432|Creative Commons Attribution b

https://doi.org/10.3311/PPci.13888

Periodica Polytechnica Civil Engineering, 63(2), pp. 432–445, 2019

Modal Sensitivity Based Sensor Placement for Damage

Identification Under Sparsity Constraint

Bartlomiej Blachowski1*

1
Institute of Fundamental Technological Research,
Polish Academy of Sciences
Pawinskiego 5b, 02-106 Warsaw, Poland
*
Corresponding author, e-mail: [email protected]

Received: 15 February 2019, Accepted: 13 March 2019, Published online: 26 March 2019

Abstract
The present study deals with a comprehensive approach for damage identification of spatial truss structures. The novelty of the
proposed approach consists of a three-level analysis. First, sensitivity of assumed modal characteristics is calculated. Second,
natural frequency sensitivity is used to determine hardly identifiable structural parameters and mode shape sensitivity is applied
to select damage-sensitive locations of sensors. Third, two sparsity constrained optimization algorithms are tested towards
efficient identification of applied damage scenarios. These two algorithms are based on ℓ1-norm minimization and non-negative
least square (NNLS) solution.
Performances of both proposed algorithms have been compared in two realistic case studies: the first one concerned a three-
dimensional truss girder with 61 structural parameters and the second one was devoted to an upper-deck arch bridge composed of
416 steel members.

Keywords
sensor placement, damage identification, ℓ1-norm minimization, sparsity constrained optimization

1 Introduction
Optimal sensor placement (OSP) and structural damage has been investigated by Liu and Chian [4]. Their conclu-
detection have been topics of intensive research for over sion was that even when strains in all elements are mea-
thirty years. The motivation behind that research is the sured it can be impossible to identify structural parame-
fact that many civil structures reach the end of their ser- ters in a complex truss structure using only a single static
vice life and structural health monitoring becomes the loading pattern.
critical aspect in assuring their safety and reliability. It was one of the reasons why researchers oriented their
One of the first papers devoted to optimization of sen- interests towards dynamic excitation based techniques.
sor placement for parameter estimation in structural sys- Among many papers from that period an interesting study
tems subjected to an earthquake excitation has been pub- has been published by Cobb and Liebst [5]. They pointed
lished by Shah and Udwadia [1]. A year later the idea of out a critical aspect of sensor placement and its influence
damage detection from changes in natural frequencies has on the ability to identify damaged elements in three-di-
been introduced by Cawley and Adams [2]. mensional trusses.
After these pioneering works a significant increase of Contrary to the studies by either Cobb and Liebst,
interest in that field of research took place and in particular or Shah and Udwadia, which used modal and response
in the 90s of the last century a number of important papers sensitivity matrices for parameter estimation, an inter-
were published. The Effective Independence (EI) technique esting novel approach has been proposed by Beck and
has been introduced by Kammer [3] for efficient placement Katafygiotis [6]. In their two-part paper they introduced
of accelerometers on large truss structures. Parametric the Bayesian inference concept to parametric estimation
identification of trusses using static strain measurements of multi degree of freedom systems and concluded that

Cite this article as: Blachowski, B. "Modal Sensitivity Based Sensor Placement for Damage Identification Under Sparsity Constraint", Periodica
Polytechnica Civil Engineering, 63(2), pp. 432–445, 2019. https://doi.org/10.3311/PPci.13888
 |433
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

