 35
5.2 Measurement Data from Experiment ...................................................................................................... 36
5.3 Comparison of FE Model Updating Performance ............................................................................. 38
CHPATER 6. CONCLUSION ............................................................................................................................................. 42
REFERENCE............................................................................................................................................................ 43
iv
List of Figures
Figure 2.1: The concept of FE model updating
Figure 2.2: Iteration process of Simplex method
Figure 3.1.1: Simply supported beam model with 10 elements Figure 3.1.2: Theoretical beam with 11 node and 22 DOFs
Figure 3.2.1: The 1st, 2nd, 3rd Translational Mode Shape
Figure 3.2.2: The 1st, 2nd, 3rd Rotational Mode Shape
Figure 3.3.1: The MAC sensitivity analysis of Material properties
Figure 3.3.2: The sensitivity analysis of Sectional properties
Figure 3.3.3: The sensitivity analysis of rotational stiffness of boundary condition and damage by
adding mass
Figure 4.1: Compared MAC sensitivity of beam Thickness – 2nd elements and 6th elements
Figure 4.2: Height decreases in 2nd, 4th, 6th element to select damage location
Figure 4.3: The sensor location to get simulation data for each case
Figure 4.4.1: The PSD of simulated for Baseline beam
Figure 4.4.2: The PSD of simulated angular velocity for Case 1
Figure 4.4.3: The PSD of simulated angular velocity for Case 2
Figure 4.5.1: Comparison of Mode Shape for Base
Figure 4.5.2: Comparison of Mode Shape for Case 1
Figure 4.5.3: Comparison of Mode Shape for Case 2
Figure 4.6.1: natural frequency by updating in Baseline
Figure 4.6.2: natural frequency by updating inCase1
Figure 4.6.3 : natural frequency by updating in Case2
Figure 4.7.1: Change of Modeshape by updating of Baseline is expressed as MAC value.
Figure 4.7.2: Change of Modeshape by updating of Case1 is expressed as MAC value.
Figure 4.7.3: Change of Modeshape by updating of Case2 is expressed as MAC value.
Figure 4.8: The updated heights of each element by FE model updating for three cases
Figure 4.9: updated parameters and error percentage
Figure 5.1: Experimental set up for measurement
Figure 5.2.1 : Angular velocity measurement
Figure 5.2.2 : Enlarged angular velocity
Figure 5.2.3 : Acceleration measurement
v
Figure 5.2.4 : Enlarged acceleration
Figure 5.3.1 : The frequency domain, PSD of angular velocity and acceleration of before-damaged
Figure 5.3.2 : The frequency domain, PSD of angular velocity and acceleration of after-damaged
Figure 5.4 Comparison of the natural frequencies
Figure 5.5.1 : Comparision MAC value of 1st mode shape for each case
Figure 5.5.2 comparision MAC value of 2nd mode shape for each case
Figure 5.6 : The updated heights of each element by FE model updating
Figure 5.7 : The error percentage of updated parameter for each case
vi
List of Tables
Table 3.1: The spec of theoretical Beam model
Table 4.1.1: RMSD value of updated parameter in damage of 2nd element
Table 4.1.2: RMSD value of updated parameter in damage of 4th element
Table 4.1.3: RMSD value of updated parameter in damage of 6th element
Table 4.1.4: RMSD value of updated parameter in multi damage
Table 4.2: Selected cases for numerical simulation using excitation
Table 4.3.1: Natural Frequency of Baseline from Numerical Simulation
Table 4.3.2: Natural Frequency of Case1 from Numerical Simulation
Table 4.3.3: Natural Frequency of Case2 from Numerical Simulation
Table 4.4: Compare RMSD value for each damage case
Table 5.1: The accelerometer and gyroscope for measuring response
Table 5.2: Natural Frequencies of from experiment
Table 5.3.1: Comparison of FE model updating result – before damaged beam
Table 5.3.2: Comparison of FE model updating result – after damaged beam
Table 5.4: Compare RMSD value to select location of damage
1
Chapter 1
1. Introduction
In the engineering and mathematics field, the finite element method is used to numerically find
approximate solution for given conditions. The FE models are made and used for system
identification and response simulation under various unexperienced conditions. Because the finite
element method finds solutions for each element, the FE model of complex structures which have
many elements have many solutions to be solved. So the simple FE model is solved manually but for
these complex structures, there are many softwares to model finite element and to solve the response
of FE model such as ABAQUS, ANSYS, MIDAS, SAP200 and etc. To simulate using FE model
under conditions, generally initial FE model should be updated to respond the existing structure. The
FE model updating is a method to update the initial model by changing parameters. In the civil
engineering, the initial FE model is updated using the field measurement data to simulate the real
existing structure. In civil engineering, the structures are mainly made of concrete which has time
dependence feature such as aging, creep and shrinkage, FE models different from the real existing
structures need to updating to be used for simulation. Without the time dependence feature of
structures, the initial FE models need to be updated due to deterioration, construction error and so on.
By updating FE model, we could make a reference after construction and many year of operation and
assess the safety of structure to various events (e.g., earthquake, flood, typhoon, tsunami etc.) for
maintenance. [1-5]. Especially, the urban infrastructures like nuclear power plant needs the perfect
modeling and model sensitive updating skill to achieve safety by simulating to various conditions. To
updated initial FE model, by the optimization updating algorithm, parameter are changed to minimize
the gap between the response from FE model and from experiment until the objective function defined
as the gap of responses meets given value. In general, in the FE model updating process, the
compared responses are mainly natural frequency and mode shape. To improve the performance of FE
model updating, the researches related with each part are ongoing to solve problem in updating
2
process. The key issues in FE model updating are field measurement, updating parameter, objective
function, optimization techniques, and interpretation of updated parameters.
When it comes to field measurement, for measuring responses from the real structure, the
accelerometers are most widely used by measuring the translational response because to measure
translational response using accelerometer is accurate and clear even easy to measure. In the most of
existing researches, only translational responses are adopted to assess or evaluate natural mode. For
instance, [6] shows the change of natural frequency and mode shape of Bernoulli-Euler beam after
crack growth. [7] is representative research which uses FE model updating method to detect damage
of simply supported beam. [8] is research of FE model updating using natural frequency, damping
ratio, and mode shape measured by accelerometer and FBG strain sensors. [6-8] and most of existing
researches adopted only translational response to assess natural frequency and mode shape. By the
result of [6], the change of mode shape due to crack growth is little. In the [6], the aim is to detect the
damage of beam structure by checking the change of mode shape after crack growth but it is hard to
detect damage because the change of mode shape is too small to distinct. Actually, in the FE model
updating, because the changes of natural frequencies and mode shapes is not sensitive as the structure
changes, to update initial model as precise FE model which has similar response with real structure is
difficult. For simply supported beam, the translational mode shapes are sine functions whereas the
rotational mode shape are cosine functions. Therefore, on the both end, the translational mode shape is
the minimum and the rotational mode shape is maximum. Based on the [6] the damaged parts on the
continuous beam have similar boundary conditions with hinge so the rotational mode shape could be
more sensitive for the damaged structures. In other words, because the change of rotational mode
shape is expected to be more sensitive than change of translational mode shape for the non-continuous
state of beam like crack or damage, using rotational mode shape in FE model updating is also
expected more effective.
So in this paper, the FE model updating using rotational response expected to be more sensitive will
be carried out to improve the performance of FE model updating process and also the result will be
compared to the result by existing method. In the chapter 2, the basic theory related with FE model
updating will be briefly covered. In the chapter 3, the sensitivity of rotational response to change of
parameter will be analyzed and based on the result of sensitivity analysis, the FE model updating to
compare the approaches by using rotational response and translational response will be carried out by
numerical simulation in the chapter 4. And in the Chanter 5, the result of lab scale experiment to
verify the FE model updating using rotational response will be reported and the conclusion will be
followed with the possibility of using rotational response in FE model updating in the chapter 6.
