Evolutionary Intelligence (2020) 13:571–591

https://doi.org/10.1007/s12065-020-00372-1

RESEARCH PAPER

Sensor placement optimization and damage identification

in a fuselage structure using inverse modal problem and firefly
algorithm
Guilherme Ferreira Gomes1 · João Vitor Purcino Pereira1

Received: 12 August 2019 / Revised: 6 February 2020 / Accepted: 20 February 2020 / Published online: 3 March 2020
© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
The structural monitoring of mechanical systems is an extremely important task for ensuring its performance and structural
health. To overcome limitations of traditional non-destructive inspections (NDIs), damage identification techniques have
been developed from global indicators, mainly those based on modal data. In this study, damages are identified by solving an
inverse problem. A fuselage model of an E190 aircraft is considered and the firefly algorithm (FA) metaheuristic is applied
to solve the inverse problem in order to identify structural damages (location and severity). The method is then solved in
two main fronts: (1) the direct problem using finite element analysis and (2) the inverse problem by minimizing an objective
function. Evaluating modal response at many points on a large-scale structure can become prohibitive. For this, a method
of optimizing sensors is performed using the Fisher information matrix (FIM). Results are compared considering the sensor
placement optimization problem. It is noticed that optimized sensors contribute to an improvement in the identification of
damages, mainly for complex and large-scale structures. The proposed optimized damage identification process using FIM-
FA has the potential to be extended to a wide range of SHM applications in complex structures. Hence, traditional NDIs
have many shortcomings due to the complexity of large-scale structures as well as modern design structures and may not
be practicable if the structure has restricted access. Accordingly, an enhanced damage identification method is developed in
order to better handle measurement data to find structural changes (or damages) in complex aerospace structures.

Keywords Damage identification · Inverse problem · Firefly algorithm · Fuselage · Sensor placement optimization

List of symbols Δkj Stiffness of the jth damaged element

M Mass matrix βj Stiffness reduction factor of the jth damaged
ÿ (t) Acceleration element
K Stiffness matrix Δ𝜆i Variation of the ith eigenvalue due to the damage
y(t) Displacement Δ𝜙i Variation of the ith eigenvector due to the damage
Y(y) Applied load f (x) Objective function
n Number of degrees of freedom x Position vector
ϕj jth eigenvector I(r) Intensity of firefly brightness as a function of
λj jth eigenvalue distance
m Total number of evaluated mode shapes I0 Firefly brightness intensity at distance equal to
ki Stiffness of the ith element zero
Kd Damaged stiffnes matrix λ Light absorption coefficient of medium
λdi jth eigenvalue of the damaged structure r Distance from the firefly
ϕdi jth eigenvector of the damaged structure β(r) Attractiveness of the firefly as a function of
distance
β0 Attractiveness of the firefly in the distance equal to
* Guilherme Ferreira Gomes zero
guilhermefergom@unifei.edu.br α Parameter of randomness
1
Mechanical Engineering Institute, Federal University β Damage intensity
of Itajubá – UNIFEI, Itajubá, Brazil Nelem Damaged element number

13
Vol.:(0123456789)
572 Evolutionary Intelligence (2020) 13:571–591

1 Introduction by several sources. In the manufacturing process, struc-

tural parts undergo strong inspection procedures to detect
Several non-destructive techniques for inspection of struc- damages, therefore, SHM techniques are more restricted to
tures, known as non-destructive testing (NDT), have been maintenance operations, aiming to detect damages caused
developed. There many traditional and well-defined meth- during the operation of mechanical systems [6–8].
ods in the industry, such as penetrant liquid, eddy current, According to Pierotti [9], the source of structural dam-
ultrasound, acoustic emission, and many others. For Farrar age can be occurred from accidents (impacts on structures),
and Worden [1], this type of inspection is of high precision environmental deterioration or fatigue. Such conditions may
and reliability, but it has the disadvantage that it is neces- lead mainly to cracking and, in the case of composite materi-
sary a priori knowledge of the existence of damages in the als, delamination. SHM techniques assume that these defects
structure and of a probable region where they are located. in the structure promote changes in the response obtained
In the aeronautical context, the collection of aircraft to by the sensors, which allows their identification. In line with
look for fragile regions in its structure is very expensive, Kaveh and Dadras [10], it is necessary that the processing
both due to the high maintenance costs and the idleness of of the data obtained experimentally reveal: (1) if there is
the vehicle. Therefore, it was necessary to search for more damage in the structure, (2) the location of the damage in
efficient methods, which do not occupy much time of the the structure and (3) the intensity of the damage identified.
machine to verify its structural integrity. In like manner, there are several types of responses of a
Motivated by the limitations of localized inspections, mechanical system that are altered due to the presence of a
researchers focused on finding global indicators in struc- structural damage. For example, the static response of the
tures and relating them to structural health monitoring and system can be analyzed, which must accumulate differences
integrity. According to Abdo [2], this type of approach can with the presence of damages in relation to its response
be classified as a response-based method, which obtains under normal conditions. Also, it can be analyzed mode
information from experimental tests in the structure, shapes, natural frequencies, modal strain energy and damp-
or model-based method, which assumes that a detailed ing coefficients [11]. Vibration-based responses are gener-
numerical model of the structure is sufficient for the loca- ally chosen due to the ease of instrumentation and also about
tion of damage. having characteristics significantly influenced by structural
A multitude of multidisciplinary techniques dedicated modification (e.g.: damages) [12–15].
to solving global structural monitoring problems has Under those circumstances, several applications in
emerged over the last few decades, and are included in a Mechanical, Aerospace and Civil [16] Engineering can be
set of expertise called structural health monitoring (SHM) formulated as optimization inverse problems. In the same
[1, 3, 4]. The procedure is based on the arrangement of way, metaheuristics are known as global methods, common
sensors along the structure and the processing of the data nature-inspired, able to address nonlinear optimization, han-
obtained to verify the structural condition of the system. dling black box objective functions for which an unambigu-
It is possible to compare these techniques with the human ous formulation is not possible [17].
nervous system, which obtains information from nerve In the case of model-based methods, it is necessary to
cells throughout the human body and sends them to the perform the numerical simulation of the structure solv-
brain, which processes and return instructions. ing a forward problem, in which the physical properties of
Nowadays, aircraft fuselage structures normally exhibit the model are well known and it is desired to predict the
a different design, which is characterized by the assem- behavior of the system given an input condition. By using
bly of multiple parts. These so-called built-up structures vibrations, it is possible to apply the Finite Element Method
originate from the philosophy of using sheet metal as a (FEM) and obtain the necessary information in each mode of
base material from which thin walled wing or fuselage vibration. In this way, it is possible to compare the response
structures can be manufactured by riveting and bonding obtained from the structure without harm to the response
[5]. According to Hansen et al. [5], efforts have been made obtained from the real structure. When a difference in the
to investigate structural alternatives are primarily moti- information is verified, an inverse problem is solved. In this
vated by potential mass reductions and savings with regard case, the system responds to an input, but both location and
to manufacturing costs and time. Indeed, the search for intensity of the damage is unknown [18]. To determine these
increasingly lightweight structures leads to a search for variables, and solve the inverse problem, optimization tech-
increasingly efficient damage identification methods. niques are used.
Equally important, with the information from the sen- A function that calculates the error between the struc-
sors, the data processing should be able to identify dam- tural responses, that is, a measure of difference between the
ages, which are defects in the structure and can be caused data, can be constructed by relating the variables to be deter-
mined: position and intensity of the damage. In this way,

