Int. J. Civil Engineering manuscript No.

(will be inserted by the editor)

Bridge damage detection through combined quasi-static influence lines

and weigh-in-motion devices

M. Breccolotti · M. Natalicchi

Received: date / Accepted: date

Acknowledgement

Not Applicable

Funding information

Not Applicable

Conflict of interest

The authors declare that they have no conflict of interest.

M. Breccolotti

Department of Civil and Environmental Engineering, University of Perugia, 06125 Perugia, Italy

Assistant Professor, Ph.D.

E-mail: [email protected]

M. Natalicchi

Sintagma S.r.l., 06132 Perugia, Italy

Civil Engineer, M.Sc.

E-mail: [email protected]
 2 M. Breccolotti and M. Natalicchi

Abstract This paper addresses a new damage detection method for simply supported and continuous

bridge structures based on combined influence lines for deflections and rotations and weigh-in-motion

devices. The real structural behaviour of the superstructure is captured by sensors that measure vertical

displacements at midspans and rotations at each support. Information on speed and weight of vehicles in

transit are provided by weigh-in-motion devices placed at the beginning and at the end of the bridge. An

optimization technique based on genetic algorithm and developed in a Matlab environment identifies the

flexural stiffness distribution that best fits the experimental data. The comparison between the identified

flexural stiffness distribution and the nominal one, or that obtained by previous identification, allows to

verify the existence and the position of single or multiple damages. The capabilities of the proposed method

was proved by means of three numerical simulations on a simply supported prestressed concrete I-beam

bridge, on a three-span continuous prestressed concrete box bridge and on a four-span continuous steel-

concrete composite bridges. Moving loads were considered to generate synthetic displacement and rotational

distributions for different load positions. Changes in the flexural stiffness as small as 20% for finite elements

with length of 0.5 m or 10% for finite elements with length of 2.0 m were successfully identified with a root-

mean-square-error of less than 2.72%. The simulations produced evidence of benefits, in terms of damage

identification, provided by the combined use of different types of influence lines with the maximum error

in the stiffness identification that reduced from 3.40% to 2.64%.

Keywords Bridges · Damage detection · Influence lines · Weigh-in-motion

1 1 Introduction

2 Many structural and infrastructural facilities are approaching the end of their original service life design.

3 For such structures which were built several decades ago, inevitable aging, effects of atmospheric agents and

4 the action of constant and time-varying stresses, have most likely affected their structural integrity. Several

5 methods can be applied to monitor and evaluate the degradation of these structures. These methods can be

6 divided into 3 types based on their purposes: Visual inspections, Event monitoring and Continuous moni-

7 toring. Visual inspections are checks carried out routinely by the managing bodies at pre-established time

8 intervals to verify the existence of evident defects on the structures. Event monitoring consists of measuring

9 certain properties of the structure for a very limited period of time whereas continuous monitoring, also

10 known as Structural health monitoring (SHM), is carried out by constant collection and recording of data

11 gained via various instruments installed on the structure over time, for a real time evaluation of integrity,
 Bridge damage detection through combined ILs and WiM devices 3

12 reliability and durability of the structure. The topic of this investigation is in line with SHM method.

13 Specific visual inspections can be eventually carried out, outside the pre-established intervals, in case the

14 event monitoring or the continuous monitoring have indicated the presence of anomalies.

15 SHM can be distinguished, for instance, in passive or active [1]. Passive monitoring is where technicians

16 performing the assessments are just monitoring the evolution of physical parameters measured through

17 embedded sensors. Conversely, active monitoring is carried out in structures equipped with both sensors

18 and actuators. In such case, perturbations are generated by actuators and the response is observed by

19 sensors.

20 For bridge structures a number of bridge monitoring systems (BMS) have been developed so far. They

21 can be classified in the following four capabilities levels [2]:

22 - Level 1: Monitoring systems providing qualitative indications about the existence of damage in the

23 structure (Detection);

24 - Level 2: Monitoring systems providing indications about the position of damage in the structure (Lo-

25 calization);

26 - Level 3: Monitoring systems providing indications about position and extent of damage in the structure

27 (Assessment);

28 - Level 4: Monitoring systems providing information on the actual safety of the damaged structure (Con-

29 sequence).

30 BMS can be distinguished from each other according to several aspects. For instance, monitoring systems

31 can use static or dynamic measurements. Generally, static measurements utilise displacements, strains and

32 deformations, whereas dynamic measurements are related to frequencies, damping ratios and modes of

33 vibration. Monitoring systems can also be distinguished based on the type, discrete or distributed, of

34 sensors. In the first case, traditional sensors are placed only at pre-determined specific points, making their

35 correct positioning imperative [3]. In the second case, continuous and real-time measurements is carried

36 out along the entire length of the structure (e.g. fibre optic cable [4]).

37 In addition to observations of deterioration state of an aging structure, it becomes crucial to identify the

38 moment in which such deterioration exceeds safety limit. This is achieved by solving an inverse problem,

39 i.e. a problem where one seeks to identify the parameters characterizing the system under investigation

40 from measurements of the effects (indirect observations). Monitoring systems that deal with this goal are
 4 M. Breccolotti and M. Natalicchi

41 divided in FEM based or FEM free [5] depending on the need to utilise finite element (FE) methods for

42 the damage identification.

43 In the last few decades, many efforts have been spent in order to develop SHM procedures, sensors and

44 technologies to evaluate the health conditions of civil engineering structures as reliably as possible. Many

45 features of structural response ranging from static [6, 7], dynamic [8–13], statistical [14–16] and acoustical

46 [17, 18] were studied for this purpose.

47 The aim of this work is to develop a simple monitoring system, consisting of only a few sensors and

48 weigh-in-motion (WiM) devices, for bridge damage detection and localization and issuing of early warning

49 notices. The possibility of using influence lines related to different deformations (deflections and rotations)

50 is investigated in order to improve damage identification which can affect a road or rail bridge. It is factually

51 proven that a better evaluation of the mechanical properties of a structure by non-destructive techniques

52 (NDT) can be obtained by combining different sources of information. This has been demonstrated with

53 regard to estimation of in-situ concrete compressive strength using NDT methods [19, 20]. Hence a bridge

54 condition monitoring methodology capable to detect and localize the damage (Level 2) is developed. It

55 makes use of deflection and rotation influence lines, weigh-in-motion devices, FE modeling and genetic

56 algorithms. The monitoring method proposed in this work is defined as static, continuous, passive and

57 FEM based.

58 2 Bridge Structural Health Monitoring using Influence lines

59 Bridge monitoring systems based on influence lines (ILs) were recently proposed by several authors. Their

60 strategies generally rely on discrete sensors layout. They differ from each in respect to the type of quantity

61 (stress, deformation, etc.) to which the influence lines are referenced.