3
Chapter 2
2. Backgrounds
The basic concepts for the FE model updating are reviewed briefly considering overall process.
Figure 2.1 illustrates the concept and process of FE model updating. To updated initial FE model, by
the optimization updating algorithm, parameter are changed to minimize the gap between the response
from FE model and from experiment until the objective function defined as the gap of responses
meets given value. In general, in the FE model updating process, the compared responses are mainly
natural frequency and mode shape. In case of comparing natural frequency, the response by FEM
analysis is directly compared with experimental response with the natural frequency value itself.
However, in case of comparing mode shape, the responses by finite element method analysis and
experimental result are compared by MAC value which means the mode assurance criteria. So in this
chapter, fist, the mode shape of simply supported beam and MAC value are review. Next, the portion
of natural frequency and mode shape decided by objective function is reviewed. Finally, the review
about the simplex method which controls the overall updating process is followed.
Figure 2.1: The concept of FE model updating
4
2.1. Mode shape of Bernoulli-Euler beam
From the Newton’s Second Law, the equation of motion of transverse vibrating beam is derived with
in assumption of elementary beam theory. Below summary of Bernoulli-Euler assumptions of
elementary beam theory are explained in chapter 6.3 of [9].
Bernoulli-Euler assumptions of elementary beam theory
1. When a beam deforms the principle plane remains plane.
2. When a beam undergoes bending due to deformation, an axis of the beam is not extended or
contracted to axial direction called neutral axis. The neutral axis is included by neutral surface
perpendicular to principle plane.
3. The cross sections, which are parallel to principle planes and perpendicular to neutral axis in
unbending remain plane and remain perpendicular to the bended neutral axis.
4. The beam is uniform because of linearly elastic material.
+
= ( , ), 0 < < (1)
is displacement of transverse vibration. ρ is mass density, E is elastic modulus, I is moment of
inertia and L is length of beam. A is area of cross section. denotes external excitation. In chapter
12.2 of [10] the above equation is derived called the differential equation of motion governing
transverse vibration. And above assumptions of elementary beam theory are applied in equation (1).
Mode Shape of Bernoulli-Euler beam
In case of free vibration, the right side of equation (1) is zero. Then the equation could be solved as;
( ) = ℎ + ℎ + ℎ + ℎ (2)
( , ) = ( ) ( − ) (3)
where V is displacement by assumption of harmonic motion and is the eigenvalue solution and four
are amplitude solution. On simply supported at both ends, the boundary condition is like below
=
= 0 (4)
The equation (2) could be solved using boundary condition expressed in equation (4). When solving
5
equation (2), ‘ =
, =
, ⋯ =
and = = =0’ is obtained. Then V is expressed as
multiple of arbitrary amplitude factor and sin function. The equation for mode shapes becomes
(x) = sin
(5)
where the mode has been normalized by C=1 and renamed as The mode shape of a uniform
simply supported beam is illustrated in Figure. The mode shape from translational DOF is sinusoidal
function based on sine function while the modes shape from rotational DOF which is based on cosine
function.
2.2. MAC value
Because the mode shape is not value but it is shape, to compare the mode shape, the MAC value has
been adopted in FE model updating. The Modal Assurance Criteria, MAC is index for correlation
between mode shape vectors introduced by Allemang and Brown [11]. (1982). Close to 1 of MAC
means the mode shape vectors have high similarity and close to 0 of MAC means the mode shape
vector has no similarity. As shown in equation (6), actually, the MAC value is cosine of angle
between two vectors - the mode for comparison and the mode for reference. In FE model updating,
the MAC value employed to compare mode shape will be updated close to 1.
= (
)
(( )
)(( )
) (6)
The MAC value employed in FEMU is used in comparing the mode shape from finite element method
of initial model or updating model and mode shape of experimental result.
2.3. Objective function
In the FE model updating process, the responses of initial model or updating model and actual model
such as natural frequencies and mode shapes are compared and updated minimize the gap of
responses by changing parameters. In this optimization process, the objective function is used. In
other words, the Fe model updating method changes the parameters which affect the modal properties
of initial FE model and minimize the difference of modal properties from initial FE mode and real
structure. On the way, frequencies and mode shapes of initial FE model are updated and the optimal
6
parameters are found. To find optimal minimization of difference, the objective function is adopted in
FE model updating algorithm. Möller and Friberg [12] presented the best objective function for civil
engineering scale as;
= ∑ , , + , , (7)
where , is normalized residual function of natural frequency and , is normalized residual
function of mode shape. , and , denote the portion of each residual function.
, = , ,
,
(8)
, =( )
(9)
where , and , are the i-th natural frequency from the real structure and FE model. The
denotes modal assurance criterion value which is covered in previous chapter 2.2 between i-th
mode shape from the real structure and FE model.
2.4. Simplex method
Simplex method is an algorithm to find optimum
solution where the gap of response of reference
and response of evaluation is smallest by iteration
as shown in figure 2.2. The objective function is
improved on each step until the objective function
cannot be improved. The entire FE model updating
process in this research is carried out Nelder-Mead
simplex method. The Nelder-Mead simplex
method [13] is widely used in FE model updating.
The principle of algorithm is simple and this
algorithm doesn’t use differential value to objective function. Comparing with Genetic Algorithm
which is also used for FE model updating the speed of calculation is prominent. Also Simplex method
is more efficient than Genetic Algorithm in estimating the system parameters.
Figure 2.2: Iteration process of Simplex method
0 20 40 60 80 100 120 1400
0.5
1
1.5x 10
-6 현재 함수 값 : 5.31465e-09
7
Chapter 3
3. Preliminary
The goal of this research is comparison of the performance of FE model updating using translational
response and the rotational response. Because the translational mode shape has limitation of
sensitivity, using translational response is difficulty in distinguishing the transformation of structure
from original shape of structure. However, using rotational mode shape is expected to be more
sensitive than using translational mode shape because rotational mode shape is type of half of cosine
function which has no zero value on the boundary even though no one tried FE model updating using
rotational mode shape ever. So, before the FE model updating is carried out, the sensitivity analysis of
translational modeshape and rotational modeshape from numerical simply supported beam model is
carried out first in this chapter following the process of Jaishi et al. [14] where the structural FE model
updating using vibration test result. In this chapter, the numerical beam model for basic system
identification is made and the sensitivity analysis to various parameters is carried out. To compare the
sensitivity, the modeshape is estimated by MAC value covered in previous chapter 2.2.
3.1. Modeling for Numerical Analysis & System Identification of Model
In theoretical beam model, the elements could be considered as a thick cross section. Then the divided
elements also satisfy the beam theory of 1st and 3rd assumption based on Newton’s Law.
Table 3.1: The spec of theoretical Beam model
Number of Elements 10
Width of Section 0.08 m
Height of Section 0.01 m
Elastic Modulus 200 GPa
Density of Steel 7850 kg/m
Boundary condition 0 N ∙ M/rad
8
Figure 3.1.1: Simply supported beam model with 10 elements.
Figure 3.1.2: Theoretical beam with 11 node and 22 DOFs.
A simply-supported beam with 2m length following the assumptions of Euler-Bernoulli beam theory
is modeled in MATLAB as shown in figure 3.1.1. The beam is composed of 10 elements with 0.2m
and rectangular section. The detailed specs of beam are on the Table 3.1. Each element has same
rectangular uniform section and same material properties such as elastic modulus and mass density.
