DINAME 2019 - Proceedings of the XV International Symposium on Dynamic Problems of Mechanics,

M.A. Savi, T.G. Ritto and W.M. Bessa (Editors), ABCM, Buzios, RJ, Brazil, March 10th to 15th, 2019

Dynamic strain analysis using spectral element method

Thiago Pereira e Silva 1, Marcela Rodrigues Machado1, Leila Khalij2

1 Universidade de Brasília, Campus Universitário Darcy Ribeiro, CEP 70910-900, Brasília-DF, Brasil.
2 Normandie Université, LMN/INSA Rouen Normandie, 76000 Rouen, France.

The spectral element method (SEM) is a computational tool that has been successfully used in the model and evaluate
the response of the wave propagation in structures. Considering that strain is an important feature of the material to a
load, be it static or cyclic and can be used in studies of static failure or fatigue phenomenon, this work used the SEM to
evaluate the dynamic strain. The SEM is used to model a healthy homogeneous and isotropic beam. The strain frequency
response function (SFRF) is applied in the study as a manner of identifying the mechanical behavior of structure such
as its dynamics properties, mode shape, natural frequencies. The efficiency of this proposed method is validated by
comparison with an experimental simulation. The comparison of the results shows that the SEM is an efficient method
considering the proximity of results, and the direct measurement of the dynamic strain.
Keywords: Spectral Element Method, Dynamic Strain, Wave propagation.

INTRODUCTION
In the industry and mechanical researching, is highly important the existence of non-destructive methods that decrease
the investigation cost when treated of the structure. The healthily of a structure is directly linked to maintenance and
dynamic analysis. As a tool to realize this analysis, the spectral element method (SEM) is an efficient method.
The spectral element method (SEM) is an alternative method to analyze wave propagation in place of finite element
method (FEM) that works with many nodes to analyze simple and complex geometries (Doyle, 1997). The formulation
of SEM begins with an exact solution to governing partial differential equations (PDE) in the frequency domain. To
interpolation function for spectral element formulation, the exact solution is used (Machado, 2012). Using an exact
solution ensures exact mass and stiffness distribution, bringing a result as element directly yields the exact dynamic
stiffness matrix. Hence, the problem dimensions are very trivial compared to conventional FE formulation even for
compelling functions having large frequency content (Lee, 2009).
Strain measurement has been commonly used for static load testing in the industry and academic environment.
Moreover, fatigue testing, durability analysis, and lifetime prediction have also been a common application where strain
gauges are used. This sort of testing is a common part of the product development process, and additional information on
product durability and dynamic performance can be assessed by obtaining the modal parameters of the system, while still
using the same instrumentation (Santos, 2011).
In this work, the SEM is going to use to model the dynamic behavior of a Euler-Bernoulli beam, with a homogeneous
and isotropic material. The results are validated by comparison with a recent work where the author did the new approach
for the normalization of strain modes for beams, and the validation of reciprocity as long as some location conditions on
the beam. The validation of SFRFs is realized comparing experimental results, obtained by literature (Santos, 2011), and
numerical results using SEM.

SPECTRAL ELEMENT METHOD

The spectral element method is like the finite element method (FEM), the main difference is in the stiffness matrix,
that is established in the frequency domain (Doyle, 2009). The inertia of the distributed mass is described exactly due to
this. Therefore, these spectrally formulated elements exactly describe the wave propagation dynamics, and this means
that elements can be distributed in decreases number of nodes without losing fidelity. There are many kinds to establish
a dynamic stiffness, but in this paper, we going to use the dynamic shape functions. They are the interpolation function
between the elements ends, but in this way, the exact displacement distributions are used instead of simple polynomials.
Dynamic strain analysis using spectral element method

Figure 1 – Spectral element of Euler-Bernoulli beam.