13
Evolutionary Intelligence (2020) 13:571–591 573

it is feasible to use optimization algorithms that minimize the optimal multi-type sensor placement determined by the
the value of the function, varying the value of the variables proposed method can avoid redundant sensors and provide
of interest. The closer to zero the value of the function, the satisfactory results for structural damage detection.
smaller the difference between the answers and the more Equally important, Dihn-Cong et al. [31] proposed an
accurate the information about the damage. However, the efficient approach for optimal sensor placement (OSP) and
search domain of the variables is large and the error func- damage identification in laminated composite structures
tion may not be very simple, which leads to the search for using iterated improved reduced system (IIRS) method using
complex algorithms that can solve the global optimization flexibility matrix responses.
problem [19]. Furthermore, Mishra et al. [32] applied an algorithm
In the same way, the search for minima of functions called ant lion optimizer (ALO) for damage detection prob-
obtained relevant studies and development throughout the lems. The results were compared with those of unified par-
twentieth century. A large portion of these algorithms uses ticle swarm optimization (UPSO). The results indicate that
the derivative of the function to minimize it. However, the the proposed algorithm required fewer parameters than other
use of these methods can restrict the algorithms to local metaheuristic algorithms to identify the location and extent
minimum points, that is, points that have a value that is mini- of damage. Correspondingly, a new metaheuristic algorithm
mum only in a specific range, and not in the entire domain was proposed by Razzaghi et al. [33] where the natural
of the function [20–23]. behavior of moth is modelled and applied in SHM problems.
To overcome problems with local minimums, many Under those circumstances, Seyedpoor et al. [34] pro-
researchers have proposed new algorithms. According to posed a new method to determine the location and severity
Yang [24], metaheuristic algorithms have gained strength of structural damages using vibration time domain responses
for global optimization problems by combining tests at ran- and differential evolution (DE) optimization algorithm. The
dom points and past performance history to look for optimal results from numerical simulation stated the efficiency of
points in neighborhoods and avoid premature stops at local the inverse problem for properly identifying damage with
minima. considering measurement noise. By the same token Wei
Recently, many metaheuristic algorithms have been pro- et al. [35] proposed an improved PSO for damage detection
posed to solve various damage identification problems. Aga- problems based on modal data as well. In like manner, Xu
thos et al. [25] developed A methodology for the detection of et al. [36] presented an iterative two-stage damage identi-
multiple cracks in 3D solids of arbitrary geometries by using fication method based on modal strain energy considering
XFEM and optimization algorithms. The method was tested multiobjective PSO.
in numerical examples involving the detection of multiple Still evolving algorithms, researchers have begun to be
cracks in solids of non-regular geometries and promising inspired by nature to improve optimization results. A pow-
results were obtained. Du et al. [26] applied the Jaya algo- erful and widely used algorithm in SHM is the Genetic
rithm to solve an optimization-based damage identification Algorithm (GA). From a set of initial test points (popula-
problem based on modal MDLAC metric with criteria based tion), the algorithm causes small changes in the position
on the difference between the flexibility matrix of the test (genetic mutations) of population points closer to the global
models and the flexibility matrix of the intact model. Fur- minimum, approaching more and more of the optimal point
thermore, Gomes et al. [27] proposed a new metaheuristic [37, 38]. In addition to the genetic algorithm, several algo-
called Sunflower optimization (SFO) applied to delamina- rithms inspired by nature have been proposed in recent dec-
tion identification in CFRP plates. The authors compared ades, such as Simulated Annealing [39, 40], Differential
SFO and obtained better results against GA. Kavek and Evolution [41–43], Particle Swarm Optimization [22, 44],
Dadras [28] presented an inverse damage identification Cuckoo Search (ref ga), Jaya Algorithm [45], SunFlower
problem study based on thermal exchange optimization Optimization [21], Ant Colony Optimization [46], Bat Algo-
algorithm (TEO). rithm [47, 48], Grey Wolf Optimization [49], Whale Opti-
Several scenarios with noise and noise-free modal data mization Algorithm [50] and many others.
are tested and the locations and extents of damages are iden- Another strong algorithm, and used in this work, the
tified with good accuracy. Moreover, Khatir et al. [29] pre- firefly algorithm (FA), or firefly algorithm, was proposed
sented a technique for damage detection and localization for by Yang [51]. It is based on the blink of light emitted by
open crack in beam- like structures using experimentally the fireflies, which has two main functions: attract potential
measured natural frequencies and Particle Swarm Optimi- breeding partners and attract prey. For the construction of
zation (PSO) algorithm. Similarly, Lin et al. [30] presented the algorithm, the author considered that each test point will
a structural damage detection-oriented multi-type sensor be considered a firefly. The lower the value of the function at
placement method considering multi-objective optimiza- the point determined by the firefly, the greater the intensity
tion using NSGA-II algorithm. Numerical results show that of its brightness, and the greater the brightness, the more

13
574 Evolutionary Intelligence (2020) 13:571–591

the neighbor firefly will be attracted. That is, for any two damage identification process using FIM-FA has the poten-
fireflies, the one that shines less will move to a point closer tial to be extended to a wide range of SHM applications
to the brightest firefly. in complex structures. Hence, traditional NDIs have many
At the end of the optimization process, it is expected to shortcomings due to the complexity of largescale structures
find the position and intensity of the damage that generate and in modern design structures and may not be practica-
the lowest value of the objective function and coincide with ble if the structure has restricted accessibility. Accordingly,
the location and intensity of the damage in the real structure. an enhanced damage identification method is developed in
In generating this information, the location of damage deter- order to better handle measurement data to find structural
mined by the SHM process may undergo a more specific changes (or damages) in complex aerospace structures.
and, if necessary, repaired inspection, ensuring the safety of This manuscript is organized as follows: Sect. 2 a general
the mechanical system structure being evaluated and drasti- bibliographic review is presented, addressing the scientific
cally reducing maintenance costs and maintenance time in innovation about the subject and the main methodologies
relation to systems monitoring systems. applied in this paper. Section 3 methodological procedure
This work aims to use the firefly algorithm to identify (direct and inverse formulation) is presented. Section 4 pre-
damages through the vibration response of an aeronautical sents the main results and discussion about the damage iden-
fuselage structure. A deep numerical study of damage iden- tification. Finally, Sect. 5 draws the conclusions.
tification applied to an E190 fuselage structure is made as
in inverse problem methodology in this present manuscript.
The results show a very good precision in terms of identify-
ing the correct damage position and its severity. 2 Backgrounds
To overcome the drawback related to the inapplicability
of the FA in discrete optimization problems, an improvement 2.1 Finite element method
is then proposed, developed and applied. The comparative
study indicates that the use of information from optimized According to Srinivas et al. [56], for a linear system without
sensor location by the Fisher information matrix (FIM), or damping and with multiple degrees of freedom, the equation
simply, FIM-FA outperforms the FA in terms of algorithm of motion is given by Eq. 1.
complexity, computational efficiency, and result quality. The
proposed optimized damage identification process using
𝐌̈y(t) + 𝐊y(t) = 𝐘(t) (1)
FIM-FA has the potential to be extended to a wide range of where M is the mass matrix of the system and K, the stiff-
SHM applications in complex structures. ness matrix. The two matrices are of order (n × n), where
According to Zhou et al. [52], the complexity of the opti- n is the number of degrees of freedom of the system. y(t)
mal sensor configuration for large and complex structures and Y(t) are the physical displacements and applied loads,
reveals the necessity for the development of efficient and respectively, and have order (n × 1). The equation for the jth
robust algorithms to accurately explore the optimum solu- eigenvalue is given by Eq. 2.
tion. Recently, a new global optimization algorithm, known
as firefly algorithm (FA), has been developed by Yang [53, 𝐊𝜙j − 𝜆j 𝐌𝜙j = 0 (2)
54]. The FA algorithm is based on the behavior of the flash-
for j = 1.2,…, m. φj is the eigenvector (mode shape), λj is
ing characteristics of fireflies. A firefly tends to be attracted
the eigenvalue (natural frequency) and m is the total number
by other fireflies with high flash intensities. Firefly algorithm
of modes to be obtained. In the finite element model of the
offers advantages of operation when searching in extensive
structure, the stiffness matrix can be represented by a stiff-
solution spaces since it does not have a starting point and it
ness matrix composition of each element (Eq. 3).
avoids falling into local optimum, improving its performance
in the global space [55]. ∑
m