62 Displacement influence lines (DILs) were frequently used for bridge damage diagnosis. He et al. [5]

63 developed a two-stage damage detection method that uses the displacement response induced by quasi-

64 static test and analytical evaluation of the differences between the displacement influence lines in the

65 undamaged and damaged states. Chen et al. proposed a damage detection method based on DILs for beam

66 structures [21]. The method decomposes the stiffness and flexibility matrices to reveal the relationship

67 between structural damage extent and DIL changes. Zeinali and Story developed a method for damage

68 localization and quantification that addresses issues with static noise deformation influence lines [22]. The

69 method fits a quadratic spline to the noisy measured deformation influence line of the beam using two
 Bridge damage detection through combined ILs and WiM devices 5

70 regularization approaches to overcome the problem of noisy measurements. Zhang and Liu developed a

71 method based on quasi-static displacement influence lines and on a new form of the Hankel matrix of the

72 damage feature to detect damage in the superstructure of a beam bridge [23]. Le et al. developed a damage

73 detection method that utilizes, similarly to those based on influence lines, the principle of virtual works

74 coupled with changes in static deflections [24]. If the number of damaged portions is known, the damage

75 severity can be calculated through a closed-form function.

76 Rotation influence lines (RILs) were recently proposed for damage identification in bridge structures by

77 several authors. Zhou et al. derived the explicit expressions of the rotational-angle influence lines (RAIL)

78 for an arbitrary section of a simply supported beam with damage by means of the principle of virtual

79 displacement and the reciprocal theorem [25]. The difference of the rotational-angle influence lines-curvature

80 (DRAIL-C) were used to developed a novel damage-identification indicator. Alamdari et al. developed a new

81 technique based on RILs established at either end of the bridge [26]. Similarly, Huseynov et al. presented a

82 bridge condition monitoring methodology where the rotation influence line difference between healthy and

83 damaged bridge states is proposed as a damage indicator [27].

84 Influence lines have also been studied for strain measurements. For instance, Wang et al. proposes a

85 fitting algorithm to extract IL information from the bridge dynamic response induced by a passing vehicle

86 [28]. It must be emphasized that in this case, as well as those relating to the curvature, influence lines for

87 statically determined structures are not affected by damage unless the strain or the curvature are measured

88 directly on the damaged element. Cantero and González used strain influence lines in combination with

89 weigh-in-motion technology to implement a Level 1 damage-detection technique for short- to medium-span

90 road bridges [29].

91 Other aspects related to damage identifications through influence lines with discrete measurements

92 were addresses by researchers. Chen et al. developed a novel approach based on adaptive B-spline basis

93 dictionary and sparse regularization to identify the bridge influence lines [30]. Zheng et al. carried out

94 research in order to compare different methods (Least Squares, Tikhonov regularization, B-spline based

95 sparse regularization, Superposition and Frequency domain) to identify bridge influence line from direct

96 measurement data [31].

97 Influence lines were also used for damage detection purposes with spatially distributed sensors. Goldfeld

98 and Klar used an iterative algorithm and Brillouin optical time domain reflectometry (BOTDR) or analysis

99 (BOTDA) to detect the configuration of cracks of a simply supported RC beam [32]. Wu et al. proposed a

100 damage identification method for continuous girder bridges based on long-gauge strain influence line [33].
 6 M. Breccolotti and M. Natalicchi

101 Brillouin scattering based distributed sensors were also used by Oskoui et al. for the development and

102 application of a method for detection of cracks in bridges [34].

103 3 Theoretical background

104 The procedure for damage identification presented in this work makes use of weigh-in-motion devices and

105 influence lines w(x, ξ), functions that represent the variation of deformation (deflection, rotation, curvature,

106 etc.) or stress (moment, shear, etc.) in a specific section ξ for a moving load placed at location x on

107 the structure. ILs are characteristic features of a bridge that depend on material properties, geometric

108 characteristics and boundary conditions.

109 They can be obtained through Betti’s theorem (first principle of reciprocity): given two sets of forces

110 acting separately on the structure, Pi (i = 1, ..., n) and Qj (j = n + 1, ..., m) the work done by the first set

111 of forces for the displacements wQi induced by the second one is equal to the work done by the forces of

112 the second set of forces for the displacements wPj induced by the first one:

n
X m
X
Pi · wQi = Qj · w P j (1)
i=1 j=n+1

113 In the present research, damage identification will be based on the combined use of displacement and

114 rotation ILs. To the best knowledge of the authors, there is no published work dealing with such method.

115 In the following a brief reference will be made to these two types of influence lines and to WiM devices.

116 3.1 Deflection influence lines

117 Based on Betti’s theorem (Eq. 1), for the structure shown in Fig. 1 the following relation holds:

FP · w P = F Q · w Q (2)

118 Assuming unit vertical loads FP and FQ , the deflection wQ of section Q produced by a unit vertical load

119 in P is equal to the deflection wP of point P for a unit vertical load applied in Q.

120 Therefore, the influence line of the deflection wQ in Q (diagram of the deflection wQ in Q as the abscissa

121 x of the point P of application of the unit force varies) corresponds to the deformed configuration of the

122 structure produced by the unit vertical force applied in Q.

123 The measurement of deflections can be carried out, for instance, by conventional systems based on linear

124 variable differential transducer (LVDT) cable systems. Robotic total stations (RTS) may also be used for
 Bridge damage detection through combined ILs and WiM devices 7

Fig. 1 First principle of reciprocity for deflection influence line.

125 displacement of few mm and up to a frequency of 4 Hz. For higher frequencies other techniques, such as

126 digital image correlation (DIC) or laser Doppler vibrometer (LDV) systems, can be adopted.