The 10 elements beam model has 11 nodes and each node is named such as ‘N1’, ‘N2’, ‘N3’… as
shown in figure 3.1.2. Because each node has 2 degree of freedom by translational displacement and
rotational displacement, the beam has 11 translational-DOFs and 11 rotational-DOFs. Note that
because the vertical displacement is 0 on the boundary node, the translational response at first node
and last node is zero. In this case, the number of total DOF is 20 not 22. To analyze the Modal
properties of translational response and rotational response, the natural frequencies and mode shapes
are obtained by solving Eigen-value problem. The first three natural frequencies of beam model are
5.792Hz, 24.015Hz, and 57.197Hz. The first three mode shapes by translational response are shown
in figure 3.2.1 and the first three mode shapes by rotational response are shown in figure 3.2.2. The
obtained 1st translational mode shape is half of sine function and the obtained 1st rotational mode
shape is half of cosine function like theory. By calculating modal properties, the system identification
is done and using this natural frequencies and mode shapes of base line, the next step, the sensitivity
analysis will be carried out.
9
Figure 3.2.1: The 1st, 2nd, 3rd Translational Mode Shape
Figure 3.2.2: The 1st, 2nd, 3rd Rotational Mode Shape
2 4 6 8 10
Node
Translational Mode shape
1st
2nd
3rd
2 4 6 8 10
Node
Rotational Mode shape
1st
2nd
3rd
10
3.2. Sensitivity Analysis of Translational and Rotational mode shapes
The transformation of structure such as damage or deterioration affects the parameters such as height
of section, elastic modulus. Because the changes of these parameters make the changes of modal
properties, inversely it is possible to detect transformation by analyzing and comparing modal
properties using MAC value which indicates the similarity of mode shapes vectors as covered in
chapter 2. Though the natural frequency from translational responses is same with the natural
frequency from rotational responses, the translational mode shapes are different from rotational mode
shapes. Therefore, the change of MAC values caused by the change of parameters is different as type
of responses. In this chapter, the analysis is carried out to compare the sensitivity of parameters vs.
MAC value in two types of response - translational mode shape and rotational mode shape. The
parameters used in this analysis are like below.
3.2.1. Material Property
The material properties are character of the materials (such as stone, steel and concrete) consisting
structures. Mass of density or elastic modulus is included in material property. In civil engineering
field, the change of material properties is caused by inner damage or change of inner state. For
example, in case of concrete structure, because of the time dependent phenomena of concrete, the
changes of external conditions and deformation by cyclic applied load, the change of material
properties is caused in concrete structure. So recently the NDE methods are adopted to detect the
inner state of concrete or damaged. In this chapter, the MAC sensitivity analysis is done by
controlling the elastic modulus and mass density.
(1) Elastic Modulus
The sensitivity analysis to elastic modulus is carried out by controlling the elastic modulus of the 6th
elements on the modeled beam whose original elastic modulus is 200 GPA and the result with in the
first three modes is shown in Figure 3.3.1. The change of MAC value from rotational mode shape is
more sensitive than the change of MAC value of translational mode shape as change of elastic
modulus. In addition, when the elastic modulus decreases, the MAC value decreases more sensitive.
(2) Mass density
The sensitivity analysis to mass density is carried out by controlling density of the 6th elements on
the modeled beam whose original mass density is 7850 kg/m and the result with in the first three
modes is shown in figure 3.3.1. In only 1st mode, the change of MAC value from rotational mode
11
shape is more sensitive than the change of MAC value of translational mode shape as change of mass
density. However from the next modes, the change of MAC value from rotational mode shape is less
sensitive than the change of MAC value of translational mode shape as change of mass density. The
difference of MAC sensitivity in 1st mode is much smaller than the difference of MAC sensitivity in
next modes.
Figure 3.3.1: The MAC sensitivity analysis of Material properties
150 200 250 30099.7
99.75
99.8
99.85
99.9
99.95
100
Elastic Modulus, E [GPa]
MA
C v
alu
e (
100%
)
MAC value of mode #1
translational
rotational
7.2 7.4 7.6 7.8 8 8.2 8.4 8.6
x 105
99.95
99.96
99.97
99.98
99.99
100
100.01
mass density, rho
MA
C v
alu
e (
100%
)
MAC value of mode #1
translational
rotational
150 200 250 30099.7
99.75
99.8
99.85
99.9
99.95
100
Elastic Modulus, E [GPa]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
7.2 7.4 7.6 7.8 8 8.2 8.4 8.6
x 105
99.95
99.96
99.97
99.98
99.99
100
100.01
mass density, rho
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
150 200 250 30099.7
99.75
99.8
99.85
99.9
99.95
100
Elastic Modulus, E [GPa]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
7.2 7.4 7.6 7.8 8 8.2 8.4 8.6
x 105
99.95
99.96
99.97
99.98
99.99
100
100.01
mass density, rho
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
12
3.2.2. Sectional Property
Sectional properties are factors to be determined by the shape of section. From the sectional
properties, the area and moment of inertia are calculated to analyze the bending in beam structure. The
sectional properties are changed when the external load is applied on structure to make
transformations or scratches. In addition, the section of concrete based structure could be changed due
to creep and shrinkage and the effective section could be changed by crack growth.
(1) Height of beam
The sensitivity analysis to height of section is carried out by controlling height of the 6th elements on
the modeled beam whose original height is 10 cm and the result with in first three modes is shown in
figure 3.3.2. The changed eight of beam is from 8cm to 12cm. The change of MAC value from
rotational mode shape is more sensitive than the change of MAC value of translational mode shape as
change of height of beam. In addition, when height of the 6th elements decreases, the MAC value
decreases more drastic.
(2) Width of beam
The sensitivity analysis to width of section is carried out by controlling width of the 6th elements on
the modeled beam whose original height is 8cm and the result with in the first three modes is shown
in figure 3.3.2. The tendency is similar with case of height of beam as expectation because both height
and width of section is related with 2nd section moment and area of section. Because the 2nd section
moment is calculated by multiplying width and cube of height, the height is more sensitive parameter
than width to MAC value. The change of MAC value from rotational mode shape is more sensitive
than the change of MAC value of translational mode shape as change of width of beam. In addition,
when width of the 6th elements decreases, the MAC value decreases more drastic.
13
Figure 3.3.2: The sensitivity analysis of Sectional properties
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)MAC value of mode #1
translational
rotational
6.5 7 7.5 8 8.5 9 9.5
x 10-4
99.7
99.75
99.8
99.85
99.9
99.95
100
Width of Beam, b [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #1
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
6.5 7 7.5 8 8.5 9 9.5
x 10-4
99.7
99.75
99.8
99.85
99.9
99.95
100
Width of Beam, b [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
6.5 7 7.5 8 8.5 9 9.5
x 10-4
99.7
99.75
99.8
99.85
99.9
99.95
100
Width of Beam, b [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
14
3.2.3. Boundary Conditions – end of beam
The boundary condition of both end of simply supported beam is mentioned in previous chapter 2.1.
As equation (6), the transverse displacement of beam is zero and the moment is also zero. However
when the damage is detected on boundary the boundary conditions are considered as changed. In this
study, the rotational change of boundary is considered. The sensitivity analysis to boundary condition
is carried out by controlling rotational of stiffness of boundaries on the modeled simply supported
beam whose original rotational stiffness is 0 and the result with in the first three modes is shown in
figure 3.3.3. The change of MAC value from rotational mode shape is more sensitive than the change
of MAC value of translational mode shape as increase of rotational stiffness of both boundaries.
3.2.4. Added Mass
The sensitivity analysis to add mass is carried out by adding mass on the 6th elements on the modeled
beam and the result with in first three modes is shown in figure 3.3.3. In only 1st mode, the change of
MAC value from rotational mode shape is more sensitive than the change of MAC value of
translational mode shape as the mass added on 6th elements. However from the next modes, the
change of MAC value from rotational mode shape is less sensitive than the change of MAC value of
translational mode shape as the mass added on 6th elements. The difference of MAC sensitivity in 1st
mode is much smaller than the difference of MAC sensitivity in next modes.