Considering a homogeneous and isotropic beam, that have constant properties along the length, then the homogeneous
differential equation is written in the spectral form as (Doyle, 2009),
𝑑 4 𝑣̂
− 𝛽 4 𝑣̂ = 0 (1)
𝑑𝑥 4

where,
𝜔2 𝜌𝐴−𝑖𝜔𝐴
𝛽2 = √ (2)
𝐸𝐼

where 𝜔 is the angular frequency, 𝜌 mass density, 𝐴 cross section area, 𝐸 Young Modulus, 𝐼 second moment of area.
Relating the two equations following, to its particular solutions we can obtain particular solutions to Eq.(1),
𝑑 2 𝑣̂
+ 𝛽 2 𝑣̂ = 0 (3)
𝑑𝑥 2

and,
𝑑 2 𝑣̂
− 𝛽 2 𝑣̂ = 0 (4)
𝑑𝑥 2

using the solution as 𝑒 −𝑖𝑘𝑥 , substituting into the equations gives

𝑘1 = ± 𝛽, 𝑘2 = ±𝑖𝛽
the complete solution is,
𝑣(𝑥, 𝑡) = ∑[𝐴𝑒 −𝑖𝛽𝑥 + 𝐵𝑒 −𝛽𝑥 + 𝐶𝑒 +𝑖𝛽𝑥 + 𝐷𝑒 +𝛽𝑥 ]𝑒 𝑖𝜔𝑡 (5)

Shape function
In the Bernoulli-Euler beam, just the flexural stress is considered, so the transverse displacement of the beam can be
written in the form,
𝑣̂(𝑥) = 𝑔̂1 (𝑥)𝑣̂1 + 𝑔̂2 (𝑥)𝐿𝜙̂1 + 𝑔̂3 (𝑥)𝑣̂2 + 𝑔̂4 (𝑥)𝐿𝜙̂2 (6)
for the two-nodded element, the frequency dependent shape functions are given as
𝑔̂1 (𝑥) = 𝑟1 ℎ̂1 (𝑥) + 𝑟2 ℎ̂2 (𝑥) (7)
𝑔̂2 (𝑥) = 𝑟1 ℎ̂3 (𝑥) + 𝑟2 ℎ̂4 (𝑥) (8)
𝑔̂3 (𝑥) = 𝑟1 ℎ̂2 (𝑥) + 𝑟2 ℎ̂1 (𝑥) (9)
𝑔̂4 (𝑥) = 𝑟1 ℎ̂4 (𝑥) + 𝑟2 ℎ̂3 (𝑥) (10)
Δ= −𝑟12 + 𝑟22 (11)
−𝑖𝑘2 𝐿 −𝑖𝑘2 𝐿 )
𝑟1 = 𝑖(𝑘1 − 𝑘2 )(1 − 𝑒 𝑒
𝑟2 = 𝑖(𝑘1 + 𝑘2 )(𝑒 −𝑖𝑘2 𝐿 − 𝑒 −𝑖𝑘2𝐿 )
where
ℎ̂1 (𝑥) = 𝑖𝑘2 [𝑒 −𝑖𝑘1𝑥 − 𝑒 −𝑖𝑘2 𝐿 𝑒 −𝑖𝑘1 (𝐿−𝑥) ] − 𝑖𝑘1 [𝑒 −𝑖𝑘2 𝑥 − 𝑒 −𝑖𝑘1 𝐿 𝑒 −𝑖𝑘2 (𝐿−𝑥) ]
ℎ̂2 (𝑥) = −𝑖𝑘2 [𝑒 −𝑖𝑘2𝐿 𝑒 −𝑖𝑘1𝑥 − 𝑒 −𝑖𝑘1(𝐿−𝑥) ] + 𝑖𝑘1 [𝑒 −𝑖𝑘1𝐿 𝑒 −𝑖𝑘2𝑥 − 𝑒 −𝑖𝑘2(𝐿−𝑥) ]
ℎ̂3 (𝑥) = [𝑒 −𝑖𝑘1 𝑥 + 𝑒 −𝑖𝑘2𝐿 𝑒 −𝑖𝑘1(𝐿−𝑥) ] − [𝑒 −𝑖𝑘2 𝐿 + 𝑒 −𝑖𝑘1𝐿 𝑒 −𝑖𝑘2(𝐿−𝑥) ]
ℎ̂4 (𝑥) = [𝑒 −𝑖𝑘2 𝐿 + 𝑒 −𝑖𝑘1 𝑥 𝑒 −𝑖𝑘1(𝐿−𝑥) ] − [𝑒 −𝑖𝑘1 𝐿 𝑒 −𝑖𝑘2 𝑥 + 𝑒 −𝑖𝑘2 (𝐿−𝑥) ]

Dynamic stiffness matrix

Considering the Euler-Bernoulli beam model with external excitation and nodal displacements given by 𝑣̂ e 𝜙̂ being the
rotational and translational degree of freedom, respectively. Thus, the solution to equation (3) is given by,
 T. Silva, M.R. Machado, L. Khalij