127 3.2 Rotation influence lines

128 Similar to the deflection influence line, the following relation can be obtained by applying Betti’s theorem

129 to the structure shown in Fig. 2:

FP · w P = M Q · ϕ Q (3)

130 Supposing unit vertical load FP and unit bending moment MQ , the rotation ϕQ of section Q produced

131 by a unit vertical load placed in P is equal to the displacement wP of point P for a unit bending moment

132 applied in Q.

133 The influence line of the rotation ϕQ in Q (diagram of the rotation ϕQ in Q as the abscissa x of the

134 point P of application of the unit force varies) corresponds to the diagram of deflections produced in the

135 structure by the unit bending moment applied in Q.

136 Rotations at specific bridge sections can be measured by using standards inclinometers with different

137 sensing technologies (accelerometer, dielectric liquid, gas bubble in liquid and pendulum). Alternatively,

138 these rotations can be derived using image processing techniques.

 8 M. Breccolotti and M. Natalicchi

Fig. 2 First principle of reciprocity for rotation influence line.

139 3.3 Determination of influence lines by FEM

140 The determination of the influence lines for deflections and rotations is straightforward by using the finite

141 element method and the static equilibrium equation:

[K] {xj } = {Fj } (4)

142 where [K] is the stiffness matrix of the system, {xj } is the vector with the nodal displacements and rotations

143 and {Fj } represents the external forces.

144 The bridge is discretized into N finite beam elements whose dimension is chosen in order to carefully

145 reproduce the structural system. In this regard, it is also necessary to consider the spatial resolution of

146 experimental data acquisition relating to displacements and rotations. This shall be further discussed and

147 illustrated in sec. 4. The stiffness matrices [Kie ] for each i − th element with i = 1 : N are initially evaluated

148 according to the Euler-Bernoulli theory with the nominal geometrical and mechanical properties:
 
2
 Ai l i 0 0 −Ai l2 0 0 
 
 
 0 12Ii 6Ii li 0 −12Ii 6Ii li 
 
 
 
E 0
 6Ii li 4Ii li2 0 2 
−6Ii li 2Ii li 
[Kie ] = 3   (5)
li 
 −A l2

 i i 0 0 Ai li2 0 0  
 
 
 0 −12Ii −6Ii li 0 12Ii −6Ii li 
 
 
 
0 6Ii li 2Ii li2 0 2
−6Ii li 4Ii li

149 E is the material elastic modulus, Ai , Ii and li the cross section, the moment of inertia and the length of

150 the i − th beam element, respectively.

 Bridge damage detection through combined ILs and WiM devices 9

151 The system stiffness matrix [K] is subsequently obtained by standard matrix structural analysis proce-

152 dures. For each different influence line, the system force vector {Fj } is a null vector with single non-null

153 element corresponding to a unit force or moment applied to the DoF for which the influence line is searched.

154 The system displacements {xj } are calculated by taking into account boundary conditions inverting Eq. 4:

{xj } = [K]−1 {Fj } (6)

155 where [K]−1 is the inverse of the stiffness matrix also known as flexibility matrix. The theoretical influence

156 line {ILj,th } can be extracted from the system displacements {xj } by multiplication with a suitable matrix:

{ILj,th } = [A] · {xj } (7)

157 where [A] is a null matrix with unit values at specific positions to extract the nodal vertical displacements

158 from the system displacement vector.

159 3.4 Derivation of bridge ILs from direct measurements

160 Real ILs can be derived from direct measurements by using the method proposed by O’Brien et al. This

161 mathematical method converts the measured response of a load effect into an influence line for that effect

162 for a vehicle of known weights and axles spacings [35]. The method was applied for strain measurements

163 but its application is generic and can thus, be applied also to displacement and rotation measurements.

164 3.5 WiM devices

165 Weigh-in-motion devices have undergone a significant technological development in recent years. A literature

166 review on WiM systems was recently published by the U.S. Federal Highway Administration [36]. Different

167 operating principles, ranging from bending plate, load cell, piezoelectric and piezoceramic sensors, are used

168 by the manufacturers. The accuracy of these devices is estimated by manufacturers and can be within the

169 ±7% range. However, there are cases where weighing errors of 2% or less, were found [37]. Their installation

170 is generally carried out by grouting the sensing device inside pavement slots, thus easily implementable

171 both in new and existing structures, involving a portion of the deck with a length of only 2 meters for each

172 lane. Recently, new systems based on free-of-axle-detector (FAD) were successfully used on a continuous

173 reinforced concrete bridge [38]. These new sensors have the advantage, compared to the traditional ones, of

174 being positioned below the deck, as opposed to externally. Hence not disturbing or hindering the movement

175 of vehicles on the bridge.

 10 M. Breccolotti and M. Natalicchi

176 4 Damage identification framework

177 The generic framework for damage identification through combined experimental influence lines and weigh-

178 in-motion evaluations is composed by the following key features:

179 - two WiM sensors embedded at the beginning and at the end of the bridge to define vehicle information

180 such as axles number, axle loads, axle spacings and vehicle speed;

181 - rotation and deflection sensors placed at suitable positions along the bridge to measure experimental

182 influence lines {δj,exp };

183 - FE model of the bridge initially built with the nominal stiffness and mass values of the structure to

184 calculate the theoretical bridge ILs {ILj,th };

185 - an optimization algorithm to update the FE model in order to minimize the differences between the

186 numerical and the experimental influence lines.

187 The operation scheme is illustrated in the flowchart of Fig. 3.

188 The readings obtained by each sensor allow to define the IL for a specific quantity (displacement and

189 rotation) and for the section where the sensor is installed. The resolution s of the IL depends on the

190 sampling frequency f of the data acquisition system and vehicle speed V :

s = V /f (8)

191 For instance, a sampling frequency of 10 Hz and a vehicle velocity of 80 km/h correspond to a resolution

192 of 2.22 m.

193 It is assumed that damage is responsible for a decrease in the flexural stiffness even if other sources

194 of changes in flexural stiffness are possible. Notwithstanding that many studies on bridge monitoring may

195 have been carried out considering vehicle dynamics, such studies were excluded in the present investigation.

196 In fact, previous researches [5, 33] demonstrated that, for normal vehicle speeds (that is, travel speeds of

197 most vehicles and trucks in the range 70 - 110 km/h), the vehicle dynamics has little effect on the damage

198 identification results. Furthermore, the objective of this monitoring procedure is to detect the presence of

199 damage by exploiting the normal traffic circulating on the infrastructure for which data on the suspension

200 system is not available. Taking into account that quasi-static conditions can be modelled as static conditions,

201 static simulations were carried out in this work.

