You are on page 1of 163

Model updating of a

‘clamped’-free beam system


using FEM TOOLS

C.S. Kraaij
DCT 2006.128

Traineeship report

Coach(es): dr.ir. R.H.B. Fey


ir. L. Kodde

Technische Universiteit Eindhoven


Department Mechanical Engineering
Dynamics and Control Technology Group

Eindhoven, January, 2007


2
Contents

Abstract 8

1 Introduction 9

2 Dynamic analysis of initial numerical beam models 11

2.1 Euler beam elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Timoshenko beam elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3 Geometrical and Physical properties of ‘beam a’ and ‘beam b’ . . . . . . . . . . . . . . . 13

2.4 Selection sensor/excitation location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.5 Eigenfrequencies and mode shapes of ‘beam a’ and ‘beam b’ . . . . . . . . . . . . . . . 16

2.6 Comparison of frequency response functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Experimental modal analysis 21

3.1 Experimental test set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Experimental test procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.3 Modal-parameter fit procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4 Model updating 29

4.1 Numerical models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.2 Comparing numerical and experimental data. . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4.2.1 Comparing initial FE model and experimental data of ‘beam a’ . . . . . . . 31

3
4.2.2 Comparing initial FE model and experimental data of ‘beam b’ . . . . . . . 32

4.3 FE Model updating process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.3.1 Update process based on the experimental data of ‘beam a’ . . . . . . . . . . 34

4.3.2 Update process based on the experimental data of ‘beam b’ . . . . . . . . . . 38

4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5 Conclusions and recommendations 41

Bibliography 43

A Sensor placement to identify 8 modes 45

B Frequency response functions FEM models 49

C Flowchart of data 55

D Frequency response function and coherency plots of Siglab 57

E Parameter Estimation from Frequency Response Measurements Using Ratio-


nal Fraction Polynomials 67

F Global Curve Fitting of Frequency Response Measurements Using Rational


Fraction Polynomials 83

G Reformulation of transfer function 93

H Frequency Response Function general viscous damping estimation 95

I Integration of Residues 105

J Derivation of the UMM mode shapes 107

K Frequency Response Function proportional damping estimation 109

L Experimental mode shapes 119

M Mode shapes of FEM TOOLS models 125

N Influence of the stiffness on the number of iterations 131

O Bayesian Parameter Estimation method in FEM TOOLS 135

P MAC matrixes after the update strategies based on ‘beam a’ 137

Q FRF of the FEM TOOLS updates with regards to ‘beam a’ 141

4
R MAC matrixes after the update strategies based on ‘beam b’ 151

S FRF of the FEM TOOLS updates with regards to ‘beam b’ 155

5
6
Abstract

Model updating is used to improve the match between the dynamic properties (eigenvalues and
eigenmodes) of a Finite Element (FE) model and test data. In this report some parameters of
initial FE models of two designs of a ‘clamped’-free beam system, are updated based on the
first five eigenfrequencies of the first five bending modes of an experimental model by using
FEM TOOLS 3.1.1. The goal of this traineeship is to learn about the possibilities of the FE model
updating process inside the commercial software package FEM TOOLS. The beam models that are
used for this purpose are a simple straight beam (‘beam a’) and a slightly different beam (‘beam
b’). The difference between the beams is the height of the beams right half.

In order to create an interesting updating situation a mismatch of approximately 10 − 15%


between the eigenfrequencies of two beam models is chosen. In M ATLAB for both beams a FE
model is built of 50 Euler beam elements. The difference between the eigenfrequencies can be
influenced by changing the height of ‘beam b’. This is done until the requirement is satisfied.
Further initially it was required that the influence of the shear stress on the eigenfrequencies is
less than 1%. Therefore, both beams model are also built using 50 Timoshenko beam elements.
The influence of the shear stress can be investigated by comparing the eigenfrequencies of the
FE models built using Timoshenko beam elements with the eigenfrequencies of the FE models
built using Euler elements. By changing the relationship between the height and the length of the
beam the influence of the shear stress on the eigenfrequencies can be reduced. Both beams are
manufactured for carrying out experimental modal analysis. Eight frequency response functions
(FRF’s) of both beam systems are determined with a Signal Analyzer and the software package
S IGLAB. Therefore the beams are excited on eight optimal locations and the acceleration response
is measured on a fixed location. These FRF’s are exported to the software package ME’ SCOPE
which is used to carry out the Orthogonal Polynomial modal-parameter fit procedure. The real
parts of the eigenvalues are very small in comparison with the imaginary parts. Therefore the
undamped angular eigenfrequencies are used in the updating process. In FEM TOOLS the initial
FE model is built (based on ‘beam a’) of fifty LINE2 elements. This element type takes the shear
stress into account. The initial FE element model is updated, with different update strategies,
based on the first five experimental eigenfrequencies corresponding to the five bending modes
of ‘beam a’ and ‘beam b’. Typical parameters which are updated are the clamping stiffnesses, the
cross-section and the second moment of area of the right half of the beam. The results show a
clear decrease in the discrepancies between the numerical and experimental eigenfrequencies.
The correctness of the updated parameters depends on the chosen strategy.

It can be concluded that FEM TOOLS is capable to improve the match between the dynamic

7
properties of a FE model and experimental data. In order to carry out a physically meaningful
model update it is important that the user has knowledge about the updating process and about
the system that is updated.

8
Chapter 1
Introduction

Model updating is used to improve the match between the dynamic properties (eigenvalues and
eigenmodes) of a Finite Element (FE) model and test data. This can be realized by making a
physically meaningful change in the values of the model parameters. The updated model can
be used for analyzing the effect of changes in the configuration or various loads, without new
experimental testing.

FEM TOOLS is a package which contains a model update utility. The goal of this traineeship
is to learn about the possibilities of the FE model updating process inside FEM TOOLS. A simple
straight beam (‘beam a’) and a similar beam model, which has a somewhat different geometry
(‘beam b’), will be used for this purpose. Before the actual model updating will be carried out,
the eigenfrequencies of both beams shall be determined experimentally and numerically. The
experimental eigenfrequencies will be compared with numerical eigenfrequencies, based on an
initial finite element model. Updating of the FE model will take place, based on differences
between the experimental and numerical eigenfrequencies.

In chapter 2 the dimensions of the beams will be determined using FE and theoretical analy-
sis. Beams ‘a’ and ‘b’ are selected so that a mismatch between 10% and 15% exists in the eigen-
frequencies. Further the influence of shear on the eigenfrequencies has to be negligibly small
(≤ 1%). The experimental tests and the modal parameter fit procedure will be described in chap-
ter 3. Both beams will be investigated experimentally. Hereby S IGLAB will be used to determine
the FRF’s and ME’ SCOPE for carrying out the modal parameter fit procedure. In chapter 4 several
update strategies for both beams will be discussed. These strategies are implemented in FEM-
TOOLS to update the FE models. The results are compared and discussed. In the last chapter
conclusions will be drawn and recommendations for future research will be given.

9
1. introduction

10
Chapter 2
Dynamic analysis of initial numerical
beam models

In Chapter 4 two simple beam models will be updated based on experimental data. For realize an
interesting update situation a mismatch of approximately 10 − 15% between the eigenfrequen-
cies of two beam models is aimed at deliberately. Further it is required that the influence of the
shear on the eigenfrequencies is less than 1%. In this chapter the dimensions of the two initial
numerical beam models will be determined. In a later stage, these beams will be manufactured
and analyzed experimentally. In this stage it is assumed that the clamp is infinitely stiff. Figure
2.1 shows the side views of the two numerical beam models. The difference between ‘beam a’ and
‘beam b’ lies in the height of the second half, i.e. free end of the beam. To determine appropri-
ate dimensions of the beams they are modeled using the finite element method to calculate the
eigenfrequencies. The height of the second part of ‘model b’ will be decreased so that the mis-
match requirement is satisfied. The 2D situation will be analyzed. Axial and torsional vibrations
will not be considered in the FE model.

(a) Beam a (b) Beam b

Figure 2.1: Numerical beam models

2.1 Euler beam elements

To determine the eigenfrequencies and modes shapes of the two beam FE models, mass and
stiffness matrices need to be build, using a finite number of beam elements. A possible beam
element is the Euler beam element.

11
2. dynamic analysis of initial numerical beam models

The mass and stiffness matrices of an Euler beam element are:

   
156 22le 54 −13le 12 6le −12 6le
ρAle 
 22le 4le2 13le −3le2 
, EI  6le 4le2 −6le 2le2 
Me = Ke = 3  
420  54 13le 156 −22le  le −12 −6le 12 −6le 
−13le −3le2 −22le 4le2 6le 2le2 −6le 4le2

where:

ρ : Density [kg/m3 ]
A : Cross-section [m2 ]
le : Element length [m]
E : Young’s modulus [N/m2 ]
I : Second moment of area [m4 ]

The mass and stiffness matrix of the system can be build by assembling the Euler beam ele-
ments. Because of the initial assumption of infinite stiff clamping, the displacement and rotation
at the clamping are zero. Hence the first two rows and columns of the mass en stiffness matrix
of the system can be deleted. Using the assembled mass matrix and the assembled stiffness ma-
trix of the total system, the eigenfrequencies and mode shapes of both beams can be determined
by solving the eigenvalue problem. In this stage we assume that there is no damping, so the
dynamics behaviour of the system can be described by:

M q̈(t) + Kq(t) = 0 (2.1)

This leads to the following eigenvalue problem:

[K − ωi2 M ]ui = 0 ⇒ [K − (2πfi )2 M ]u = 0 (2.2)

where,
ωi
fi = 2π : Eigenfrequency of mode i [Hz]
ui : Mode shape of mode i [−]

For the straight beam (‘beam a’) based on Euler theory also an analytical solution of the eigen-
value problem is available. The following formulae describe the eigenfrequencies and mode
shapes of a clamped-free beam system [Blevins, 1979]:

µ ¶1/2
λ2i EI
fi = (2.3)
2πL2 ρA

µ ¶ µ ¶ µ µ ¶ µ ¶¶
λi x λi x λi x λi x
ui (x) = cosh − cos − σi sinh − sin , i = 1, 2, 3...(2.4)
L L L L

12
2. dynamic analysis of initial numerical beam models

where [Blevins, 1979]:

i λi σi
1 1.87510407 0.734095514
2 4.69409113 1.018467319
3 7.85475744 0.999224497
4 10.99554073 1.000033553
5 14.13716839 0.999998550

2.2 Timoshenko beam elements

Influence of shear stress is neglected in the Euler formulation. To determine the effect of the
shear stress on the eigenfrequencies and mode shapes, Timoshenko beam elements can be used.
The mass matrix of the Timoshenko beam element is identical to the mass matrix of the Euler
beam element. The element stiffness matrices differ. In the Timoshenko beam element stiffness
matrix the dimensionless coefficient Φ is added, to correct for the shear influence.

 
12/le2 6/le −12/le2 6/le
EI  6/le 4 + Φ −6/le 2 − Φ 12EI
Ke =   ,Φ =
le (1 + Φ) −12/le −6/le 12/le2 −6/le 
2
GAs le2
6/le 2 − Φ −6/le 4 + Φ

where:

G : Shear modulus [N/m2 ]


As : Shear-resisting area [m2 ]

2.3 Geometrical and Physical properties of ‘beam a’ and ‘beam b’

The material and geometric properties are selected based on the eigenfrequencies. The dimen-
sions of ‘beam a’ and ‘beam b’ will be chosen so that the difference in eigenfrequencies is between
10% and 15%. Furthermore it is required that influence of the shear on the eigenfrequencies is
less than 1%. The analyses of the eigenfrequencies will follow in section 2.5. The selected material
properties (the beams are made of S235) and geometrical properties are collected in table 2.1.

