0DQXVFULSW

&OLFNKHUHWRGRZQORDG0DQXVFULSW0DQXVFULSW6WUHVVHV'\QDPLF5HYLHZ9GRF[

1
Modal Analysis Based Stress Estimation for Structural Elements
2
3
Subjected to Operational Dynamic Loadings
4
5
6
7
8 Pelayo Fernández*1, Anders Skafte2, Manuel L. Aenlle1 and Rune Brincker3
9
10 1
11 Department of Construction and Manufacturing Engineering,
12 University of Oviedo, Gijón, Spain.
13 *[email protected], [email protected],
14 2
Department of Engineering, Aarhus University, Denmark.
15
[email protected]
16
17 3 Center for Oil and Gas – Technical University of Denmark
18 [email protected]
19
20
21
22
23
24
Abstract:
25
26
27 The calculation of dynamic stresses plays a crucial role in the fatigue design and fatigue life
28
29 prediction of structures. However, the uncertainty in the structural dynamic properties, i.e.,
30
31
32 stiffness, mass and damping, as well as the necessary use of simplified loading models, are the
33
34 main sources of uncertainty in fatigue. In this work, a procedure for estimating dynamic stresses
35
36
37
is proposed based on the real responses of the structure under dynamic loadings and the modal
38
39 parameters of the structure, i.e., natural frequencies, damping ratios and mode shapes. The
40
41 proposed methodology was validated by experimental tests in a simply-supported glass beam
42
43
44 and a rectangular glass plate pinned-supported at 3 corners, both subject to dynamic random
45
46 loads. The estimated stresses are compared with those measured with strain gages attached to
47
48
49 some points of the structures. The estimated stresses are in good agreement with those provided
50
51 by the strain gages which demonstrates that the methodology can be used to estimate more
52
53
54 accurate stress time-histories.
55
56
57
58
59
60 Keywords: experimental stresses; fatigue; Operational conditions; Strain Mode Shapes.
61
62 1
63
64
65
 1
2
3
4
1 Introduction
5
6
7 Fatigue is one of the ultimate limit states that has to be verified in structures and mechanical
8
9
10 components. In fatigue design, remaining fatigue life and fatigue retrofitting, the main sources of
11
12 uncertainty are the fatigue material characterization and the real hot spot stress time histories.
13
14
15 Fatigue analysis can be required at the design stage, during construction or in actual use. When
16
17 the structure is in service, fatigue analysis can be applied to calculate performance, to compare
18
19
20 with earlier calculations or to make decisions about the serviceability of the structure (fatigue
21
22 retrofitting, out of service). Planning, testing and evaluation of fatigue data is a recursive subject
23
24
appearing in the existing literature on which laboratories and research groups dealing with
25
26
27 fatigue design are interested on.
28
29
30 As regarding the fatigue material characterization, no specific model seems to be generally
31
32 accepted by the community, although the S-N field and the Miner Rule finds good acceptance
33
34
35 both by the academic milieu as well as by research groups concerned with industrial applications
36
37 [1, 2].
38
39
40 With respect to the stress time-histories, the uncertainty is considerably high because the results
41
42
43 are mainly influenced by inaccuracies of the analytical or numerical model of the structure and
44
45 by the use of simplified load models which does not represent, with the needed accuracy, the
46
47 load characteristics (variable amplitude, random nature, frequency bandwidth, sequence effect,
48
49
50 etc.) A numerical model of the structure, e.g. a Finite Element Model (FE;), is generally used to
51
52 estimate the stresses but the stiffness, mass and damping properties of the structure can only be
53
54
55 estimated in an approximate way. With respect to the loading, the real loads acting on structures
56
57 such as wind, traffic, sea waves, etc., usually excites several modes of the structure so a broad
58
59
60 band frequency fatigue process occurs and the simplified loading models commonly used are not
61
62 2
63
64
65
 able to reproduce with accuracy the real loading [3, 4, 5]. Moreover, sometimes the load
1
2 spectrum is changed after a number of years by a modified use of the structure, which differs
3
4
5 from the initial expectations [6] and the load spectrum must be evaluated again.
6
7
8 Since the main difficulties in fatigue design are due to inaccuracies in the dynamic properties of
9
10 the structure and its broad band response, it seems reasonable to combine modal analysis with
11
12
13 analytical models in order to reduce the uncertainties in the time stress-histories of the fatigue
14
15 process [7, 8, 9, 10].
16
17
18 Modal analysis techniques are commonly used to determine the modal parameters of a system
19
20 and have been applied for many years in mechanical and civil engineering applications [11, 12,
21
22
23 13]. Operational modal analysis (OMA) is now an alternative to Classical modal analysis (CMA)
24
25 since in OMA the testing is normally performed by just measuring the responses under the
26
27
28 natural or operational conditions, i.e. the structure is excited by loads such as wind loads, wave
29
30 loads, traffic loads, etc., which can be considered an important advantage in medium and large
31
32
33 structures where the use of artificial excitation devices may be expensive or impractical [14, 15,
34
35 16].
36
37
38 Due to the fact that testing of the structure is carried out under its operational loads conditions,
39
40
the normal service of the structure does not need to be interrupted during the data measurements
41
42
43 so it is an adequate technique for health monitoring and, it allows as a better understanding of the
44
45 random loading fatigue process and, subsequently, a reliable stresses estimation.
46
47
48 In this work, it is presented a methodology to estimate stresses in any arbitrary point of the
49
50