improperly located sensors can result in incorrect esti- Damage identification using three different objective
mation of stiffness parameters. Further exploration of the functions based on residuals obtained from static and
Bayesian approach and information theory and their appli- modal measurements has been proposed by Kourehli [21].
cation to the sensor placement problem has been done by He applied simulated annealing to verify his method on a
Papadimitriou [7]. He used the concept of information four-story steel frame proposed as IASC-ASCE structural
entropy (IE) and proposed a computationally efficient health monitoring benchmark. Bayer et al. [22] proposed
sequential sensor placement technique giving results sim- damage localization using a criterion defined as the prod-
ilar to those yielded by a genetic algorithm. uct of selected mode flexibility and the modal curvature.
In 2009 an excellent overview of existing techniques for Experimental setup consisted of 3-meter long simply sup-
optimal sensor placement was published by Barthorpe and ported beam with rectangular hollow section. Six mode
Worden [8]. Recently, Zhou et al. [9] have proposed a soft- shape have been used to properly localize damage within
ware architecture implementing four approaches for OSP, the beam. An efficient damage index for multiple damage
namely: mode shape difference, Effective Independence, identification using the concepts of frequency response
information entropy and modal energy. Besides sensor function (FRF) and strain energy has been proposed by
placement optimization, an important aspect of any modal Bagherahmadi and Seyedpoor [23].
characteristic based SHM system is the solver used for At the end of the overview of damage identification
parametric estimation. methods one has to mention the problems with the deploy-
Early works from the 90-ties suggested ℓ2-norm min- ment of Structural Health Monitoring systems for dam-
imization, an example being the distinguished book by age identification. An interesting alternative for acceler-
Friswell and Mottershead [10]. However, researchers are ation or strain measurements for evaluation of dynamic
currently reporting the ℓ1-norm minimization technique structural response has been proposed by Beben [24]. He
as a most accurate in structural parameter identification applied an interferometric radar for dynamic testing of
(Hernandez [11]). Additionally, SHM techniques can be various bridges. However, as he concluded the method is
implemented as off-line or on-line algorithms. Among suitable for small-to-medium scale bridges.
the latter it is worth mentioning the works by Ebrahimian As one can conclude from the above literature over-
et al. [12] utilizing the Extended Kalman Filter (EKF) view, there is continuous progress towards efficient SHM
approach or by Grinsberg et al. [13] utilizing sparsity con- systems for damage identification. However, there are still
strained EKF. relatively few comprehensive studies of the design of such
Currently used methods for damage identification can a SHM system taking into account all aspects of the issue,
be classified as either time or modal domain based ones. namely: sensor placement, sensitivity to structural param-
Representative examples belonging to the first group are eter changes caused by damage and finally the frequency
the work by An et al. [14], where curvature between two range of excitation. This work is a preliminary attempt to
neighboring measurement points has been used to iden- fill this gap.
tify damage in a simple beam structure, by Pnevmatikos The present study proposes a comprehensive method-
et al. [15], where a wavelet based technique has been ology for damage quantification in large truss structures.
used to localize damage in a planar frame structure, or The methodology is composed of modal sensitivity based
by Blachowski et al. [16], in which changes in axial strain sensor placement together with sparsity constrained dam-
accelerations have been applied to damage localization in age identification. For the purpose of sparsity constrained
a spatial truss structure. Methods from the second group optimization two solvers will be considered. The first one
have been investigated by An et al. [17], who utilized QR is based on the non-negative least squares (NNLS) algo-
decomposition to detect changes in the flexibility matrix rithm and the second one utilizes ℓ1-norm minimization
based on measured modes, or by Blachowski et al. [18] for subject to equality constraints as implemented in the
detection of damage in a bolted lap connection. Damage l1-magic toolbox. Effectiveness of the proposed method-
in flanged connections of tall steel towers was the topic of ology will be verified in two case studies. The first one
the paper by Blachowski and Gutkowski [19]. Recently, a employs a laboratory-scale truss structure with 61 struc-
time domain technique for simultaneous sensor placement tural elements and the second one deals with real-scale
and damage identification in truss structures has been pro- arch bridge used in Japan as a benchmark for verification
posed by Blachowski et al. [20]. of seismic design quality.
434|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

2 Proposed methodology for damage-sensitive Table 1 Sensor placement for sparsity constrained damage
sensor placement and sparsity constrained damage identification

identification Algorithm:

Before detailed description of the proposed methodol- 1. Choose a proper parametrization for the expected type of damage
(Eq. (2))
ogy three important aspects of any Structural Health
2. Calculate frequency and mode shape sensitivity matrices for the
Monitoring system devoted to damage identification will
modes within the excitation range (Eq. (4) and (5))
be briefly presented. They are: a) damage type and loca-
3. Create a rating of damage-sensitive degrees of freedom for
tion, b) number, type and distribution of allowable sensors potential sensor placement (Eq. (6))
and finally c) source of excitation (Fig. 1). Below a short 4. Select uncorrelated locations for optimal sensor placement (Eq. (7))
description of the above aspects is given: 5. Identify damage using sparsity constrained optimization
a) Damage type and location algorithms (Eq. (8))
First of all, when designing a successful SHM system 6. Evaluate the overall damage identification system
one should be aware that there is no universal SHM sys- 7. If the required criteria are not satisfied, return to 1. and depending
tem able to detect all possible damage scenarios. The main on the available resources perform one of the following:
(a) Change the assumed parameterization
difficulty is to properly characterize which of all possible
(b) Modify the source of excitation to increase the number of
scenarios can occur first and then parametrize the struc- modes in the structural response
ture under investigation in such a way that this damage (c) Increase the number of sensors
scenario is included in the given parametrization. otherwise save the optimal sensor placement and finish the algorithm
b) Number, type and distribution of sensors
The second important aspect is related to acquired Initialize
measurement data. It is well known that poorly distrib-
uted sensors are not able to capture crucial characteristics Parametrize the expected
of the structure. Therefore, it is critical to consider this type of damage

aspect before installing a monitoring system on the inves-

Determine modal
tigated structure.
sensitivities
c) Source of excitation
Similarly to the sensor placement aspect, location and Create a rating of
frequency range of excitation can have significant influ- damage-sensitive DoFs
ence on the performance of any SHM system.
Taking into account the above issues the present work Select uncorrelated
proposes a comprehensive algorithm for solving dam- locations for sensor
placement
age identification by means of low-frequency vibrations.
Outline of the proposed methodology is presented in a step- Perform one
Identify damage using
by-step manner as Algorithm (Table 1) or in a graphical optimization under
of the options
of step 7 of
form as flowchart (Fig. 2). sparsity constraints
the algorithm

Damage
location
Are performance No
criteria satisfied?