For the calculation of the Timoshenko element stiffness matrix the following parameter val-
ues are used:
1
Shear modulus (G) : G = 2(1+ν) E [N/m2 ]
Effective shear area (As ) : As = κA [m2 ]
In table 2.1 also the mass of one accelerometer is given. The number and selection of the axial
location of the accelerometers will be discussed in section 2.4. The masses of the accelerometers
are taken into account by the calculations of the eigenfrequencies and mode shapes.

13
2. dynamic analysis of initial numerical beam models

Table 2.1: Beam properties


Material properties
Young’s modulus (E) 2.1 × 1011 [N/m2 ]
3
Poisson number (ν) 10 [−]
Density (ρ) 7850 [kg/m3 ]
Geometrical properties
Timoshenko shear coefficient (κ) 5/6[−]
Height (h) 10 [mm]
Height 2nd half ‘beam b’ (h2 ) 7.5 [mm]
Width (w) 60 [mm]
Length (l) 500 [mm]
Accelerometer
Mass (sensor+wire) M ±40 [kg]

2.4 Selection sensor/excitation location

The experiments to experimentally determine the frequency response functions are carried out
using hammer excitation and three accelerometers. These three accelerometers will be positioned
at the same axial position of the beam. They will be distributed over the width of the beam in or-
der to detect possible torsional modes (which are absent in the FE models) in the experiment.
The added mass and the stiffness of the wire of the accelerometers have some influence on the
eigenfrequencies. The influence of the added mass can be taken in to account in the models.
Therefore the sensor location needs to be determined in this fase. The sensor locations are se-
lected based on the linear independency of the mode shapes Ue of interest. These mode shapes
are calculated using the parameter values of table 2.1. Obviously, the masses of the accelerometers
were not taken into account in this calculation. The selection method is known as the Effective
Independence Method (EIM). The first step is to delete from Ue all the rotations and inaccessible
dof’s. After this there remain m candidate dof’s. Inaccessible dof’s for ‘beam b’ from figure2.1(b)
are on the end of the beam and in the middle where the height is reduced. On these locations it is
not possible to place the sensors. For ‘beam a’ only the dof on the end of the beam is inaccessible.
The next step is to select the number of mode shapes (e) which are of interest and need to be
detected. Based on this information the ‘Fisher Information Matrix’ Fee [de Kraker, 2004] can be
determined:

t
Fee = Ume Ume (2.5)

The ‘Fisher Information Matrix’ can be used to calculate the idempotent matrix Gmm [de Kraker, 2004]:

−1 t
Gmm = Ume Fee Ume (2.6)

The idempotent matrix has the property that G2mm = Gmm and its trace equals its rank. The
diagonal terms represent the partial contribution to the rank of Gmm of each measurement dof.

14
2. dynamic analysis of initial numerical beam models

The selection method is an iterative process of subsequently deleting the dof with the smallest
term on the diagonal, followed by updating Ume , Fee and recomputing Gmm . This will be done
until the number of sensors (s = 8) is reached (m ≥ e). The selected mode shapes are the first 5
bending modes(e = 5).

Actually, it was decided that in the experiments only the transversal response was measured
at the last node but one at the free end of the beams. As stated before, at this axial position three
accelerometers are placed over the width of the beam in order to detect possible torsional modes.
Hammer excitation was carried out at the 8 selected sensor locations. The reason for this simply
was that the dynamics of the system are not changed because the three sensors stay fixed. Figure
2.2 shows the result of the selection procedure.

1 1 1 1 1 1

0.8 0.8
0.5 0.5 0.5 0.5
0.6 0.6
0 0 0 0
0.4 0.4
−0.5 −0.5 −0.5 −0.5
0.2 0.2

0 −1 −1 0 −1 −1
0 50 0 50 0 50 0 50 0 50 0 50
DOF−nr DOF−nr DOF−nr DOF−nr DOF−nr DOF−nr

1 1 1 1
Mode Shape Mode Shape
0.5 0.5 Excitation location 0.5 0.5 Excitation location

0 0 0 0

−0.5 −0.5 −0.5 −0.5

−1 −1 −1 −1
0 50 0 50 0 50 0 50
DOF−nr DOF−nr DOF−nr DOF−nr

(a) Model a (b) Model b

Figure 2.2: Excitation location

It may seem a bit striking that some excitation places are located close to each other. This,
however, may be due to the fact that 8 positions are used to identify only 5 modes. In case 8
positions would have been used to identify 8 modes, a more uniform axial distribution of the
locations would have been obtained (Appendix A). This is a result of the selected method, which
only looks at the partial contribution of each measurement dof to the linear independency. The
excitation locations, which are measured relative to the clamping, are given in table 2.2.

Locations ‘Beam a’ [mm] ‘Beam b’ [mm]


1 100 110
2 110 120
3 200 220
4 210 230
5 300 320
6 400 410
7 410 420
8 490 490

Table 2.2: Excitation locations

15
2. dynamic analysis of initial numerical beam models

The location numbers in table 2.2 will be used in section 2.6 to indicate the response location
i and the excitation location j of Hij . The accelerometers are located 490 mm of the clamping,
this location is chosen because the response on this point is large in general. Figure 2.3 shows the
positions of the three accelerometers (si ).

Figure 2.3: Location of accelerometers on the beam (top view)

2.5 Eigenfrequencies and mode shapes of ‘beam a’ and ‘beam b’

In this section the eigenfrequencies and mode shapes are calculated using the material and geo-
metric properties of table 2.1. First it is verified if the mean difference in the eigenfrequencies of
‘beam a’ and ‘beam b’, is between 10% and 15%. The calculations are carried out with a FE model
of 50 elements, built in M ATLAB. The FE models contain the masses of the 3 accelerometers (40
grams for each accelerometer+wire). The results are shown in table 2.3.

Table 2.3: Eigenfrequencies Euler beam


fi Analytic (Hertz) ‘Beam a’ (Hertz) ‘Beam b’ (Hertz) difference a,b (%)
1 33.4 33.1 36.3 9.7
2 209.4 207.7 178.2 −14.2
3 586.4 582.4 506.3 −13.1
4 1149.2 1142.5 962.0 −15.8
5 1899.7 1890.4 1632.2 −13.7

Herein the last column is derived as follow:

fb −fa
fa × 100%

The small difference (order 1%) between the analytic eigenfrequencies and the eigenfrequen-
cies of ‘beam a’ can be explained by the addition of the accelerometer masses in ‘beam a’, which
results in a small decrease in the eigenfrequencies. The mean difference between the eigenfre-
quencies between both models is 13.3%. This satisfies the previously mentioned requirement.

The influence of the shear on the eigenfrequencies can be analyzing by comparing the eigen-
frequencies of the Euler FE models with those of the Timoshenko FE models. Hence a FE model
based on the Timoshenko beam elements is build, of 50 elements, in M ATLAB. The eigenfre-
quencies of ‘beam a’ and ‘beam b’ calculated with the Timoshenko FE model are compared with
the results of the Euler FE model in table 2.4.

16
2. dynamic analysis of initial numerical beam models

Table 2.4: Eigenfrequentie Euler and Timoshenko models


fi ‘Beam a’ ‘Beam b’
Timoshenko Euler diff. (%) Timoshenko Euler diff. (%)
1 33.1 33.1 0 36.3 36.3 0
2 207.4 207.7 0.1 177.9 178.2 0.2
3 580.1 582.4 0.4 504.2 506.3 0.4
4 1134.1 1142.5 0.7 955.5 962.0 0.7
5 1868.5 1890.4 1.2 1612.6 1632.2 1.2

Herein:

fE −fT
diff.= fT × 100%

The mean difference between the eigenfrequencies of ‘beam a’ and ‘beam b’ is respectively
−0.48 and −0.50%, which satisfy the requirement (≤ 1%). This means that the influence of the
shear stress, for the indicated beam dimensions, is insignificant. Hence the analyses of the mode
shapes will carried out with the Euler FE model.

The lowest 5 bending mode shapes of ‘beam a’ and ‘beam b’ are shown in figure 2.4. These
mode shapes are normalized by dividing the mode elements by the element with the largest
modulus. The difference between ‘beam a’ and ‘beam b’ lies in the height of the second half of
the beam with the free end. So for ‘beam b’ the first half is stiffer and heavier in relation to the
second part. This explains the difference in the mode shapes.

1 1 1

0.8
0.5 0.5
0.6
0 0
0.4
−0.5 −0.5
0.2

0 −1 −1
0 50 0 50 0 50
DOF−nr DOF−nr DOF−nr

1 1
Mode Shape Beam a
0.5 0.5 Mode Shape Beam b

0 0

−0.5 −0.5

−1 −1
0 50 0 50
DOF−nr DOF−nr

Figure 2.4: Mode shapes (Euler beam models)

17
2. dynamic analysis of initial numerical beam models

To compare the mode shapes several techniques are available. The Modal-Assurance-Criterion
(MAC) is a widely use method. The MAC-number is a representation of the correlation between
two mode shapes. The MAC-number is defined as follows [de Kraker, 2004]:

H
|uie unj |2
M ACij = H H (2.7)
(uie uei )(unj unj )

where,

uei : Numerical mode shape ‘beam b’


uni : Numerical mode shape ‘beam a’
H
uei : Hermitian transpose of the numerical mode shape ‘beam b’
H
uni : Hermitian transpose of the numerical mode shape ‘beam a’

Although the MAC criterion originally was developed for comparison of experimental and nu-
merical mode shapes, here modes resulting from two different numerical models are compared.
Because the mode shapes are real in this case, the Hermitian transposed of the mode shapes can
be replaced by the normal transposed of the mode shapes. Figure 2.5 shows a graphical represen-
tation of the MAC-matrix.

0.8

0.6

0.4

0.2

0
6
6
4
4
2
2
Beam a 0 0
Beam b

Figure 2.5: Graphical representation MAC-matrix Euler models

The graphical representation (figure 2.5) of the MAC-matrix shows that the correlation be-
tween the corresponding mode shapes is high (diagonal elements close to 1). This means that
the applied change in the height of the free end of the beam, in ‘beam b’, doesn’t radically change
the mode shapes compared to ‘beam a’.

18
2. dynamic analysis of initial numerical beam models

2.6 Comparison of frequency response functions

In this section the FRF’s of ‘beam a’ and ‘beam b’ are compared. All models are undamped.
Therefore both the Euler and Timoshenko beam models are used. All FRF’s are shown in appen-
dix B. To give an example H6,8 , based on Euler beam elements, is shown in figure 2.6. In all FRF’s
in this section and in appendix B, 0 dB = 1 m/N .

|H6,8|
−50
Beam a Euler beam model
Beam b Euler beam model
(displacement/force) [dB]

−100

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H )
6,8
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

Figure 2.6: H6,8 , Euler beam models

The FRF’s clearly show the difference in dynamic behaviour between both beams. From the
2nd mode onwards the eigenfrequencies of ‘beam b’ have a lower value then the eigenfrequencies
of ‘beam a’ (see also table 2.3). The first eigenfrequentie, however, is clearly smaller for ‘beam a’
compared to ‘beam b’. The influence of inclusion of shear in the beam model is investigated in
figure 2.7 for H6,8 . Here, FRF’s based on an Euler beam model are compared with FRF’s based on
a Timoshenko beam model.Its clear that the discrepancies between the FRF’s based on the Euler
and Timoshenko models are very small in the low frequency range but grow for larger excitation
frequencies. This can be explained because shear modeling becomes more important for shorter
wavelengths, i.e. higher modes. However, for the frequency range of intrest (first five modes) the
differences can be neglected.