51 structure which combines a numerical model, correlated and updated using operational modal
52
53 analysis, and the experimental accelerations measured at several discrete points of the structure.
54
55
56
57
58
59
60
61
62 3
63
64
65
 The fundamentals of the modal based stress estimation method lie, on one hand, in the correct
1
2 modal identification and expansion of the mode shapes of the structure and, on the other hand, in
3
4
5 the measurement of the responses of the structure under the real loading conditions.
6
7
8 The technique can be applied to a finite time segment or it can be implemented to estimate
9
10 strains in real time. The proposed methodology was validated by experimental test in a simply-
11
12
13 supported glass beam and a rectangular glass plate pinned-supported at 3 corners. The estimated
14
15 time-history stresses were compared with those measured with strain gages attached to some
16
17
points of the structures. The estimated stresses are in good agreement with the experimental
18
19
20 results which demonstrates that the methodology can be used to improve the fatigue design
21
22 reliability.
23
24
25
26
27 2 State of the Art
28
29
30 The first studies in this field appear in the 80’s when modal analysis strain techniques, i.e. using
31
32 strain gage sensors, began to be used to improve or complement the knowledge about the
33
34
35 dynamic behaviour of a structure [17, 18]. The benefits of modal strain based methods are
36
37 significant since the results can be directly related with stresses and, consequently, with the
38
39
40 integrity evaluation of a structure subject to random dynamic loading processes such as fatigue
41
42 [19].
43
44
45 The improvement of numerical analysis applications and codes, such as the finite element
46
47 method, allowed to combine strain modal analysis with numerical models. In this way, the
48
49
50 studies conducted by Massé and Pastorel [20] report the analysis of wind turbines blades where
51
52 aerodynamic loads were, firstly, frequency decomposed and, then, applied to a numerical model
53
54
55 to obtain the stress distributions. Okubo and Yamaguchi [21] proposed a different methodology
56
57 for the estimation of the dynamic operational strains based on the use of a transformation matrix
58
59
60 that relates strain and displacement mode shapes; identified from a previous modal analysis of
61
62 4
63
64
65
 the structure. A similar process was proposed by Gun-Myung [22] but expanding the strain
1
2 estimation to un-measured points.
3
4
5 Although nowadays numerical models are widely used in structural calculation, analytical
6
7
8 models can be used in some simple cases such as beams where, e.g. bending Euler-Bernoulli
9
10 equations can be combined with acceleration responses in order to estimate the modal stresses or
11
12
13 strains. In this area, Ciora [23] presented a method for the determination of pipeline dynamical
14
15 stress using the signals of two accelerometers and the Euler-Bernoulli bending theory where a
16
17
periodical approximation for the continuous transversal motion under vibration conditions in the
18
19
20 pipeline was used. A similar process was proposed by Chan-Jung et al. [24] where dynamic
21
22 strain responses are predicted using a method based on the classical beam theory and the
23
24
25 measured accelerations.
26
27
28 At the end of the 90’s, OMA adds a new dimension to the modal based stress methods since only
29
30 the responses of a structure are needed to identify the modal parameters [25]. This can be
31
32
33 considered an advantage in the stress estimation problem of random fatigue processes where, in
34
35 general, the forces are unknown (wind, sea waves, traffic, temperature changes, etc.).
36
37
38 The studies of Graugaard et al. [26] and Hjelm et al. [27] introduced operational modal analysis
39
40
in the modal stress based methods. The authors proposed a method that combines a finite
41
42
43 element model and the modal parameters of the structure identified with OMA, to predict the
44
45 stresses on a structure. This methodology was applied and validated successfully by the authors
46
47
48 in lattice structures. In this direction, Fernández et al. [28] also presented a similar method for
49
50 beam structures where the modal strain mode shapes are obtained using the Euler-Bernoulli
51
52
53 theory and the mode shapes identified with OMA. More recently, Papadioti et al. [29] proposed a
54
55 modal based method using the operational responses of the structure for estimating damage
56
57
58
accumulation due to fatigue. The authors predicted the stress responses with a high fidelity finite
59
60 element model using vibration measurements from a limited number of sensors and different
61
62 5
63
64
65
 prediction methods, i.e. Kalman filter or modal expansion. The proposed stress estimations
1
2 methods were validated using simulated data from a chain-like spring-mass model and a small-
3
4
5 scale model of a vehicle structure.
6
7
8 With respect to the measurement techniques, nowadays a great number of different sensors such
9
10 as accelerometers, LASER sensors, LVDT, etc. can be used to measure the response of the
11
12
13 structure under different loading conditions. If the modal testing is carried out with
14
15 accelerometers, the sensors used for modal identification can also be utilized to measure the
16
17
response. The mode shape accuracy can be improved using surface response methods such as
18
19
20 digital image correlation [30] or laser scanner [31].
21
22
23
24
3 Theory
25
26
27 In this work, the methodology to estimate stresses is presented for beam and Kirchoff plate finite
28
29 elements.
30
31
32
33 3.1 Bending Elements Formulation
34
35 A variety of structures working in bending conditions can be modeled with beam elements, see
36
37
38
Fig. 1. If the Euler-Bernoulli theory is assumed, the bending moment and the curvature can be
39
40 related by the equation:
41