Yes
Terminate
Structural health
monitoring for Source of Fig. 2 Flowchart of the proposed algorithm
excitation
damage ID
The subsequent sections of this paper are devoted to
these three aspects. The next section deals with a param-
etrization suitable for stiffness degradation in structural
Sensor members of trusses. Section 2.2 will present a damage
placement
sensitive sensor placement methodology. And finally,
Fig. 1 Three critical aspects of vibration-based SHM Section 2.3 will describe a sparsity constrained damage
 |435
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

identification method based on knowledge of a set of nat- where K0 is the nominal stiffness matrix (obtained by cal-
ural frequencies and mode shapes. Wrong excitation will ibration of the initial FEM using experimental data from
not allow proper identification of the assumed modes and a healthy structure), K j is the stiffness matrix correspond-
can cause the whole SHM system to not work correctly. ing to modification of the j-th stiffness parameter. In this
study such a parameter will be related to the stiffness of
2.1 Damage parametrization in modal sensitivity based an individual truss member, but generally it can describe
methods stiffnesses of a set of truss elements constituting the j-th
Measurement of low frequency vibrations of an engineer- substructure.
ing structure is the most frequently utilized technology in Having parametrized the structure under investigation
designing a health monitoring system. This is because such we can express the modal parameter of the damaged struc-
vibrations provide a global characteristic of the structure’s ture in terms of the healthy one using Taylor expansion
condition. Two competitive approaches for damage iden-
∂b
tification are distinguished here: the first one is performed bd = b0 − ∆θ j + O ( ∆θ j2 ) . (3)
∂θ j
in time-domain and is based on dynamic response modifi- b =b0

cation and the second one in modal-domain uses changes In Eq. (3) bd denotes the modal parameter (a natural fre-
in natural frequencies and mode shapes. Although, the lat- quency or a single component of the mode shape vector)
ter approach is criticized for its lower sensitivity as com- of the damaged structure, b 0 is the corresponding modal
pared to the former, its unquestionable advantage is an eas- parameter of the healthy structure, and ∂b/∂θj denotes the
ier formulation for random excitation, which is frequently derivative of this modal parameter with respect to the j-th
ambient and unknown. Therefore, in the case of large structural parameter. In the case of many modal and struc-
scale civil structures modal parameters based approaches tural parameters the above equation takes vector form and
are very often the only option. corresponding derivative is a whole matrix called sensitiv-
The derivation of the proposed methodology for dam- ity. A detailed derivation of this matrix will be the topic of
age identification is initiated by recalling the fundamen- the subsequent section.
tal equation involved in modal parameter extraction. The
undamped eigenvalue problem of a multi-degree of free- 2.2 Damage-sensitive sensor placement methodology
dom structure is described by the well-known set of alge- The derivative of modal Eq. (1) with respect to the struc-
braic equations tural parameter takes the following form

( K − ω M )φ
2
k k = 0, k = 1, 2,…nm , (1) ∂
( K − ωk2M )φk + ( K − ωk2M ) ∂∂θ φk = 0 . (4)
∂θ j j
where K and M are the stiffness and mass matrix, respec-
tively, ωk and ϕk denote the k-th natural frequency and Multiplying on the left by ϕkT and using the undamped
mode shape, respectively, finally nm is the number of modes eigenvalue solution the derivative can be written as
under investigation.
∂
The above equation serves as a tool to predict modal φkT
∂θ j
( K − ωk2M )φk = 0 , (5)
pairs (natural frequency and corresponding mode) of the
structure, for which structural parameters might differ from or more explicitly
the nominal ones. For that reason, at this stage it is neces-
 ∂K ∂ωk ∂M 
sary to introduce the parametrization describing structural φkT  − 2ωk M − ωk2  φk = 0 . (6)
 ∂θ ∂θ j ∂θ j 
modifications caused by damage. Additionally, it is reason-  j
able to assume that in a deteriorating structure only stiffness Assuming that there is no change in the mass matrix
parameters will change and mass properties, identified at and utilizing the orthogonal property of the eigenvalue
an early stage, will not change during the service life of the problem, ϕkTMϕl = 0 for k ≠ l, with mass-normalized eigen-
structure. Taking the above assumptions into account the vectors ϕkTMϕk the above equation can be further simpli-
parametrization used in this study takes the following form fied to
np

K = K 0 − ∑θ j K j , (2) ∂ωk ∂K
= φkT φk (7)
j =1 ∂θ j ∂θ j
436|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

Determining the derivative of the k-th natural fre- parameter

quency with respect to the j-th structural parameter allows (3)