19
2. dynamic analysis of initial numerical beam models

|H6,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model

(displacement/force) [dB]
−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H6,8)
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

Figure 2.7: H6,8 , Euler and Timoshenko beam models

2.7 Conclusion

The models of ‘beam a’ and ‘beam b’ satisfy the requirements. Assuming that the clamping is
infinitely stiff, the mean difference between the eigenfrequencies between both models is 13.3%.
In reality the clamping of the experiments will have a finite stiffness, so in reality the eigenfre-
quencies probably will decrease somewhat. Further the influence of the shear stress is negligibly
small in the frequency range of intrest.

20
Chapter 3
Experimental modal analysis

This chapter describes the experimental test set-up, the experimental test procedure and the
modal parameter fit procedure to determine the frequency response functions, the eigenvalues
and corresponding mode shapes of ‘beam a’ and ‘beam b’ in an experimental way. The tests are
done in the DCT-lab of the Department of Mechanical Engineering at Eindhoven University of
Technology. There an experimental set-up is realized. Data acquisition will be done with S IGLAB.
The frequency response functions, determined in S IGLAB, will be fitted with the Orthogonal Poly-
nomial modal-parameter fit procedure of the experimental modal analysis package ME’ SCOPE.
Appendix C describes the data flow between the used packages.

3.1 Experimental test set-up

The experimental set-up is shown in figure 3.1. The excitation hammer that is used has a synthetic
tip and a mass of 154 gram. The hammer can generate an impact signal up to approximately
2500 Hz.

(b) Clamp

(a) Set-up (c) Sensors


Figure 3.1: Experimental set-up

21
3. experimental modal analysis

The beam is oriented in vertical direction (figure 3.1(c)). The upper end of the beam is clamped
between two metal blocks. The blocks are equipped with holes, perpendicular measurement to
the direction, for fixation to the wall. Three accelerometers are located 490 mm from the clamping
(figure 3.1(b)). The accelerometer in the middle will be used to determine the required FRF’s. The
other two FRF’s, measured on the other accelerometers, will be used for recognizing possible
torsional modes. The beam on the picture is the so called ‘beam a’. Obviously, the same set-
up is used for ‘beam b’. Although it is tried to make the clamping as stiff as possible, it will
never be infinitely stiff. Of course the finite clamping stiffness will have some influence on the
eigenfrequencies. The experimental situation is schematically shown in figure 3.2.

(a) Beam a (b) Beam b

Figure 3.2: Experimental Beams

3.2 Experimental test procedure

The excitation and response sensors which are used during the experiments are shown in figure
3.3. The hammer is used to excite the beam. This will be done on eight locations, see table 2.2. The
accelerometers are connected to the beam using wax. The hammer and accelerometers are used
to determine FRF’s of the system. The digitized time signals are collected in a Signal Analyzer
which is connected to a computer.

(a) Hammer with force transducer (b) Accelerometers

Figure 3.3: Excitation and response sensors

22
3. experimental modal analysis

The software that is used to process the measured time signals which are collected by the
Signal Analyzer, is S IGLAB. With the Dynamics Signal Analyzer (vna) utility, the FRF’s can be
determined. The first step is setting S IGLAB data acquisition parameters. This can be done in the
S IGLAB measurement setup window (see figure 3.4, [Sig, 1999]).

Figure 3.4: Siglab Measurement Setup Window

S IGLAB measures volts, hence a scaling unit has to be set for each measurement channel.
The scaling units for the used sensors can be found in table 3.1. The accelerometer in the middle
is of type 303A3, the accelerometers at the outside are of type 303A2.
Table 3.1: Scaling units
Sensors Scaling unit
Force sensor 435 [N/V ]
Accelerometer 303A2 954 [(m/s2 )/V ]
Accelerometer 303A3 885 [(m/s2 )/V ]

The sample frequency to be used in S IGLAB depends on the chosen bandwidth (frequency
range of intrest). The first 5 eigenfrequencies are lower than 2.0KHz, which is chosen as the
bandwidth. The sample frequency and the frequency resolution can be determined by the follow-
ing formulae [Sig, 1999]:

fs = 2.56 × BW (3.1)

2.56 × BW
∆f = (3.2)
N

The record length (N) is set on the maximum value, 8192 points. The beams are lightly
damped, hence the transient signal is not damped out in the measurement time (N/fs = 1.6 [s]).

23
3. experimental modal analysis

Therefore numerical damping is introduced, via application of a so-called exponential window


(figure 3.5(a)). The measured hammer impulse signal will be followed by a small noise signal.
To eliminate this noise a rectangular force window (figure 3.5(b)) is used. This is a rectangular
window which is multiplied with the measured force signal. By using these windows signal
leakage is avoided as much as possible. Figure 3.5 shows the principle of the force and exponential
window. To elucidate up the principle a simulation signal is added.

2.5 1.6
Transient signal Impact signal
2 Simulated transient signal 1.4 Simulated impact signal
Exponential window Force window
1.2
1.5
1
1
0.8
0.5
x(t)

x(t)
0.6
0
0.4
−0.5
0.2
−1
0

−1.5 −0.2

−2 −0.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
time [s] time [s]

(a) Exponential window (b) Force window

Figure 3.5: Windows

So, in S IGLAB a combination of the exponential and force windows is applied. The force win-
dow is used on channel 1 (hammer) and the exponential window is used on the other (response)
channels. There are two exponential windows available, ‘Exponential 0.1’ and ‘Exponential 0.01’.
The first one has decayed to 10% of the maximum value at the end of the frame. The second one
has decayed to 1% of the maximum value. The experiments are carried out with the ‘exponential
0.1’ window. The force window has value 1 for the first 20% of the measured time. Hereafter the
impact signal is set to zero. The trigger delay (1%) ensures that the complete excitation signal and
response signal are captured. By transforming the time signals to the frequency domain, using
FFT, the experimental FRF’s and the corresponding coherence can be determined. To increase
the reliability, the estimates will be done based on the average of 4 measurements. Estimates of
these functions are calculated using the auto power spectrum (Pxx ) of the excitations and the
cross spectrum (Pxy ) of the excitation and response ([Sig, 1999]):

P̂xy (ω)
Ĥ(ω) = (3.3)
P̂xx (ω)

The accuracy of the measured FRF’s can be evaluated partly using the coherence function:

|P̂xy (ω)|2
Ĉ(ω) = (3.4)
P̂xx (ω)P̂yy (ω)

The plots of the measured FRF’s and the corresponding coherences are shown in Appendix
D. H6,8 of ‘beam a’ is shown in figure 3.6 as an example. The plots make clear which resonance

24
3. experimental modal analysis

peaks belong to torsional modes. The influence of the torsional modes on the FRF’s of intrest is
negligible. Furthermore the coherences show that the measurement results are reliable.

|H6,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H6,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

Figure 3.6: H6,8 ‘beam a’ and corresponding coherence

3.3 Modal-parameter fit procedure

The experimental frequency response functions are used to determine the eigenvalues and mode
shapes of the system. The Orthogonal Polynomial modal-parameter fit procedure, is carried out,
in the package ME’ SCOPE. The FRF’s are imported in ME’ SCOPE using an interface with S IGLAB.
After importing the FRF’s the excitation and response points have to be indicated for each FRF.
The fit procedure starts with an indication of the number of modes (5) in the frequency range of
interest. The program selects all modes whose magnitude in the FRF is above a adaptable thresh-
old. After the modes of interest are selected the eigenvalue Pk = µk + jωk (ωk is the damped
angular eigenfrequency, µk represents the amount of damping) of each mode can be determined.
Therefore the Orthogonal Polynomial method is used. The Orthogonal Polynomial method is a
global fit procedure that fits a set of FRF’s. This methode makes use of the orthogonality prop-
erty. Appendices E [Richardson and Formenti, 1982] and F [Richardson and Formenti, 1985] contain
more information about the Orthogonal Polynomial method. If the eigenvalues of the modes
are known, the corresponding residues can be calculated. Therefore also the Orthogonal Polyno-
mial method is used. The Orthogonal Polynomial fit procedure in ME’ SCOPE assumes general
viscous damping. Moreover, light damping is assumed so that the eigenvalues and eigenmodes
occur in complex conjugate pairs. The matrix of FRF’s for a general viscously damped system

25
3. experimental modal analysis

with undercritically damped modes is formulated as follows [MEs, 2005]:

1 X³ Rk Rk∗ ´
n
H(ω) = − (3.5)
2j jω − Pk jω − Pk∗
k=1

Where,

Rk : Residue matrix (complex)


Pk : Complex eigenvalue
Rk∗ : Complex conjugate of the residue matrix
Pk∗ : Complex conjugate of the eigenvalue
Equation 3.5 can also be written in the form:

n
X X n
Ak A∗k
H(ω) = + (3.6)
−µ + j(ω − ωk ) −µ + j(ω + ωk )
k=1 k=1

Which is used in [de Kraker, 2004], see Appendix G

The plots of these fit results are shown in Appendix H. ME’ SCOPE makes a distinction be-
tween residues in displacement/force unit, residues in velocity/force unit and residues in accel-
eration/force unit. In this case the unit of the residues is acceleration/force. ME’ SCOPE contains
a function that integrates the residues to change the dimension of an FRF. The results are dis-
cussed in Appendix I. An expression for the FRF of a proportionally damped system is:

n
X uOk uTOK
H(ω) = 2 + 2jξ ωω 2
(3.7)
mk (ωOk k Ok − ω )
k=1

Where,
q
−µ
ωOk = µ2k + ωk2 , ξk = ωOk

Herein µk is the real part of the eigenvalue Pk , ωk is the imaginary part of the eigenvalue
and ωOk is the real undamped angular eigenfrequency of the undamped system, ξk is the dimen-
sionless damping coefficient and mk = uTOk M uOk is the modal mass. ME’ SCOPE calculates the
residues and the Unit Modal Mass (UMM) mode shape, i.e. mk = 1. The following formula (for
the derivation, see appendix J) is used to calculate the UMM mode shapes based on the residue
in ME’ SCOPE [Richardson, 2000]:

r
ω(k)
u(k) = × r(k) (3.8)
rij (k)

Herein rij (k) is the driving point residue. The modal parameter fit is done on the basis of
the assumption of viscous damping. Hence the UMM mode shape are complex. To use the FRF

26
3. experimental modal analysis

estimation, for an proportionally damped system, the imaginary part of the mode shapes will be
deleted. The plots of these FRF’s are shown in Appendix K. The real and imaginary part of the
UMM mode shape are plotted in Appendix L. Notice that the UMM mode shapes in ME’scope are
normalized on the mass matrix (uTOk M uOk = 1) but in the graphical representing of Appendix L
the mode shapes are normalized by dividing the mode elements by the element with the largest
modulus . The frequencies and damping of the modes are shown in table3.2.