𝑣̂(𝑥, 𝑡) = 𝐴1 𝑒 −𝑖𝑘𝑥 + 𝐴2 𝑒 −𝑘𝑥 + 𝐴3 𝑒 𝑖𝑘(𝐿−𝑥) + 𝐴4 𝑒 𝑘(𝐿−𝑥) (12)

Where 𝑥 is the analysis point and 𝐿 is the beam length. Applying the boundaries conditions to 𝑥 = 0 and 𝑥 = 𝐿 in the
eq.18 𝐴𝑖 (𝑖 = 1: 4) can be obtained in function of spectral nodal displacement. The boundaries conditions can be used to
write the eq.18 in matrix form,

1 1 𝑒 −𝑖𝑘𝐿 𝑒 −𝑘𝐿 𝐴1 𝑣̂1

−𝑖𝑘
[ −𝑖𝑘𝐿 −𝑘 𝑖𝑒 −𝑖𝑘𝐿 𝑒 ] {𝐴2 } = 𝜙̂1
−𝑘𝐿
(13)
𝑒 𝑒 −𝑘𝐿 1 1 𝐴3 𝑣̂2
⏟−𝑖𝑒 −𝑖𝑘𝐿 −𝑒 −𝑘𝐿 𝑖𝑘 𝑘 𝐴4 {𝜙̂2 }
𝜓

The spectral forces (shear force and bending moment) in the beam nodes are given by,
2𝑣
̂
̂ (𝑥) = +𝐸𝐼 𝑑
𝑀 (14)
𝑑𝑥 2
𝑑3 ̂
𝜙
𝑉̂ (𝑥) = −𝐸𝐼 (15)
𝑑𝑥 3

To the left side of the beam (𝑥 = 0) and the right side (𝑥 = 𝐿) and the nodal spectral forces can be written in the matrixial
form,
𝑣̂1 −𝑖𝑘 3 𝑘3 𝑖𝑒 −𝑖𝑘𝐿 𝑘 3 −𝑒 −𝑘𝐿 𝑘 3 𝐴1
̂
𝜙1 2
𝑘2 −𝑒 −𝑖𝑘𝐿 𝑘 2 −𝑒 −𝑘𝐿 𝑘 2 ] {𝐴2 },
= [ −𝑘 (16)
𝑣̂2 𝑖𝑒 −𝑖𝑘𝐿
𝑘3 −𝑒 −𝑘𝐿 𝑘 3 −𝑖𝑘 3 𝑘3 𝐴3
{𝜙̂2 } ⏟𝑒 −𝑖𝑘𝐿
𝑘2 −𝑒 −𝑘𝐿 𝑘 2 𝑘2 −𝑘 2 𝐴4
Γ

Considering the Eq. (13) and (16), the frequency dependent dynamic stiffness which relates the nodal spectral forces with
the nodal spectral displacement, can be written as,
{𝐹} = [𝑆(𝜔)]{𝑑(𝜔)}, (17)
where {𝐹} is the vector content of the forces, {𝑑(𝜔)} is the displacement vector and the dynamic stiffness matrix is given
by,
[𝑆(𝜔)] = Γ 𝜓 −1 (18)

Stress-strain relation
The Bernoulli Euler beam model adopts that the deflection of the centerline 𝑣(𝑥, 𝑡) is small and only transverse. While
this theory assumes the existence of a transverse shear force, it neglects any shear deformation due to it (Doyle, 2009).

Figure 2 –Slender beam in flexure and typical loaded element.

The displacements are expanded in a Taylor series about the mid-plane displacements 𝑢̂ (𝑥, 0) and 𝑣̂ (𝑥, 0) as
̂
𝜕𝑢
𝑢̂(𝑥, 𝑦) ≈ 𝑢̂(𝑥, 0) + 𝑦 | + ⋯ = 𝑢(𝑥) − 𝑦𝜙(𝑥) + ⋯ (19)
𝜕𝑦 𝑦=0
𝜕𝑣̂
𝑣̂(𝑥, 𝑦) ≈ 𝑣̂(𝑥, 0) + 𝑦 | + ⋯ = 𝑣(𝑥) − 𝑦𝜓(𝑥) + ⋯ (20)
𝜕𝑦 𝑦=0