(3) s14 mode
one to create the whole matrix, which will be referred to (3)
(3)
s12
s13
s11 (3)
here as the sensitivity matrix ω S. (3) s
(3)
(3) s
s23 24
s21 22 (3)
Having determined the sensitivity of a natural fre- (3)
(3) s
s33 34 (2)
(2)
(3) s (2)
s14
quency one can proceed to obtain the derivative of the DoF s31 32 (3) s
(3) (2)
s11
s12
s13
(2)
(3) s43 44 (2) s24
mode shape with respect to structural parameters. For that (3) s
s41 42 (3) (2) s
(2) s23
(3) s s21 22 (2)
purpose we return to the derivative of the eigenvalue prob- (3) s
(3) s53 54 (2) s34 (1)
(2) s33 (1)
s51 52 (2) s (1) s13
s14
lem (Eq. (1)) and omit the left multiplication step. Next, s31 32 (2) s
(2) (1) s
(2) s43 44 s11 12 (1)
(1)
s24
(2) s
using the Nelson technique [25] we can calculate the mode s41 42 (2) s
(2) (1) s
(1) s23
(2) s53 54 s21 22 (1)
(1)
shape sensitivity as follows: (2) s
s51 52 (1) s
(1) s33
s34
s31 32 (1)
First, we define the fk vector (1)
(1)
s43
s44
(1) s
s41 42 (1)
(1)
 ∂K ∂ωk  (1) s53
s54
fk =  − 2ωk M  φk . (8) (1) s
s51 52
 ∂θ ∂θ j 
 j
Fig. 3 Mode shape sensitivity matrix
Next, we find the index of the maximum component of
the absolute value of the k-th mode shape. Using this index, so one should provide additional information about the
we make the following partitioning of the dynamic stiffness mutual correlation of the corresponding rows of the
matrix sensitivity matrix. If these rows are highly correlated,
another location with smaller metrics should be chosen.
 v1   ( K − ωk M )11 0 (K − ω M )
−1
2 2
  f1 
k 12
 Mathematically, we can express this issue using the fol-
v  =    0  . (9)
 0  0 1 0   lowing co-linearity matrix
 v 2  k ( K − ω 2 M ) 
 k 0 ( K − ωk M )22 
2 f 2  k
1 nm φ ( k ) φ ( k ) T
φ
ψ= ∑ S S , (12)
21

nm k =1
Finally, mode shape sensitivity denoted here as ϕ S can
be determined using the following formula where ϕ S̅ (k) denotes the row normalized sensitivity matrix
for the k-th mode shape.
∂φk
= v k − φkT Mv kφk . (10) Application of the presented modal sensitivity analysis
∂θ j
to both sensor placement and damage identification will
It should be noted here that the mode shape sensitivity be the topic of subsequent sections.
matrix is a three dimensional object, which is presented
schematically in Fig. 3. Such an object is described by 2.3 Sparsity constrained optimization for damage
three indices φ si(, j) . The meaning of the individual indi-
k
identification
ces is as follows: the first index i represents degrees of While the previous section introduced the sensitivity
freedom, the second index j is associated with structural matrix for sensor placement purposes, the present section
parameters, and finally the third index k corresponds to utilizes this matrix for damage identification. Since the
subsequent modes of the structure. most effective method for damage identification should
For the above mode shape sensitivity matrix we can be able to detect damage-induced parametric modifica-
define cumulative metrics for the i-th degree of freedom tion as early as possible in this study it is assumed that a
1
n p nm change in structural parameters can be determined using
di = ∑∑ φ si(, j) . (11)
φ k
the first order term of the Taylor expansion of modal char-
j =1 k =1
acteristics. The corresponding set of equations is usually
Such metrics give us information about the sensitivity expressed in the following form
of an individual DoF to changes in structural parameters.
∆b = S∆θ , (13)
The DoFs with the highest sensitivity are good candidates
for sensor locations. The method is very simple and effi- where ∆b represents the difference between measured and
cient, but it has one drawback. The DoFs selected in this calculated modal characteristics, S = [ω S, ϕ S (1), ϕ S (2), …,
way are often sensitive to the same structural parameters, S (nm)]T is the overall sensitivity matrix composed of two
 |437
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

matrices representing natural frequency and mode shape by Candes and Romberg. As it will be demonstrated in the
sensitivities, respectively, and finally ∆θ is the difference section devoted to numerical examples, the above formu-
between nominal and modified structural parameters. lation can also provide large estimation error of the struc-
Generally, the approach for finding the solution to the tural parameters.
above set of equations depends on dimensions of the sensi- Additionally, the most accurate parameter estimation
tivity matrix S  n × n , where np is the number of avail-
d p
has been obtained using the least squares solution with
able measurement data and np is the number of unknown non-negativity constraints. The mathematical formulation
parameters. In the case of an over-determined set of equa- of the optimization problem is as follows:
tions ie. nd ≥ np most frequently a solution with the minimal
min ∆θ ||S∆θ –∆b||22 subject to ∆θ > 0 . (16)
ℓ2-norm of the residual vector ||S∆θ – ∆b||2 is satisfactory.
However, in the low frequency based damage identifica- The non-negative least squares (NNLS) algorithm was
tion the situation is rather different and we have fewer data published originally by Lawson and Hanson. This algo-
than parameters to identify, ie. nd < np. In such a situation rithm has been implemented in MATLAB and can be
infinitely many solutions can exist and one has to specify invoke using lsqnonneg command.
additional constraints to find an appropriate solution.
In the case of an under-determined set of equations 3 Numerical examples
the simplest solution is called basis solution and can be The theoretical considerations presented in the previous
obtained in MATLAB by applying the matrix left divi- sections will be illustrated with two case studies. The first
sion operator " \ " . As a result we get a solution in the one is a laboratory-scale three-dimensional truss struc-
form of a vector with r non-zero elements, where r is the ture shown in Fig. 4. and the second one is a full-scale
rank of the sensitivity matrix. Another option is the min- steel arch bridge used as a benchmark problem for seismic
imum ℓ2-norm solution, but in the under-determined case design (Fig. 14).
the norm for the unknown parameters is minimized and
such a solution can be obtained using a pseudo-inverse 3.1 MERO truss - 61 member structure
of the sensitivity matrix pinv(S). Unfortunately, neither The case study presented in this subsection is based on a
matrix left division nor pseudo-inverse can provide a good commercially available exhibit system, known as MERO
estimate of the structural parameters in general case of a M12 [27]. The structure spans 3.5 m and consists of 61 steel
damage scenario. members with circular hollow sections. The dimensions of
To overcome the above issue nowadays researchers fre- the cross section are: 22 mm in diameter, 1 mm of wall thick-
quently use approaches based on regularization, where a ness. Young’s modulus E, Poisson ratio ν and mass density
certain parameter λ is used to make a trade-off between ρ are equal to 205 GPa, 0.3 and 7850 kg/m3, respectively.
two norms: the first one for residuals and the second one The structure is simply supported with one end fixed
for uncertain parameters and the other resting on a roller, which allows movement
in the longitudinal direction. Additionally, the structure is
min ∆θ S∆θ − ∆b 2 + λ ∆θ 2 . (14)
2 2