"# $
42

• ! # = &!
43

" %
44 (1)
45

where • is the Young’s modulus, is the second moment of the cross section about y axis, $ =
46
47 !
48
$(') is the vertical displacement (see Fig. 1) and &! is the bending moment. As we are
49
50

interested in stresses, Navier’s Law can be applied and in a section located at distance % the
51
52
53
54
55 stresses can be determined by:
56
57
58
59
60
61
62 6
63
64
65
 !"
1 •( ) = % (2)
#$
2
3
4 where % is the distance from the neutral axis to the point of interest in the section. Eqs. (1) and
5
6
(2) can be combined and the following relation between stress and curvature is obtained:
7
8
9
10 *+ ,
•( ) = −'% (3)
11 *+
12
13
14 Due to the practical difficulties to measure the real curvature, an approximation can be
15
16 introduced using the finite element method, so that the displacement in any arbitrary point of the
17
18
19
beam element, see Fig. 1, can be obtained as [32]:
20
21
22 ,( ) = {- / ( )} {,/ } (4)
23
24
25
where {- / ( )} and {,/ } are vectors containing the element shape functions, e.g. Hermité
26
27
28 polynomials, and the nodal displacements corresponding to the element 0, respectively.
29
30
31 Taking into account that the second derivative of Eq. (4) is the curvature, and introducing the
32
33
34 time dependent displacement vector, {,/ (1)} in Eq. (3), the following expression is obtained:
35
36
37
38
•( , 1) = −' {- / "( )}{,/ (1)} % (5)
39
40 where superscript “ 44 ” indicates second derivative with respect to . An useful consequent step is
41
42
43 to introduce modal-superposition [33], so that the vector {,/ (1)}, can be expressed in terms of
44
45 the mode shapes of the structure, [5 / ], by means of the modal coordinates, {6(1)}, as follows:
46
47
48
49 {,/ (1)} = [5 / ]{6(1)} (6)
50
51
52 The introduction of the the modal coordinates in the stress estimation process is of great interest
53
54
55 because they can be used to expand the displacements to non-measured points. Finally, if Eq. (6)
56
57 is substituted in Eq. (5), the expression to determinate the stress time-histories at each point of
58
59
60 the beam element is given by:
61
62 7
63
64
65
 •( , !) = −# {% & "( )}[* & ]{+(!)} - (7)
1
2
3 If several elements are used to model the structure, an assembly process should be used to obtain
4
5
6
the whole structure stress distribution.
7
8
9 3.2 Kirchoff Plates Formulation
10
11 A similar methodology as that proposed for beams can be used for plates. The strains and the

small transverse (out-of-plane) displacement .( , /) in a Kirchoff thin plate (see Figure 2) are
12
13
14
15
16 related by the equation [34]:
17

: ;.
18

⎧ ⎫
19

⎪ :
⎪ : ;. ⎪
;
12 ⎪
20

1
21

0 3 5 = −-
⎨ :/ ⎬
22

423
; (8)
23

⎪
⎪2 : . ⎪
;
⎪
24

⎩ : :/⎭
25
26
27
28
29 where z is the distance to the neutral surface.
30
31
32 If a finite element model is introduced, the displacement in any arbitrary point of a plate element
33
34
35 is given by [34]:
36

.( , /) = [% & ( , /)] {.& }

37
38 (9)
39

Where [N& ( , /)] is a matrix containing the shape functions and {.& } is a vector containing the
40
41
42

displacements and rotations of the nodes A.B , C2 , C3 D. If expression (10) is now substituted in Eq.
43
44
45
46
47 (9), the strain vector in the element is given by:

12 ( , /)
48
49

F 13 ( , /) G = −- [H & ( , /)] {.& }

50

423 ( , /)
51 (10)
52
53

Where [H & ( , /)] is the strain matrix of the element which contains the second derivatives of the
54
55
56
57
58 shape functions, i.e.:
59
60
61
62 8
63
64
65
 *+
⎧ ⎫
⎪ *" + ⎪
⎪ *+ ⎪
1

[ (", #)] =
! [0 ! (", #)]
2

⎨ *# +
⎬
3
(11)
⎪
⎪2 * + ⎪
⎪
4
5

⎩ *"*#⎭
6
7
8
9
10
Now, if modal-superposition is considered, the strains of the plate element can be expressed as:

45 (", #, 6)
11
12

3 7 (", #, 6) 9 = −; [
4 !
(", #)] [< ! ] {>(6)}
13

857 (", #, 6)
14 (12)
15
16

where the product [ ] [<! ] represents the curvatures of the mode shapes. Finally, the stresses at
17
18
19
20
21 any point of the plate element can be obtained by:

?5 (", #, 6)
22
23

3 ?7 (", #, 6) 9 = −; [A] [ !
(", #)] [< ! ] {>(6)}
24

@57 (", #, 6)
25 (13)
26
27

where A is the constitutive matrix that depends on the material properties, which for an isotropic
28
29
30
31
32 material is given by:

1 D 0
33

B
34

D 1 0
[A] = E DG
35

1−D + 1 −
36

0 0
(14)
2
37
38

with B and D being the Young’s modulus and the Poisson ratio, respectively.
39
40
41
42
43
44
45
46 4 Methodology
47
48 In this section we detailed the different steps needed to apply successfully the proposed stresses
49
50
based method.
51
52
53
54 4.1 Experimental Mode Shapes of the Structure
55
56 As it was detailed in the introduction, the modal parameters of the structure that define its
57
58
59 dynamic behaviour must be identified, specially the experimental mode shapes of the structure,
60
61
62 9
63
64
65
 [ϕ]•• , since the responses of the structure must be decomposed using the modal-superposition
1
2 technique (see Eq. (6)). Either classical modal analysis or operational modal analysis can be used
3
4
5 to identify the modes. Taking into account that random loads, such as natural forces, e.g. wind or
6
7 sea waves, are not easily measured, operational modal analysis (OMA) results in an
8
9
10 advantageous technique to obtain the modal parameters of the structure. Frequency Domain
11
12 Decomposition (FDD) [35] or Stochastic Subspace Identification (SSI) [36] techniques can be
13
14
15
used to identify the modes. Although modal identification implies to obtain not only the mode
16
17 shapes but also natural frequencies and damping ratios, it is remarkable that in this stress
18
19 estimation process, these modal parameters (natural frequencies and damping ratios) are not
20
21
22 necessary since they will be implicitly in the response of the structure by means of the modal
23
24 coordinates. In any case, the identification of frequencies and damping ratios provides always
25
26
27 relevant information about the dynamic behaviour that could be used for additional stages, e.g.
28
29 finite element model up-dating or health monitoring of the structure.
30
31
32
33
4.2 Modal Coordinates
34
35 Once the experimental mode shapes of the structure are obtained, the modal coordinates can be
36
37 calculated using modal-superposition, see Eq. (6). Since the responses of the structure are
38
39
40 usually measured with accelerometers, the acceleration modal coordinates are obtained from
41