The hypothesis of the Bernoulli-Euler beam is that vertical deflection is nearly constant through the thickness while
the horizontal displacement follows the “plane sections remain plane” assumption. Only one term in each expansion is
retained and obtain the approximate displacements as
𝑢̂(𝑥, 𝑦) ≈ −𝑦𝜙(𝑥), 𝑣̂(𝑥, 𝑡) ≈ 𝑣(𝑥) (21)
The axial strain equivalent to these deformations are
̂
𝜕𝑢 𝜕𝜙 𝜕2 𝑣̂
𝜖𝑥𝑥 = = −𝑦 = −𝑦 (22)
𝜕𝑥 𝜕𝑥 𝜕𝑥 2
Dynamic strain analysis using spectral element method

Entering a time-harmonic forcing in the point x=0 of an infinite beam, tow waves are formed. The forward and
backward waves, that are characterized by subscripts i and r. The two groups can be denoted as
𝑣𝑖 = ∑ 𝐴𝑒 −𝑖(𝛽𝑥−𝜔𝑡) + ∑ 𝐵𝑒 −𝛽𝑥+𝑤𝑡 (23)
𝑣𝑟 = ∑ 𝐶𝑒 𝑖(𝛽𝑥−𝜔𝑡) + ∑ 𝐷𝑒 𝛽𝑥+𝑤𝑡 (24)
assuming the wave propagates symmetrically outward, then at x=0:
𝜕𝑣
= 0, 𝑣𝑖 = 𝑣𝑟
𝜕𝑥

for the forward disturbance, the strain quantities are:

𝜖𝑖 = 𝑦 ∑ 𝛽 2 𝐴[𝑒 −𝑖𝛽𝑥 + 𝑖𝑒 −𝛽𝑥 ]𝑒 𝑖𝜔𝑡 (25)

STRAIN MODAL ANALYSIS

Nodal displacement and strain relationship

According to (Kranjc, 2014) the strain response of a dynamical system is derived from the motion response. The motion
steady-state response 𝑿(𝜔) of a histerestic damped dynamical system can be written as (Ewins, 1984) and (Maia, 1997),
𝑿(𝜔) = 𝚽[𝜔𝑟2 (1 + 𝑖𝜂𝑟 ) − 𝜔2 ]−1 𝚽 𝑇 𝑭(𝜔) = 𝑯(𝜔)𝑭(𝜔) (26)
where 𝚽 is the modal matrix, 𝜔𝑟 are the natural frequencies, 𝜂𝑟 are the damping loss factors, 𝑭(𝜔) is the vector of the
excitation force, 𝑯(𝜔) is the receptance matrix.
The modal solution to a Euler Bernoulli beam has the form:
Φ𝑟 = 𝐴1 sin(𝜆𝑟 𝑥) + 𝐴2 cos(𝜆𝑟 𝑥) + 𝐴3 sinh(𝜆𝑟 𝑥) + 𝐴4 cosh(𝜆𝑟 𝑥) (27)
where the constants 𝐴1 , 𝐴2 , 𝐴3 and 𝐴4 depend directly of the beam boundaries conditions, the subscript 𝑟 relates the 𝑟 𝑡ℎ
mode. The value of 𝜆𝑟 depends on the beam properties and the natural frequency 𝜔𝑟 of its related mode,
𝜔 𝐸𝐼
𝜆𝑟 = √ 𝑟 , and 𝑐 = √ (28)
𝑐 𝜌𝐴

The mode curvature 𝑘 can be used to relate the displacement with strain for a bending beam. In the bending of a beam,
the strain can be directly related to mode curvature as,
Φ𝑟𝜀 = ℎ𝑘, 𝑘 = Φ′′ and Φ𝑟𝜀 = ℎΦ′′ (35)
where h is the distance from the neutral line to the surface of the beam, 𝜓 is the strain mode and the ′′ is the second spatial
differentiation. Applying the strain mode in the Eq.(24) result to the strain steady-state response 𝑿𝜀 (𝜔) (Bernasconi,
1989),
𝑿𝜀 (𝜔) = 𝚽 𝜀 [𝜔𝑟2 (1 + 𝑖𝜂𝑟 ) − 𝜔2 ]−1 𝚽 𝑻 𝑭(𝜔) = 𝑯𝜀 (𝜔) 𝑭(𝜔) (29)
𝜀 𝜀
where 𝑯 (𝜔) is the strain Frequency Response Function (FRF) matrix and Φ is the matrix of mass-normalized strain
mode shapes. 𝑯𝜀 (𝜔) can be written as (Yam et.al, 1996),
𝑨𝜀𝑟
𝑯𝜀 (𝜔) = ∑𝑁
𝑟=1 (30)
𝜔𝑟2 −𝜔2 +𝑖𝜂𝑟 𝜔𝑟2