15
Figure 3.3.3: The sensitivity analysis of rotational stiffness of boundary condition and damage by
adding mass
1 2 3 4 5 6 7 8 9 10
x 104
100
100
100
100
100
Rotational Stiffness(Boundary) [N*m/rad]
MA
C v
alu
e (
100%
)MAC value of mode #1
translational
rotational
0.2 0.4 0.6 0.8 199.85
99.9
99.95
100
100.05
Added mass on 2th Elements [kg]
MA
C v
alu
e (
100%
)
MAC value of mode #1
translational
rotational
1 2 3 4 5 6 7 8 9 10
x 104
100
100
100
100
100
100
Rotational Stiffness(Boundary) [N*m/rad]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
0.2 0.4 0.6 0.8 199.85
99.9
99.95
100
100.05
Added mass on 2th Elements [kg]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
1 2 3 4 5 6 7 8 9 10
x 104
100
100
100
100
100
100
Rotational Stiffness(Boundary) [N*m/rad]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
0.2 0.4 0.6 0.8 199.85
99.9
99.95
100
100.05
Added mass on 2th Elements [kg]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
16
3.3. FE model updating process approached by using rotational response
So far before simulation or experiment, the MAC sensitivity is analyzed as damage parameters based
on the theoretical modeshape. As the result, among the parameters used in sensitivity analysis, elastic
modulus, the size of section and rotational stiffness of boundary are more sensitive when using
rotational response than when using translational response. Therefore in case of damage of elastic
modulus, the size of cross section or rotational stiffness of boundary using gyroscope to measure
angular velocity is recommended in FE model updating process. In case of mass density change and
change of mass by adding, those are not appropriate parameters to employ in FE model updating
using rotational mode because using rotational response is more sensitivity only at the first mode.
Among the factors which decide the section, the height of section is more sensitive than the width of
section. The change of the height and the width affect the area of section and inertia of section
because the inertial of moment is proportion to cube of height. In summary, measuring the angular
velocity using gyroscope sensors is expected to improve the performance of the FE model updating in
several damage cases. From the next chapters, the FE model updating process is carried out based on
the responses of numerical simulation or experiment measurement. In each step, the performance and
error of updated modal properties and parameters are compared as which sensors are used.
17
Chapter 4
4. Numerical Simulation
From the numerical simulation, the transformation of original beam is defined as damage of beam.
Before verifying the FE model updating using rotational modeshapes from numerical simulation
result, appropriate damaged parameter was selected by considering the possibility of realization in
laboratory.
4.1. Selecting parameter
In the sensitivity analysis of the previous chapter 3, rotational modeshape has more sensitive
changes than translational modeshape especially to elastic modulus, size of section and boundary
rotational stiffness. Sung et al. [15] proposed new sensing method by data fusion of acceleration and
angular velocity based on that to the rotational stiffness of boundary changes such as hinge, using
angular velocity could be supplement the disadvantage of using accelerometer especially on edge or
hinge of beam. The sensitivity analysis result completed in the previous chapter 3.2.3 is accord with
the result of Sung et al.[15].
In civil engineering field most of structure is composed of concrete or steel e.g., tall buildings using
reinforced concretes, prestressed concrete bridges and tower using steel truss. The most widely used
material, concrete is not homogeneous material and for high external stress, concrete has nonlinear
behavior. In addition concrete has time dependent characteristics such as strain due to relaxation,
creep and shrinkage. The damage of concrete structure is highly related with material property –
elastic modulus. Therefore covering elastic modulus as parameter in FE model updating process is
highly recommended to show the superiority of using rotational modeshape. However for simply
supported beam made of steel in laboratory making damage cases similar to real reduces of elastic
modulus is difficult. In this paper, height of section is selected as parameter FE model updating
18
process to compare the efficiency of FE model updating processes by using rotational responses and
translational responses. In real field, the increase or decrease of height of section could be depicted as
local damage by scratch or surface deformation by cyclic load. Considering the size of section
determines the area and moment of inertia change of sectional properties suggests damage of rebar
and connection parts near the rebar in reinforced concrete structure because the section of RC is
expressed as virtual section by amount of concrete and rebar. For example, the decrease of connection
force between rebar and concrete or decreases of ratio of reinforcement bring the decrease size of
section. From the sensitivity analysis of section height in previous chapter 3.2.2, selecting the location
of damage is started by comparing the MAC sensitivity for each damage case as shown in figure 4.1.
On the next page, for the selected damage parameter, height of section, the details such as magnitude
of damage and location of damage is analyzed.
Figure 4.1: Compared MAC sensitivity of beam Thickness – 2nd elements and 6th elements
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #1
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)MAC value of mode #1
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #2
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
0.8 0.85 0.9 0.95 1 1.05 1.1 1.1598
98.5
99
99.5
100
Heigh of Beam, h [cm]
MA
C v
alu
e (
100%
)
MAC value of mode #3
translational
rotational
19
4.2. Numerical Simulation using Exact Values
The figure 4.1 compares the MAC sensitivity of 2 case of damage – damaged on 2nd element and
damaged on 4th element. As the height of 2nd element or 4th decreases, the MAC value also decreases.
On the first mode, the sensitivity is almost same but on 2nd and 3rd mode, case of damaged 2nd element
is more sensitive to change of height. Based on the difference of sensitivity of MAC, the location of
damage is selected by dividing the cases into near the boundary and near the middle of beam. As
mentioned in chapter 3.1, the natural frequency and modeshape of theoretical simply supported beam
are provided by solving Eigen value problem. In same way, the theoretical natural frequency and
modeshape of damaged beam are calculated first. The height of section is reduced from initial value,
from 1cm to 0.8cm in damaged elements. The FE model updating process using exact value found by
theoretical model is carried out to decide the locations of damage. As shown in figure 4.2, the
decreased height of 2nd element, 4th element and 6th element is assumed in each case.
Figure 4.2: Height decreases in 2nd, 4th, 6th element to select damage location
For each case, the FE model updating using theoretical frequencies and modeshape is carried out
twice. First, translational frequency and modeshapes are used in process and second rotational
frequency and modeshapes are used in process. In this analysis, the magnitude of damage and the
portion of objective function covered in chapter 2.3 are determined as scenario of damage case.