The work presented here assesses the potential of this 𝐊= ki (3)

technique as the main tools in the inverse problem solution. i=1

To the authors’ best knowledge, there are no (or very scarce) where ki represents the stiffness matrix of the ith element.
studies in the literature investigating the use of optimized Often, damage causes loss of stiffness in one or more
vibration data acquisition in fuselage structures by means elements, but not a loss of mass. Using the superscript d
of finite element update and FIM-FA. In addition, the com- for the properties of the damaged structure, the stiffness
parative study indicates that the use of information from matrix ­Kd, the ith modal eigenvalue λid and the ith eigen-
optimized sensor location by FIM, or simply, FIM-FA out- vector φid can be written as shown in Eqs. 4–6.
performs the FA in terms of algorithm complexity, computa-
tional efficiency, and result quality. The proposed optimized

13
Evolutionary Intelligence (2020) 13:571–591 575

∑
m
∑
m
where 𝐊 ̃ is the mass matrix, 𝜆̃j the jth eigenvalue, q̃ j the jth
𝐊d = 𝐊 + Δkj = 𝛼j kj (4) eigenvector and m the number available or calculated modes
j=1 j=1
of the damaged structure.
Here, the mass matrix M of the structure is maintained
𝜆di = 𝜆i + Δ𝜆i (5) constant even after the insertion of the structural damage,
which translates into a mathematical character the behav-
𝜙di = 𝜙i + Δ𝜙i (6) ior of a delamination, which is a common and dangerous
damage in composite materials, where there is detachment
Therefore, the stiffness matrix for the damaged structure without loss of material (mass).
can be found through a stiffness matrix composition of the The matrices K and M are assumed to be definite positive
elements multiplied by a reduction factor αj (j = 1,2,…, m) symmetries and thus the eigenvalues are positive and the
associated with each of the m elements. eigenvectors can be taken as K-ortogonal. Similar conditions
apply to 𝐊 ̃ , 𝜆̃j and q̃ j . Considering the mass normalization
of the modes, orthogonality conditions are defined by Eq. 9.
2.1.1 Damaged finite element {
T ̃ ̃ 𝛿ij = 0 for i ≠ j
q̃ i 𝐊̃qj = 𝛿ij 𝜆i with
𝛿ij = 1 for i = j (9)
It has been previously argued that there is an increase in
the reduction of stiffness due to the increase in the size of
Since the stiffness matrix of the damaged structure is
the delamination [57]. Still, several works [58–60] have
given by 𝐊 ̃ = 𝐊 − 𝛿𝐊 , the perturbation in the matrix cor-
studied, addressed and made evident the reduction of stiff-
responding to the element e is given by 𝛿𝐊e = 𝛼𝐊e where
ness due to delamination damages. For this, the modeling
0 ≤ α ≤ 1 is a local stiffness multiplier. The α parameter
of delamination by means of a numerical approach of the
translates the severity of the structural damage.
percentage reduction of local stiffness is valid.
It is well known that the presence of damage in a struc-
2.2 Firefly optimization algorithm
tural system will induce changes in the modal parameters,
which can be used as signals to infer the state of damage.
Based on the behavior of fireflies attracted by light sources,
Natural frequencies and mode shapes, the most common
Xin-She Yang [54] proposed the firefly algorithm that has
parameters, have been widely used for damage identification
its pseudo-code shown in Fig. 1.
[26]. A promising approach is to determine the location and
In this algorithm, there are two central questions: the
severity (extent) of structural damage by solving an inverse
variation of the brightness intensity and the attractiveness
optimization problem. This solution can be only achieved
of each individual. In a simple way, one can assume that
by global metaheuristics optimization algorithms due to the
the attractiveness of a firefly is determined by its bright-
multimodal characteristics. In this formulation the objective
ness, which is directly related to the value of the objective
function is defined in terms of modal data, i.e., modal shift
function. As in the real case, it will be considered that the
between the undamaged numerical model and real damaged
structure.
According to Dos Santos et al. [61] and Gomes et al. [62]
for a discrete, undamaged structure, the eigenvalue equation
can be written as shown in Eq. 7.
𝐊qi = 𝜆i 𝐌qi for i = 1, … , n (7)
where K and M are the stiffness and mass matrices, respec-
tively. λi the ith eigenvalue, ­qi the ith eigenvector, and n is
the number of modes evaluated or available for the structure
in question.
If the structure is then subjected to some kind of dam-
age, its stiffness is then altered, so Eq. 5 can be rewritten as
shown in Eq. 8.
̃ qj = 𝜆̃j 𝐌̃qj
𝐊̃ for m = 1, … , n (8)

Fig. 1  Pseudo code of the firefly algorithm (FA) [24]

13
576 Evolutionary Intelligence (2020) 13:571–591

medium absorbs the intensity of the light emitted by the

fireflies, based on the distance between each individual. It
is called the light absorption coefficient. The expression for
intensity I is given by Eq. 10.

(10)
2
I(r) = I0 e−𝜆r

where I0 is the original intensity of the light and r is the

distance between the fireflies.
The attractiveness β, which is proportional to the intensity
of light and can be defined and written as shown in Eq. 11.

(11)
2
𝛽(r) = 𝛽0 e−𝜆r

where β0 is the attractiveness at r = 0.

The distance between two fireflies i and j in xi and xj,
respectively, is the Cartesian distance given by Eq. 12.
√
√ d
√
‖ ‖ √∑ ( )2
(12)
rij = ‖xi − xj ‖ = xi,k − xj,k
‖ ‖
k=1

where xi,k is the k-th component of the space coordinate xi

of the ith firefly. The motion of a firefly i is determined by
Fig. 2  Cross section of the fuselage of the E190 aircraft (adapted
a more attractive (bright) firefly j and is defined by Eq. 13.
from [73, 74])
2( ) ( )
1
xi = xi + 𝛽0 e−𝛾rij xj − xi + 𝛼 𝐫𝐚𝐧𝐝 − (13)
2 Regarding the pre-processing by FEM, the structure was
where the second term represents attractiveness while the analyzed by means of free vibration (no boundary condi-
third term determines randomness, α being a parameter of tions). The fuselage was set at both ends and a shell element
randomness. rand is a random number generator evenly dis- was applied in a mapped manner. The mesh in question has 8
tributed in [0,1]. nodes and 6 degrees of freedom per node (three translations
In addition, some studies indicate that the FA is particu- and three rotations).
larly suited for parallel implementation and may outperform The fuselage is one of the main structural assemblies of
existing algorithms, such as PSO, GA, SA, and differential an aircraft. It accommodates passengers, crew and loads
evolution, in terms of efficiency and success rates [54, 63]. as well as being the link between other aircraft structures
This fact qualifies and justify the choice of FA as the global such as wing and horizontal and vertical stabilizers. Due
optimizer due to the complex inverse optimization problem to flight altitude, it is necessary to pressurize the fuselage
in aeronautical structure. for the well-being of passengers and crew. At each flight of
the aircraft, a pressurizing and depressurizing cycle will be
completed and large pressure differences provided by these
events require special dedication to the verification of fatigue
3 Numerical modelling phenomena in this structure. Thus, constant structural moni-
toring of the fuselage is required and SHM techniques can
3.1 Direct problem modelling be of great help.
For the computational cost reduction, the fuselage rein-
Initially, computational modeling of the geometry of an forcements were not modeled, only their coating being
aeronautical structure was carried out. In order to work, a evaluated. In preparation for the application of the FEA
fuselage coating section was modeled based on the Embraer shell element was chosen to work the geometry. The struc-
E190 aircraft. The geometry modeling, mesh creation ture was modeled on six layers of carbon fiber, specified as
and finite element simulation and analysis. The geometry [0/± 45/90]S considering the thickness t = 1 mm per layer. It
adopted in this study is shown in Fig. 2. Only a section part is emphasized here that the focus is purely on the geometry
of the fuselage in question with a total length L = 3 m was of an E-Jet model. The data of the materials used here give
considered. rise to a new fuselage (composite material) which does not
correspond to the real-world case, since the actual fuselage