Table 3.2: Frequencies and Damping

³ ´Beam a ³ ´Beam b
ωk ωk
Mode Frequency 2π [Hz] Damping (ξk ) [-] Frequency 2π [Hz] Damping (ξk ) [-]
1 31.9 0.0107 35.2 0.0012
2 197.9 0.0025 167.4 0.0035
3 553.0 0.0013 473.2 0.0013
4 1082.2 0.0020 898.3 0.0014
5 1781.5 0.0028 1535.6 0.0044

3.4 Conclusions

In this chapter the dynamics characteristics of the beam systems are experimentally determined.
The FRF’s which are measured in the centre of the beam are reliable. The influence of the torsion
modes can be neglected. The modal-parameter fit procedure is carried out with the Orthogonal
Polynomial method. The FRF’s of the modal-parameter fit are compared with the experimental
data, assuming viscous and proportional damping. Comparison of the FRF’s shows that propor-
tional damping is a legitimate assumption for this system. Because the damping values ξk are
very small, the update process can be carried out based on the undamped eigenfrequencies.

27
3. experimental modal analysis

28
Chapter 4
Model updating

Model updating is used to improve the match between the dynamic properties (eigenfrequencies
and mode shapes) of a FE model and test data. In this traineeship the update is carried out with
the commercial software package FEM TOOLS 3.1.1, and will be based on the eigenfrequencies.
In this chapter the FE models will be build in FEM TOOLS. First the dynamic properties (eigen-
frequencies and mode shapes) of the infinitely stiff clamped FE model, build in FEM TOOLS, will
be compared with the dynamic properties of the infinitely stiff clamped FE models build in M AT-
LAB . Subsequently the initial FE model used in FEM TOOLS is introduced. Difference update
strategies will be discussed. The results of the update strategies are analyzed and compared with
dynamics properties of the experimental beams.

4.1 Numerical models

Before the update process is started the infinitely stiff clamped FE models from Chapter 2, ‘beam
a’ and ‘beam b’ build in M ATLAB of Timoshenko elements, will be compared with the infinitely
stiff clamped FE models build in FEM TOOLS. The FE models, in FEM TOOLS, are build up from
50 one-dimensional LINE2 elements. This element type is taken the shear stress into account.
More complex element types are available in FEM TOOLS. Nevertheless this simple element type
is chosen. FEM TOOLS used the Lanczos Subspace Method (LSM) to obtain the dynamics proper-
ties of the FE models. LSM solves the following eigenvalue problem [FEM, 2005]:

[K − ωi2 M ]ui = 0 (4.1)

The same eigenvalue problem is applied in Chapter 2 for the infinitely stiff clamped FE models
in M ATLAB. In FEM TOOLS the first 5 eigenfrequencies and corresponding mode shapes are
determined. The eigenfrequencies are shown in (table 4.1) and the mode shapes are shown in
Appendix M.

29
4. model updating

Table 4.1: Eigenfrequencies Matlab and FEMtools FE models


Matlab FEMtools
fi Model a (Hz) Model b (Hz) Model a (Hz) Model b (Hz)
1 33.1 36.3 33.1 36.5
2 207.4 177.9 207.6 179.5
3 580.1 504.5 581.8 507.5
4 1134.1 955.5 1140.9 966.0
5 1868.5 1612.6 1887.1 1641.6

The eigenfrequencies of the FE models, built in FEM TOOLS, are slightly different from the
eigenfrequencies determined with the FE model in M ATLAB. This may be caused by application
of different eigenvalue solvers or by application of a slightly different beam element.

4.2 Comparing numerical and experimental data

A schematic drawing of the initial FE model built in FEM TOOLS is shown in figure 4.1

Figure 4.1: Initial FE model

The unknown parameters in the initial FE model are the radial and transversal stiffness. The
number of update iterations strongly depends on the initial guess of the clamping stiffnesses
(Ky and Kr ), see appendix N. Therefore it is essential to choose a reasonable initial radial and
transversal clamping stiffness. The updates are carried out for the experimental ‘beam a’ and
‘beam b’, using the initial FE model. The chosen parameters (Ax ,Iz ,M ) in the initial FE model
are based on the measured properties of ‘beam a’ and ‘beam b’. The measurement results are
shown in table 4.2. In the initial FE model Ax and Iz are set equal to the measured geometric
properties of the first half of the beam. In other words the initial FE model is based on ‘beam a’.

30
4. model updating

Table 4.2: Properties experimental beams


Geometric properties Beam a Beam b
h [mm] (1st half) 9.9 10
h2 [mm] (2th half) − 7.3
b [mm] 59.9 60
Ax [m2 ] (1st half) 5.93 × 10−04 6 × 10−04
Iz [m4 ] (1st half) 4.8434 × 10−09 5 × 10−09
Ax,2 [m2 ] (2th half) − 4.83 × 10−04
Iz,2 [m4 ] (2th half) − 1.95 × 10−09
Mass 3 accelerometers Beam a Beam b
M [kg] 0.012 0.012

In this traineeship only the first 5 undamped angular eigenfrequencies corresponding to the
first 5 bending mode shapes are updated. The damping of the modes is eliminated p in the up-
dating process. So, only the undamped angular eigenfrequencies (ωOK = µ2 + ω 2 ) of the
damped eigenvalues (Pk = µk + jω ³ k ) of the imported
q experimental
´ data is used in the updating
process. The damping coefficient ξk = −µ/ µ2k − ωk2 of the experimental data will used later
on, in order to determine of the FRF’s of the updated models.

4.2.1 Comparing initial FE model and experimental data of ‘beam a’

First the initial FE model (in FEM TOOLS) and the experimental data of ‘beam a’ will be compared.
Therefore an initial guess for the radial and transversal stiffness is necessary.

Ky = 107 : Initial transversal stiffness [N/m]


Kr = 105 : Initial radial stiffness [N/rad]

The eigenfrequencies of the initial FE model and the experimental data of ‘beam a’ are com-
pared in table 4.3. Herein:

Table 4.3: Eigenfrequencies of the experimental and initial FE model


fi Experiments [Hz] FEMtools FE model [Hz] diff. [%]
1 31.9 33.1 3.8
2 197.9 207.7 5.0
3 553.0 582.4 5.3
4 1082.2 1142.5 5.6
5 1781.5 1890.4 6.1

fF EM −fExp
fExp. × 100%

The results show a mismatch of approximately 5%. The difference between the corresponding
mode shapes can be analysed with the MAC matrix, see figure 4.2 and table 4.4.

31
4. model updating

Table 4.4: MAC-matrix


Experimental mode nr.
FE mode nr.
1 2 3 4 5
1 99.7 2.5 3.3 0.2 0.4
2 3.2 99.7 1.7 2.7 5.0
3 2.9 0.0 96.2 0.2 10.4
4 0.6 0.0 14.0 86.4 0.1
5 0.2 6.0 0.4 65.3 42.5

Figure 4.2: Graphical representation MAC-


matrix

The MAC-matrix shows the correlation between the mode shapes. Obviously, the diagonal
of the MAC-matrix is representing the correlation between the corresponding mode shapes. The
correlation on the diagonal is clearly decreasing after the 3rd mode shape. Apparently, the 5th
numerical modes has a higher correlation with the 4th experimental mode than with the 5th
experimental mode. A model updating procedure has been carried out in FEM TOOLS to improve
the clamping stiffnesses leading to Ky = 1.25 × 108 N/m and Kr = 8.86 × 104 N/rad. More
information on the model updating process will be given in section 4.3.

4.2.2 Comparing initial FE model and experimental data of ‘beam b’

In the initial FE model the transversal and radial clamping stiffnesses are chosen based on the
updating results with regard to ‘beam a’ as discussed in the previous subsection:

Ky = 1.25 × 108 : Initial transversal stiffness [N/m]


Kr = 8.86 × 104 : Initial radial stiffness [N/rad]

The eigenfrequencies of the initial FE model and the experimental data of ‘beam b’ are com-
pared in table 4.5. The results show a mismatch of approximately 16.5%. This mismatch is due
to the (deliberate) bad initial guesses of A2 and I2 for the right half of ‘beam b’. For A2 and I2 the
same values are taken as for the left part of ‘beam b’. So the (bad) initial FE model for ‘beam b’ is
in fact a good model for ‘beam a’.

32
4. model updating

Table 4.5: Eigenfrequencies of the experimental and FE model


fi Experiments [Hz] FEMtools FE model [Hz] diff. [%]
1 35.1 31.6 -10
2 167.1 198.5 18.8
3 473.2 555.1 17.3
4 898.2 1082.9 20.6
5 1535.5 1775.5 15.7

Herein:
fF EM −fExp
fExp. × 100%

The correlation between the experimental mode shapes and the mode shapes of the initial FE
model can be analyzed with the MAC matrix, see figure 4.3 and table 4.6

Table 4.6: MAC-matrix


Experimental mode nr.
FE mode nr.
1 2 3 4 5
1 99.1 7.2 1.5 0.1 0.4
2 4.4 96.7 0.0 0.6 1.5
3 3.9 2.1 97.2 0.7 9.0
4 0.6 0.8 5.3 95.8 0.0
5 0.3 2.7 12.1 12.6 97.2

Figure 4.3: Graphical representation MAC-


matrix

The MAC-matrix shows a surprisingly good correlation between the experimental and nu-
merical corresponding mode shapes.

4.3 FE Model updating process

In this section the FE model updating process, carried out with FEM TOOLS, is discussed. Up-
dating is carried out using a Bayesian Parameter Estimation (BPE). The BPE error which is to be

33
4. model updating

minimized is defined as follows:

E = ∆RT CR ∆R + ∆P T CP ∆P (4.2)

where,

E : Error
CR : Weighting matrix experimental data (eigenfrequencies, eigenmodes, MAC values, etc)
CP : Weighting matrix updated model parameters

The minimization is realized by the iteratief method which is discussed in appendix O. The
stop criterion in FEM TOOLS is based on the Correlation Coefficients (CC). Several CC are
available in FEM TOOLS. Because the updates in this traineeship are only based on the eigenfre-
quencies the chosen stop criterion is based on CC_ABS which is defined as follows:

N
1 X |∆fi |
CC_ABS = Cri (4.3)
N fi
i=1

where,

N : Number of eigenfrequencies (N=5)


Cri : Expected relative error
∆fi = fF Ei − fei : Difference between the (updated) FE frequency and the experimental frequency
fi = fei : Experimental frequency
For the expected relative error (Cri = 100cri ) the standard value (1%) is used for all eigenfre-
quencies. An absolute or relative stop criterion may be used in FEM TOOLS: CCt < ²1 and
|CCt+1 − CCt | < ²2 . In our case CCt = CC_ABSt . Here, a relative stop criterion is used (so,
²1 = 0 and ²2 = 0.001%) because this gives the best impression of the degree of convergence.

4.3.1 Update process based on the experimental data of ‘beam a’

In this subsection the update strategies of ‘beam a’ are discussed. The parameters of the initial
FE model of ‘beam a’ are given in the 3rd column of table 4.8. The two parameter values of the
clamping stiffness can be considered to be inaccurate whereas the other geometrical properties
can be considered to be accurate. Also the physical properties can be considered to be accurate,
see table 2.1. The experimental target eigenfrequencies of ‘beam a’ and the eigenfrequencies of
the initial FE model are given in table 4.3.