assembled using screws located at the end of each tubular

The disadvantage of the above formulation is the lack of member. This in turn causes transfer of bending moments
a general method for appropriate selection of the regular- between neighboring tubes.
ization parameter λ [26]. This fact was reflected in the numerical model by
Recently, a new idea has appeared, on the basis of assuming that the whole structure behaves as a spatial
which the solution can be obtained assuming sparsity of frame. The first four mode shapes of the structure and cor-
the vector of unknown parameters. In this formulation we responding natural frequencies are presented in Fig. 5.
solve the following optimization problem:
min ∆θ ||∆θ||1 subject to S∆θ –∆b = 0 . (15)

The sparsity is enforced by the ℓ1-norm, which is noth-

ing else than sum of absolute values of the individual com-
ponents of ∆θ. An implementation of this optimization
problem can be found in the l1-magic toolbox authored Fig. 4 MERO truss – 3D view.
438|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

c) 1st torsional mode, ω3 = 950 rad/s

a) 1st vertical mode, ω1 = 368 rad/s

b) 1st lateral mode, ω2 = 604 rad/s d) 2nd vertical mode, ω4 = 1100 rad/s
Fig. 5 Mode shapes of MERO truss

Fig. 6 Element numbering of the MERO truss

Fig. 7 Cumulative element sensitivity of the first six natural frequencies to stiffness modification

Beam elements of this kind of structure can be classified into stiffness in diagonals has a smaller effect on this change.
one of the two groups depending on their function (Fig. 6): The situation is even worse when we look at the horizontal
• main elements, transferring loads to the supports, elements which have marginal influence not only on the
• auxiliary elements. first or second, but all first six natural frequencies. This
Within the first group we can further distinguish upper fact should be taken into account when designing an SHM
(no. 7 to 20) and lower chord elements (no. 56 to 61). The system for this kind of structures.
second group consists of horizontal elements (no. 1 to 6) The next step in the proposed methodology for dam-
and diagonals (remaining elements). age identification is the selection of damage-sensitive
The rationale behind the above classification can be locations for potential placement of sensors. For that pur-
better understood by looking at the cumulative sensitivity pose one needs to determine the sensitivity of individual
of the first six natural frequencies to stiffness modifica- degrees of freedom of the investigated structure. This can
tion in individual groups (Fig. 7). In this figure we can see be done using the mode shape sensitivity matrix intro-
that, except element no. 14 attached to the roller support, duced in Section 2.2. Sensitivity analysis provides infor-
lower and upper chord elements have strong influence on mation presented in Fig. 9 and a node numbering helpful
the change in modal characteristics, while modification of in sensor positioning is shown in Fig. 8.
 |439
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

Fig. 8 Node numbering of the MERO truss

Fig. 9 Mode shape sensitivity at individual DOFs (6 modes taken into consideration)

Fig. 10 Sensor placement using Effective Independence method

Fig. 11 Sensor placement using damage-sensitive method

Based on Fig. 9, the following conclusions can be drawn: For comparison of the damage-sensitive sensor place-
the most sensitive locations are at nodes no. 6 with 7 and ment methodology presented here with one of the most
then 15 with 16. This is manifested by the highest values frequently used methods called Effective Independence,
of sensitivities in the vertical direction (yellow bars). For both strategies have been tested and the results are pre-
positioning of the next two sensors, the candidate location sented in Figs. 10 and 11.
would be 12 and 10.
440|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

Fig. 12 Multiple damage identification in the MERO truss (6 modes with 6 sensors)

Table 2 Single damage element identifiability in 3D truss structure Another important observation is related to the num-
No. of modes ber of sensors versus number of excited modes. Increasing
1 2 3 4 5 the number of excited modes results in better identifiabil-
1 6 (4) 31 (8) 38 (17) 55 (23) 59 (35) ity of the system parameters than increasing the number
2 19 (6) 45 (15) 53 (25) 61 (34) 61 (49) of sensors.
No. of sensors