13
Evolutionary Intelligence (2020) 13:571–591 577

Table 1  Mechanical properties Property Value Fig. 3c. Note that the region near the stress concentrating
applied the FE preprocessing windows receives special care, with the mesh a little more
Density ρ (kg/m3) 1400.00 refined.
Ex (GPa) 125.00 Once the material and geometry are defined, it is possible
Ey (GPa) 8.50 to perform the simulation and obtain the vibration response
Ez (GPa) 8.50 of the global system. The evaluated structure has 2064 ele-
νxy 0.25 ments. The structure is considered as clamped–clamped
νyz 0.31 in term of boundary condition (real case of the aircraft
νxz 0.02 operation).
Gxy (GPa) 3.10 To simulate the damaged structure, it was assumed that
Gyz (GPa) 3.10 damage would weaken the material in the vicinity of its
Gxz (GPa) 3.10 location. Thus, to insert a damage in the structure, a previ-
ously chosen element had its stiffness affected by a factor
α, as commented in Sect. 2.1.1. For this study, the induced
damage is chosen as the element number Ne = 1732 and its
severity factor α = 0.15.
Note that the chosen element corresponds to the location
of the damage and the factor α is inversely proportional to its
stiffness reduction, i.e., β = (1–α) intensity. Comparing with
a crack, the larger its size, the greater the damage intensity
and more the stiffness of the next region is affected. It is
necessary to choose these parameters in advance to simulate
the damaged structure, but the algorithm of resolution of
the inverse problem is not aware of this information, on the
contrary, it wishes to find them.
The FEA will provide several types of responses after
the modal simulation, in each node generated in the mesh.
In an experimental test, getting this kind of information
would only be possible through sensors. Piezoelectric sen-
sors can be used to obtain the displacement of the structure
at the point where it is coupled, so it was chosen to use dis-
placement information in only a few nodes, which simulate
the existence of sensors in the structure. For this work, 12
nodes were chosen to obtain the displacement in only one
direction—z direction—which could be performed in a real
structure through unidirectional accelerometers. Then the
study is subdivided into two parts; (1) considering sensors
uniformly distributed throughout the structure and (2) opti-
mized sensors.
It is important to note that there is a difference in the
detection, location and identification of damages. In this
study, the damage is induced to the structure so that the abil-
ity to locate it in an already known position can be verified.
Therefore, the focus of this study is restricted to the identifi-
cation (location) of the damage, starting from the hypothesis
Fig. 3  Meshed E190 fuselage section modelled by FEM in a front that the damage is already present in the structure.
view, b side view and c isometric view
This study is dedicated to the identification of structural
damages in laminated composite structures, especially
of the E190 is made of isotropic material. The properties of regarding the identification of stiffness reduction or delami-
the laminate in question employed in this study are shown nation. The damage was modeled as a local loss of structural
in Table 1. stiffness, since this is the main failure modes of compos-
The modeled fuselage section is shown in Fig. 3a, b. The ite materials. Simulated damage as a reduction of stiffness
geometry corresponding to the fuselage part is shown in faithfully models the mechanically structural behavior of

13
578 Evolutionary Intelligence (2020) 13:571–591

delamination, where there is in fact a very strong relation- Maximizing the determinant of the information matrix will
ship between the size of delamination and the reduction of maximize a combination of the spatial independence of the
stiffness [64]. target modal partitions and their signal strength in the sen-
Equally important, it is true that there is an increase in sor data [21, 67, 68]. It has been shown that the FIM can be
the reduction of stiffness due to the increase in the size of decomposed into the contributions of each candidate sensor
the delamination. The delamination, is an important failure location in the form shown in Eq. 15.
mode in composite materials that may not be visible on the
∑
nc
∑
nc
structural surface, but is capable of affecting strength and Q= 𝜙Tsi 𝜙si = Qi (15)
stiffness [65]. i=1 i=1
The damage can be translated as a geometric variation
(e.g.: cracks) or in the mechanical properties (e.g.: Young’s being φsi the ith line of the mode partition array associated
modulus). In this context, the damage here models were due with the ith candidate sensor location, nc is the number of
to a variation in the local property of stiffness. The damage candidate sensors.
then mathematically can be modeled as a location (finite Then, the sensors must be placed so as to provide the best
element number damaged Ne) and the severity/intensity of estimate of the target modal response. The maximization of
the damage (parameter α) such as αKNe. When α is equal to the determinant of the information matrix is chosen as the
1 there is no damage present in the structure. On the other criterion of the positioning of the sensor, since it results in the
hand, α equal to zero the entire region obtained Ne has maxi- maximization of the signal intensity and the independence of
mum damage. Any value between zero and one exhibits a the main directions [69].
severity present. The objective of this part of the work is to apply the evolu-
tionary method (FA) of optimization of positioning of sensors
for the precise modal identification in mechanical structures.
3.2 Inverse problem modelling A discrete-type optimization problem using FA is formulated
by defining the positions of the sensor according to the criteria
An inverse problem is a method that seeks to obtain a physi- quoted in the previous paragraphs.
cal data from observed measurements by mathematical Equally important, finding the optimal sensor locations
framework. It’s a very useful and researched area due the under a certain evaluation criterion is a complicated nonlinear
fact that data can be measured indirectly. In this study, a optimization problem, traditional optimization methods often
total of 12 sensors is used in the evaluated fuselage section encounter many insurmountable difficulties in solving this
distributed on its area. These sensors collect vibration data problem Intelligence optimization algorithms such as the GA
from the free vibration condition in the normal area condi- and the FA provide powerful approaches to overcome these
tion (uniaxial). obstacles [52]. Furthermore, a discrete optimization problem
The quality of the information obtained depends on the in implemented in the traditional FA.
point of acquisition. For this, knowing where to attach sen- The mathematical optimization formulation is shown in
sors is crucial for efficient damage identification. Eq. 16 where the best sensor position vector X = {x1, ­x2, …,
­x12} is obtained by the maximization of J­ FIM. In order to avoid
3.2.1 Sensor placement optimization sensors, place in the same location, a constraint is added to
spread the positions.
As shown by Kammer [66] the array of sensors can be given ( 12 )
in the form of a estimation problem with a corresponding ∑
maximize JFIM (𝐗) = det 𝜙Tsi 𝜙si
Fisher information matrix (FIM) given by Eq. 14.
i=1
(16)
‖ ‖
Q = 𝜙Ts W𝜙s (14) subject to: g(𝐗) ∶ ‖pos(xi ) − pos(xj )‖ ≤ d
‖ ‖
1 ≤ xi ≤ 6592
where W is a weighting matrix. The modal response is
estimated based on the data measured by the sensors. The The constraint g(X) is included to induce a minimal dis-
maximization of Q results in the minimization of the cor- tance d = 10 cm between all the sensors. The design space
responding error covariance matrix, which results in the best is defined by the total number of nodes, i.e., 6592. Figure 4
estimate. The sensors must be placed in such a way that Q is display the design space for the SPO problem.
maximized in an appropriate matrix norm.
The maximization of the FIM determinant is a commonly
used criterion for the estimation of optimal parameters.