where 𝑨𝜀𝑟 is the strain modal constants matrix, equivalent to the rth mode and can be written as,
𝜀 𝜀 𝜀
𝜙1𝑟 𝜙1𝑟 ⋯ 𝜙1𝑟 𝜙𝑘𝑟 … 𝜙1𝑟 𝜙𝑑𝑟
⋮ ⋱ ⋮ ⋱ ⋮
𝜀 𝜀 𝜀
𝑨𝜀𝑟 = 𝜙𝑗𝑟 𝜙1𝑟 ⋯ 𝜙𝑗𝑟 𝜙𝑘𝑟 ⋯ 𝜙𝑗𝑟 𝜙𝑑𝑟 (31)
⋮ ⋱ ⋮ ⋱ ⋮
𝜀
[𝜙𝑁𝑠𝑟 𝜙1𝑟 ⋯ 𝜙𝑁𝜀 𝑠𝑟 𝜙𝑘𝑟 ⋯ 𝜙𝑁𝜀 𝑠𝑟 𝜙𝑁𝑑𝑟 ]
𝑁𝑑 𝑥𝑁𝑠
𝜀
where 𝜙𝑗𝑟 and 𝜙𝑘𝑟 are the components of 𝚽𝑟𝜀 and 𝚽𝑟 , respectively. 𝑁𝑑 and 𝑁𝑠 are the dimensions of 𝚽𝑟 and 𝚽𝑟𝜀 ,
respectively (Yam et.al, 1996).

Simulated Results
The parameters used in the simulation are given in Table (1), the beam is made of a steel material, free-free boundaries
conditions were considered. The beam is excited by a unitary force at some point along of its length.
 T. Silva, M.R. Machado, L. Khalij

Table 1 – Material and geometrical properties of beam

Properties Value [SI]

Length 1m
Width 0,03 m
Thickness 0,005 m
Young’s modulus 210x109 Pa
Density 7860 kg/m3

The beam was discretized was into eight points (see fig.3), where each point has one accelerometer and one strain
gauge, less the point three has just strain gauge (Santos, 2011). The experimental data was extracted from the work of
Santos et al. (Santos, 2011) and the simulated results were obtained by using similar material and geometrical properties
presented in their paper.

Figure 3 – Beam discretization

In the fig. 4 (a) the SFRF shows the dynamic response of a beam using the theoretical modal analysis, SEM and
experimental modal analysis. The beam was excited in point 3 and measured in point 5, figure 4(a), and excited in 3 and
measured in point 8, figure 4(b). By comparing the numerical methods, SEM and theoretical modal analysis presented
similar resonance picks and close mode shapes with exception of the amplitudes at the two firsts resonance frequencies.
By comparing the numerical methods with the experimental SFRF obtained by the experimental modal analysis, there is
a small discrepancy at the first and second resonance frequencies, it is related with the randomness in boundary condition
that would not be considered in the numerical model. From frequency, 100Hz is seeing a lot of noise in the experimental
data. However, the methods showed a small difference, the SEM had a good response in obtained the SFRF. That
advantage in using the proposed technique to numerically estimated de SFRS is because, SEM calculates directly the
SFRF and, once it is formulated based on the analytical wave solution the dynamic response has an accurate solution.

(a) (b)
Figure 4 (a) – Beam SFRF using three different methods, modal analytical, SEM and Modal experimental.
In (a), the excitation was inserted in the point 3 and measurement done the point 5 and in (b) the
excitation was maintained in point 3 and the measurement was done in the point 8.

The experimental modal data to calculate the SFRF using the modal analysis were taken from (Santos, 2011). The
strain modal shape was obtained by measured the strain deformation through 8 strain gauges set along the beam. To obtain
the modal shape, the dimension, material properties and boundaries conditions of the experimental test and the numerical
simulations are the same.
Dynamic strain analysis using spectral element method

Figure 5 –1º to 6º displacement modes shapes (--) and strain modes shapes (˗̶̶ ), obtained
through the analytical modal analyses.