202 The experimental data extracted from the bridge refer to vertical displacements and rotations. This

203 data was contaminated with random noise, with intensity of up to ±5% to simulate real-world conditions
 Bridge damage detection through combined ILs and WiM devices 11

Start

Experimental investigation Theoretical model

Define sensors Define nominal input data:

types and positions geometry, materials, ...

Realization of Realization of

experimental setup FE model

Data acquisition: vehicles Determination of

weight, displacements and displacements and

rotations rotations ILs

Calculate experimental Calculate factored

deformations δj,exp ILs αj · ILj,th

Compare δj,exp and αj · ILj,th

Optimize fitness function Z (Eq. 9)

by changing flexural stiffness through GA

Identification completed

End

Fig. 3 Flowchart for determination of flexural stiffness distribution through ILs and WiM.

204 and take into account WiM devices accuracy. Position of the sensors and data points were located where

205 maximum values were expected. For displacements, sensors were placed at midpoint of each span. As far as

206 rotations is concerned, sensors were located on each supports according to the findings of Huseynov et al.

207 [27]. Although installation of more sensors would have been beneficial to obtain a more accurate damage

208 detection [21], for economic reasons positioning of sensors in each span and at each support was deemed to

209 be adequate.
 12 M. Breccolotti and M. Natalicchi

210 Genetic Algorithms (GA) were used to implement the optimization procedure. GA have been used in

211 structural damage detection since the beginning of the century [39]. This optimization method was chosen

212 in the present investigation for its capacity to deal with problems with a large number of unknown param-

213 eters to identify and for the ability to track down global minimum rather than local minimum. For each

214 simulation, the genetic algorithm searched the best solution with 3000 generations. In each generation 500

215 different populations of the unknown stiffness distribution were analyzed. In order to reduce the computa-

216 tional time, GA starts searching the flexural stiffness distribution that minimize the fitness function from

217 initial random populations in the range ±10% from the undamaged nominal distribution of the flexural

218 stiffness. Subsequently, the bending stiffness distributions were sought in the range of 50% - 110% of the

219 nominal bending stiffness. Each new population is created by selecting the individuals with three different

220 strategies:

221 - individuals that obtained the best fitness function values;

222 - individuals generated from the current population by making random changes to a single individual

223 (mutation);

224 - individuals generated from the current population by combining two individuals (crossover).

225 At present, investigation was chosen to generate 5% of a new population by selecting the best individuals,

226 50% with the crossover function and the remaining 45% by mutation.

227 Unknown flexural stiffness distribution is evaluated by minimising the value of the fitness function Z.

228 This is calculated as the sum of squared differences between the experimental displacement or rotation

229 time histories {δj,exp } and the corresponding theoretical influence lines {ILj,th } multiplied by a suitable

230 coefficient αj :
p
X
Z= (δj,exp − αj · ILj,th )2 (9)
j=1

231 where p is the number of sensors placed on the monitored bridge for which an influence line can be

232 determined for each vehicle passage. The coefficients αj are weighting factors that take into account two

233 aspects:

234 - the experimental displacement or rotation time histories {δj,exp } measured on the bridge refers to

235 non-unit vehicle loads while theoretical ILs {ILj,th } are calculated assuming unitary loads in Eq. 7;

236 - the different scales of the theoretical ILs {ILj,th } for different types of sensors (displacement and rota-

237 tion). These corrections factors ensure that deflection and rotations measurements contribute similarly

238 in the value of the fitness function Z.

 Bridge damage detection through combined ILs and WiM devices 13

239 It should be noted that the proposed method allows for the influence of elements stiffness on the stress

240 distribution, which is always present in statically indeterminate structures.

241 5 Numerical validation

242 Three case studies found in the literature were used to validate the proposed methodology. The first

243 refers to a prestressed, simply supported interior bridge beam. The other case studies refer to continuous

244 bridges, one made of structural concrete and the other one with a composite steel and concrete deck.

245 These different case studies were carried out to investigate whether the proposed method can be used

246 interchangeably for different layouts (simply supported bridges or multi-span continuous bridges) and for

247 different construction technologies (prestressed RC bridges or composite steel-concrete bridges). For each

248 case study a 1-D bridge/beam model was developed in a Matlab environment through matrix structural

249 analysis. Linear elastic behaviours were assumed for each material in service conditions of the analyzed

250 structures. Euler-Bernoulli beam elements were used to model the structures. The adoption of Timoshenko

251 beam elements was deemed to be unnecessary, considering the modest relevance of shear deformability in

252 the investigated cases. Numerical simulations were used to generate pseudo-experimental data assuming

253 that a conventional vehicle travels on the bridge. Simplified load conditions were considered in these first

254 investigations. A single axis vehicle is assumed with a total weight of 100 kN for the first case study and

255 300 kN for the other two cases. These values were chosen within a range of values considering the following

256 factors:

257 - real weights of vehicles transiting over the bridges (in Europe the maximum allowable weight of four

258 axle truck is 400 kN);

259 - total weight of “frequent” lorries listed in EN 1991-2;

260 - ability of the vehicle in transit to activate any non-linear mechanisms due to damage in the bridge;

261 - low precision of displacements and rotation sensors for small values of the quantities to be measured.

262 For the purposes of this monitoring scheme, it is not necessary to perform the identification procedure for

263 each vehicle passing over the bridge. It will be carried out only in cases where time histories of displacements

264 and rotations corresponding to the passage of a single vehicle of sufficiently high weight on the bridge are

265 available. In case of multi-axle vehicles, it can be addressed by considering the principle of superposition

266 and procedures such as that proposed by [35]. Different damage scenarios were speculated for each case
 14 M. Breccolotti and M. Natalicchi

267 study. An influence line was calculated for each different sensor placed along the bridge and for each damage

268 configuration.

Fig. 4 First case study [40]: Elevation (left, dimension in m) and cross section (right, dimension in mm) of the prestressed

concrete beam. Blue and black numbers in the elevation view represent sensors and elements numbering, respectively.

3
Sens. 1 Sens. 2 Sens. 3

2
Vertical displacement (mm)

-1

-2

-3

-4
0 5 10 15 20 25
Coordinate X (m)

Fig. 5 First case study, products αj ·ILj,th . Solid lines for displacements, dashed lines for rotations. For sensors numbering

see Fig. 4.