13
Evolutionary Intelligence (2020) 13:571–591 579

Table 2  FA biological operators

FA parameter Value

Population size N = 40
Light absorption λ = 1.0
Attractiveness β0 = 2.0
Step size scaling factor α = 0.2
Iterations (stopping criterion) it = 100

is different from the intensity of the damage previously

quoted [24]. For this study, biological operators/param-
eters are setup as shown in Table 2 where the stop criterion
is defined as the maximum number of iterations it = 100.
Defining the displacements obtained in the simulations
without damage of “base value” and those obtained in
the simulations with damage generated by the individu-
Fig. 4  Configuration of possible sensor positions—design space als of the algorithm of “calculated value”, it was possible
to establish the function to be minimized. The algorithm
search domain for variable 1, damage element, ranges
3.2.2 Damage identification from 1 to 2064. Since the element number does not accept
decimal digits, a procedure has been added to the code
With the finite geometry and damage modeling process, the for the algorithm to always provide whole numbers to the
inverse problem is solved and the optimization techniques simulation software. For variable 2, harm intensity, there
are used. The objective function to be minimized requests is no need to use integer programming and its value var-
information from the simulation of the structure without ies from 0 to 1. For 5 mode shapes and 12 sensors, the
damage and from another simulation with a damage gener- optimization problem was determined.
ated in the element and with intensity defined by the opti- Since the desired structural response is obtained, i.e.
mization algorithm. modal parameters in this study, an index to assess dam-
The number of design variables is equal to 2, i.e., age conditions must be constructed. In this case, an index
X = {Ne, α}. The algorithm generates an initial popula- will be evaluated through an objective function that will
tion, where each individual represents a value of each of the later be minimized by means of an optimization algorithm.
variables and, through the FA-FEA coupling, the FEA is Equation 8 shows the objective function or damage index
executed and performs a numerical simulation for each indi- (DI). The function in question is the root mean square
vidual generated. From the value of the function obtained error (RMSE) of the data evaluated by the optimizer and
by each individual, a new population is generated and the the actual response (structure with induced damage in the
process starts again. scenarios discussed).
Above all, many methodologies can be employed for non- The algorithm will search to minimize the function in
linear system modeling. Each method has its own advan- question. The objective function has a minimum when
tages or drawbacks. The need to determine both the structure J(X) = 0 and this point occurs when 𝜑basej
= 𝜑calculated
j
. In
and the parameters of the engineering systems makes the other words, the function will be minimal when X = X and *

modeling of these systems a difficult task [70]. The existing at this point the damage is found accurately.
damage identification methods can be divided into model- The number of search points by the minimum of the
based and model-free [71]. Model-based methods presup- function, that is, the population size, influences the opti-
pose an accurate finite element model of the target structure, mization result. The more fireflies were arranged in the
following that one obvious advantage of this approach is search, the greater the chances of the global minimum
that the damage detected and identified has a direct physical being reached. However, more simulations are required
interpretation. and computational cost increases. To verify the effect of
FA requires some parameters, as described in Sect. 2.2, the number of individuals on the final result, optimiza-
and were chosen according to the recommendations of tions were performed considering the population size of
the author of the algorithm and are shown in Table 2. It 40 individuals (20 times the size of the design vector).
is important to remember that the parameter β 0 shown In all cases, 100 iterations were adopted as a stopping
below is necessary for the execution of the algorithm and

13
580 Evolutionary Intelligence (2020) 13:571–591

criterion. Since the algorithm is metaheuristic and depends where X is the design vector X = {x1, ­x2} = {Ne, α} com-
on random initial choices, the result of one optimization posed by the element number and damage severity. The
may be different from another, being able to find the global subscript n = 5 is the total of evaluated mode shapes and
minimum or not. Therefore, for each population size, 15 j indicates the sensor number. φbase is the modal data from
optimizations were performed. With this information, it the real damaged structure for s sensors and φcalculated is the
is possible to verify the influence of the number of search modal data obtained from FA corresponding the damage
individuals and to determine if the proposed method com- vector X. In addition, Fig. 5 shows the process flow diagram
bined with the FA is efficient in the task of identifying highlighting the iterative FA process.
damages. As can be seen from the Fig. 6, a priori are collected
� modal data of the actual damaged structure (mode shapes
⎛�� � �2 ⎞
� ⎜� 1 �
5 12 𝜙calculated (𝐗) ⎟ at specific positions). This data is constant throughout the
⎜� n damage identification process. Then, this modal data set will
s,j
minimize Jk (𝐗) = 1− ⎟
n=1 ⎜ s=1 𝜙base
s,j
(𝐗) ⎟ feed the optimization algorithm, where an objective func-
⎝ ⎠
tion is formulated and then minimized. The algorithm will
subject to: 1 ≤ Ne ≤ 2064 create an initial population that has the damage informa-
0≤𝛼≤1 tion (position and severity). This vector with damage data,
(17)
known as design vector, will be the input to obtain a new
modal response by FEA. From this it is possible to evaluate

Fig. 5  Flowchart of the finite

element update process by
firefly algorithm

13
Evolutionary Intelligence (2020) 13:571–591 581

Fig. 6  Flowchart of the dataset

preparation for the damage
identification process consider-
ing objective function minimi-
zation

Table 3  Natural frequencies [Hz] of the undamaged and damaged The presence of a single structural damage in the fuselage
fuselage model considering Ne = 1732 and α = 0.15 was considered in this study. The structure was discretized
Pristine Damaged Frequency into 2064 finite elements (Fig. 3c) with known fixed number-
variation ing. This numbering reflects the position of the damage in
the three-dimensional physical space.
Mode 1 63.865 63.863 0.002
Table 3 shows the influence of a small structural damage
Mode 2 65.108 65.103 0.005
on the modal response of the structure. In fact, the damage
Mode 3 65.743 65.597 0.146
modifies the structural matrices which promote a variation in
Mode 4 66.181 66.143 0.380
the dynamic responses of the system. By making intelligent
Mode 5 70.750 70.675 0.760
use of modal responses, a robust inverse method that is able
to accurately identify the location and intensity of structural
damages can be formulated.
the objective function J. In a utopic sense, it is expected to In addition, natural frequencies are excellent damage
obtain J = 0, where at this point the real mode shapes will be detection metrics (whether a damage exists or not). How-
equal to those obtained by the algorithm. This point is only ever, they are not sufficiently capable of promoting the iden-
possible when the damage identified is equal to the actual tification of damages (locating). For this, mode shapes are
damage. The efficiency of this process will be compared employed, because they are more local than global metrics.
using the Φbase matrix from optimal locations by SPO. Figures 7 and 8 show the first 5 mode shapes for the
intact and damaged structure. It can be seen that the mode
shapes of the damaged structure (Fig. 8) does not reveal to
4 Results the naked eye the position of the inserted damage. This is a
complex task, however, using an appropriate inverse method,
4.1 Damage effect the damage can be revealed (identified).
As a consequence, the insertion of a structural damage
In this study, the damage was adopted as a local reduction modifies the dynamic response and from this order of varia-
of stiffness. The damage is then considered by the multipli- tion, it is possible to treat it by means of intelligent process-
cation of the Ne element by a loss of stiffness α. The fuse- ing until the location of the expected damage is obtained.
lage section is then composed of 2064 elements. A damage The purpose of a damage identification system is the
was induced at the element number N ­ e = 1732 with damage solution of an inverse problem, that is, the identification
severity α = 0.15 (85% of stiffness reduction).