Figure 5 shows the 1st to 6th displacement and strain modal shapes. For each mode shape the related to correspondent
ressonance frequency is showed in Table 2 and Table 3.

Table 2 – Resonance frequency by mode for SEM, modal analytical and modal experimental methods.

Frequency [Hz]
Mode
SEM Modal Analytical Modal Experimental
1 26,54 26.36 17,10
2 73,24 73,24 64,37
3 143,50 143,56 136,32
4 237,40 237,31 234,4
5 354,40 354,50 354,3
6 495,55 495,13 495,8

Table 3 – Resonance frequency by mode comparing SEM and modal analytical methods.

Frequency [Hz]
Mode Error [%]
SEM Modal Analytical
1 26,54 26.36 0,68
2 73,24 73,24 0
3 143,50 143,56 0,04
4 237,40 237,31 0,03
5 354,40 354,50 -0,02
6 495,15 495,13 0,01
 T. Silva, M.R. Machado, L. Khalij

Table 4 – Resonance frequency by mode comparing SEM and modal experimental methods.

Frequency [Hz]
Mode Error [%]
SEM Modal Experimental
1 26,54 17,10 55,20
2 73,24 64,37 13,77
3 143,50 136,32 5,26
4 237,40 234,4 1,27
5 354,40 354,3 0,02
6 495,55 495,58 -0,01

In Table 2 the six resonance frequency correspondent to each mode obtained by SEM, theoretical modal analysis, and
experimental modal analysis are compared. Table 3 and 4 comparing SEM with theoretical modal analysis, and SEM
compared to experimental modal analysis, respectively. The maximum relative error between the two numerical methods
was 0.68%. By comparing the SEM with the experimental data, Table 4, for the first and second frequencies, the error
was 55% and 13% respectively, for the 3rd to 6th the maximum relative error was 5%. The high error in the first resonance
frequency is due the boundary condition.

CONCLUSIONS
In this research, we considered the SFRF using SEM, as an alternative method to analyses the strain dynamic behavior.
The SFRF usually is obtained by modal analytical, in numerical cases and modal experimental at experimental cases. In
modal experimental is used the strain gauges to measure the structure response. These works proposed the use of SEM in
the SFRF formulation.
Approximated results between the SEM and the modal analytical. In all frequencies the response of both methods is
similar showing that the theoretical model for both methods are equivalents. In the SEM and modal experimental
comparations, the results had a high error in low frequency, that can be caused by boundaries conditions of the experiment
or simply the imprecision of both methods in low frequency. The precision of the SEM increases with the frequency.

REFERENCES
Doyle, J.F., 1997. “Wave propagation in structures: spectral analysis using fast discrete Fourier transforms”. Mechanical
engineering. Springer-Verlag New York, Inc., New York, 2nd edition.
Machado, M.R. and DosSantos, J.M.C., 2012. “Wave propagation in a cracked beam spectral element including parameter
uncertainty”. In Proceedings of the 1st International Symposium on Uncertainty Quantification and Stochastic
Modeling.
Lee, U., 2009. “Spectral element method in structural dynamics” John Wiley & Sons, Inha University, Republic of Korea.
Santos, F.L.M, 2011. “strain-based experimental modal analysis: use of mode curvature and strain-to-displacement
relations”, Siemens Industry Software, Leuven, Belgium.
D. Ewins, 1984. “Modal testing: Theory and practice, Mechanical engineering research studies: Engineering dynamics
series”, Research Studies Press.
N. Maia, J. Silva, 1997 “Theoretical and Experimental Modal Analysis, Mechanical Engineering Research Studies:
Engineering Dynamics Series”, John Wiley& Sons, Limited.
L. Yam, T. Leung, 1996. D. Li, K. Xue, “Theoretical and experimental study of modal strain analysis”, Journal of Sound
and Vibration 191 (2) 251 – 260.
Kranjc, T., Slaviˇc, J. and Bolteˇzar, M., “The mass normalization of the displacement and strain mode shapes in a strain
experimental modal analysis using the mass-change strategy”, Journal of Sound and Vibration, Volume 332, Issue
26, 23 December 2013, Pages 6968-6981, ISSN 0022-460X, http://dx.doi.org/10.1016/j.jsv.2013.08.015.
O. Bernasconi, D. J. Ewins, 1989. “Application of strain modal testing to real structures”, In: Proceedings of the 7th
International Modal Analysis Conference (IMAC-VII), pp. 1453–1464.