34
4. model updating

To update the initial FE model the following update strategies are selected:

Strategy 1a : Update the initial values of parameters Ky and Kr .

Strategy 1b : Update the initial values of parameters Ky , Kr and M (accelerometer mass).

Strategy 1c : Update the initial values of parameters Ax and Iz (Ax = Ax2 and Iz = Iz2 ).

Strategy 1d : Update in two steps, first the initial values of parameters Ky , Kr and M are
updated. Subsequently, the initial values of parameters Ax and Iz (Ax = Ax2 and Iz = Iz2 )
are updated.

Strategy 1e : Update in two steps, first the initial values of parameters Ax and Iz (Ax = Ax2
and Iz = Iz2 ) are updated. Subsequently, the initial values of parameters Ky , Kr and M
are updated.

Strategy 1f : Update the initial values of parameters Ax , Iz , Ky , Kr and M (Ax = Ax2 and
Iz = Iz2 ).

In the updates of ‘strategie 1d ’, ‘Strategie 1e ’ and ‘Strategie 1f ’ the values of Ax and Iz are
bounded (±5%). The geometrical parameters b and h of ‘beam a’ are also measured (2nd column
of table 4.8). The results of the applied strategies can be evaluated and compared by considering
the final values of CC_ABS, the final differences in the eigenfrequencies and the resulting up-
dated parameter values, when the stop criterion is satisfied. The results are shown in table 4.7
and table 4.8.

Table 4.7: Updating results of the applied strategies for ‘beam a’


Strategy 1a Strategy 1b Strategy 1c Strategy 1d Strategy 1e Strategy 1f
N1 ∗ 17 17 12 17 2 44
N2 ∗ − − − 2 44 −
∆f1 [%] −0.68 −0.68 −1.33 −0.85 −0.83 −0.78
∆f2 [%] 0.45 0.45 0.69 0.28 0.31 0.36
∆f3 [%] 0.50 0.50 1.07 0.33 0.36 0.41
∆f4 [%] 0.14 0.14 0.49 −0.05 −0.01 0.04
∆f5 [%] −0.29 −0.29 −0.93 −0.52 −0.46 −0.41
²2 [%] 0.001 0.001 0.001 0.001 0.001 0.01
CC_ABS [%] 0.41 0.41 0.9 0.41 0.39 0.40
* N is the number of iterations

35
4. model updating

The stop criterion of ‘Strategy 1f ’ (²2 = 0.01) is chosen higher compared with the other
strategies (²2 = 0.001). The reason is that the stop criterion ²2 = 0.001 can not be satisfied.
This is illustrated in figure 4.4. After reaching a minimum the correlation continuous increasing.

Figure 4.4: Correlation CC_ABS ’Strategie f ‘

Table 4.7 shows no perceptible difference between the results of ‘strategy 1a ’ and ‘strategy
1b ’. This is because the influence of changing the accelerometer mass on the eigenfrequencies
is small. The correlation between the experimental and numerical mode shapes can be analysed
with the MAC matrices (appendix P). All the MAC matrices of the updated models give a sat-
isfying result. The goal of the update is to improve the match between the dynamic properties
(eigenvalues and eigenmodes) of the FE model and the experimental setup. However, a physi-
cally meaningful change in the values of the model parameters is required for a useful update
result. Hence the updated parameters are compared with the average measuring results of the
experimental ‘beam a’. The updated beam height and width can be calculated by formulae 4.4.
The measurements results and the updated parameter values are shown in table 4.8.


Iz = 1 3 r
12 bh  12Iz Ax
⇒ h= ; b= (4.4)
 Ax h
Ax = bh

36
4. model updating

Table 4.8: Measured parameters of ‘beam a’ compared with the results of the update strategies
Parameters Measuring Initial Updated values
result model a b c
Transversal stiffness (Kt ) [N/m] − 1 × 107 1.23 × 108 1.23 × 108 ∗
Radial stiffness (Kr ) [N/rad] − 1 × 105 9.24 × 104 8.97 × 104 ∗
Width ‘beam a’ (b) [mm] 59.9 60 ∗ ∗ 7.9
Height ‘beam a’ (h) [mm] 9.9 10 ∗ ∗ 10.1
Mass accelerometer (M ) [g] 12 12 ∗ 10.3 ∗
Parameters Measuring Initial Updated values
result model d e f
Transversal stiffness (Kt ) [N/m] − 1 × 107 1.23 × 10 8 1.25 × 108 1.26 × 108
Radial stiffness (Kr ) [N/rad] − 1 × 105 8.97 × 10 4 9.17 × 104 9.30 × 104
Width ‘beam a’ (b) [mm] 59.9 60 63 63 63
Height ‘beam a’ (h) [mm] 9.9 10 10 10 10
Mass accelerometer (M ) [g] 12 12 10.3 11.4 11.5
* means that this parameter is not updated

The different update strategies, with the exception of ‘strategy c’, give approximately the same
results. This could be expected because the only incorrect initial parameters are the clamping
stiffnesses. The FRF’s of the updated FE models are shown in Appendix Q (herein 0 dB =
1 m/N ). Based on the FRF’s there can be concluded that all strategies, with the exception of
‘strategy c’, gives a good approximation of the experimental determined FRF’s. The results of
‘strategy c’ attract the attention because the divergent value of the width. This can be explained by
looking at the tracking of the parameters (Ax and Iz ). The track of the relative parameter change
of the cross area and the 2nd moment of area are approximately equal. So, only the width of the
beam is seriously changing.

0
Cross−section
Second moment of area
−10

−20
−86

−30 −88

−90
Parameter change (%)

−40

−92

−50
−94

−60 −96

−98
−70
2 4 6 8 10 12

−80

−90

−100
0 2 4 6 8 10 12
Iteration

Figure 4.5: Parameter tracking

37
4. model updating

The width b, however, does not influence the eigenfrequencies of a clamped-free beam, as
follows from the analytical expression[Blevins, 1979]:

s s s
λ2i EI λ2i Ebh3 λ2i Eh2
fi = = = fi = (4.5)
2πL2 ρA 2πL2 12ρbh 2πL2 12ρ

So, the modal updating procedure may given an arbitrary value for b as a result!

4.3.2 Update process based on the experimental data of ‘beam b’

In this subsection the update strategies of ‘beam b’ are considered. The experimental eigenfre-
quencies of ‘beam b’ and the eigenfrequencies of the initial FE model are given in table 4.5. The
parameters of the initial FE model of ‘beam b’ are given in the 3rd column of table 4.10. For,
the initial values of the clamping stiffnesses the initial values obtained in the previous subsection
4.3.1 are taken (for decrease the necessary iterations the Kr is chosen somewhat lower). These
values can be considered to be reasonably accurate. However, nuts may be tightened by different
moments for ‘beam a’ and ‘beam b’. Please note that the initial height of the right half of ‘beam
b’ has been chosen inaccurate on purpose in order to see if the model updating procedure works
well. To update the initial model the following update strategies are selected:

Strategy 2a : Update the initial values of parameters Ky and Kr .

Strategy 2b : Update the initial values of parameters Ax2 and Iz2 .

Strategy 2c : Update the initial values of parameters Ax2 , Iz2 , Ky and Kr .

Strategy 2d : Update in two steps, first the initial values of parameters Ky , Kr and M are
updated. Subsequently, the initial values of parameters Ax2 and Iz2 are updated.

Strategy 2e : Update in two steps, first the initial values of parameters Ax2 and Iz2 are updated.
Subsequently, the initial values of parameters Ky , Kr and M are updated.

Strategy 2f : Update the initial values of parameters Ax2 , Iz2 , Ky , Kr and M .

Notice that the parameters Ax2 and Iz2 for the right half of the beam, in these update strate-
gies, are uncoupled of respectively Ax and Iz for the left half of the beam. Also the boundary on
the parameters Ax2 and Iz2 are deleted. The results are shown in table 4.9 and table 4.10.

38
4. model updating

Table 4.9: Updating results of the applied strategies for approaching ‘beam b’
Strategy 2a Strategy 2b Strategy 2c Strategy 2d Strategy 2e Strategy 2f
N1 ∗ 14 6 47 14 6 44
N2 ∗ − − − 12 6 −
∆f1 [%] −10.57 0.01 −0.01 −0.85 −0.56 −0.02
∆f2 [%] 14.49 0.44 −0.06 0.28 0.00 −0.09
∆f3 [%] 5.84 0.44 0.33 0.33 0.13 0.37
∆f4 [%] −1.06 −0.04 0.00 −0.05 −0.17 0.00
∆f5 [%] −8.41 −0.91 −0.23 −0.52 −0.79 −0.24
²2 [%] 0.001 0.001 0.001 0.001 0.001 0.001
CC_ABS [%] 8.1 0.37 0.13 0.40 0.33 0.14
* N is the number of iterations

The results of updating strategies ‘2a ’ and ‘2d ’ are unsatisfactorily. The reason is clear, the
difference in the eigenfrequencies depended on the height of the right half of the beam. In
the update strategy ‘2a ’ only the radial and transversal clamping stiffnesses are updated. This
also explains why ‘strategy 2d ’ is not a suitable update strategy. The correlation between the
numerical and experimental mode shapes can be analysed with the MAC matrices (appendix
R). The MAC matrices confirm the comments about the update ‘strategie 2a ’ and ‘strategie 2d ’.
To verify physical meaning of the updated parameters these are compared with the measured
parameters of the experimental ‘beam b’ in table 4.10.

Table 4.10: Measured parameters of ‘beam b’ compared with the results of the update strategies
Parameters Measuring Initial Updated values
result model 2a 2b 2c
Radial stiffness (Kr ) [N/rad] − 1.25 × 108 9.92 × 10 6 ∗ 2.70 × 108
Transversal stiffness (Kr ) [N/m] − 8.86 × 104 8.13 × 10 4 ∗ 8.49 × 104
Width ‘beam b’ (b) [mm] 60 60 ∗ 61.4 61.5
Height ‘beam b’ (h) [mm] 10 10 ∗ ∗ ∗
Height right half ‘beam b’ (h2 ) [mm] 7.3 10 ∗ 7.2 7.1
Mass accelerometers (M ) [g] 12 12 ∗ ∗ ∗
Parameters Measuring Initial Updated values
result model 2d 2e 2f
Radial stiffness (Kr ) [N/rad] − 1.25 × 108 9.65 × 106 1.54 × 108 2.47 × 108
Transversal stiffness (Kr ) [N/m] − 8.86 × 104 7.68 × 104 7.86 × 104 9.59 × 104
Width ‘beam b’ (b) [mm] 60 60 53.1 61.4 61.5
Height ‘beam b’ (h) [mm] 10 10 ∗ ∗ ∗
Height right part ‘beam b’ (h2 ) [mm] 7.3 10 7.8 7.2 7.1
Mass accelerometers (M ) [g] 12 12 5.3 12.6 16.4
* means that this parameter is not updated

Good results are obtained for updating strategies 2b ,2c ,2e and 2f . The FRF’s of the updated
FE models are shown in Appendix S (herein 0 dB = 1 m/N ).