20
2nd element Magnitude of Damage
RMSD [%] / Error [%] J 1mm 2mm 3mm 4mm
Rotational
High
M.S
1.5043 /
0.46
2.5298 /
0.15
5.4771 /
0.07
7.1460 /
1.5
Low
M.S
2.8360 /
2.44
9.8396 /
2.47
10.8099 /
1.09
10.8400 /
7.32
Translational
High
M.S
2.7668 /
1.10
7.7739 /
5.44
10.0795 /
5.67
9.5180 /
2.08
Low
M.S
3.9560 /
7.06
9.8424 /
8.03
15.4028 /
22.33
20.5333 /
9.56
Table 4.1.1: RMSD value of updated parameter in damage of 2nd element
4th element Magnitude of Damage
RMSD [%] / Error [%] J 1mm 2mm 3mm 4mm
Rotational
High
M.S
3.3690 /
2.09
4.5219 /
1.11
6.9117 /
0.81
8.8759 /
1.12
Low
M.S
3.4628 /
6.16
7.4129 /
11.32
15.3223 /
33.05
35.5033 /
104.41
Translational
High
M.S
3.5997 /
3.29
6.2896 /
1.61
15.3841 /
19.96
16.4815 /
0.06
Low
M.S
4.4829 /
9.09
7.7796 /
13.52
12.9439 /
27.92
14.3720 /
8.35
Table 4.1.2: RMSD value of updated parameter in damage of 4th element
6th element Magnitude of Damage
RMSD [%] / Error [%] J 1mm 2mm 3mm 4mm
Rotational
High
M.S
2.4806 /
0.54
5.9875 /
1.82
6.6780 /
0.06
7.3096 /
0.02
Low
M.S
2.4818 /
0.56
5.9874 /
1.82
6.8917 /
0.43
8.8569 /
0.40
Translational
High
M.S
1.5813 /
0.66
4.5987 /
0.74
7.1237 /
1.98
6.9570 /
0.26
Low
M.S
2.3257 /
0.46
4.1500 /
0.47
5.9910 /
0.04
10.4440 /
1.74
Table 4.1.3: RMSD value of updated parameter in damage of 6th element
21
2nd & 5th elements Magnitude of Damage
RMSD [%] / Error [%] J 1mm 2mm 3mm
Rotational
High
M.S
3.8236 /
1.45
11.4634 /
23.23
16.0623 /
0.10
Low
M.S
4.0408 /
54.33
12.9180 /
12.85
12.4720 /
14.15
Translational
High
M.S
4.4349 /
29.26
8.0078 /
34.65
27.8197 /
13.32
Low
M.S
3.5909 /
12.42
6.9711 /
13.44
33.1946 /
156.55
Table 4.1.4: RMSD value of updated parameter in multi damage
The table 4.1.1~4.1.4 shows the percentage RMSD value of updated parameters comparing with the
assumed value of height depend on magnitude of damage, type of used modeshape in updating, the
portion of objective function to respond modeshape and the location of damage. A percentage RMSD
(root mean square deviation) method is used to quantify the accuracy of the updated heights of section:
(10)
where ℎ denotes updated height of element and ℎ denotes initial value of assumed height of
each element. The Error [%] is calculated by comparing updated parameter of damaged element and
assumed value of damaged element. The J represents objective function to inform the degree of
weight on modeshapes. High M.S means the higher weight on modeshape and Low M.S means the
higher weight on natural frequency. In the above table 4.1.1~4.1.4, the cases whose RMSD or Error
value is more than 10(%) are marked as red color because this updating process adopts exact values
from theoretical modeshapes and natural frequencies. As sensitivity analysis, using rotational
responses in model updating is more efficient than using translational responses in all analyzed
location. Considering the accuracy of updated parameters, using rotational responses is efficient for
detecting the damage of boundary of beam and using translational responses is efficient for detecting
the damage of middle of beam. The magnitude of damage is also related with the performance of
model updating because the simplex method is especially dependent on the initial point. If the height
loss is more than 4mm, adjusting the initial value from 1.0cm to 0.9, 0.8 or 0.7 or changing the
portion of objective function to respond mode shape part more is recommended. For damage on 2nd
element, the 4mm damage could be also discovered in the process of updating parameter by using the
(%) = ∑ (ℎ − ℎ )
∑ (ℎ )
× 100
22
objective function which has high weight on modeshape as shown in table 4.1.1. For damaged on 6th
element case, using rotational response is more efficient for precise updating of parameters than using
rotational response as shown in table 4.1.3. For the multiple damages as shown in table 4.1.4, though
the exact values from theoretical modal properties are employed, it is hard to update the parameters as
assumed height.
4.3. Numerical Simulation using Excitation
In the analysis of previous chapter 4.2, the overall FE model updating result is good because exact
value, the theoretical solution of initial beam, was used. Theoretical natural frequency and the
theoretical modeshape are used in this updating process to predict theoretical beam. Even though
overall updated result is good, the advantage and disadvantage of using rotational response and using
translational response are turned out. From the comparing the updated parameters, height, the
updating process based on the rotational mode has superior performance than the updating process
based on the translational mode especially in damage of 2nd element case as shown in table 4.1.1. The
FE model updating approached by rotational modeshape using high weight on modeshape sensitive to
height loss is successful for all damage case – height decrease of 2nd, 4th and 6th element. Rotational
modeshape based FE model updating is predominant to detect damage near the end of beam. Because
the goal of this research is to verify the possibility of using rotational response in FEMU, the FE
model updating using rotational response will be carried out and be compared with the translational
response for the three cases like below table 4.2.
Table 4.2: Selected cases for numerical simulation using excitation
Scenario
Base No damaged on the beam
Case1 2mm Damaged on 2nd
element
Case2 4mm damaged on 2nd element
23
Figure 4.3: The sensor location to get simulation data for each case
4.3.1 Getting numerical simulation data
Before using the modal properties measured in experiment, the FE model updating using the modal
properties from Numerical simulation with Matlab Simulink was carried out to verify the FE model
updating using rotational modeshape. From numerical simulation in this chapter, the acceleration is
measured for translational response and angular velocity is measured for rotational response. By the
Matlab Simulink, a random input excitation is created and the position of excited input is illustrated in
figure 4.3. Also Figure 4.3 shows the location of sensors to measure acceleration and angular velocity.
In each case, the position of sensors and input is same with base line. To avoid symmetric placement
of sensor, the distance between sensors are different e.g., the distance between 1st location and 2nd
location is 40cm but the distance between 2nd location and 3rd location is 60cm. In this simulation, the
24
measured responses have 5% of noise. Sampling rate of simulation is about 2.56× 70Hz ⋍180Hz and
the simulated responses are cut off within 70Hz because in this research, only the first three modes are
considered. The responses of acceleration and angular velocity by random excitation are shown in
figure 4.4.1~4.4.3. The first mode of simply supported beam is half of sine function approached by
translational-DOF while when approached by rotational-DOF the first mode is half of cosine function.
Therefore the 2nd location of sensor has high PSD in 1st mode in acceleration measurement and the 3rd
location of sensor has high PSD in 1st mode in angular velocity measurement. The Frequency
Response Function of Acceleration is distinguished from the FRF of Angular velocity. The result
gained from chapter 4. 2 could be also related to it.
Figure 4.4.1: The PSD of simulated for Baseline beam
Figure 4.4.2: The PSD of simulated angular velocity for Case 1
Figure 4.4.3: The PSD of simulated angular velocity for Case 2
0 10 20 30 40 50 60 70 80 9010
-10
10-8
10-6
10-4
10-2
freqeuncy [Hz]
PS
D
Angualar velocity measurement
1st location
2nd location
3rd location
0 10 20 30 40 50 60 70 80 9010
-6
10-4
10-2
100
102
freqeuncy [Hz]
PS
D
Acceleration measurement
1st location
2nd location
3rd location
0 10 20 30 40 50 60 70 80 9010
-8
10-6
10-4
10-2
freqeuncy [Hz]
PS
D
Angualar velocity measurement
1st location
2nd location
3rd location
0 10 20 30 40 50 60 70 80 9010
-6
10-4
10-2
100
102
freqeuncy [Hz]
PS
D
Acceleration measurement
1st location
2nd location
3rd location
0 10 20 30 40 50 60 70 80 9010
-8
10-6
10-4
10-2
freqeuncy [Hz]
PS
D
Angualar velocity measurement
1st location
2nd location
3rd location
0 10 20 30 40 50 60 70 80 9010
-6
10-4
10-2
100
102
freqeuncy [Hz]
PS
D
Acceleration measurement
1st location
2nd location
3rd location
25
The natural frequencies from simulation result are shown in table 4.3.1~3. Though the natural
frequencies of first three modes are not perfectly coincide with the theoretical value, the over all
measured natural frequencies from numerical simulartion are appropirate considering the erorr of
measurement. The natural frequency of beam is changed as damage which adjust the height of section.
In FE model updating process, the initial model is updated by changing the parameters to update the
natural frequency for reducing the gap between the natural frequcy from measurement and updated
natural frequency. The modeshape is also updated following the algorithm which reduces the gap
between MAC value of modeshape vectors from measurment and mode shape vectors of updated
model. Likewise natural frequency, the mode shape is also changed as decrease of the height of
section defined as damage in this paper. In figure 4.5.1~4.5.3, the greenline depicts the the modeshape
of initial beam and the redline depicts the theoretically transformed modeshape of damaged beam.
The transfromation of modeshape comparing with the initial modeshape is outstading in case of
rotational mode as expected with the result of sensitiviry analysis in the previous chapter 3.