3 25 (8) 50 (21) 59 (38) 61 (49) 61 (55) Finally, identification of multiple damage scenarios has
4 34 (13) 53 (25) 61 (43) 61 (54) 61 (57) been tested using both algorithms. A large number of ran-
5 36 (18) 56 (28) 61 (44) 61 (54) 61 (59) domly selected damage scenarios have been analyzed and
6 44 (19) 57 (35) 61 (46) 61 (56) 61 (61) again the NNLS algorithm gave a higher accuracy of esti-
7 47 (24) 57 (40) 61 (48) 61 (56) 61 (61) mation. One example of a multiple damage scenario with
8 47 (30) 57 (51) 61 (51) 61 (59) 61 (61) the applied modification and those identified is shown in
Fig. 12. The values in parentheses in this figure represent
In the analyzed MERO structure, both methods gave estimation error.
similar results as concerns selecting vertical directions in
nodes 6, 7 and 16. The remaining sensor locations dif- 3.2 Influence of measurement noise and modelling
fer and highlight the difference between methods. While errors
Effective Independence tries to maximize the determi- Identification of damage scenario presented at the end of
nant of the Fisher information matrix for mode shapes previous subsection has been performed under assump-
itself, the damage-sensitive method selects locations with tion that measured modal parameters are free of measure-
the highest mode shape sensitivity metrics. ment noise and the mathematical model of structure under
The third step in the proposed methodology is verification investigation is known accurately. In reality, however,
of the damage identifiability. For that purpose we assume measurements are often corrupted by noise and model of
the number of excited modes and number of sensors. Then, the structure contains some modelling errors. The source
we are using a sparsity constrained optimization to identify of these errors can be related to uncertain boundary con-
damage in every structural member. The damage is intro- ditions or unknown stiffness of semi-rigid connections.
duced by reducing stiffness of an element by 1 % of its ori- Taking into account the above aspects Eq. (13) should be
ginal value. The effectiveness of the two algorithms descri- modified to the following form
bed in Section 2.3, namely: ℓ1-norm minimization (L1min)
∆b = S∆θ + ε , (17)
and non-negative least squares (NNLS), is compared.
The results are presented in Table 2. Values in paren- where ε represents errors caused by measurement noise
theses correspond to the L1min method. Generally, NNLS and modeling simplifications.
demonstrates better performance in damage identification The difference between these two type of errors is
than L1min and the remaining analysis will refer to that related to the fact that measurement noise is random and
algorithm. Additionally, we can observe that using only independent at every sensor location. In contrary model-
measurements from 1 sensor for a structure excited only ling errors have systematic character and exhibit certain
with the first mode, both algorithms are able to identify correlation.
correctly at most 6 damaged elements. Identification of To investigate the influence of the above errors on the
all 61 elements is possible when at least three modes are damage identification process with aid of proposed meth-
excited and the structure is equipped with 4 sensors, or odology, the truss structure presented in previous subsec-
one excites 4 modes and uses more than 2 sensors. tion has been analyzed. The first identification has been
 |441
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

a) Case 1 - Numerical errors only

b) Case 2 - Numerical and measurement errors

c) Case 3 - Numerical and modelling errors

Fig. 13 Influence of measurement noise and modelling errors on accuracy of damage identification

performed for ideal case which was free of measurement identification procedure localize damaged elements prop-
and modelling errors (Fig. 13a). The estimation errors pre- erly (Fig. 13c). However, in this case false positive element
sented in this figure are related to numerical accuracy of no. 34 is reported by both algorithms.
the applied algorithm only. The comparison of these three cases allows to conclude
The next case takes into account the measurement that the proposed methodology for sparsity constrained
errors (Fig. 13b). It was assumed that the measured nat- damage identification is relatively insensitive to measure-
ural frequencies and mode shapes are corrupted by 5 % ment and modelling errors and should be suitable for real
error. As can be observed in the figure NNLS and L1min engineering applications.
algorithms still correctly localize damaged elements.
However, the estimation errors are higher than in previous 3.3 Steel arch bridge - 416 member structure
case and their values exceed 10 %. The second case study consists of a full-scale arch bridge
The third analyzed case included modelling errors com- composed of 416 steel members. The material and geomet-
ing from unmodelled elastic support with stiffness equal rical data for this structure have been taken from the paper
to 500 N/m. by Lu et al. [28]. While in this paper seismic performance
This additional support acts at node no. 1 in x-direc- of the bridge is evaluated, here we consider the required
tion and makes the structure stiffer than the nominal number of sensors and modes in order to assure full iden-
one. Similarly as in the case of measurement noise the tifiability of this structure.
442|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

173,00 m

18,00 m

16,87 m
114,00 m

Fig. 14 Elevation of the investigated steel arch bridge

(a) 1st lateral mode, ω1 = 10.6 rad/s

(b) 1st vertical mode, ω2 = 13.9 rad/s

(c) 2nd vertical mode, ω3 = 17.9 rad/s

Fig. 15 First three mode shapes and natural frequencies of the arch bridge

The first three mode shapes of the structure are pre- used in that sensitivity analysis. Then, again Effective
sented in Fig. 15. Independence based sensor placement has been compared
The first step of the damage identification, similarly with that obtained by the damage-sensitivity method.
as in the case of the MERO truss, was the calculation of As in the case of the MERO truss, sensor distributions
modal sensitivities for both parametric sensitivity ver- obtained using both methods are quantitatively quite sim-
ification and sensor placement. The first sixteen modal ilar with the preferred sensor location on the steel arch in
frequencies and corresponding mode shapes have been the vertical and horizontal directions.
 |443
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