13
582 Evolutionary Intelligence (2020) 13:571–591

Fig. 7  First five mode shapes

for the undamaged E-190 fuse-
lage model

of a system that describes the relationship between an 4.2 Sensor placement optimization results
unknown input and a known output. This means that the
purpose of damage identification is to describe an existing In the development of the inverse optimization method
structural model, based on data obtained experimentally or two cases were considered: i) evaluation of the response in
simulated. Given the above, it is necessary to investigate if non-optimized sensors and ii) evaluation of the response in
the insertion of a localized damage, the response evaluated optimized sensors. The quality of a method of identifying
in this study are really affected significantly. damage depends heavily on the quality of the response eval-
uated. In this case, the evaluated response is mode shape. It

13
Evolutionary Intelligence (2020) 13:571–591 583

Fig. 8  First five mode shapes

for the damaged E-190 fuselage
model

is logical then not to evaluate nodal points (null amplitude of 12 sensors was defined a priori and this value was fixed.
of vibration). This seems to be trivial, however, if and only Figure 9a shows the non-optimized configuration, i.e., the 12
if a single mode of vibration is considered. By adding sev- sensors placed uniformly distributed along the entire surface
eral modes of vibration, an onerous task known as sensor of the fuselage structure. In this study, no sensitivity analysis
placement optimization (SPO) should be addressed. A total was performed. However, in view of the applicability in an

13
584 Evolutionary Intelligence (2020) 13:571–591

(a) (b)

Fig. 9  Fuselage structure with uniform distributed (a) and optimized (b) sensors

aircraft structural where weight is a limiting factor. In the result is much superior when considering optimized sensors.
analyzes performed, 12 sensors were chosen by the “rule of The computational cost for both cases was exactly the same
the thumb” of previous simulations. where no parameter was modified, only the position of the
Figure 9b shows directly the result of the optimization sensors. This fact led to a significant improvement in the
process defined by Eq. 16 by the minimization of J­ FIM. The damage identification result.
sensors are then selected at optimal points so as to obtain Overall damage was assessed when compared to the
the maximum determinant by the Fisher criterion. There is whole structure dimension. The vibration mode alone
also a spacing defined by the established mining process. is not capable of revealing this small induced damage,
Once these optimized sensors are obtained, the reverse opti- however, by formulating an inverse problem by means of
mization process can be performed by aggregating optimal suitable objective functions, real damage can be identified
sensor characteristics placed at these key points. In addition, with some degree of confidence. The graphical process
Fig. 10 shows the convergence of the sensors throughout the results are shown in Fig. 12 for the two cases being evalu-
generations and Fig. 11 shows the fast convergence consid- ated. As it can be seen, the damage was properly located
ering FIM. considering the proposed formulation by Fisher informa-
tion matrix. In addition, the computational cost was sig-
4.3 Damage identification results nificantly reduced.
Moreover, there is a challenge to develop an accurate
In this section the overall results of the inverse method model for complex structure and it can be intricate as well
of damage identification will be presented. After knowing to obtain and update the parameters defining the structure
the influence of the damage parameters on the structural if the modelling is incorrect. In the same way, physics-
response, one can construct a correct objective function. based (i.e., model-based) approaches make use of the sys-
The formulation of the optimization problem is shown in tem models to estimate relevant metrics, in this case the
Eq. 17. damage location and its severity. The main advantage of
Since the algorithm FA is a meta-heuristic, several runs physics-based approaches is that the physical knowledge
must be considered due to the random nature of the method. of the system is incorporated into the SHM process, which
In this case, at least 10 independent runs were considered. is especially useful for predicting system responses to new
Also, due to the fact that meta-heuristic algorithm does not conditions and system configurations such as the damage
have proven mathematical convergence, although in practice, status. Nevertheless, physics-based approaches are typi-
yes. cally more computationally intensive than data-based tech-
Table 4 shows the summarized results of the methodol- niques [72].
ogy used to solve the inverse optimization problem. The Besides, visual inspections and traditional non-destruc-
damage was induced with known parameters (location and tive inspections such as ultrasound, acoustic emission are
severity) and the ability of the algorithm to identify this real regarded as the main techniques in the aeronautical field.
and unknown damage was evaluated. It is evident that the Nevertheless, these techniques have many shortcomings due

13
Evolutionary Intelligence (2020) 13:571–591 585

Fig. 10  Convergence results for sensors s­ 1 to ­s12

13
586 Evolutionary Intelligence (2020) 13:571–591

Fig. 11  Convergence results for

the objective function J­ FIM 1.8 Best penalty value
Mean penalty value
1.7

1.6

1.5

Penalty value
1.4

1.3

1.2

1.1

1
0 10 20 30 40 50 60 70 80 90 100
Generation

Table 4  Overall results Objective Uniform sensors Optimized sensors

of damage identification
considering the objective Ne α J Ne α J
function Jφ
1732 0.1500 0 1732 0.1500 0

Run 1 1620 0.2763 1.6342 1711 0.1244 0.1361

Run 2 1731 0.0057 0.6066 1732 0.1536 0.1457
Run 3 1711 0.1242 0.1365 1711 0.1241 0.1358
Run 4 1731 0.0064 0.6071 1731 0.0074 0.5516
Run 5 1732 0.1499 0.0044 1732 0.1489 0.0450
Run 6 1711 0.1245 0.1326 1732 0.1496 0.0163
Run 7 1731 0.0059 0.6056 1711 0.1241 0.1362
Run 8 1711 0.1250 0.1332 1732 0.1496 0.0163
Run 9 1711 0.1193 0.3310 1711 0.1223 0.1762
Run 10 1732 0.1492 0.0353 1732 0.1537 0.1457
Mode/average 1711 0.1086 0.4227 1732 0.1258 0.1505
St. deviation – 0.0800 0.4639 – 0.0415 0.1447

to the complexity of large-scale structures and in modern the severity parameter, that is, the intensity of the damage is
design structures. a variable of extreme difficulty to be identified.
The convergence results are shown in Figs. 13 and 14. For In addition, Fig. 15 shows the convergence result of the
both convergence results of decision variables (Figs. 13 and objective function over the 100 iterations evaluated. Again, a
14), it can be seen that the use of optimized sensors led to a significant improvement in results can be clearly seen when
more correct identification of damage-induced parameters using sensors located in optimal positions. The decay of
(objective Ne = 1732 and α = 0.15). In general, the dam- Fig. 15a is much more interesting than that of Fig. 15b.
age severity parameter was not improved in relation to the It could be mentioned some disadvantages that tradi-
optimized sensors. On the other hand, the damage location tional NDI techniques may not be practicable if the struc-
received a significant improvement. In fact, it is known that ture has restricted accessibility. Therefore, the operator
must know a damaged region a priori to be able to do

13
Evolutionary Intelligence (2020) 13:571–591 587

Fig. 12  Graphical results for the

damage identification method
considering non-optimized sen-
sors (a) and optimized sensors
(b). (legend: red dot—induced
damaged, blue dot—identified
damage) (color figure online)

the inspection, otherwise much time will be spent on the The superiority of the proposed technique is that it is
assessment. In this sense, many efforts have been made in developed based on a combination of only the first five
developing advanced and optimized damage identification modes considering optimal sensing points obtained by
techniques that handle measurement data to find structural Fisher information matrix. This feature turns the proposed
changes in complex structures, especially in the aerospace method into an efficient tool for damage identification
field to avoid time standing for aircrafts. techniques and it is suitable in practical cases due to the
limited modal acquisition points.
By taking the modal experiments for the Embraer E190
5 Conclusion fuselage structure the experimental results verify that the
optimized positioning identification can eliminate some
An efficient damage identification method consisting of experimental errors. Compared with the uniform measur-
an optimal vibration data from optimized sensor location ing points on the structure, the optimal sensor placement
and a firefly optimization algorithm has been presented in
this research for damage site identification in aeronautical
fuselage structure.

13
588 Evolutionary Intelligence (2020) 13:571–591

(a) (b)

Fig. 13  Convergence results for the decision variable N

­ e considering non-optimized sensors (a) and optimal sensor locations (b)

(a) (b)

Fig. 14  Convergence results for the decision variable α considering non-optimized sensors (a) and optimal sensor locations (b)

was defined more reliable, which greatly improves the effi- that some aspects of the method could not be covered in the
ciency of damage identification, modelled as a local stiffness present study. As ready mentioned, the simulation is free of
reduction. noise. At the same time, simulated noise cannot properly
Regarding the numerical aspect, the optimized algorithm represent real conditions. Therefore, modelling errors related
was able to perform very well despite the great difficulty to acquisition and instrumentation system should be further
arising from the large search space with more than 2000 investigated. Additionally, in the present study, the structural
elements combined with the severity value. This leads to an damage model is generic (stiffness reduction or delamination
extremely complex functional (many local optima points). in CFRP), whereas in reality, the damage can have many
Developing a method, or improving an algorithm for large- specific shapes (cracks, voids, etc.).
scale structures, is crucial for reliably assessing regions sub- Consequently, further investigation of the use of complex
ject to structural damage. damage models (arbitrary shape) is necessary to examine
Despite the efficiency of the proposed method for struc- the general applicability of the proposed method. Equally
tural damage assessment in the numerical simulations con- important, further analysis of sensor sensitivity should be
sidering optimal sensor location, it should be pointed out performed to optimize both position and quantity.