39
4. model updating

4.4 Conclusions

In this chapter FE model updates have been carried out in FEM TOOLS for two beam systems
‘beam a’ and ‘beam b’. Therefore first initial FE models were built. These initial FE models are
updated using different strategies in order to approximate the experimental eigenfrequencies of
‘beam a’ and ‘beam b’. The number of iterations depends on the quality of the initial guesses
of the parameters. The quality of the update depends on the chosen strategy. Obviously, the
most important point of the strategy is that the right parameters are selected. If two-step model
updating is carried out, it is wise to first update the parameters whose expected mismatch is
highes.

40
Chapter 5
Conclusions and recommendations

The software package FEM TOOLS is capable to improve the match between an initial FE model
and experimental data by updating the values of selected parameters of the FE model. The analy-
ses in this report are based on FE models of a ‘clamped’-free beam system and the first 5 un-
damped angular eigenfrequencies of two slightly different experimental beams. The updates are
carried out by using different updating strategies. The iterative updating process is based on the
difference between the eigenfrequencies of the FE model and the experimental data. Obviously,
it is important that the user has knowledge about the updated system and the updating process.
By selecting the right parameters and a well considered updating strategy, FEM TOOLS is capable
to approach the parameters of the experimental model. The initial FE element model is built in
FEM TOOLS using fifty LINE2 beam elements. This element takes the shear stress into account.
The results show that for the used beam dimensions the influence of the shear stress on the
eigenfrequencies is negligibly small. The modes of the experimental models are weakly damped.
Therefore undamped angular eigenfrequencies are used in the updating process.

In this report we only discuss updates based on the experimental eigenfrequencies. In the
future updates based on experimental eigenfrequencies, experimental mode shapes and MAC-
values could be considered. Also the influence of using different confidence values for the eigen-
frequencies and model parameters would be investigated.

41
5. conclusions and recommendations

42
Bibliography

[Sig, 1999] (1999). Siglab User Guide, Section 5. Spectral Dynamics, San Jose, CA, USA.

[FEM, 2005] (2005). FEMtools Theoretical Manual Version 3.1. Dynamic Design Solutions, Leu-
ven, Belgium.

[MEs, 2005] (2005). ME’scopeVES Application Note #28. Vibrant Technology, Inc., Scotts Valley,
CA, USA.

[Blevins, 1979] Blevins, R. D. (1979). Formulas for natural frequency and modeshape. Van Nostrand
Reinhold Company.

[de Kraker, 2004] de Kraker, B. (2004). A Numerical - Experimental Approach in Structural Dy-
namics. Shaker Publishing B.V.

[Richardson, 2000] Richardson, M. H. (2000). Modal mass, stiffness and damping. Technical
report, Vibrant Technology, Inc.

[Richardson and Formenti, 1982] Richardson, M. H. and Formenti, D. L. (1982). Parameter esti-
mation from frequency response measurements using rational fraction polynomials. 1st IMAC
Conference, Orlande, FL, USA. (Appendix E).

[Richardson and Formenti, 1985] Richardson, M. H. and Formenti, D. L. (1985). Global curve
fitting of frequency response measurements using rational fraction polynomials. 3rd IMAC
Conference, Orlande, FL, USA. (Appendix F).

43
BIBLIOGRAPHY

44
Appendix A
Sensor placement to identify 8 modes

45
a. sensor placement to identify 8 modes

1 1
Mode Shape
Excitation location 0.5

0.5 0

−0.5

0 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
1 2

0.5
1
0
0
−0.5

−1 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
1 2

0 1

−1 0

−2 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr

Figure A.1: Sensor placement beam a

46
a. sensor placement to identify 8 modes

1 1
Mode Shape
Excitation location 0.5

0.5 0

−0.5

0 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
1 1

0.5 0.5

0 0

−0.5 −0.5

−1 −1
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
1 2

0.5 1

0 0

−0.5 −1

−1 −2
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr
2 2

1 1

0 0

−1 −1

−2 −2
0 10 20 30 40 50 0 10 20 30 40 50
DOF−nr DOF−nr

Figure A.2: Sensor placement beam b

47
a. sensor placement to identify 8 modes

48
Appendix B
Frequency response functions FEM
models

This Appendix shows an overview of the frequency response function of four FEM models. The
Timoshenko beam models and Euler beam models are built of 50 elements. The resolution (∆f )
of the plots is 5Hz and 0dB is equal to 1m/N .

|H1,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model
(displacement/force) [dB]

−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H )
1,8
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

49
b. frequency response functions fem models

|H2,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model

(displacement/force) [dB]
−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H2,8)
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

|H3,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model
(displacement/force) [dB]

−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H )
3,8
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

50
b. frequency response functions fem models

|H4,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model

(displacement/force) [dB]
−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H4,8)
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

|H5,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model
(displacement/force) [dB]

−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H )
5,8
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

51
b. frequency response functions fem models

|H6,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model

(displacement/force) [dB]
−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H6,8)
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

|H7,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model
(displacement/force) [dB]

−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H )
7,8
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

52
b. frequency response functions fem models

|H8,8|
−50
Beam a Euler beam model
Beam b Euler beam model
Beam a Timoshenko beam model

(displacement/force) [dB]
−100 Beam b Timoshenko beam model

−150

−200

−250
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

∠(H8,8)
200

150
angle [o]

100

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
frequency [Hz]

53
b. frequency response functions fem models

54
Appendix C
Flowchart of data

The Flowchart is started in S IGLAB,where the experimental FRF’s are determined. From
S IGLAB these data can be exported to M ATLAB. The data that will be exported can be selected

55
c. flowchart of data

on the ‘File Storage’ drop-down menu in the plotting window. To extract the acquired data to
a M ATLAB workspace, type the following command in the workspace: name=vna(‘get’,‘meas’).
The interface between S IGLAB and ME’ SCOPE is realized by using the dynamically linked library
WrtBlk32.dll which is used to write ME’ SCOPE compatible Data Blocks disk files. For the in-
terface between ME’ SCOPE and FEM TOOLS some adaptation will be carried out. At first, the
UMM mode shapes and the DOF can be saved in two separated .uff files which can be joined
together in one .uff file. Subsequently, the imaginary part of the eigenvalue is replaced by the
undamped natural frequency. The data of the updated FE model can exported in a .uff file.
To convert .uff to .mat some M ATLAB script can be found on the internet site of Mathworks
(http://www.mathworks.com/matlabcentral/fileexchange/loadFileList.do).

56
Appendix D
Frequency response function and co-
herency plots of Siglab

This Appendix shows an overview of the frequency response function of the experimental models.
Herein 0dB is equal to 1(m/s2 )/N .

57
d. frequency response function and coherency plots of siglab

|H1,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H1,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H1,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H1,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.1: H1,8

58
d. frequency response function and coherency plots of siglab

|H2,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H2,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H2,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H2,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.2: H2,8

59
d. frequency response function and coherency plots of siglab

|H3,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H3,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H3,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H3,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.3: H3,8

60
d. frequency response function and coherency plots of siglab

|H4,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H4,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H4,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H4,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.4: H4,8

61
d. frequency response function and coherency plots of siglab

|H5,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H5,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H5,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H5,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.5: H5,8

62
d. frequency response function and coherency plots of siglab

|H6,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H6,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H6,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H6,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.6: H6,8

63
d. frequency response function and coherency plots of siglab

|H7,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H7,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H7,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H7,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.7: H7,8

64
d. frequency response function and coherency plots of siglab

|H8,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H8,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H8,8|
50
(acceleration/force) dB

−50
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
∠(H8,8)
200

100
[o] angle

−100

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

1
Center
0.8 Left
Coherence

Right
0.6

0.4

0.2

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure D.8: H8,8

65
d. frequency response function and coherency plots of siglab

66
Appendix E
Parameter Estimation from Frequency
Response Measurements Using Ra-
tional Fraction Polynomials

67
e. parameter estimation from frequency response measurements using rational
fraction polynomials

68
e. parameter estimation from frequency response measurements using rational
fraction polynomials

69
e. parameter estimation from frequency response measurements using rational
fraction polynomials

70
e. parameter estimation from frequency response measurements using rational
fraction polynomials

71
e. parameter estimation from frequency response measurements using rational
fraction polynomials

72
e. parameter estimation from frequency response measurements using rational
fraction polynomials

73
e. parameter estimation from frequency response measurements using rational
fraction polynomials

74
e. parameter estimation from frequency response measurements using rational
fraction polynomials

75
e. parameter estimation from frequency response measurements using rational
fraction polynomials

76
e. parameter estimation from frequency response measurements using rational
fraction polynomials

77
e. parameter estimation from frequency response measurements using rational
fraction polynomials

78
e. parameter estimation from frequency response measurements using rational
fraction polynomials

79
e. parameter estimation from frequency response measurements using rational
fraction polynomials

80
e. parameter estimation from frequency response measurements using rational
fraction polynomials

81
e. parameter estimation from frequency response measurements using rational
fraction polynomials

82
Appendix F
Global Curve Fitting of Frequency
Response Measurements Using Ra-
tional Fraction Polynomials

83
f. global curve fitting of frequency response measurements using rational fraction
polynomials

84
f. global curve fitting of frequency response measurements using rational fraction
polynomials

85
f. global curve fitting of frequency response measurements using rational fraction
polynomials

86
f. global curve fitting of frequency response measurements using rational fraction
polynomials

87
f. global curve fitting of frequency response measurements using rational fraction
polynomials

88
f. global curve fitting of frequency response measurements using rational fraction
polynomials

89
f. global curve fitting of frequency response measurements using rational fraction
polynomials

90
f. global curve fitting of frequency response measurements using rational fraction
polynomials

91
f. global curve fitting of frequency response measurements using rational fraction
polynomials

92
Appendix G
Reformulation of transfer function

In this Appendix the reformulation of the transfer function estimation for generally viscously
damped systems with undercritically damped modes is discussed. The follow formulation is
used in [MEs, 2005]:

1 X³ Rk Rk∗ ´
n
H(ω) = − (G.1)
2j jω − Pk jω − Pk∗
k=1

Where,

Rk : Residue matrix (complex)


Pk : Complex eigenvalue
Rk∗ : Complex conjugate of the residue matrix
Pk∗ : Complex conjugate of the eigenvalue
The residue matrix and complex eigenvalue can be split up in a real and imaginary part:

Rk = RkR + jRkI
Pk = µk + jωk

By using this in equation G.1 the follow formula can be obtained:

n ³
X 1/2(RkI − jRkR ) 1/2(RkI + jRkR ) ´
H(ω) = +
jω − (µk + jωk ) jω − (µk − jωk )
k=1

By rewriting the denominator and changing the order, this expression can be written in the follow
form:

n ³
X 1/2(RkI + jRkR ) 1/2(RkI − jRkR ) ´
H(ω) = +
−µk + j(ω − ωk ) −µk + j(ω + ωk )
k=1

93
g. reformulation of transfer function

By replacing numerators by Ak and A∗k the transfer function estimation for generally viscously
damped systems with undercritically damped modes is:

n
X Xn
Ak A∗k
H(ω) = + (G.2)
−µ + j(ω − ωk ) −µ + j(ω + ωk )
k=1 k=1

This formula is used in [de Kraker, 2004].

94
Appendix H
Frequency Response Function gen-
eral viscous damping estimation

This Appendix shows an overview of the frequency response function assuming general viscous
damping compared with the experimental data. Herein 0dB is equal to 1(m/s2 )/N .