Table 4.3.1: Natural Frequency of Baseline from Numerical Simulation
[Hz] 1st mode 2nd mode 3rd mode
Rotational mode 5.781 24.028 57.283
Translational mode 5.780 23.997 57.078
Theoretical 5.792 24.015 57.197
Table 4.3.2: Natural Frequency of Case1 from Numerical Simulation
[Hz] 1st mode 2nd mode 3rd mode
Rotational mode 5.702 22.934 54.478
Translational mode 5.715 23.084 54.289
Theoretical 5.705 23.012 54.271
Table 4.3.3: Natural Frequency of Case2 from Numerical Simulation
[Hz] 1st mode 2nd mode 3rd mode
Rotational mode 5.496 20.741 49.891
Translational mode 5.447 20.721 49.753
Theoretical 5.447 20.760 49.842
26
Figure 4.5.1: Comparison of Mode Shape for Base
2 4 6 8 10
Node
Rotational Mode Shape #1 base
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #1 base
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #2 base
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #2 base
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #3 base
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #3 base
initial
damaged
measure
updated
27
Figure 4.5.2: Comparison of Mode Shape for Case 1
2 4 6 8 10
Node
Rotational Mode Shape #1 case1
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #1 case1
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #2 case1
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #2 case1
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #3 case1
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #3 case1
initial
damaged
measure
updated
28
Figure 4.5.3: Comparison of Mode Shape for Case 2
2 4 6 8 10
Node
Rotational Mode Shape #1 case2
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #1 case2
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #2 case2
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #2 case2
initial
damaged
measure
updated
2 4 6 8 10
Node
Rotational Mode Shape #3 case2
initial
damaged
measure
updated
2 4 6 8 10
Node
Translational Mode Shape #3 case2
initial
damaged
measure
updated
29
4.3.2 Comparison of FEMU Performance using excitation
To achieve goal of this research, the performance of FE model updating using rotational response
from simulation response is evaluated. To compare the performance of FE model updating using
rotational response and translational response, the damage is on 2nd element as result of FE model
updating using exact value as chapter 4.2. Though the performance of FE model updating using
translational response could be improved by changing objective function to consider natural
frequency more, the adjusting the objective function to consider natural frequency more is excluded
because, in this research, based on the MAC value sensitivity of rotational response to damage, the
objective function is designed to consider modeshape. By changing the parameters, the initial FE
model is updated to have modal property proximate with the modal property from simulation. To
reduce the gap between modal property of FE model and modal property from simulation the FE
model is repeatedly updated following simplex method algorithm mentioned in previous chapter 2.4.
The updated natural frequencies optimizing the objective function by simplex method are shown in
figure 4.6. The natural frequencies are updated with measured natural frequencies for all case. The
natural frequencies in both translational response and rotational response are updated from initial
value to almost same value of natural frequency from measurement. The updated mode shape for each
case is marked as shown figure 4.5.1~4.5.3 and the figure 4.7 shows the updated mode shapes for
each case using MAC value. Initial and updated MAC value is calculated by comparing with the
modeshape vector from simulation measurement. The initial mode shapes are updated to have MAC
value near 1 for all case whether using rotational response or translationa response. In the case 1 and
case2, when using rotational response, the MAC value is lower than when using translational response
for same damage before updating but after updating the MAC value get close to 1 in all of cases –
updated to measurement. The figure 4.8 shows the updated parameter, height for each case. For all
cases, the updated parameters has less error using rotational response the updated parameters than
using translational response as shown in table 4.4 and figure 4.9. For baseline, updating process using
rotational response updated parameters within 5% error but using translational response updated with
large error especially to one parameter-9th element. For case 1 and case 2, both methods could detect
the location of damage and magnitude of damage in the damaged element, but using rotational
responses has less error than using translational responses also.
30
Figure 4.6.1: natural frequency by updating in Baseline
Figure 4.6.2: natural frequency by updating inCase1
Figure 4.6.3 : natural frequency by updating in Case2
1 2 30
10
20
30
40
50
60frequency updated in baseline
mode
Natu
ral
Fre
quen
cy
initial
t-measurment
t-updated
r-measurment
r-updated
theoretical
1 2 30
10
20
30
40
50
60frequency updated in case1
mode
Natu
ral F
requ
en
cy
initial
t-measurment
t-updated
r-measurment
r-updated
theoretical
1 2 30
10
20
30
40
50
60frequency updated in case2
mode
Natu
ral
Fre
quen
cy
initial
t-measurment
t-updated
r-measurment
r-updated
theoretical
31
Figure 4.7.1: Change of Modeshape by updating of Baseline is expressed as MAC value.
Figure 4.7.2: Change of Modeshape by updating of Case1 is expressed as MAC value.
Figure 4.7.3: Change of Modeshape by updating of Case2 is expressed as MAC value.
1 2 30.7
0.75
0.8
0.85
0.9
0.95
1MAC value of Baseline
mode
MA
C v
alu
e
t-initial
t-updated
r-initial
r-updated
1 2 30.7
0.75
0.8
0.85
0.9
0.95
1MAC value of case1
mode
MA
C v
alu
e
t-initial
t-updated
r-initial
r-updated
1 2 30.7
0.75
0.8
0.85
0.9
0.95
1MAC value of case2
mode
MA
C v
alu
e
t-initial
t-updated
r-initial
r-updated
32
Figure 4.8: The updated heights of each element by FE model updating for three cases
2 4 6 8 100
0.5
1
1.5
heig
ht
[cm
]
Updated height - Baseline
assumed acc gyro
2 4 6 8 100
0.5
1
1.5
he
ight
[cm
]
Updated height - Case1
assumed acc gyro
2 4 6 8 100
0.5
1
1.5
heig
ht
[cm
]
Updated height - Case2
assumed acc gyro
33
Table 4.4: Compare RMSD value for each damage case
RMSD(%) Baseline Case1 Case2
Rotational mode 1.627 5.484 7.142
Translational mode 6.496 8.037 9.960
Figure 4.9: updated parameters and error percentage
2 4 6 8 100
10
20
30
40
50
perc
enta
ge
[%
]
Error of updated height - Baseline
acc gyro
2 4 6 8 100
10
20
30
40
50
pe
rce
nta
ge
[%
]
Error of updated height - Case1
acc gyro
2 4 6 8 100
10
20
30
40
50
pe
rcen
tag
e [
%]
Error of updated height - Case2
acc gyro
34
4.4 Validation of FE Model Updating using Rotational Response
The FE model updating to find the damaged location and the damaged height using rotational
modeshape and translational modeshape from the simulation using matlab Simulink is carried out. In
this process, location of damaged is confined as 2nd element of beam and the objective function is
escaped from only focusing on the natural frequency. The magnitude of damage is only controlled as
case. As a result, the FE model updating performance using rotational responses is more efficient than
the FE model updating performance using translational responses. The percentage RMSD of updated
parameter is lower, overall error percentage of each element is lower and the modeshape is also
updated closer to modeshape from measurement when using rotational response. From numerical
simulation result, the validation of using rotational response in FE model updating is completed.
Actually the accelerometer is most widely used sensor to measure structure because of sensitive to all
of responses and the response is clear and accurate, the validation using rotational sensor is
challenging. In next chapter, The FE model updating is carried out from the experimental
measurement to verify FE model updating using rotational responses. Because using gyroscope sensor
to measure angular velocity is more sensitive to damage of each parameter, the result and performance
of model updating could be also improved as simulation result.
35
Chapter 5
5. Experiment
In this chapter, the FE model updating is carried out using modal properties from the experimental
measurement to verify the FE model updating using rotational response. The transverse responses of
before-damaged beam and after-damaged beam are measured by accelerometers and gyroscopes
sensors.