Table 3 Single damage element identifiability in arch bridges 4 Conclusions

No. of modes A comprehensive methodology for damage localiza-
1 5 10 tion and quantification has been proposed. It consists of
8 109 (49) 369 (212) 402 (299) three levels of investigation. Firstly, it utilizes sensitiv-
sensors
No. of

16 148 (71) 391 (272) 403 (339) ity analysis to determine the influence of stiffness mod-
24 163 (87) 404 (338) 406 (354) ification of individual structural members on changes in
modal characteristics of the structure. The second level
The results of the second step of the proposed method- of the proposed methodology uses the sensitivity matrix
ology, which is identification of a single damaged element, of the natural frequencies to determine hardly identifi-
are presented in Table 2. Similarly as in the case of the able parameters. Simultaneously, sensitivity of the cor-
MERO truss, better results have been obtained from the responding mode shapes helps to select damage-sensitive
NNLS algorithm. Using 16 sensors and exciting 10 modes positions of sensors. The third level applies two sparsity
it was possible to identify correctly a single damage in 406 constrained optimization algorithms to efficiently iden-
out of 416 structural members. The remaining 10 elements tify the applied damage scenarios. These two algorithms
were hardly identifiable even using a larger number of are based on ℓ1-norm minimization and non-negative least
sensors and modes. However, investigation of the results square (NNLS) solution, respectively. In both presented
revealed that these were the elements connected to sup- case studies of a three-dimensional truss girder and a
ports and to assure their proper identification they should steel arch bridge better results have been achieved using
be possibly equipped with dedicated monitoring systems. the NNLS algorithm. Moreover, the performed analyses

Fig. 16 Sensor placement on the arch bridge using Effective Independence method

Fig. 17 Sensor placement on the arch bridge using damage-sensitive method

444|Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

revealed that excitation of more modes is more advanta- [7] Papadimitriou, C. "Optimal sensor placement methodology for para-
geous than measurement of more degrees of freedom. This metric identification of structural systems", Journal of Sound and
Vibration, 278(4–5), pp. 923–947, 2004.
observation can be useful in designing and implementing
https://doi.org/10.1016/j.jsv.2003.10.063
the SHM system in real complex engineering structures. [8] Barthorpe, R. J., Worden, K. "Sensor Placement Optimization". In:
Finally, robustness of the proposed methodology in the Boller, C., Chang, F.-K., Fujino, Y. (eds.) Encyclopedia of Structural
presence of measurement and modelling errors has been Health Monitoring, 1st ed., John Wiley & Sons, Hoboken, New
demonstrated. Jersey, United States, 2009.
https://doi.org/10.1002/9780470061626.shm086
[9] Zhou, K., Wu, Z. Y., Yi, X. H., Zhu, D. P., Narayan, R., Zhao, J.
Acknowledgement
"Generic Framework of Sensor Placement Optimization for Structural
The support of the National Science Centre, Poland, granted Health Modeling", Journal of Computing in Civil Engineering,
under grant agreement DEC2017/25/B/ST8/01800, is grate- 31(4), 2017.
fully acknowledged. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000662
[10] Friswell, M., Mottershead, J. "Finite Element Model Updating
Abbreviations in Structural Dynamics", 1st ed., Springer Nature, Dordrecht,
Netherlands, 1995.
The following abbreviations are used in this manuscript:
https://doi.org/10.1007/978-94-015-8508-8
DoF Degree of freedom [11] Hernandez, E. M. "Identification of isolated structural damage from
EI Effective Independence incomplete spectrum information using l1-norm minimization",
EKF Extended Kalman Filter Mechanical Systems and Signal Processing, 46(1), pp. 59–69, 2014.
FEM Finite element method https://doi.org/10.1016/j.ymssp.2013.12.009
FIM Fisher information matrix [12] Ebrahimian, H., Astroza, R., Conte, J. P. "Extended Kalman
filter for material parameter estimation in nonlinear structural
IE Information entropy
finite element models using direct differentiation method",
MAC Modal assurance criterion Earthquake Engineering & Structural Dynamics, 44(10), pp.
NNLS Non-negative least squares 1495–1522, 2015.
OSP Optimal sensor placement https://doi.org/10.1002/eqe.2532
SHM Structural Health Monitoring [13] Ginsberg, D., Fritzen, C.-P., Loffeld, O. "Sparsity-constrained
Extended Kalman Filter concept for damage localization and iden-
tification in mechanical structures", Smart Structures and Systems,
References
21(6), pp. 741–749, 2018.
[1] Shah, P. C., Udwadia, F. E. "A Methodology for Optimal Sensor
https://doi.org/10.12989/sss.2018.21.6.741
Locations for Identification of Dynamic Systems", Journal of
[14] An, Y., Blachowski, B., Ou, J. "A degree of dispersion-based dam-
Applied Mechanics, 45(1), pp. 188–196, 1978.
age localization method", Structural Control and Health Monitoring,
https://doi.org/10.1115/1.3424225
23(1), pp. 176–192, 2016.
[2] Cawley, P., Adams, R. D. "The location of defects in structures
https://doi.org/10.1002/stc.1760
from measurements of natural frequencies", The Journal of Strain
[15] Pnevmatikos, N. G., Blachowski, B., Hatzigeorgiou, G. D., Swiercz,
Analysis for Engineering Design, 14(2), pp. 49–57, 1979.
A. "Wavelet analysis based damage localization in steel frames
https://doi.org/10.1243/03093247V142049
with bolted connections", Smart Structures and Systems, 18(6), pp.
[3] Kammer, D. C. "Sensor placement for on-orbit modal identifica-
1189–1202, 2016.
tion and correlation of large space structures", Journal of Guidance,
https://doi.org/10.12989/sss.2016.18.6.1189
Control, and Dynamics, 14(2), pp. 251–259, 1991.
[16] Blachowski, B., An, Y., Spencer Jr., B. F., Ou, J. "Axial Strain
https://doi.org/10.2514/3.20635
Accelerations Approach for Damage Localization in Statically
[4] Liu, P.-L., Chian, C.-C. "Parametric Identification of Truss Structures
Determinate Truss Structures", Computer-Aided Civil and
Using Static Strains", Journal of Structural Engineering, 123(7), pp.
Infrastructure Engineering, 32(4), pp. 304–318, 2017.
927–933, 1997.
https://doi.org/10.1111/mice.12258
https://doi.org/10.1061/(ASCE)0733-9445(1997)123:7(927)
[17] An, Y., Blachowski, B., Zhong, Y., Holobut, P., Ou, J. "Rank-
[5] Cobb, R. G., Liebst, B. S. "Sensor Placement and Structural Damage
revealing QR decomposition applied to damage localization in truss
Identification from Minimal Sensor Information", AIAA Journal,
structures", Structural Control and Health Monitoring, 24(2), 2017.
35(2), pp. 369–374, 1997.
https://doi.org/10.1002/stc.1849
https://doi.org/10.2514/2.103
[18] Blachowski, B., Swiercz, A., Gutkiewicz, P., Szelazek, J.,
[6] Katafygiotis, L. S., Beck, J. L. "Updating Models and Their
Gutkowski, W. "Structural damage detectability using modal and
Uncertainties. II: Model Identifiability", Journal of Engineering
ultrasonic approaches", Measurement, 85, pp. 210–221, 2016.
Mechanics, 124(4), pp. 463–467, 1998.
https://doi.org/10.1016/j.measurement.2016.02.033
https://doi.org/10.1061/(ASCE)0733-9399(1998)124:4(463)
 |445
Blachowski
Period. Polytech. Civ. Eng., 63(2), pp. 432–445, 2019