95
h. frequency response function general viscous damping estimation

|H1,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H1,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H1,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
1,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.1: H18

96
h. frequency response function general viscous damping estimation

|H2,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H2,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H2,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
2,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.2: H28

97
h. frequency response function general viscous damping estimation

|H3,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H3,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H3,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
3,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.3: H38

98
h. frequency response function general viscous damping estimation

|H4,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H4,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H4,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
4,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.4: H48

99
h. frequency response function general viscous damping estimation

|H5,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H5,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H5,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
5,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.5: H58

100
h. frequency response function general viscous damping estimation

|H6,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H6,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H6,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
6,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.6: H68

101
h. frequency response function general viscous damping estimation

|H7,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H7,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H7,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
7,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.7: H78

102
h. frequency response function general viscous damping estimation

|H8,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H8,8)
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H8,8|
80

60
(acceleration/force) dB

40

20

−20

−40

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
8,8
200
Experiment
150 Fit (viscous damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure H.8: H88

103
h. frequency response function general viscous damping estimation

104
Appendix I
Integration of Residues

Because during the traineeship it is not succeed to plot a right FRF of the integrated residues.
The integrate method of ME’ SCOPE is discussed in this Appendix. In ME’ SCOPE distinction is
made between residues in displacement/force unit, residues in velocity/force unit and residues
in acceleration/force [Richardson, 2000].

Rv (k) Ra (k)
Rd (k) = = (I.1)
Pk Pk2

where,

Rd (k) = residue matrix in displacement/force units.


Rv (k) = residue matrix in velocity/force units.
Ra (k) = residue matrix in acceleration/force units.
Pk = µk + jω = pole location.

On the basis of the poles and the residues the frequency response function can be determined.
Therefore the following formula is used:

1 X³ Rk Rk∗ ´
n
H(ω) = − (I.2)
2j jω − Pk jω − Pk∗
k=1

Its clear that the dimension of these frequency response function depends on the dimension
of the residue.

105
i. integration of residues

The frequency response function can also be integrated ([Richardson, 2000]):

Hv (s) Ha (s)
Hd (s) = = (I.3)
s s2

where,

Hd (s) = residue matrix in displacement/force units.


Hv (s) = residue matrix in velocity/force units.
Ha (s) = residue matrix in acceleration/force units.
s = jω

These two methods can be compared with each other. In this report the integration of residues
is called ‘Method 1’ and the integration of the frequency response function ‘Method 2’. The results
are presented in Figure I.1. In first plot shows the result of the modal parameter fit procedure,
H6,8 . The dimension of the frequency response function can be changed with ‘Method 1’ and
‘Method 2’. The results of these dimension change are shown in the second (velocity/force units)
and third (displacement/force units) plot of figure I.1.

|H6,8|
60
(acceleration/force) dB

MEscope fit
40

20

−20
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
|H6,8|
0
Method 1
velocity/force) dB

Method 2

−50

−100
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz
|H6,8|
0
displacement/force) dB

Method 1
−50 Method 2

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

Figure I.1: Comparison of the FRF’s

The mismatch between the methods it clear. This is the reason that ‘Method 1’ is not used in
this report.

106
Appendix J
Derivation of the UMM mode shapes

The j th column (or row) of residue matrix for mode (k) can be described by the following formula
[Richardson, 2000]:

     

 r1j (k) 
 
 u1k ujk 
 
 u1k 


 r (k) 
 
 u2k ujk 
 
 u2k 


 2j 
 
 
 
 


 
 
 
 
 

 .   .   . 
. = Ak . = Ak ujk . , k = 1, ..., m (J.1)

     
 rjj (k) 
 



 (ujk )2 



 ujk





 
 
 
 
 


 . 
 
 . 
 
 . 

     
rnj (k) unk ujk unk

Notice that the residue, which reflects the physical properties of the structure, is unique in
value, while the mode shapes are only unique in shape. The mode shape can be scaled on dif-
ferent manners, therefore the scaling factor Ak will be used. In ME’ SCOPE the mode shapes are
scaled to the unit modal masses [Richardson, 2000]:

     
. . .
 .   .   . 
     
UT MU = 
 mk =
 
1
Ak ωk
=
  1 
 (J.2)
 .   .   . 
. . .

Equation (J.2) is made clear that the UMM mode shape can be obtained by setting the scaling
factor as follow [Richardson, 2000]:

1
Ak = (J.3)
ωk

107
j. derivation of the umm mode shapes

The UMM mode shape (k) can be obtained by substitute (J.3) in (J.1) [Richardson, 2000]:

     

 u1k 
 
 r1j (k)  
 r1j (k) 


 u2k 
 
 r (k) 
 
 r2j (k) 


 
 
 2j 
 
 


 
 
 
 r 
 

 .  1  .  ωk 
. 
. = . = . , k = 1, ..., m (J.4)

 
 Ak ujk 
 
 rjj (k) 
 


 u jk 
 
 r jj (k) 
 
 rjj (k) 


 
 
 
 
 

 .
 
 
 . 
 
 . 

     
unk rnj (k) rnj (k)

Here, rij is the driving point residue. The driving point residue is define in equation (J.1) as
follow [Richardson, 2000]:

rjj (k) = Ak (ujk )2

where,
1
Ak = ωk

108
Appendix K
Frequency Response Function propor-
tional damping estimation

109
k. frequency response function proportional damping estimation

|H1,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H1,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H1,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
1,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.1: Proportional damping H18

110
k. frequency response function proportional damping estimation

|H2,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H2,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H2,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
2,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.2: Proportional damping H28

111
k. frequency response function proportional damping estimation

|H3,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H3,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H3,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
3,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.3: Proportional damping H38

112
k. frequency response function proportional damping estimation

|H4,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H4,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H4,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
4,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.4: Proportional damping H48

113
k. frequency response function proportional damping estimation

|H5,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H5,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H5,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
5,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.5: Proportional damping H58

114
k. frequency response function proportional damping estimation

|H6,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H6,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam a

|H6,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
6,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam b
Figure K.6: Proportional damping H68

115
k. frequency response function proportional damping estimation

|H7,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H7,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam 1

|H7,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
7,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam 2
Figure K.7: Proportional damping H78

116
k. frequency response function proportional damping estimation

|H8,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H8,8)
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Beam 1

|H8,8|
0
(displacement/force) dB

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠(H )
8,8
200
Experiment
150 Fit (proportional damping)
100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Beam 2
Figure K.8: Proportional damping H88

117
k. frequency response function proportional damping estimation

118
Appendix L
Experimental mode shapes

119
l. experimental mode shapes

Real part mode shape 1


1
Experimental mode shapes
Numerical mode shapes
0.8

0.6
Respons

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 1


0

−0.01

−0.02
Respons

−0.03

−0.04

−0.05
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(a) Beam a

Real part mode shape 1


1
Experimental mode shapes
Numerical mode shapes
0.8

0.6
Respons

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 1


0.14

0.12

0.1

0.08
Respons

0.06

0.04

0.02

−0.02
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(b) Beam 2b
Figure L.1: Mode shape 1

120
l. experimental mode shapes

Real part mode shape 2


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 2


0.02

0.01

0
Respons

−0.01

−0.02

−0.03
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(a) Beam a

Real part mode shape 2


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 2


0.02

0.015

0.01
Respons

0.005

−0.005

−0.01
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(b) Beam b
Figure L.2: Mode shape 2

121
l. experimental mode shapes

Real part mode shape 3


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 3


0.1

0.05
Respons

−0.05

−0.1
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(a) Beam a

Real part mode shape 3


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

−3 Imaginary part mode shape 3


x 10
5

−5
Respons

−10

−15

−20
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(b) Beam b
Figure L.3: Mode shape 3

122
l. experimental mode shapes

Real part mode shape 4


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 4


0.02

−0.02
Respons

−0.04

−0.06

−0.08

−0.1

−0.12
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(a) Beam a

Real part mode shape 4


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 4


0.015

0.01

0.005
Respons

−0.005

−0.01

−0.015
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(b) Beam b
Figure L.4: Mode shape 4

123
l. experimental mode shapes

Real part mode shape 5


1
Experimental mode shapes
Numerical mode shapes
0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

Imaginary part mode shape 5


0.04

0.02

0
Respons

−0.02

−0.04

−0.06

−0.08
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(a) Beam a

Real part mode shape 5


1
Experimental mode shapes
Numerical mode shapes

0.5
Respons

−0.5

−1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

−3 Imaginary part mode shape 5


x 10
5
Respons

−5
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Distance of clamp [m]

(b) Beam b
Figure L.5: Mode shape 5

124
Appendix M
Mode shapes of FEMtools models

In this appendix the first 5 mode shapes of the FEM TOOLS models are plotted in the same figure
as the first 5 mode shapes of the numeric model.

125
m. mode shapes of femtools models

Mode shape 1
1
FEMtools mode shape
Numeric mode shape
0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(a) Beam a

Mode shape 1
1
FEMtools mode shape
Numeric mode shape
0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(b) Beam b
Figure M.1: Mode shape 1

126
m. mode shapes of femtools models

Mode shape 2
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(a) Beam a

Mode shape 2
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(b) Beam b
Figure M.2: Mode shape 2

127
m. mode shapes of femtools models

Mode shape 3
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(a) Beam a

Mode shape 3
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(b) Beam b
Figure M.3: Mode shape 3

128
m. mode shapes of femtools models

Mode shape 4
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(a) Beam a

Mode shape 4
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(b) Beam b
Figure M.4: Mode shape 4

129
m. mode shapes of femtools models

Mode shape 5
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(a) Beam a

Mode shape 5
1
FEMtools mode shape
Numeric mode shape

0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 50 100 150 200 250 300 350 400 450 500
Distance [mm]

(b) Beam b
Figure M.5: Mode shape 5

130
Appendix N
Influence of the stiffness on the num-
ber of iterations

The number of update iterations strongly depends on the initial guess of the radial stiffness. This
can be shown based on an example. Two updates strategies, with an difference initial transversal
and radial stiffness, are implemented in FEM TOOLS. Both update strategies (‘strategy X’ and
‘strategy Y’ ) only update the radial and transversal stiffness. For this example the stop criterion
²1 > CC is set on 0.41 (²2 is set off). The Correlation Coefficients (CC) that is used in this
example is define as follow:

N
1 X ∆fi
CC_ABS = Cri (N.1)
N fi
i=1

The following initial guesses are used for the radial and transversal stiffness:

KyX = 1 × 107 [N/m] KyY = 1 × 108 [N/m]


KrX = 1 × 105 [N/rad] KrY = 9 × 106 [N/rad]
Both updating strategies give the following results:

Ky = 1.25 × 108 [N/m]


Kr = 9.20 × 104 [N/rad]
Only the number of iteration is differed. ‘Strategy X’ needs 28 iterations for satisfy the stop
criterion where ‘strategy Y’ needs 94 iterations. Figure N.1 shows the correlation tracking for the
‘Strategy X’ and ‘Strategy Y’.

131
n. influence of the stiffness on the number of iterations

(a) ‘Strategy X’

(b) ‘Strategy Y’
Figure N.1: Correlation tracking

In the plot of ‘strategy Y’ between the ±2nd and ±55th iterations the correlation is nearly
constant. This is a result of the fact that influence of the radial stiffness on the eigenfrequencies
is very small. This is shown in figure N.2.

132
n. influence of the stiffness on the number of iterations

Figure N.2: Parameters tracking

Where:

Parameter 1 = Ky
Parameter 2 = Kr
accordingly, there can be stated that the number of iteration depends of the quality of the ini-
tial guess for the radial stiffness.