{ ̈ (")}#$ = [%]&'
42
#$ {+̈ }#$
43 (15)
44

indicates experimental data and [ϕ]••

45
•• represents the inverse matrix of the
46 Where subscript “ex”
47

experimental mode shapes. The pseudoinverse must be used if the matrix [ ]!" is not square.
48
49
50
51 Then, the displacement modal coordinates, {q(t)}•• , are determined using time or frequency
52
53
54 domain integration techniques [37].
55
56
57 Although the stress estimation method was developed in the theory section using the mass
58
59 normalized mode shapes of the structure, [ ], arbitrary normalized or un-scaled mode shapes,
60
61
62 10
63
64
65
 [•], can also be used without loss of accuracy. The mass normalized mode shapes [ϕ] and the
1
2 un-scaled mode shapes [ψ] are related by [38]:
3
4
5
6 ["] = [•] [$] (16)
7

where [•] is a diagonal matrix containing the scaling factors [39, 40].
8
9
10
11
12
13
Now, if Eq. (16) is substituted in Eq. (6), it results in:
14
15
" (#)}
16 { = [$ " ][•]{%(#)} (17)
17
18
19 or, alternatively:
20
21
22
" (#)}
23 { = [$ " ]{% ∗ (#)} (18)
24
25
26
27 where {q∗ (t)} = [α]{q(t)} are called here un-scaled modal coordinates, which can be estimated
28
29 by means of the expression:
30
31
32
33 {% ∗ (#)} = [$" ]*+ { , (#)} (19)
34
35
36 4.3 Mode Shapes Expansion
37
38
39 The experimental mode shapes are only known at the measured DOF's, usually translational
40
41 DOF's. If we want to estimate stresses in any arbitrary point of a finite element using Eqs. (7)
42
43
44 and (13), the mode shape components at all the DOF's of the element (translations and rotations)
45
46 must be known. This implies that the experimental mode shapes have to be expanded to the un-
47
48
49 measured DOF’s. As an example, in the bending case, we can measure all or part of the DOF’s
50
51 corresponding to the displacements of the structure (see Fig. 1), however it is not possible (or it
52
53
is very difficult) to measure the rotations. Thus, at least an expansion to the rotational DOF’s
54
55
56 will be necessary.
57
58
59
60
61
62 11
63
64
65
 The expansion process can be carried out by different ways depending on the type of the
1
2 structure (beam, plate, three dimensional, etc.), the number of DOF's measured in the
3
4
5 experiments, the type of structure as well as the complexity of the model (analytical or
6
7 numerical) considered to model the structure (beam or plate elements in opposite to hexaedric
8
9
10 elements, etc.) [12]. The following expansion techniques have been proposed in the literature:
11
12
13 a) Fitting of experimental mode shapes. For simple cases such as beam-like structures (or
14
15 even plates), mathematical functions that represents adequately the mode shapes such as spline
16
17
18 curves or Fourier series [41] can be used to fit the experimental mode shapes. According to Eqs.
19
20 (7) and (13), the displacements and rotations at the nodes of each element are needed and they
21
22
23 can be obtained directly from the fitted mathematical function in those necessary points. The
24
25 continuity of the function and of its first derivative must be assured. An alternative consists of
26
27
28 determining the strain mode shapes directly from the fitted function (second derivate) at the
29
30 points where the stresses are going to be estimated.
31
32
33 In general, these methods require a regular fine measurement grid in the testing process, mainly
34
35
36 if high order modes are going to be fitted. With this technique, the use of analytical models or
37
38 the assembly of a finite element model of the structure is avoided [41].
39
40
41
42
b) Analytical mode shapes. Another way of avoiding the assembly of a finite element model is
43
44 to use the mode shape equations corresponding to analytical models, which are reported in the
45
46 literature for the most common one-dimensional (beams) and two-dimensional (shell-plates)
47
48
49 under different support conditions [42], from which the strain mode shapes are derived.
50
51
52 c) Numerical mode shapes. A more general method consists of using a finite element model
53
54
55 of the structure. With this technique, the numerical mode shapes are used to expand the
56
57 experimental mode shapes to the un-measured DOF’s. It is assumed that the experimental mode
58
59
shapes, can be expressed as a linear combination of the numerical mode shapes [43]:
60
61
62 12
63
64
65
 [ $
!"# ] =[ $
%& ] ∙ [(] (20)
1
2
3
4 where subscripts ‘ex’ and ‘FE’ indicates experimental and numerical mode shapes, respectively,

superscript ‘m’ indicates measured DOF’s and [•] is a transformation matrix. Due to the fact that,
5
6
7
8
9 in general, a low number of modes are identified in the experimental model compared with those

that can be extracted from the numerical model, the matrix [•] has to be estimated from the
10
11
12
13
14 limited information given by the truncated set of experimental modes by:
15

[•] = [
16
17 # $ #
18 !" ] [ &' ] (21)
19
20
+ $ +
21 where ( )* , is the pseudo-inverse of [ )*
]. The pseudo-inverse can be computed by singular
22
23
24 value decomposition or by the corresponding least square solution:
25
26
27 # $ # - # ./ #
28 [ !" ] = ([ !" ] [ !" ], [ !" ] (22)
29
30
31 Then, the experimental mode shapes can be expanded to the un-measured degrees of freedom by
32
33
34 means of the expression:
35
36
0# 0#
37 [ &' ] =[ !" ] ∙ [•] (23)
38
39
40 ‘um’
where superscript indicates un-measured DOF’s. The same expression can be used with un-
41
42
43 scaled mode shapes resulting in a different [T] matrix.
44
45
46 4.4 Stress Estimation
47
48
49 Finally, the stresses can be estimated using equations (7) and (13) when mass scaled mode
50
51 shapes are used in beam and Kirchoff plates elements, respectively. If un-scaled mode shapes are
52
53
54 going to be used, the stresses can be determined after substituting Eq. (19) in Eqs. (7) and (13),
55
56 respectively, i.e.:
57
58
59
60
61
62 13
63
64
65
 •( , !) = −# {% & "( )} [*& ]{+ ∗ (!)} ℎ (24)
1
2
3
4 for bending elements and
5

•0 ( , 1)
6

/ •2 ( , 1) 4 = −5 [6] [7 & ( , 1)] [*& ] {+ ∗ (!)}

7
8

302 ( , 1)
9 (25)
10
11
12
13
14 for kirchoff plate elements.
15
16
17 5 Experimental program
18
19
20 5.1 Simply supported glass beam
21
22
23 A monolithic glass beam with a 100 x 6 mm rectangular section and 1 m long was tested in the
24

lab in simply supported configuration. A Young’s modulus # = 72 GPa and a Poisson ratio : =
25
26

0.22, respectively, were considered as mechanical properties for the glass.