269 5.1 Case study 1

270 The first case study is a simply supported interior bridge beam [40] with an overall length of approximately

271 25 m (Fig. 4). It is made up of a precast pretensioned AASHTO type IV girder and a 20 cm thick cast-

272 in-place slab. Its constant cross section is shown at the bottom part of Fig. 4. Concrete with 48 MPa

273 compressive strength and 35 GPa elastic modulus was used for the beam. The slab was considered to be
 Bridge damage detection through combined ILs and WiM devices 15

274 made of concrete with 35 MPa compressive strength and 30 GPa elastic modulus. Further details on the

275 geometrical features and the materials mechanical properties can be found in the cited reference.

276 The structure was discretized with 50 beam elements having 0.5 m length. It is assumed that 3 sensors

277 measuring one vertical displacement (at midspan) and two rotations (at each end) are placed on the bridge

278 (Fig. 4).

279 A single damage scenario with 20% flexural stiffness reduction in element 24 was considered for this case

280 study. Position of the damaged element was chosen close to the midspan section where the maximum stress

281 is expected. Examples of the products αj · ILj,th between the coefficient αj and the theoretical influence

282 lines ILj,th for the damaged configuration and for each sensors are shown in Fig. 5. Solid line refers to the

283 displacement influence line for the sensor placed at midspan. Conversely, dashed lines refer to rotation ILs

284 for sensors placed on the supports.

285 The flexural stiffness distributions resulting from the optimization procedure is shown in Fig. 6. In this

286 figure, the histogram bars represent the identified moment of inertia values. Black solid line represents the

287 true inertia moment distribution in the damaged case whilst the red dashed line represents the undamaged

288 one.

289 5.2 Case study 2

290 The second case study is a prestressed concrete two cell box bridge taken from the Appendix B of NCHRP

291 Project 12-71 [41]. It has an overall length of 210 m with three continuous spans of 60, 90 and 60 m

0.35
Identified
True
0.3 Undamaged

0.25
Moment of inertia (m4)

0.2

0.15

0.1

0.05

0
0 5 10 15 20 25
Coordinate X (m)

Fig. 6 First case study, comparison between identified (blue-gray bars), true (damaged, solid black lines) and undamaged

(dashed red lines) flexural stiffness distributions.

 16 M. Breccolotti and M. Natalicchi

292 constructed on site by cast on falsework. Its elevation is shown in Fig. 7. In the same figure, the bridge

293 constant cross section with overall width and height of 12.9 m and 3.6 m, respectively are shown. Concrete

294 with compressive strength fc of 28 MPa and 32 GPa elastic modulus was assumed. Further details on the

295 geometrical features and the materials mechanical properties can be found in the cited reference. In the

296 current investigations, the influence of prestressing on the flexural stiffness of the prestressed elements [42]

297 is neglected. The structure was discretized with 105 beam elements having 2.0 m length. It is assumed that

298 7 sensors that measures 3 vertical displacements and 4 rotations are placed on the bridge according to the

299 schematic shown in Fig. 7.

Fig. 7 Second case study [41]: Elevation (top, dimension in m) and cross section (bottom, dimension in mm) of the

prestressed concrete bridge. Blue and black numbers in the elevation view represent sensors and elements numbering,

respectively.

300 The following two damage scenarios were assumed for this case study:

301 - 20% flexural stiffness reduction in element 53;

302 - 10% flexural stiffness reduction in elements 31 and 75.

303 These scenarios simulate damages in the most stressed elements of the bridge. It is useful to point out

304 that each finite element has a length corresponding to 0.95% of the length of the entire bridge and 3.33%

305 of the length of the shortest span.

 Bridge damage detection through combined ILs and WiM devices 17

306 Examples of the products αj · ILj,th determined for the first damaged configuration and for each sensors

307 are shown in Fig. 8 with the same line types used for case study 1.

308 Flexural stiffness distributions resulting from the optimization procedure for the two damage scenarios

309 are shown in Fig. 9. Also in this case the same graphical representation used in the previous case study is

310 used.

5
Sens. 1 Sens. 2 Sens. 3
4 Sens. 4 Sens. 5 Sens. 6
Sens. 7
3
Vertical displacement (mm)

-1

-2

-3

-4

-5
0 50 100 150 200
Coordinate X (m)

Fig. 8 Second case study, products αj · ILj,th for damage scenario 1. Solid lines for displacements, dashed lines for

rotations. For sensors numbering see Fig. 7.

311 5.3 Case study 3

312 The third case study is a composite concrete - steel bridge described in the SCI Publication 140 [43]. It

313 is a continuous box girder bridge with 4 different spans and a total length of 235 m. The general layout

314 and the typical cross section are shown in Fig. 10. Concrete class 40 with compressive strength fc of 40

315 MPa and elastic modulus Ec equal to 31 GPa was assumed for the deck. Steel grade S355 with yielding

316 strength fy of 355 MPa and elastic modulus Es equal to 205 GPa was used for the box girder. The height

317 and the thickness of the steel plates that make up the box sections vary according to the stress distribution

318 along the bridge. The minimum and maximum height of the steel box girder are equal to 1.8 m and 3.2

319 m, respectively. Variations of the heights and thicknesses of flanges and webs along the development of the

320 bridge are indicated in Tab. 1. Entire length of the bridge is divided in consecutive zones having flanges

321 and webs with constant thickness (rows 3, 4 and 5 of Tab. 1). The lengths of these zones are listed in the
 18 M. Breccolotti and M. Natalicchi

322 first row of Tab. 1. Height of the steel box at the beginning and at the end of each zone is shown in row 2

323 of the same table. A linear variation is assumed for the box height inside each zone.

324 The concrete deck is constant along the bridge length. It has a width of 5.45 m and its thickness varies

325 in the transversal direction from 0.23 m at the border to 0.30 m in the central part. Further details on the

326 geometrical features and the materials mechanical properties can be found in the cited reference.