5.1 Experimental Set up
As shown in figure 4.3, the experiment to get responses of transverse vibration is set up. Before-
damaged beam is made following the baseline and after-damaged beam is made following the case2
in figure 4.3. A steel beam, an impact hammer, 3 accelerometers, 3 gyroscope sensors and DAQ
system are prepared as shown in figure 5.1. The 2m steel beam is manufactured as numerical model
made in chapter 3 and to depict the decrease of height (damage), the part of steel beam is cut off as
shown in figure 5.1. Table 5.1 shows the spec of accelerometers and gyroscopes sensors to measure
the acceleration and angular velocity. The sensors are installed on the three location 40cm, 100cm and
Table 5.1: The accelerometer and gyroscope for measuring response
Accelerometer Gyroscope
Model PCB 353B33 ADXRS646
Sensitivity 100mV/g 9mV/°/sec
Measurement range ±50 g pk ±300°/sec
Figures
36
Figure 5.1: Experimental set up for measurement
190cm far from the left boundary of beam. PXIE-1082 of National instrument is used for data
acquisition. The sampling rate is 2048Hz and the duration time of each measurement is 180s. Each
case is measured 3times repeatedly.
5.2 Measurement data from experiment
The measured transverse responses using accelerometer and gyroscope sensor are shown as below
table 5.2 and figure 5.2 ~ 5.3. Because it is hard to find optimization solution using first three modes,
the first mode and second mode (third mode is not included unlike the simulation) are used in the FE
Table 5.2: Natural Frequencies of from experiment
Before-damaged 1st mode 2nd mode
Rotational mode 5.647 21.844
Translational mode 5.594 21.843
Theoretical 5.792 24.015
After-damaged 1st mode 2nd mode
Rotational mode 5.333 18.360
Translational mode 5.240 18.360
Theoretical 5.447 20.760
37
Figure 5.2.1 : Angular velocity measurement Figure 5.2.2 : Enlarged angular velocity
Figure 5.2.3 : Acceleration measurement Figure 5.2.4 : Enlarged acceleration
Figure 5.3.1 : The frequency domain, PSD of angular velocity and acceleration of before-damaged
Figure 5.3.2 : The frequency domain, PSD of angular velocity and acceleration of after-damaged
0 20 40 60 80 10010
-10
10-8
10-6
10-4
10-2
freqeuncy [Hz]
PS
D
Angualar velocity measurement
1st location
2nd location
3rd location
0 20 40 60 80 10010
-6
10-4
10-2
100
102
freqeuncy [Hz]
PS
D
Acceleration measurement
1st location
2nd location
3rd location
0 20 40 60 80 10010
-10
10-8
10-6
10-4
10-2
freqeuncy [Hz]
PS
D
Angualar velocity measurement
1st location
2nd location
3rd location
0 20 40 60 80 10010
-6
10-4
10-2
100
102
freqeuncy [Hz]
PS
D
Acceleration measurement
1st location
2nd location
3rd location
0 1 2 3 4 5 6 7 8
x 104
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
time[s]
an
gula
r ve
loc
ity [
theta
/s]
1st location 2nd lodation 3rd location
1.004 1.006 1.008 1.01 1.012 1.014 1.016 1.018 1.02
x 104
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
time[s]
accele
rati
on [
m/s
2]
1st location 2nd lodation 3rd location
3.04 3.041 3.042 3.043 3.044 3.045 3.046 3.047 3.048 3.049 3.05
x 104
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
x 10-3
time[s]
an
gula
r ve
loc
ity [
theta
/s]
1st location 2nd lodation 3rd location
0 1 2 3 4 5 6 7 8
x 104
-10
-8
-6
-4
-2
0
2
4
6
8
10
time[s]
accele
rati
on [
m/s
2]
1st location 2nd lodation 3rd location
38
FE model updating process from measurement in this experiment. Figure 5.2 is time domain
measurement data and figure 5.3 is frequency domain data. To imitate stationary signal, the impulse is
repeated on input point by hammer as shown in figure 5.2.
5.3 Comparison of FE Model Updating Performance
From the measurement, the translational response and rotational response are updated as shown in
table 5.3 and figure 5.4~5. By the FE model updating, the natural frequency is updated close to
measured natural frequency reduced by damage (decrease of height of section). The mode shape is
also updated by FEMU process using responses from experiment as shown in figure 5.5 where the
MAC values are compared to measurement data – translational modeshape of initial state, updated
translational modeshape, theoretical translational modeshape, rotational modeshape of initial state,
updated rotational modeshape, and theoretical rotational modeshape. In figure 5.5.1 shows the
comparison data of 1st modeshape. The initial values are updated to have the similar response of
measurement. In figure 5.5.2 shows the comparison data of 2nd modeshape. The initial values of 2nd
mode are also updated. The MAC between measured mode shape and initial mode shape by rotational
response is 0.7321 which means quite different vectors but in this case, the mode shape is updated by
using rotational mode shape verified that has more sensitivity than translational response. Figure 5.6
illustrates the updated heights with much errors compared to the result of using simulation responses.
Even though the updated heights have much error than simulation result, using rotational response has
lower RMSD than using translational response as shown in table 5.4 and figure 5.7. For the non-
damaged beam, the RMSD (%) error of parameters by FE model updating using rotational response is
11.072% and the RMSD (%) error of parameters by FE model updating using translational response is
23.161%. For the non-damaged beam, the RMSD (%) error of parameters by FE model updating
using rotational response is 17.169% and the RMSD (%) error of parameters by FE model updating
using translational response is 27.344%. By the FE model updating using response from experiment,
using rotational response is turned out to be more efficient in model updating.
For the before damaged beam, actually the initial state of response is same with the theoretical
response but there is difference between initial response and measured response. It could be caused by
not considered conditions such as weight of sensors, the exact position of sensors. In this experiment,
the first mode and second mode are used to find the solution of simplex method because of
optimization problem. To decide the not considered conditions previous to FE model updating is one
of method to use 1st, 2nd and 3rd mode and to get the more accurate updated result.