27
28
29
30
31
32 The experimental modal parameters were identified with operational modal analysis. The beam
33
34 was excited applying many small hits along the beam with an impact hammer, random in time
35
36
37 and space [44]. The responses were measured using seven uniformly distributed accelerometers
38
39 (models: B&K 4508B and PCB 333B32) with a sensitivity of 100 mV/g and two strain gage
40
41 sensors HBM LY11-6/350 during a period of approximately 4 minutes. The test setup is shown
42
43
44 in Figure 3 where the arrows indicate the measured direction with the accelerometers. The
45
46 responses were recorded with a sampling frequency of 1632 Hz and using a National Instruments
47
48
49 Compact DAQ acquisition system equipped with NI9234 and NI9236 modules.
50
51
52 The modal parameters were estimated with the ARTeMIS MODAL software using the
53
54
55
frequency-domain decomposition (EFDD) [44] and the stochastic subspace iteration (SSI) [46]
56
57 methods. The two techniques provide similar results, and therefore only the modal parameters
58
59 estimated with the FDD technique are presented in the paper. The first five modes for the glass
60
61
62 14
63
64
65
 beam were considered in the analysis. The singular value decomposition (SVD) of the responses
1
2 is presented in Figure 4. On the other hand, a finite element model was also assembled in
3
4
5 ABAQUS and the beam was discretized using 8 Euler-Bernoulli beam elements (see Figure 1).
6
7
8 Table 1 shows the experimental natural frequencies, identified with the EFDD technique,
9
10
11
together with those obtained from the numerical model. It can be inferred that a very good
12
13 agreement exists between the experimental and the numerical natural frequencies, the error being
14
15 less than 3%. With respect to the mode shapes, the MAC (modal assurance criterion) between
16
17
18 the experimental and numerical mode shapes is also presented in Table 1. A very good
19
20 correlation exists between the numerical and experimental mode shapes, the MAC being very
21
22
23 close to one for all the modes considered in the investigation.
24
25
26 Then, the experimental mode shapes were expanded to the un-measured DOF’s using the
27
28 numerical mode shapes extracted from the FEM and applying the methodology presented in
29
30
31 section 4.3.c)
32
33
34 The stresses at the points where the strain gages were attached were estimated using the
35
36
37 methodology described in sections 3 and 4. In order to excite the structure, several hits were
38
39 applied to the beam in random positions, using an impact hammer. In Figure 5 is shown the
40
41
42
power spectral density (PSD) of the experimental stresses measured with strain gage 1 (g1, see
43
44 Figure 3). The peaks corresponding to the first five modes identified by OMA can easily be seen
45
46 in Figure 5 together with some peaks at 50, 150, 250, 350 and 450 Hz corresponding to electrical
47
48
49 noise. In Figure 5 the PSD of the un-scaled displacement modal coordinates is also presented
50
51 which were calculated from the experimental accelerations and the experimental mode shapes
52
53
54 (see section 4.2). From Figure 5, it can also be inferred that the first two modes contribute the
55
56 most to the stress magnitude (••∗ (•) and • ∗ (•)). This fact is in good agreement with the stresses
57
58
59
60
61
62 15
63
64
65
 recorded with the strain gages where the higher values correspond to modes 1 and 2 (see Figure
1
2 5).
3
4
5
6 In Figures 6 and 7 are presented the experimental stresses (measured with strain gages) and those
7
8 estimated with Eq. (24) at points 1 and 2, (strain gages 1 and 2, respectively, see Figure 3), at
9
10
11
different time periods. In both cases, it can be seen that a good correlation is obtained, being the
12
13 average error less than 8%. However, there is also a contribution of the zero-load noise level in
14
15 the strain gages to the errors between the measured and the estimated stresses. On the other hand,
16
17
18 the accelerometers are also more sensitive to impacts than the strain gages (see Figures 6 and 7)
19
20 and therefore there is a significant discrepancy between the estimated and the measured stresses
21
22
23 at the beginning of the impact loading. This effect also contributes to increase the error.
24
25
26 The dynamic stress distribution for the entire beam can also be calculated with the proposed
27
28 methodology. In Figure 7, the estimated stresses at a specific time interval for the complete glass
29
30
31 beam are also presented. As expected from the experimental stresses and the displacement modal
32
33 coordinates (Figure 5) where modes 1 and 2 are predominant, it can be observed in Figure 7
34
35
36 (right) how these mode shapes appear in the stress 3D profile: 2 peaks corresponding to mode
37
38 shape 2 at the beginning of each impact acting simultaneously with mode shape 1 (one peak) that
39
40
41 can be clearly seen after vanishing mode 2 (due to its higher damping).
42
43
44 5.2 A glass plate
45
46
47 The second experiment was performed on a rectangular monolithic glass plate measuring 1400 x
48
1000 x 10 mm, with a Young´s modulus • = 72 GPa and a Possion ratio • = 0.22. The glass
49
50
51
52 plate was pinned supported in 3 corners, allowing only translational movement perpendicular to
53
54
55 the plate. 13 accelerometers and 6 strain gauge where mounted to the plate. Figure 8 illustrates
56
57 the test setup.
58
59
60
61
62 16
63
64
65
 For the experimental modal analysis, load was induced to the plate by gently tapping it with the
1
2 finger. Two persons were exciting the plate at the same time to ensure that the load was applied
3
4
5 all over during measurements. The accelerations were measured using B&K4508B
6
7 accelerometers and strains were measured using HBM LY11-120 strain gauges. Both
8
9
10 accelerations and strains were acquired using the same National Instruments Compact DAQ
11
12 acquisition systems used for the glass beam.
13
14
15 A FE model was constructed in MSC Nastran consisting of 100 DOF´s and 81 QUAD4
16
17
18 elements. Frequencies, translational and rotational mode shapes components where extracted,
19
20 and the mode shapes in terms of stains were calculated using the principles described in section
21
22
23 3.2 in the DOF´s where the strain gauges were placed.
24
25
26 Experimental mode shape expansion was carried out using the same procedure as described with
27
28 the simply supported beam. A plot of the singular values can be seen in Figure 9. The first 8
29
30
31 natural frequencies and mode shapes were estimated and used as a basis for the analysis. The
32
33 MAC values between experimental and FE mode shapes are presented in Table 2, as well as the