[19] Blachowski, B., Gutkowski, W. "Effect of damaged circular flange- [24] Beben, D. "Application of Interferometry Method for Dynamic
bolted connections on behaviour of tall towers, modelled by multi- Continuous Testing of Bridges", Periodica Polytechnica Civil
level substructuring", Engineering Structures, 111, pp. 93–103, 2016. Engineering, 60(3), pp. 387–395, 2016.
https://doi.org/10.1016/j.engstruct.2015.12.018 https://doi.org/10.3311/PPci.8679
[20] Blachowski, B., Swiercz, A., Jankowski, L. "Virtual Distortion [25] Nelson, R. B. "Simplified calculation of eigenvector derivatives",
Method based optimal sensor placement for damage identification", AIAA Journal, 14(9), pp. 1201–1205, 1976.
presented at ISMA 2018 / USD 2018, International Conference https://doi.org/10.2514/3.7211
on Noise and Vibration Engineering / International Conference [26] Hou, R., Xia, Y., Bao, Y., Zhou, X. "Selection of regularization
on Uncertainty in Structural Dynamics, Leuven, Belgium, Sept. parameter for l1-regularized damage detection", Journal of Sound
17–19, 2018. and Vibration, 423, pp. 141–160, 2018.
[21] Kourehli, S. S. "Damage Diagnosis of Structures Using Modal Data https://doi.org/10.1016/j.jsv.2018.02.064
and Static Response", Periodica Polytechnica Civil Engineering, [27] Suwała, G., Jankowski, Ł. "Nonparametric identification of struc-
61(1), pp. 135–145, 2017. tural modifications in Laplace domain", Mechanical Systems and
https://doi.org/10.3311/PPci.7646 Signal Processing, 85, pp. 867–878, 2017.
[22] Bayer, J., Král, J., Urushadze, S. "Localization of Simulated Damage https://doi.org/10.1016/j.ymssp.2016.09.018
on a Steel Beam from Random Vibrations", Periodica Polytechnica [28] Lu, Z., Ge, H., Usami, T. "Applicability of pushover analysis-based
Civil Engineering, 62(1), pp. 112–116, 2018. seismic performance evaluation procedure for steel arch bridges",
https://doi.org/10.3311/PPci.10625 Engineering Structures, 26(13), pp. 1957–1977, 2004.
[23] Bagherahmadi, S. A., Seyedpoor, S. M. "Structural damage detec- https://doi.org/10.1016/j.engstruct.2004.07.013
tion using a damage probability index based on frequency response
function and strain energy concept", Structural Engineering and
Mechanics, 67(4), pp. 327–336, 2018.
https://doi.org/10.12989/sem.2018.67.4.327