13
Evolutionary Intelligence (2020) 13:571–591 589

(a) (b)

Fig. 15  Convergence results for the objective function J considering non-optimized sensors (a) and optimal sensor locations (b)

Acknowledgements The authors are grateful to the Brazilian Fund- 12. Deraemaeker A, Worden K (eds) (2012) New trends in vibration
ing Institutions CAPES, CNPq (Grant Number 431219/2018-4) and based structural health monitoring, vol 520. Springer, Berlin
FAPEMIG (Grant Number APQ-00385-18) for the financial supports. 13. Farrar CR, Doebling SW, Nix DA (2001) Vibration–based struc-
tural damage identification. Philos Trans R Soc Lond Ser A
Math Phys Eng Sci 359(1778):131–149
14. Gomes GF, da Cunha SS, Alexandrino PDSL, de Sousa BS,
References Ancelotti AC (2018) Sensor placement optimization applied to
laminated composite plates under vibration. Struct Multidiscipl
1. Farrar CR, Worden K (2006) An introduction to structural Optim 58(5):2099–2118
health monitoring. Philos Trans R Soc A Math Phys Eng Sci 15. Gomes GF, Mendez YAD, Alexandrino PDSL, da Cunha SS,
365(1851):303–315 Ancelotti AC (2018) A review of vibration based inverse meth-
2. Abdo MAB (2014) Structural health monitoring: history, appli- ods for damage detection and identification in mechanical struc-
cations and future, 1st edn. Open Science Publishers, Nova York tures using optimization algorithms and ANN. Arch Comput
3. Balageas D, Fritzen CP, Güemes A (eds) (2006) Structural Methods Eng 26(4):883–897. https​: //doi.org/10.1007/s1183​
health monitoring, vol 493. ISTE, London 1-018-9273-4
4. Sohn H et al (2003) A review of structural health monitoring 16. Yoo DG, Kim JH, Geem ZW (2014) Overview of Harmony
literature: 1996–2001. Los Alamos National Laboratory, New Search algorithm and its applications in Civil Engineering. Evol
Mexico Intel 7(1):3–16
5. Hansen LU, Häusler SM, Horst P (2008) Evolutionary mul- 17. Barontini A, Masciotta MG, Ramos LF, Amado-Mendes P,
ticriteria design optimization of integrally stiffened airframe Lourenço PB (2017) An overview on nature-inspired optimiza-
structures. J Aircr 45(6):1881–1889 tion algorithms for structural health monitoring of historical
6. Di Sante R (2015) Fibre optic sensors for structural health buildings. Procedia Eng 199:3320–3325
monitoring of aircraft composite structures: recent advances 18. Tarantola A (2005) Inverse problem theory and methods for
and applications. Sensors 15(8):18666–18713 model parameter estimation, 1st edn. Society for Industrial and
7. Staszewski W, Tomlinson G, Boller C, Tomlinson G (2004) Applied Mathematics, Philadelphia
Health monitoring of aerospace structures. Wiley, Chichester 19. Alkayem NF, Cao M (2018) Damage identification in three-
8. Zhao R, Yan R, Chen Z, Mao K, Wang P, Gao RX (2019) Deep dimensional structures using single-objective evolutionary
learning and its applications to machine health monitoring. algorithms and finite element model updating: evaluation
Mech Syst Signal Process 115:213–237 and comparison. Eng Optim. https​: //doi.org/10.1080/03052​
9. Pierotti MJ (2005) Aircraft maintenance engineering: develop- 15X.2017.14142​06
ing an aircraft maintenance program using reliability centred 20. Begambre O, Laier JE (2009) A hybrid Particle Swarm Optimiza-
maintenance/MSG3 analysis and taking into consideration tion-Simplex algorithm (PSOS) for structural damage identifica-
ETOPs and low utilisation. City University London, Londres tion. Adv Eng Softw 40(9):883–891
10. Kaveh A, Dadras A (2018) Structural damage identification 21. Gomes GF, da Cunha SS, Ancelotti AC (2019) A sunflower opti-
using an enhanced thermal exchange optimization algorithm. mization (SFO) algorithm applied to damage identification on
Eng Optim 50(3):430–451 laminated composite plates. Eng Comput 35(2):619–626
11. Kim N, Kim S, Lee J (2019) Vibration-based damage detection 22. Kang F, Li JJ, Xu Q (2012) Damage detection based on improved
of planar and space trusses using differential evolution algo- particle swarm optimization using vibration data. Appl Soft Com-
rithm. Appl Acoust 148:308–321. https​: //doi.org/10.1016/j. put 12(8):2329–2335
apaco​ust.2018.08.032