34
35
36 differences in frequency. The results show that the mode shapes are well correlated. The
37
38 frequencies, however, show a larger deviation compared with the results from the glass beam,
39
40
41 which is mainly explained by inaccuracies in the modelling of the supports in the FE model.
42
43 However, the information needed from the FE model is limited to the mode shapes
44
45
46 (used in the expansion of the experimental mode shapes) because the information corresponding
47
48
49 to the natural frequencies and the damping is contained in the experimental modal coordinates.
50
51 Thus, a good accuracy can be obtained in the stress estimation if a good correlation exists in the
52
53
54 mode shapes even though there is not a good correlation in the natural frequencies. For the plate,
55
56 the experimental mode shapes were expanded using the technique described in section 4.3.c.
57
58
59
60
61
62 17
63
64
65
 For the strain estimation a series of small impulses was applied to the plate using an impact
1
2 hammer. Both the measured accelerations and strains were highpass filtered at 5 Hz to get rid of
3
4
5 any noise at DC, and lowpass filtered at 110 Hz so only 8 modes were included in the response.
6
7 The following analysis is based on the bandpass filtered response.
8
9
10
11
Translations were obtained by integrating the accelerations twice with respect to time and the
12
13 strains where estimated using the methodology described in section 3.2. Figure 10 shows the
14
15 measured (bandpass filtered) and estimated signal for strain gauge 5 (according to Figure 8). The
16
17
18 errors for all 6 strain gauges are shown in Table 3.
19
20
21 The estimation shows good results with errors between 1.4 % and 6.1 %. The results are further
22
23
24 illustrated by comparing the auto spectral densities (see Figure 11), and rainflow flow count (Fig.
25
26 12) for strain gauge 5. At figure showing the rainflow count the error is measured as the relative
27
28 error between the individual bins. Naturally we see an increase in error for the bins with large
29
30
31 strain values, because the fewer counts increases the uncertainty in the calculation of the error.
32
33
34
35
8 Conclusions
36
37
38 The material characterization and the stress-time history are the major sources of uncertainty in
39
40 fatigue. This uncertainty comes from inaccuracies in the numerical models and from the use of
41
42
43 simplified loading models.
44
45
46 The purpose of this paper is to propose an efficient method for calculating stress-time histories in
47
48
49
structures subjected to random loadings. The technique can be applied to short time periods of in
50
51 real time to structures in operation.
52
53
54 The proposed methodology combines the mode shapes identified by modal analysis (classical or
55
56
57 operational), expanded using a finite element model or an alternative method, and the
58
59 experimental accelerations measured at several discrete points of the structure.
60
61
62 18
63
64
65
 The technique has been validated by experimental tests carried out on a simply supported glass
1
2 beam and on a rectangular glass plate pinned at 3 corners, comparing the estimated stresses with
3
4
5 those recorded with strain gages.
6
7
8 It has been demonstrated that the stresses can be estimated with an error less than 10 % if an
9
10
11
accurate modal identification is performed and a good finite element model is assembled.
12
13
14 Acknowledgments
15
16
17 The authors would like to acknowledge the financial support given by the Spanish Ministry of
18
19
20 Science and Innovation through the project BIA 2011-28380-C02-01 and the CajAstur
21
22 Fellowship-University of Oviedo 2011 program.
23
24
25
26
27
28
29
30
31
32
33 References
34
35 [1] Lee YL, Hathaway, Pan J, Hathaway R, Barkey M (2005) Fatigue Testing and Analysis:
36
37
38 Theory and Practice. Elsevier Butterworth-Heinemann
39
40 [2] Dowling NE, Thangjithan S (2001) An overview and discussion of basic methodology for
41
42
fatigue. Symposium on Fatigue and Fracture Mechanics, 31:3-36
43
44
45 [3] Bishop NWM (1999) Vibration fatigue analysis in the finite element environment. Anales de
46
47 Mecánica de la Fractura, 18:8-23
48
49
50 [4] Gao Z, Torgeir M (2008) Frequency-domain fatigue analysis of wide-band stationary
51
52 Gaussian processes using a trimodal spectral formulation. Int J Fatigue 30:1944-1955
53
54
55 [5] Benasciutti D, Tovo R (2007) On fatigue damage assessment in bimodal random processes.
56
57 Int J Fatigue 29:232-244
58
59
60
[6] Schijve J (2001) Fatigue of Structures and Materials. Kluwer Academic Publishers
61
62 19
63
64
65
 1
2 [7] Townley GE, Klahs JW (1986) Using test and system dynamic analysis for component life
3
4
5 prediction. In Proceedings of the 4th International Modal Analysis Conference (IMAC). Los
6
7 Angeles CA. Vol. 1 pp 656-662
8
9
10 [8] Verdonck E, Snoeys R (1984) Life time prediction based on the combined use of finite
11
12 element and modal analysis data. In Proceedings of the 2nd International Modal Analysis
13
14
15
Conference (IMAC). Orlando FL. pp 572-579
16
17 [9] Liefooghe D, Sas P. (1991) Optimizing the fatigue lifetime of structural components by using
18
19 dynamic analysis methods. In Proceedings of the 9th International Modal Analysis Conference
20
21
22 (IMAC) pp 344-351
23
24 [10] Liefooghe D, Leuridan J, Van der Auweraer H (1992) Integration of structural dynamics
25
26
27 into fatigue prediction. Noise Vib Control Word 6-8
28
29 [11] Ewins D. (1984) Modal Testing: Theory and Practice. Research Studies Press LTD. John
30
31
32
Wiley & Sons INC
33
34 [12] Maia N, Silva J, He J, Lieven N, Lin R-M, Skingle, G, To W, Urgueira A (1997)
35
36 Theoretical and Experimental Modal Analysis. Research Studies Press Ltd
37
38
39 [13] Heylen W, Lammens S, Sas P (1998) Modal Analysis: Theory and Testing. Katholieke
40
41 Universiteit Leuven
42
43
44 [14] Cantieni R. (2005) Experimental Methods Used in System Identification of Civil
45
46 Engineering Structures. In Proceedings of the International Operational Modal Analysis
47
48
49 Conference (IOMAC). Copenhagen pp 249-260
50
51 [15] Cunha A, Caetano E (2005) From input-utput to output-only modal identification of civil
52
53
54
engineering. In Proceedings of the 1st International Operational Modal Analysis Conference
55
56 (IOMAC), Copenhagen, Denmark, pp 11-27
57
58
59
60
61
62 20
63
64
65
 [16] Møller N, Brincker R. Herlufsen H, Andersen P (2001) Modal testing of mechanical
1
2 structures subject to operational excitation forces. In Proceedings of the 19th International
3
4
5 Modal Analysis Conference (IMAC). Orlando FL. pp 262-269