39
Table 5.3.1: Comparison of FE model updating result – before damaged beam
Translational Rotational
Initial Updated Initial Updated
Natural
frequency
1st freq [Hz] 5.792 5.594 5.792 5.647
2nd freq [Hz] 24.015 21.844 24.015 21.844
Mode
shape
1st MAC 0.9885 0.9890 0.9991 0.9994
2nd MAC 0.9987 0.9993 0.9993 0.9999
Table 5.3.2: Comparison of FE model updating result – after damaged beam
Translational Rotational
Initial Updated Initial Updated
Natural
frequency
1st freq [Hz] 5.447 5.240 5.447 5.332
2nd freq [Hz] 20.760 18.360 20.760 18.364
Mode
shape
1st MAC 0.9890 0.9902 0.9749 0.9869
2nd MAC 0.9421 0.9927 0.7321 0.9948
40
Figure 5.4 Comparison of the natural frequencies
Figure 5.5.1 : Comparision MAC value of 1st mode shape for each case
Figure 5.5.2 comparision MAC value of 2nd mode shape for each case
0
1
2
3
4
5
6frequency updated - 1st mode
before damaged after damaged
Natu
ral
Fre
qu
enc
y
initial
t-measurment
t-updated
r-measurment
r-updated
0
5
10
15
20
25frequency updated - 2nd mode
before damaged after damaged
Natu
ral
Fre
qu
enc
y
initial
t-measurment
t-updated
r-measurment
r-updated
10.9
0.92
0.94
0.96
0.98
1MAC value of before damaged
mode
MA
C v
alu
e
t-initial
t-updated
t-theo
r-initial
r-updated
r-theo
10.7
0.75
0.8
0.85
0.9
0.95
1MAC value of after damaged
mode
MA
C v
alu
e
t-initial
t-updated
t-theo
r-initial
r-updated
r-theo
20.9
0.92
0.94
0.96
0.98
1MAC value of before damaged
mode
MA
C v
alu
e
t-initial
t-updated
t-theo
r-initial
r-updated
r-theo
20.7
0.75
0.8
0.85
0.9
0.95
1MAC value of after damaged
mode
MA
C v
alu
e
t-initial
t-updated
t-theo
r-initial
r-updated
r-theo
41
Figure 5.6 : The updated heights of each element by FE model updating
Figure 5.7 : The error percentage of updated parameter for each case
Table 5.4: Compare RMSD value to select location of damage
RMSD(%) Translational Rotational
Before damaged 23.161 11.072
After damaged 27.344 17.169
2 4 6 8 100
0.5
1
1.5
he
ight
[cm
]
Updated height - before damaged
assumed gyro acc
2 4 6 8 100
0.5
1
1.5
heig
ht
[cm
]
Updated height - after damaged
assumed gyro acc
2 4 6 8 100
20
40
60
80
100
perc
enta
ge
[%
]
Error of updated height - before damaged
acc
gyro
2 4 6 8 100
20
40
60
80
100
perc
enta
ge
[%
]
Error of updated height - after damaged
acc
gyro
42
Chapter 6
6. Conclusion
So far, the Finite Element model updating using rotational response and translational response is
carried out to retain validity of FE model updating using rotational responses. Started from the
analysis the MAC sensitivity of rotational response to damage, FE model updating using rotational
response is verified by controlling the thickness of beam in numerical simulation and in lab-scale
experiment. In the preliminary analysis, by using the rotational response in FE model updating, the
updated parameter is expected to have more accurate than using translational response especially
when the damage is occurred near the boundary and high mode-shape responded objective function is
useful in FEMU when the updated model is far from the initial. In numerical simulation, using
rotational response is also evaluated to have more efficient performance in FE model updating
compared to using translational response. In the numerical simulation, the thicknesses of beam of
each element were accurately assessed with error less than 8% when using rotational modeshape. In
the experiment, however, the performance in FE model updating when using rotational responses was
short of expectation from preliminary work and simulation result although the validity of using
rotational response is achieved by comparing the performance of using translational response in
experiment. In the experiment, natural frequencies and modeshapes are updated close to measured
data and the RMSD error of updated thicknesses is less than 18%. Comparing the error is about 28%
when using translational response, using rotational response is considerable to use in finite element
model updating. To reduce error of experiment first using the high performance gyroscope is needed.
And not only thickness of each element but also the mass of each element with sensors is selected as
parameters to be updated in finite element model updating.
43
Reference
[1] C. P. Fritzen, D. Jennewein, and T Kiefer, “Damage detection based on model updating
methods,” Mechanical Systems and Signal Processing, vol. 12, no. 1, pp. 163-186,
1998.
[2] J. M. W. Brownjohn, and P. Q. Xia, “Dynamic assessment of a curved cable-stayed
bridge by model updating,” Journal of Structural Engineering-ASCE, vol. 126, no. 2, pp.
252-260, 2000.
[3] A. Teughels, J. Maeck, and G. De Roeck, “Damage assessment by FE model updating
using damage functions,” Computers & Structures, vol. 80, no. 25, pp. 1869-1879, 2002.
[4] J. M. W. Brownjohn, P. Moyo, P. Omenzetter, and Y. Lu, “Assessment of highway bridge
upgrading by dynamic testing and finite-element model updating,” Journal of Bridge
Engineering-ASCE, vol. 8, no. 3, pp. 162-172, 2003.
[5] B. Jaishi, and W. X. Ren, “Damage detection by finite element model updating using
modal flexibility residual,” Journal of Sound and Vibration, vol. 290, no. 1, pp. 369-387,
2006.
[6] Shen, M. H., & Pierre, C. (1990). Natural modes of Bernoulli-Euler beams with
symmetric cracks. Journal of sound and vibration, 138(1), 115-134.
[7] Jaishi, B., & Ren, W. X. (2005). Structural finite element model updating using ambient
vibration test results. Journal of Structural Engineering, 131(4), 617-628.
[8] Moaveni, B., He, X., Conte, J. P., & De Callafon, R. A. (2008). Damage identification of
a composite beam using finite element model updating.Computer‐Aided Civil and
Infrastructure Engineering, 23(5), 339-359.
[9] R.R. Craig, Jr.,(2000) Mechanics of of Materials, 2nd ed., New York, Wiley.
[10] R.R. Craig, Jr., A. J. Kurdila, (2006). Fundamentals of Structural Dynamics, 2nd ed.,
New York ,Wiley.
[11] Allemang, R. J. (2003). The modal assurance criterion–twenty years of use and
abuse. Sound and vibration, 37(8), 14-23.
44
[12] P. W. Möller and O. Friberg, “Updating large finite element models in structural
dynamics,” AIAA Journal, vol. 36, no. 10, pp. 1861-1868, 1998
[13] J. A. Nelder, and R. Mead, “A simplex method for function minimization,” The
Computer Journal, vol. 7, no. 4, pp. 308-313, 1965.
[14] Jaishi, B., & Ren, W. X. (2005). Structural finite element model updating using ambient
vibration test results. Journal of Structural Engineering, 131(4), 617-628.
[15] Sung, S. H., Park, J. W., Nagayama, T., & Jung, H. J. (2014). A multi-scale sensing
and diagnosis system combining accelerometers and gyroscopes for bridge health
monitoring. Smart Materials and Structures, 23(1), 015005.
45
Acknowledgement
제일 먼저, 부족한 제게 UEE에서 공부할 수 있는 기회를 주신 윤정 수님께 감사
드립니다. 리고 졸업논문을 비롯해 석사과정 동안 도해주신 조수 사님께도 감사
의 말씀을 전합니다. 난 2년간 완 히 이해 못하는 부족한 학생에게도 친절하게 설명
을 해주신 김재홍 수님과 이영주 수님께도 감사 드립니다. 졸업논문 표 평가를 해
주신 신명수 수님, Torbol 수님께도 감사 드립니다. 요 자주 찾아 뵙 는 못했 만
정신적인 멘토가 되어주신 문수현 수님, 윤새라 수님께도 늦게나마 감사 드립니다.
2년 동안 같은 수업을 많이 들어 수업동료라고 할 수 있는 민영이와 성우 리고 Fritz
에게도 많은 도움을 았던 것 같아 고맙고 항상 많이 도와 던 연 실 동료인 현 이,
화, 은 이, R.P.에게 나는 잘해 주 못한 것 같아 고마움과 미안한 마음이 듭니다.
작년 여름 ANCRISST 학회 행 스태프로 일할 때 많이 도와주신 금은 학 에 안 계시
는 조해영 선생님께도 감사의 말씀 드립니다. 또 작년 여름, 엄청 고생했던 하얼빈
에서 만나 이것저것 도움 많이 주신 임승현 사님께도 감사의 말을 전합니다. 대학원생
활을 힘들게 한 것은 아니 만 나고 나니 때때로 스트레스를 거나, 운동을 하고 싶
을 때 같이 축 를 했던 위대 도시락 리고 참석율이 높 는 않았 만 말만
폭탄주 전문 모임인 급형에도 고맙다는 말을 전합니다. 위대 노땅들만 모인
경로당 멤 , 특히 6년넘게 함께 해온 09학 동기들에게도 고맙다는 말을 전
하고 싶습니다.
남들에게는 쉬울 도 모르는 석사졸업이 만 다리를 다쳐 수술하는 개인사정도 있고
여러 가 로 힘든 점이 있어 개인적으로 힘든 과정이었다고 생각하는데 무사히 끝마쳐서
기쁘게 생각하고 있고, 앞으로 어떤 일이 있어도 열심히 도전해봐야겠다는 생각이 듭니
다. 끝으로, 학문 연 뿐만 아니라 생활 전 에 걸쳐서 저를 2년동안 도해주신
도 수님, 심성한 수님께 깊은 감사의 뜻을 표합니다.
2015.01.16
명환드림