327 Structure has been discretized with 94 beam elements having 2.5 m length corresponding to 1.06% of

328 the length of the entire bridge. It is assumed that 9 sensors that measures 4 vertical displacements and 5

329 rotations are placed on the bridge according to the schematic shown in Fig. 10. The following three damage

330 scenarios were assumed for this case study:

20
Identified
18 True
Undamaged
16

14
Moment of inertia (m4)

12

10

0
0 50 100 150 200
a) Coordinate X (m)
20
Identified
18 True
Undamaged
16

14
Moment of inertia (m4)

12

10

0
0 50 100 150 200
b) Coordinate X (m)

Fig. 9 Second case study, comparison between identified (blue-gray bars), true (damaged, solid black lines) and undamaged

(dashed red lines) flexural stiffness distributions: a) damage scenario 1, b) damage scenario 2.
 Bridge damage detection through combined ILs and WiM devices 19

Zone length 10 27.5 10 7.5 7.5 15 30 15 7.5 7.5 10 27.5 15 12.5 22.5 10

Height 1.8 1.8 2.35 2.75 3.2 2.6 1.8 1.8 2.6 3.2 2.75 2.35 1.8 1.8 1.8 1.8 1.8

Top flange 15 15 25 45 45 25 20 15 25 25 15 15 20÷45 45÷20 15 15

Webs 15 15 15 20 20 15 15 15 20 20 15 15 15÷20 20÷15 15 15

Bottom flange 25 40 50 60 60 50 40 50 60 60 50 50 35÷60 60÷35 40 25

Table 1 Length and height of the steel box girder (m) and thickness (mm) of webs, top and bottom flanges along the

bridge.

331 - 20% flexural stiffness reduction in element 22;

332 - 20% flexural stiffness reduction in element 37;

333 - 10% flexural stiffness reduction in elements 23 and 52.

334 In this case, these scenarios simulate damages in the most stressed elements of the bridge.

335 Products of αj · ILj,th for damage scenario 1 and for different sensors are shown in Fig. 11. Similarly

336 to previous case studies, the distributions of the moment of inertia identified through the optimization

337 procedure for the three damage scenarios are shown in Fig. 12.

338 5.4 Comments on obtained results

339 Results presented in the previous sections show that the proposed procedure is capable of identifying the

340 effective distribution of the bending stiffness along a bridge. It would also be possible to achieve better

341 identifications by performing GA optimizations with larger populations and with a greater number of

342 generations. Irregularities of this distribution, such as those shown in Figs. 6, 9 and 12 or differences with

343 previously obtained distributions can highlight the presence and the location of structural damage. This

344 can support and identify locations for performing visual inspection to better evaluate the damage severity.

345 It must be stated that the proposed procedure should be carried out during periods without relevant

346 wind velocities and ground accelerations. In fact, it is not possible to reliably distinguish the deformations

347 produced by wind and earthquakes from those caused by the vehicular traffic. For this purpose, anemometers

348 and accelerometers should be used to check the environmental conditions and validify these values.
 20 M. Breccolotti and M. Natalicchi

349 The advantages deriving from the combined use of different quantities with which build the influence

350 lines can be inferred from Tab. 2. In this table errors in the bending stiffness evaluations coming from the

351 above-mentioned simulations and from further simulations where displacement and rotation influence lines

352 were used separately are demonstrated. In detail, root-mean-square-errors (RMSE) were calculated taking

353 into account the relative differences between the identified cross section moments of inertia Ij,id and the

354 true ones Ij,t :

N  2
X (Ij,id − Ij,t )
RM SE = 100 · (10)
Ij,t
j=1

355 where N is the number of beam elements used to model the bridge.

356 Furthermore, absolute errors committed in the evaluation of the flexural rigidity of the damaged elements

357 are reported in the table (Err1 and Err2). From the data analysis shown in the table, it can be seen that:

358 - use of combined influence lines provided in most analysis, produce a more precise assessment of the

359 damage. In the results there are few exceptions probably due to the method of random generation of

360 stiffness distributions used by genetic algorithm;

361 - the average values of the RMSE and of the errors incurred in relation to the damaged elements are

362 always smaller for the analysis carried out with combined influence lines.

Fig. 10 Third case study [43]: Elevation (top, dimension in m) and cross section (bottom, dimension in mm) of the

composite bridge. Blue and black numbers in the elevation view represent sensors and elements numbering, respectively.
 Bridge damage detection through combined ILs and WiM devices 21

363 Thus, the combined use of displacement and rotation influence lines makes it possible to determine a

364 more reliable structural damage identification compared to that achieved by using displacement and rotation

365 ILs separately. Nevertheless, this method also has certain disadvantages. In fact, the use of combined

366 influence lines attracts higher cost, associated with monitoring system and greater computational time

367 to identify the unknown parameters with the optimization algorithm. Fortunately, these drawbacks are

368 of modest relevance. Higher cost of the monitoring system is limited as it is linked to the number of

369 sensors, hence system and hardware acquisition for post-processing of data remains practically unchanged.

370 The major computational time is only relevant for the first identification, when all the parameters of the

371 structure are largely unknown. Once the first identification is completed, the evidence of damage can be

372 investigated, reaching the same value as the fitness function, in a much smaller time. For these reasons, the

373 use of combined influence lines can be considered generally beneficial.

374 6 Conclusions

375 The work described herein proposes a new damage identification method for simply supported and con-

376 tinuous bridges that exploits weigh-in-motion devices and the combined use of different types of influence

377 lines. WiM apparatus placed at the beginning and at the end of the bridge provide information on vehicle

378 axles, weight and speed. Displacements at each midspan and rotations at each support are measured by

379 appropriate sensors. Sampling frequency is chosen to obtain a suitable spatial resolution for typical vehicles

380 speed traveling on the bridge. An influence line can be drawn for each sensor and for each vehicle traveling

25
Sens. 1 Sens. 2 Sens. 3
20 Sens. 4 Sens. 5 Sens. 6
Sens. 7 Sens. 8 Sens. 9
15
Vertical displacement (mm)

10

-5

-10

-15

-20
0 50 100 150 200 250
Coordinate X (m)

Fig. 11 Third case study, products αj · ILj,th for damage scenario 1. Solid lines for displacements, dashed lines for

rotations. For sensors numbering see Fig. 10.