6
7 [17] Bernasconi 0, Ewins DJ (1989) Modal strain/stress fields. J Modal Anal 68-76
8
9
10 [18] Staker C. (1985) Modal analysis efficiency improved via strain frequency response
11
12 functions. In Proceedings of the 3rd International Modal Analysis Conference (IMAC),
13
14
15
Orlando FL. Vol. 1 pp 612-617
16
17 [19] Vári LM, Heyns PS (1997) Strain modal testing – a critical appraisal. Res Dev J 13(3):83-
18
19 90
20
21
22 [20] Masse B, Pastorel H (1989) Stress calculation on the Sandia 34-meter Wind turbine with the
23
24 local circulation method and turbulent wind. Institut de Reserche d’Hydro-Quebec, Varennes,
25
26
27 QC, Canada
28
29 [21] Okubo N, Yamaguchi K (1995) Prediction of dynamic strain distribution under operating
30
31
32
conditions by use of modal analysis. In Proceedings of the 13th International Modal Analysis
33
34 Conference (IMAC), vol 1 pp 91-96
35
36 [22] Lee G-M (2007) Prediction of strain responses from the measurements of displacements
37
38
39 responses. Mechanical systems and Signal Processing 21:1143-1152
40
41 [23] Ciora Gh T (1998) Experimental method for pipeline dynamic stress estimation by vibration
42
43
44 measurement. In Proceedings of the 16th International Modal Analysis Conference (IMAC).
45
46 Santa Barbara CA. pp 457-463
47
48
49 [24] Chan-Jung K, Bong-Hyun L, Hyun-Cheol J, Hyeon-Ho J, Yeon June K (2012) Estimation
50
51 of strain at elastic systems using acceleration response. Transactions of the Korean Society for
52
53
54
Noise and Vibration Engineering 22:9-14
55
56
57
58
59
60
61
62 21
63
64
65
 [25] Brincker R, Ventura C, Andersen P (2003) Why output-only modal testing is a desirable
1
2 tool for a wide range of practical applications. In Proceedings of the 21st International Modal
3
4
5 Analysis Conference (IMAC). Orlando FL. paper 265.
6
7 [26] Graugaard-Jensen J, Hjlelm H, Much K (2004) Modal based fatigue estimation. M. Sc.
8
9
10 Thesis. Aalborg University.
11
12 [27] Hjelm H, Brincker R, Graugaard-Jensen J, Munch K (2005) Determination of stress
13
14
15
histories in structures by natural input modal analysis. In Proceedings of the 21th International
16
17 Modal Analysis Conference (IMAC) Orlando FL.
18
19 [28] Fernández P, Lopéz-Aenlle M, Brincker R, Fernández-Canteli A (2009) stress estimation in
20
21
22 structures using operational modal analysis. In Proceedings of the 3rd International
23
24 Operational Modal Analysis Conference (IOMAC) Porto Novo Ancona. Paper 225.
25
26
27 [29] Papadioti DC, Giagopoulos D, Papadimitriou C (2014) Fatigue monitoring in metallic
28
29 structures using vibration measurements. In Proceedings of the 23rd International Modal
30
31
32
Analysis Conference (IMAC). Orlando FL. paper 329
33
34 [30] Bing P, Kermao Q, Huimin X, Anand A (2009) Two-dimensional digital image correlation
35
36 for in-plane displacement and strain measurement: a review. Meas Sci Technol, 20 (17pp),
37
38
39 doi:10.1088/0957-0233/20/6/062001.
40
41 [31] Lee HM, Park HS (2011) Gage-free stress estimation of a beam-like structure based on
42
43
44 terrestrial laser scanning. Comput-Aided Civ Inf, 26:647-658
45
46 [32] Yang TY (1986) Finite element structural analysis. Prentice Hall PTR.
47
48
49 [33] Clough Ray W, Penzien J (1993) Dynamics of structures, 2nd edition, McGraw-Hill, New
50
51 York.
52
53
54
[34] Zienkiewicz OC, Taylor RL (2005) The finite element method for solid and structural
55
56 mechanics. 6th ed. Elsevier, Oxford
57
58
59
60
61
62 22
63
64
65
 [35] Brincker R, Zhang L, Andersen P (2001) Output-only modal analysis by frequency domain
1
2 decomposition. Smart Mater Struct 10:441-445
3
4
5 [36] Van Overschee P, De Moor B (1996) Subspace identification for linear systems: Theory,
6
7 implementation, applications. Kluwer Academic Publishers
8
9
10 [37] Brandt A (2011) Noise and Vibration Analysis: Signal analysis and Experimental
11
12 Procedures. John Willey and Sons, LTD. United Kingdom
13
14
15
[38] Parloo E, Verboven P, Guillaume P, Van Overmeire (2002) M. Sensitivity - based
16
17 operational mode shape normalization. Mech. Systems and Signal Proc., 16:757-767
18
19 [39] Brincker, R. and Andersen, P. (2003) A way of getting scaled mode shapes in output only
20
21
22 modal analysis. In Proceedings of the 21st International modal analysis conference (IMAC).
23
24 Santa Barbara CA.
25
26
27 [40] López-Aenlle M, Brincker R, Pelayo F, Canteli, A.F. (2012) On exact and approximated
28
29 formulations for scaling-mode shapes in operational modal analysis by mass and stiffness
30
31
32
change. J Sound Vib 331:622-637
33
34 [41] Pelayo, F. (2010) Scaling factors and stresses: Estimation in structural components by
35
36 means of Operational Modal Analysis. Ph.D. Thesis. University of Oviedo
37
38
39 [42] Blevins R.D. (2001) Formulas for natural frequency and mode shape (2nd Edition). Krieger
40
41 Publishing Company. Florida
42
43
44 [43] Friswell MI, Mottershead JE (1995) Finite element model updating in structural dynamics.
45
46 Kluwer Academic Publishers.
47
48
49 [44] Pelayo F, López-Aenlle M, Brincker R, Fernández-Canteli A (2009) Artificial Excitation in
50
51 Operational Modal Analysis, In Proc. Of the 3th International Operational Modal Analysis
52
53
54
Conference (IOMAC). Portonovo Ancona. Paper 223
55
56
57
58
59
60
61
62 23
63
64
65
 Figure Captions
1
2
3 Fig. 1 Example of Euler-Bernoulli beam (left) and bending element with two nodes (right)
4
5
6
7
Fig. 2 Plate element
8
9
10 Fig. 3 Test setup or Sensor and strain gage location for the glass beam
11
12
13 Fig. 4 Singular value decomposition of the experimental responses for the glass beam
14
15
16
Fig. 5 Power spectral density of experimental stress at point g2 (left) and displacement modal
17
18
19 coordinates (right) for the glass beam
20
21
22 Fig. 6 Experimental and estimated stresses for strain gage 1 at different specific time intervals
23
24
25
26
Fig. 7 Experimental and estimated stresses for strain gage 2 (left) and estimated stress for the
27
28 complete glass beam (right) at specific time intervals
29
30
31 Fig. 8 Sensor and strain gauge location for the rectangular glass plate
32
33
34
35 Fig. 9 Singular value decomposition of the experimental responses for the glass plate
36
37
38 Fig. 10 Measured and estimated strains for the glass plate
39
40
41 Fig. 11 Auto spectral density for measured and estimated strains at SG5 for the glass plate
42
43
44
45 Fig. 12 Comparison between measured and estimated strain in form of rainflow count (SG5) for
46
47 the glass plate
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62 24
63
64
65
 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62 25
63
64
65
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHGRF[