13
590 Evolutionary Intelligence (2020) 13:571–591

23. Seyedpoor SM (2012) A two stage method for structural damage 43. Vo-Duy T, Ho-Huu V, Dang-Trung H, Nguyen-Thoi T (2016) A
detection using a modal strain energy based index and particle two-step approach for damage detection in laminated composite
swarm optimization. Int J Non-Linear Mech 47(1):1–8 structures using modal strain energy method and an improved dif-
24. Yang X-S (2014) Nature-inspired optimization algorithms, 1st ferential evolution algorithm. Compos Struct 147:42–53
edn. Elsevier Inc., Londres 44. Cancelli A, Laflamme S, Alipour A, Sritharan S, Ubertini F (2020)
25. Agathos K, Chatzi E, Bordas SP (2018) Multiple crack detec- Vibration-based damage localization and quantification in a preten-
tion in 3D using a stable XFEM and global optimization. Comput sioned concrete girder using stochastic subspace identification and
Mech 62(4):835–852 particle swarm model updating. Struct Health Monit 19(2):587–605.
26. Du DC, Vinh HH, Trung VD, Hong Quyen NT, Trung NT (2018) https​://doi.org/10.1177/14759​21718​82001​5
Efficiency of Jaya algorithm for solving the optimization-based 45. Dinh-Cong D, Vo-Duy T, Ho-Huu V, Nguyen-Thoi T (2019) Dam-
structural damage identification problem based on a hybrid objec- age assessment in plate-like structures using a two-stage method
tive function. Eng Optim 50(8):1233–1251 based on modal strain energy change and Jaya algorithm. Inverse
27. Gomes GF, de Almeida FA, Alexandrino PDSL, da Cunha SS, Probl Sci Eng 27(2):166–189
de Sousa BS, Ancelotti AC (2019) A multiobjective sensor 46. Xu HJ, Liu JK, Lu ZR (2016) Structural damage identification based
placement optimization for SHM systems considering Fisher on cuckoo search algorithm. Adv Struct Eng 19(5):849–859
information matrix and mode shape interpolation. Eng Comput 47. Deqing G, Yuanwei B, Zhou T (2019) Damage identification of
35(2):519–535 frame structure based on wavelet analysis-bat algorithm. J Civ Eng
28. Kaveh A, Dadras A, Montazeran AH (2018) Chaotic enhanced Archit 1(1):1–6
colliding bodies algorithms for size optimization of truss struc- 48. Yang XS, Hossein Gandomi A (2012) Bat algorithm: a novel
tures. Acta Mech 229:2883–2907. https​://doi.org/10.1007/s0070​ approach for global engineering optimization. Eng Comput
7-018-2149-8 29(5):464–483
29. Khatir S, Dekemele K, Loccufier M, Khatir T, Wahab MA (2018) 49. Zare Hosseinzadeh A, Ghodrati Amiri G, Jafarian Abyaneh M,
Crack identification method in beam-like structures using changes Seyed Razzaghi SA, Ghadimi Hamzehkolaei A (2019) Baseline
in experimentally measured frequencies and particle swarm opti- updating method for structural damage identification using modal
mization. Comptes Rendus Mécanique 346(2):110–120 residual force and grey wolf optimization. Eng Optim. https​://doi.
30. Lin JF, Xu YL, Law SS (2018) Structural damage detection-ori- org/10.1080/03052​15X.2019.15934​00
ented multi-type sensor placement with multi-objective optimiza- 50. Azizi M, Ejlali RG, Ghasemi SAM, Talatahari S (2019) Upgraded
tion. J Sound Vib 422:568–589 Whale Optimization Algorithm for fuzzy logic based vibration con-
31. Dinh-Cong D, Dang-Trung H, Nguyen-Thoi T (2018) An efficient trol of nonlinear steel structure. Eng Struct 192:53–70
approach for optimal sensor placement and damage identification 51. Yang X-S (2010) Firefly algorithms for multimodal optimization.
in laminated composite structures. Adv Eng Softw 119:48–59 In: International symposium on stochastic algorithms, pp 169–178.
32. Mishra M, Barman SK, Maity D, Maiti DK (2019) Ant lion opti- https​://doi.org/10.1007/978-3-642-04944​-6_14
misation algorithm for structural damage detection using vibration 52. Zhou GD, Yi TH, Zhang H, Li HN (2015) A comparative study
data. J Civ Struct Health Monit 9(1):117–136 of genetic and firefly algorithms for sensor placement in struc-
33. Razzaghi S, Ali S, Adel Sanjideh B, Zare Hosseinzadeh A (2019) tural health monitoring. Shock Vib 2015:1–10. https​://doi.
Vibration-based updating method for structural health monitoring org/10.1155/2015/51869​2
using Moth-Flame optimization algorithm. Modares Civ Eng J 53. Yang XS (2008) Nature-inspired Metaheuristic algorithm. Luniver
19(3):95–109 Press, Beckington
34. Seyedpoor SM, Ahmadi A, Pahnabi N (2019) Structural damage 54. Yang X-S (2009) Firefly algorithms for multimodal optimization.
detection using time domain responses and an optimization method. Stochastic algorithms: foundations and applications. Springer, Ber-
Inverse Probl Sci Eng 27(5):669–688 lin, Heidelberg, pp 169–178. https​://doi.org/10.1007/978-3-642-
35. Wei Z, Liu J, Lu Z (2018) Structural damage detection using 04944​-6_14
improved particle swarm optimization. Inverse Probl Sci Eng 55. Escobar CM, González-Estrada OA, Acevedo HGS (2017) Damage
26(6):792–810 detection in a unidimensional truss using the firefly optimization
36. Xu M, Wang S, Jiang Y (2019) Iterative two-stage approach for algorithm and finite elements. arXiv preprint arXiv​:1706.04449​
identifying structural damage by combining the modal strain energy 56. Srinivas V, Ramanjaneyulu K, Jeyasehar CA (2010) Multi-stage
decomposition method with the multiobjective particle swarm opti- approach for structural damage identification using modal strain
mization algorithm. Struct Control Health Monit 26(2):e2301 energy and evolutionary optimization techniques. Struct Health
37. Greco A, D’Urso D, Cannizzaro F, Pluchino A (2018) Damage Monit 10(2):219–230. https​://doi.org/10.1177/14759​21710​37329​1
identification on spatial Timoshenko arches by means of genetic 57. Lagace PA (1989) Composite materials: fatigue and fracture, vol 2.
algorithms. Mech Syst Signal Process 105:51–67 ASTM
38. Chou JH, Ghaboussi J (2001) Genetic algorithm in structural dam- 58. Zhang J, Fan J, Herrmann KP (1999) Delaminations induced by
age detection. Comput Struct 79(14):1335–1353 constrained transverse cracking in symmetric composite laminates.
39. Haznedar B, Kalinli A (2018) Training ANFIS structure using Int J Solids Struct 36(6):813–846
simulated annealing algorithm for dynamic systems identification. 59. Takeda N, Ogihara S (1994) Initiation and growth of delamination
Neurocomputing 302:66–74 from the tips of transverse cracks in CFRP cross-ply laminates.
40. He RS, Hwang SF (2006) Damage detection by an adaptive real- Compos Sci Technol 52(3):309–318
parameter simulated annealing genetic algorithm. Comput Struct 60. Kashtalyan M, Soutis C (2000) The effect of delaminations induced
84(31–32):2231–2243 by transverse cracks and splits on stiffness properties of composite
41. Dinh-Cong D, Vo-Duy T, Ho-Huu V, Dang-Trung H, Nguyen-Thoi laminates. Compos Part A: Appl Sci Manufact 31(2):107–119
T (2017) An efficient multi-stage optimization approach for damage 61. Dos Santos JA, Soares CM, Soares CM, Pina HLG (2000) Develop-
detection in plate structures. Adv Eng Softw 112:76–87 ment of a numerical model for the damage identification on com-
42. Lu K, Li YY (2019) A robust locating multi-optima approach for posite plate structures. Compos Struct 48(1–3):59–65
damage identification of plate-like structures. Appl Soft Comput 62. Gomes GF, de Almeida FA, Junqueira DM, da Cunha Jr SS,
75:508–522 Ancelotti AC Jr (2019) Optimized damage identification in CFRP

13
Evolutionary Intelligence (2020) 13:571–591 591

plates by reduced mode shapes and GA-ANN methods. Eng Struct 70. Gandomi AH, Yang XS, Talatahari S, Alavi AH (2013) Metaheuris-
181:111–123 tic algorithms in modeling and optimization. In: Metaheuristic appli-
63. Gandomi AH, Yang X-S, Alavi AH (2011) Mixed variable cations in structures and infrastructures, pp 1–24
structural optimization using firefly algorithm. Comput Struct 71. Neves AC, González I, Leander J, Karoumi R (2017) Structural
89(23–24):2325–2336 health monitoring of bridges: a model-free ANN-based approach
64. Talreja R (1999) Damage Mechanics and Fatigue Life Assessment to damage detection. J Civ Struct Health Monit 7(5):689–702
of Composite Materials. Int J Damage Mech 8(4):339–354. https​:// 72. Kong X, Cai CS, Hu J (2017) The state-of-the-art on framework
doi.org/10.1177/10567​89599​00800​404 of vibration-based structural damage identification for decision
65. Garg AC (1988) Delamination a damage mode in composite struc- making. Appl Sci 7(5):497
tures. Eng Fract Mech 29(5):557–584 73. EMBRAER 190 (2005) Airport Planning Manual. Retrieved June
66. Kammer DC (1991) Sensor placement for on-orbit modal identifica- 26, 2019 from https​://www.embra​ercom​merci​alavi​ation​.com/wp-
tion and correlation of large space structures. J Guid Control Dyn conte​nt/uploa​ds/2017/02/APM_E190.pdf
14(2):251–259 74. EMBRAER E190 [Online image] (2019) Retrieved June 26, 2019
67. Gomes GF, de Almeida FA, da Cunha SS, Ancelotti AC (2018) An from https​://www.embra​ercom​merci​alavi​ation​.com/comme​rcial​
estimate of the location of multiple delaminations on aeronautical -jets/e190/
CFRP plates using modal data inverse problem. Int J Adv Manuf
Technol 99(5–8):1155–1174 Publisher’s Note Springer Nature remains neutral with regard to
68. Kammer DC, Tinker ML (2004) Optimal placement of triaxial jurisdictional claims in published maps and institutional affiliations.
accelerometers for modal vibration tests. Mech Syst Signal Process
18(1):29–41
69. Rao ARM, Lakshmi K, Kumar SK (2015) Detection of delamination
in laminated composites with limited measurements combining PCA
and dynamic qpso. Adv Eng Softw 86:85–106

13