133
n. influence of the stiffness on the number of iterations

134
Appendix O
Bayesian Parameter Estimation method
in FEMtools

The Bayesian Parameter Estimation methode is based on the sensitivity matrix. The relationship
between the modal characteristics (in this case the eigenfrequencies) and the parameters can be
writing as [FEM, 2005]:

Re = Ra + S(Pu − Po ) ⇒ ∆R = S∆P (O.1)

Where,

Re : Vector containing the experimental eigenfrequencies


Ra : Vector containing the predicted eigenfrequencies (for Po )
S : Sensitivity matrix
Pu : Vector containing the updated parameters values
Po : Vector containing the old parameters values

Bayesian Parameter Estimation (BPE) error is defined as follow [FEM, 2005]:

E = ∆RT CR ∆R + ∆P T CP ∆P (O.2)

Where,

E : Error
CR : Weighting matrix exponential data
CP : Weighting matrix model parameters

The error (E) can be minimized using the follow formula [FEM, 2005]:

Pu = Po + G(−∆R) (O.3)

135
o. bayesian parameter estimation method in femtools

Where,

G = (CP + S T CP S)−1 S T CR : Gain matrix

The weight matrixes are used to express the confidence in the updated parameters and fre-
quencies. The follow expression are used in FEM TOOLS for the diagonal of the weight matrixes
[FEM, 2005]:

µ ¶2 µ ¶2
1 1
CRii = (O.4)
Ri cri

µ ¶2 µ ¶2
1 1
CPjj = (O.5)
Pj cpj

Where,

Ri : Response value i
Pi : Parameter value i
cri : Scatter value response i
cpj : Scatter value parameter j

In this case the standaard values are used for cri (0.01) and cpj (0.25).

136
Appendix P
MAC matrixes after the update strate-
gies based on ‘beam a’

In this appendix the MAC matrixes of the update strategies based on ‘beam a’ are shown. To give
a complete overview the MAC matrixes after updating are plotted next to the MAC matrix of the
initial FE model (before updating).

(a) Initial model (b) After ‘strategie 1a ’

Figure P.1: Graphical representation of the MAC matrixes

137
p. mac matrixes after the update strategies based on ‘beam a’

(a) Initial model (b) After ‘strategie 1b ’

Figure P.2: Graphical representation of the MAC matrixes

(a) Initial model (b) After ‘strategie 1c ’

Figure P.3: Graphical representation of the MAC matrixes

138
p. mac matrixes after the update strategies based on ‘beam a’

(a) Initial model (b) After ‘strategie 1d ’

Figure P.4: Graphical representation of the MAC matrixes

(a) Initial model (b) After ‘strategie 1e ’

Figure P.5: Graphical representation of the MAC matrixes

139
p. mac matrixes after the update strategies based on ‘beam a’

(a) Initial model (b) After ‘strategie 1f ’

Figure P.6: Graphical representation of the MAC matrixes

140
Appendix Q
FRF of the FEMtools updates with
regards to ‘beam a’

141
q. frf of the femtools updates with regards to ‘beam a’

|H18|
−60
Modal−Parameter Fit
Strategie 1a
−80
Strategie 1b
(displacement/force) dB

Strategie 1c
−100

−120

−140

−160
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H18
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H18|
−60
Modal−Parameter Fit
Strategie 1d
−80
Strategie 1e
(displacement/force) dB

Strategie 1f
−100

−120

−140

−160
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
18
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.1: FRF updated FE models

142
q. frf of the femtools updates with regards to ‘beam a’

|H28|
−60
Modal−Parameter Fit
−80 Strategie 1a
Strategie 1b
(displacement/force) dB

−100 Strategie 1c

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H28
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H28|
−60
Modal−Parameter Fit
−80 Strategie 1d
Strategie 1e
(displacement/force) dB

−100 Strategie 1f

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
28
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.2: FRF updated FE models

143
q. frf of the femtools updates with regards to ‘beam a’

|H38|
−50
Modal−Parameter Fit
Strategie 1a
Strategie 1b
(displacement/force) dB

−100 Strategie 1c

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H38
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H38|
−60
Modal−Parameter Fit
−80 Strategie 1d
Strategie 1e
(displacement/force) dB

−100
Strategie 1f
−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
38
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.3: FRF updated FE models

144
q. frf of the femtools updates with regards to ‘beam a’

|H48|
−50
Modal−Parameter Fit
Strategie 1a
Strategie 1b
(displacement/force) dB

−100 Strategie 1c

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H48
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H48|
−60
Modal−Parameter Fit
−80 Strategie 1d
Strategie 1e
(displacement/force) dB

−100
Strategie 1f
−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
48
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.4: FRF updated FE models

145
q. frf of the femtools updates with regards to ‘beam a’

|H58|
−40
Modal−Parameter Fit
−60
Strategie 1a
Strategie 1b
(displacement/force) dB

−80
Strategie 1c
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H58
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H58|
−40
Modal−Parameter Fit
−60
Strategie 2d
Strategie 2e
(displacement/force) dB

−80
Strategie 2f
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
58
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.5: FRF updated FE models

146
q. frf of the femtools updates with regards to ‘beam a’

|H68|
−40
Modal−Parameter Fit
−60
Strategie 1a
Strategie 1b
(displacement/force) dB

−80
Strategie 1c
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H68
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H68|
−50
Modal−Parameter Fit
Strategie 1d
Strategie 1e
(displacement/force) dB

−100 Strategie 1f

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
68
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.6: FRF updated FE models

147
q. frf of the femtools updates with regards to ‘beam a’

|H78|
−40
Modal−Parameter Fit
−60
Strategie 1a
Strategie 1b
(displacement/force) dB

−80
Strategie 1c
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H78
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H78|
−50
Modal−Parameter Fit
Strategie 1d
Strategie 1e
(displacement/force) dB

−100 Strategie 1f

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
78
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.7: FRF updated FE models

148
q. frf of the femtools updates with regards to ‘beam a’

|H88|
−40
Modal−Parameter Fit
−60 Strategie 1a
Strategie 1b
(displacement/force) dB

−80
Strategie 1c
−100

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H88
0

−50
[o] angle

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 1a , 1b and 1c

|H88|
−40
Modal−Parameter Fit
−60 Strategie 1d
Strategie 1e
(displacement/force) dB

−80
Strategie 1f
−100

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
88
0

−50
[o] angle

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 1d , 1e and 1f


Figure Q.8: FRF updated FE models

149
q. frf of the femtools updates with regards to ‘beam a’

150
Appendix R
MAC matrixes after the update strate-
gies based on ‘beam b’

In this appendix the MAC matrixes of the update strategies based on ‘beam b’ are shown. To give
a complete overview the MAC matrixes after updating are plotted next to the MAC matrix of the
initial FE model (before updating).

(a) Initial model (b) After ‘strategie 2a ’

Figure R.1: Graphical representation of the MAC matrixes

151
r. mac matrixes after the update strategies based on ‘beam b’

(a) Initial model (b) After ‘strategie 2b ’

Figure R.2: Graphical representation of the MAC matrixes

(a) Initial model (b) After ‘strategie 2c ’

Figure R.3: Graphical representation of the MAC matrixes

152
r. mac matrixes after the update strategies based on ‘beam b’

(a) Initial model (b) After ‘strategie 2d ’

Figure R.4: Graphical representation of the MAC matrixes

(a) Initial model (b) After ‘strategie 2e ’

Figure R.5: Graphical representation of the MAC matrixes

153
r. mac matrixes after the update strategies based on ‘beam b’

(a) Initial model (b) After ‘strategie 2f ’

Figure R.6: Graphical representation of the MAC matrixes

154
Appendix S
FRF of the FEMtools updates with
regards to ‘beam b’

155
s. frf of the femtools updates with regards to ‘beam b’

|H18|
−60
Modal−Parameter Fit
−80 Strategie 2a
Strategie 2b
(displacement/force) dB

−100
Strategie 2c
−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H18
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H18|
−60
Modal−Parameter Fit
−80 Strategie 2d
Strategie 2e
(displacement/force) dB

−100 Strategie 2f

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
18
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.1: FRF updated FE models

156
s. frf of the femtools updates with regards to ‘beam b’

|H28|
−60
Modal−Parameter Fit
−80 Strategie 2a
Strategie 2b
(displacement/force) dB

−100
Strategie 2c
−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H28
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H28|
−60
Modal−Parameter Fit
−80 Strategie 2d
Strategie 2e
(displacement/force) dB

−100 Strategie 2f

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
28
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.2: FRF updated FE models

157
s. frf of the femtools updates with regards to ‘beam b’

|H38|
−50
Modal−Parameter Fit
Strategie 2a
Strategie 2b
(displacement/force) dB

−100 Strategie 2c

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H38
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H38|
−50
Modal−Parameter Fit
Strategie 2d
Strategie 2e
(displacement/force) dB

−100 Strategie 2f

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
38
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.3: FRF updated FE models

158
s. frf of the femtools updates with regards to ‘beam b’

|H48|
−50
Modal−Parameter Fit
Strategie 2a
Strategie 2b
(displacement/force) dB

−100 Strategie 2c

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H48
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H48|
−50
Modal−Parameter Fit
Strategie 2d
Strategie 2e
(displacement/force) dB

−100 Strategie 2f

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
48
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.4: FRF updated FE models

159
s. frf of the femtools updates with regards to ‘beam b’

|H58|
−50
Modal−Parameter Fit
Strategie 2a
Strategie 2b
(displacement/force) dB

−100 Strategie 2c

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H58
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H58|
−40
Modal−Parameter Fit
−60
Strategie 2d
Strategie 2e
(acceleration/force) dB

−80
Strategie 2f
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
58
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.5: FRF updated FE models

160
s. frf of the femtools updates with regards to ‘beam b’

|H68|
−40
Modal−Parameter Fit
−60
Strategie 2a
Strategie 2b
(displacement/force) dB

−80
Strategie 2c
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H68
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H68|
−40
Modal−Parameter Fit
−60
Strategie 2d
Strategie 2e
(displacement/force) dB

−80
Strategie 2f
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
68
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.6: FRF updated FE models

161
s. frf of the femtools updates with regards to ‘beam b’

|H78|
−40
Modal−Parameter Fit
−60
Strategie 2a
Strategie 2b
(displacement/force) dB

−80
Strategie 2c
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H78
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H78|
−40
Modal−Parameter Fit
−60
Strategie 2d
Strategie 2e
(displacement/force) dB

−80
Strategie 2f
−100

−120

−140

−160

−180

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
78
200

150

100

50
[o] angle

−50

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.7: FRF updated FE models

162
s. frf of the femtools updates with regards to ‘beam b’

|H88|
−40
Modal−Parameter Fit
−60 Strategie 1a
Strategie 1b
(displacement/force) dB

−80
Strategie 1c
−100

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H88
0

−50
[o] angle

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(a) Strategie 2a , 2b and 2c

|H88|
−40
Modal−Parameter Fit
−60 Strategie 1d
Strategie 1e
(displacement/force) dB

−80
Strategie 1f
−100

−120

−140

−160

−180
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

∠H
88
0

−50
[o] angle

−100

−150

−200
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Hz

(b) Strategie 2d , 2e and 2f


Figure S.8: FRF updated FE models

163

You might also like