22 M. Breccolotti and M. Natalicchi

7
Identified
True
6 Undamaged

Moment of inertia (m4)

4

0
0 50 100 150 200
a) Coordinate X (m)
7
Identified
True
6 Undamaged

5
Moment of inertia (m4)

0
0 50 100 150 200
b) Coordinate X (m)
7
Identified
True
6 Undamaged

5
Moment of inertia (m4)

0
0 50 100 150 200
c) Coordinate X (m)

Fig. 12 Third case study, comparison between identified (blue-gray bars), true (damaged, solid black lines) and undamaged

(dashed red lines) flexural stiffness distributions: a) damage scenario 1, b) damage scenario 2, c) damage scenario 3.
 Bridge damage detection through combined ILs and WiM devices 23

Case Damage Displacement Rotation Disp. and Rot.

study Scenario RMSE Err1 Err2 RMSE Err1 Err2 RMSE Err1 Err2

1 1 3.25 4.30 - 2.95 3.14 - 2.02 1.63 -

2 1 2.29 4.20 - 2.99 3.28 - 1.90 1.23 -

2 2 2.21 3.97 4.87 1.95 4.16 4.35 1.84 4.45 1.74

3 1 3.08 0.75 - 2.45 0.16 - 2.81 0.35 -

3 2 2.52 4.57 - 3.03 6.19 - 2.58 6.21 -

3 3 2.63 2.64 0.35 2.98 1.15 1.92 2.78 1.96 0.33

mean 2.66 3.40 2.61 2.72 3.01 3.13 2.32 2.64 1.03

Table 2 Root-mean-square-errors (RMSE, %) and absolute errors (%) in damaged elements stiffness evaluation for different

case studies and different damage scenarios.

381 on the bridge. It is considered that damage modifies displacement and rotation ILs since it reduces the

382 flexural stiffness of the beam elements. Displacements and rotations measured experimentally are compared

383 with those obtained by theoretical ILs and FE analysis assuming the nominal bending stiffness distribu-

384 tion or that obtained from previous identification procedures. An optimization technique that uses genetic

385 algorithms updates the hypothetical flexural stiffness distribution of the FE model in order to minimize a

386 least squares based fitness function between the theoretical and the experimental displacements and rota-

387 tions. The proposed method has been validated with application to three case studies: a simply supported

388 prestressed concrete I-beam bridge, a continuous three-span prestressed RC box bridge and a four-span

389 composite steel-concrete bridge, found in the literature. Principal findings of the investigation are summed

390 up as follows:

391 – Proposed method results in enablement to identify single or multiple damaged elements with dimension

392 equal to 0.95% of the total bridge length and stiffness reduction as small as 10%;

393 – Numerical simulations allowed emphasizing the improvement in bridge damage identification achieved

394 by the combined use of different types of influence lines;

 24 M. Breccolotti and M. Natalicchi

395 – Maximum root-mean-square-errors decreased from 3.25% and 3.03%, obtained respectively with only

396 displacements and rotations measurements, to 2.78% with the use of combined displacements and ro-

397 tations measurements;

398 – – Maximum errors in determination of flexural stiffness of damaged elements decreased from 4.87% and

399 6.19%, obtained respectively with only displacements and rotations measurements, to 4.45% with the

400 use of combined displacements and rotations measurements.

401 It can, thus, be noted that relevant advantages can be gained in structural monitoring by means of

402 combining information from different data sources. This approach, which is already in use in other fields (e.g.

403 SonReb tests on hardened concrete) it is not yet fully exploited in the field of structural monitoring. In the

404 proposed procedure the combined use of weight-in-motion devices, displacement and rotation influence lines

405 make it possible to produce a more reliable and precise structural identification. The proposed monitoring

406 scheme can be used for remote detection of ongoing structural changes and issuing of early warning notices

407 without the need for visual inspections and with a very small number of sensors. The procedure can also

408 be used for different types of bridges (cable-stayed, integral-abutment, etc.). In these cases, however, it

409 will be necessary to identify the optimal positions where place the specific sensors. For example, in an

410 integral-abutment bridge it could be of little benefit to have inclinometers on the shoulders. In fact, due to

411 the structural scheme, the very modest rotations expected at the supports could be outside the measuring

412 range of the sensors currently available on the market.

413 Conflict of interest

414 The authors declare that there is no conflict of interest.

415 References

416 1. Balageas D. Introduction to Structural Health Monitoring. In: Structural Health Monitoring. John

417 Wiley & Sons, Ltd, 2010. Chap. 1:13–43.

418 2. Rytter A. Vibrational Based Inspection of Civil Engineering Structures. Fracture and Dynamics, Dept.

419 of Building Technology and Structural Engineering, Aalborg University 1993;R9314.

420 3. Sun H and Büyüköztürk O. Optimal sensor placement in structural health monitoring using discrete

421 optimization. Smart Materials and Structures 2015;24.

 Bridge damage detection through combined ILs and WiM devices 25

422 4. Casas JR and Cruz PJS. Fiber Optic Sensors for Bridge Monitoring. Journal of Bridge Engineering

423 2003;8:362–73.

424 5. He WY, Ren WX, and Zhu S. Damage detection of beam structures using quasi-static moving load

425 induced displacement response. Engineering Structures 2017;145:70 –82.

426 6. Robert-Nicoud Y, Raphael B, Burdet O, and Smith I. Model identification of bridges using measure-

427 ment data. Computer-Aided Civil and Infrastructure Engineering 2005;20:118–31.

428 7. Nguyen V, Schommer S, Maas S, and Zürbes A. Static load testing with temperature compensation

429 for structural health monitoring of bridges. Engineering Structures 2016;127:700–18.

430 8. Zhao J and Dewolf J. Dynamic monitoring of steel girder highway bridge. Journal of Bridge Engineer-

431 ing 2002;7:350–6.

432 9. Gentile C and Bernardini G. Output-only modal identification of a reinforced concrete bridge from

433 radar-based measurements. NDT and E International 2008;41:544–53.

434 10. Magalhães F, Cunha A, and Caetano E. Vibration based structural health monitoring of an arch

435 bridge: From automated OMA to damage detection. Mechanical Systems and Signal Processing

436 2012;28:212–28.

437 11. Bayraktar A, Altunişik A, and Türker T. Structural condition assessment of Birecik highway bridge

438 using operational modal analysis. International Journal of Civil Engineering 2016;14:35–46.

439 12. Bayraktar A, Altunişik A, and Türker T. Structural health assessment and restoration procedure of

440 an old riveted steel arch bridge. Soil Dynamics and Earthquake Engineering 2016;83:148–61.