z
y

1 2
x
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHHSV
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHGRF[

Accelerometers

g1 g2

Strain gages
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHHSV
)LJXUHOHIW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH/HSV
)LJXUHULJKW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH5HSV
)LJXUHULJKW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH5HSV
)LJXUHOHIW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH/HSV
)LJXUHULJKW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH5HSV
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHGRF[

Acc. 13 Acc. 10 Acc. 6 PS

0.33 SG6 SG4 SG1

Acc. 12 Acc. 9 Acc. 5 Acc. 2

SG5 SG2
1.00
0.33

Acc. 11 Acc. 8 Acc. 4 Acc. 1

0.33

SG4

PS Acc. 7 Acc. 3 PS
0.46 0.46 0.46
1.40

PS: Pinned Support

Acc.: Accelerometer
SG: Strain Gauge
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHHSV
)LJXUHOHIW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH/WLI
)LJXUHULJKW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH5WLI
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHHSV
)LJXUH
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUHHSV
)LJXUHOHIW
&OLFNKHUHWRGRZQORDG)LJXUH)LJXUH/HSV
7DEOH
&OLFNKHUHWRGRZQORDG7DEOH7DEOHGRF[

Table 1 Experimental and Numerical natural frequencies for the glass beam

Frequency
Mode Experimetal Numerical Error MAC
[Hz] [Hz] [%]
1 15.26 14.89 2.42 0.9999
2 59.11 58.58 0.90 0.9998
3 132.3 134.1 1.36 0.9995
4 233.4 238.7 2.27 0.9999
5 363.3 373.9 2.92 0.9988
7DEOH
&OLFNKHUHWRGRZQORDG7DEOH7DEOHGRF[

Table 2 Experimental and Numerical natural frequencies for the rectangular glass plate

Frequency
Mode Experimetal Numerical Error MAC
[Hz] [Hz] [%]
1 6.44 5.97 7.30 0.9999
2 16.41 15.94 2.86 0.9994
3 29.30 28.18 3.82 0.9989
4 38.67 36.92 4.53 0.9987
5 52.15 49.91 4.30 0.9982
6 73.83 69.39 6.01 0.9980
7 76.76 73.09 4.78 0.9971
8 94.92 89.23 6.00 0.9971
7DEOH
&OLFNKHUHWRGRZQORDG7DEOH7DEOHGRF[

Table 3 Error between measured and estimated strains for the glass plate

Strain
SG1 SG2 SG3 SG4 SG5 SG6
gage
Error [%] 6.05 3.18 2.26 5.14 2.58 1.37