441 13. Altunişik A, Kalkan E, Okur F, Karahasan O, and Ozgan K. Finite-Element Model Updating and

442 Dynamic Responses of Reconstructed Historical Timber Bridges using Ambient Vibration Test Results.

443 Journal of Performance of Constructed Facilities 2020;34.

444 14. Breccolotti M, Materazzi A, and Venanzi I. Identification of the nonlinear behaviour of a cracked

445 RC beam through the statistical analysis of the dynamic response. Structural Control and Health

446 Monitoring 2008;15:416–35.

447 15. Breccolotti M and Materazzi A. Identification of a non-linear spring through the Fokker-Planck equa-

448 tion. Probabilistic Engineering Mechanics 2008;23:146–53.

449 16. Fiore A and Marano G. Serviceability Performance Analysis of Concrete Box Girder Bridges Under

450 Traffic-Induced Vibrations by Structural Health Monitoring: A Case Study. International Journal of

451 Civil Engineering 2018;16:553–65.

 26 M. Breccolotti and M. Natalicchi

452 17. Colombo S, Main I, and Forde M. Assessing damage of reinforced concrete beam using ”b-value”

453 analysis of acoustic emission signals. Journal of Materials in Civil Engineering 2003;15:280–6.

454 18. Nair A and Cai C. Acoustic emission monitoring of bridges: Review and case studies. Engineering

455 Structures 2010;32:1704–14.

456 19. Ploix MA, Garnier V, Breysse D, and Moysan J. NDE data fusion to improve the evaluation of concrete

457 structures. NDT & E International 2011;44:442 –448.

458 20. Breccolotti M and Bonfigli MF. I-SonReb: an improved NDT method to evaluate the in situ strength

459 of carbonated concrete. Nondestructive Testing and Evaluation 2015;30:327–46.

460 21. Chen ZW, Cai QL, and Zhu S. Damage quantification of beam structures using deflection influence

461 lines. Structural Control and Health Monitoring 2018;25.

462 22. Zeinali Y and Story B. Impairment localization and quantification using noisy static deformation

463 influence lines and Iterative Multi-parameter Tikhonov Regularization. Mechanical Systems and Signal

464 Processing 2018;109:399–419.

465 23. Zhang S and Liu Y. Damage detection in beam bridges using quasi-static displacement influence lines.

466 Applied Sciences (Switzerland) 2019;9.

467 24. Le N, Thambiratnam D, Nguyen A, and Chan T. A new method for locating and quantifying damage

468 in beams from static deflection changes. Engineering Structures 2019;180:779 –792.

469 25. Zhou Y, Di S, Xiang C, Li W, and Wang L. Damage Identification in Simply Supported Bridge Based

470 on Rotational-Angle Influence Lines Method. Transactions of Tianjin University 2018;24:587–601.

471 26. Alamdari M, Kildashti K, Samali B, and Goudarzi H. Damage diagnosis in bridge structures using

472 rotation influence line: Validation on a cable-stayed bridge. Engineering Structures 2019;185:1–14.

473 27. Huseynov F, Kim C, O’Brien E, Brownjohn J, Hester D, and Chang K. Bridge damage detection

474 using rotation measurements – Experimental validation. Mechanical Systems and Signal Processing

475 2020;135.

476 28. Wang NB, He LX, Ren WX, and Huang TL. Extraction of influence line through a fitting method

477 from bridge dynamic response induced by a passing vehicle. Engineering Structures 2017;151:648–64.

478 29. Cantero D and González A. Bridge Damage Detection Using Weigh-in-Motion Technology. Journal of

479 Bridge Engineering 2015;20.

480 30. Chen Z, Yang W, Li J, Yi T, Wu J, and Wang D. Bridge influence line identification based on adaptive

481 B-spline basis dictionary and sparse regularization. Structural Control and Health Monitoring 2019;26.
 Bridge damage detection through combined ILs and WiM devices 27

482 31. Zheng X, Yang DH, Yi TH, and Li HN. Development of bridge influence line identification methods

483 based on direct measurement data: A comprehensive review and comparison. Engineering Structures

484 2019;198.

485 32. Goldfeld Y and Klar A. Damage Identification in Reinforced Concrete Beams Using Spatially Dis-

486 tributed Strain Measurements. Journal of Structural Engineering 2013;139.

487 33. Wu B, Wu G, Yang C, and He Y. Damage identification method for continuous girder bridges based

488 on spatially-distributed long-gauge strain sensing under moving loads. Mechanical Systems and Signal

489 Processing 2018;104:415–35.

490 34. Oskoui E, Taylor T, and Ansari F. Method and monitoring approach for distributed detection of

491 damage in multi-span continuous bridges. Engineering Structures 2019;189:385–95.

492 35. O’Brien EJ, Quilligan MJ, and Karoumi R. Calculating an influence line from direct measurements.

493 Proceedings of the Institution of Civil Engineers - Bridge Engineering 2006;159:31–4.

494 36. Federal Highway Administration. FHWA-HRT-16-024, LTBP Program’s Literature Review on Weigh-

495 in-Motion Systems. U.S. Department of Transportation, 2016:1–36.

496 37. Group K. Weigh In Motion. url: https://www.kistler.com/en/solutions/traffic-solutions/weigh-in-motion/.

497 (accessed: 20.05.2021).

498 38. Wu H, Zhao H, Liu J, and Hu Z. A filtering-based bridge weigh-in-motion system on a continuous

499 multi-girder bridge considering the influence lines of different lanes. Frontiers of Structural and Civil

500 Engineering 2020;14:1232–46.

501 39. Chou JH and Ghaboussi J. Genetic algorithm in structural damage detection. Computers & Structures

502 2001;79:1335 –1353.

503 40. Naaman AE. Prestressed concrete analysis and design: fundamentals. Techno Press 3000, Ann Arbor,

504 Michigan, 2004.

505 41. National Academies of Sciences, Engineering, and Medicine. Development of Design Specifications and

506 Commentary for Horizontally Curved Concrete Box-Girder Highway Bridges. The National Academies

507 Press. Washington, DC, 2008.

508 42. Breccolotti M. On the Evaluation of Prestress Loss in PRC Beams by Means of Dynamic Techniques.

509 International Journal of Concrete Structures and Materials 2018;12.

510 43. Steel Construction Institute. SCI Publication P140, Design Guide for Composite Box Girder Bridges

511 (Second edition). SCI, 2004:63–116.