You are on page 1of 16

Engineering Structures 166 (2018) 46–61

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Fatigue life evaluation of a composite steel-concrete roadway bridge T


through the hot-spot stress method considering progressive pavement
deterioration

Guilherme Alencara, , Abílio M.P. de Jesusb, Rui A.B. Calçadaa, José Guilherme S. da Silvac
a
CONSTRUCT, Faculty of Engineering (FEUP), University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
b
LAETA-INERGI, Faculty of Engineering (FEUP), University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
c
Structural Engineering Department, State University of Rio de Janeiro, Rua São Francisco Xavier, No. 524, Rio de Janeiro, Brazil

A R T I C LE I N FO A B S T R A C T

Keywords: Steel and composite steel-concrete bridges are subjected to random traffic loads along their life cycle which
Distortion-induced fatigue generate significant dynamic impacts. The road-roughness of asphalt pavements is one of the most important
Structural dynamics aspects that contributes to the significant increase of the stress amplitudes and hence to serious load-induced
Fatigue behaviour fatigue concerns. In this context, welded joints are well known as the weakest points in bridges, since they are
Web-gap cracking
prone to stress concentrations leading to initiation of fatigue cracks being in current design codes evaluated
Road-roughness
using primarily the nominal stress method (NSM). However, a more accurate stress definition, which considers
Hot-spot stress method
Composite bridges the complexity of the stress field at the welds, becomes necessary mainly in old roadway steel bridges that are
often subjected to out-of-plane stresses. In this paper, the hot-spot stress method (HSM) is used to evaluate the
fatigue life of a welded joint subjected to distortion induced-fatigue, considering the vehicle speeds and a
progressive deterioration model for the road pavement. The welded joint was modelled by solid elements and
was integrated with a 3D dynamic bridge model using the sub modelling technique. Besides the importance of
considering the bridge-vehicle dynamic interactions with the pavement road-roughness, the dynamic amplifi-
cation effects on local stresses and the relatively high scatter found in the fatigue lives considering global and
local approaches show that a detailed local stress definition is fundamental to evaluate the fatigue performance
of existing roadway bridges. The effects of the annual traffic increase rate on the fatigue life are also discussed.

1. Introduction cracking is the result of out-of-plane distortion or other unanticipated


secondary stresses at fatigue-sensitive details. For this reason, infra-
During the life cycle of a bridge, dynamic impacts due to random structure maintenance bodies of many countries are concerned with the
traffic loads and deteriorated road surface conditions can induce serious assessment of distortion-induced fatigue cracks in old steel bridges with
fatigue issues for bridge components. For roadway bridges, Nishikawa heavy traffic, in order to ensure their serviceability. This fatigue-sen-
et al. [1] have reported that such fatigue cracks are mainly due to both sitive detail is present in the investigated case study bridge of the
the relative deflection of main girders and the reinforced concrete deck present work (see Fig. 1a), hence the present work is concerned with
by the wheel load of heavy traffic, becoming more frequent in roadway the fatigue cracks that may develop at weld ends located at the con-
steel girder bridges. In fact, according to Fisher & Roy [2] until the junction of the girder web and the transverse stiffener.
1970’s the steel bridge construction practice had not used welding in Basically, the fatigue requirement for existing steel bridges can be
the transverse web stiffeners to the tension flange of welded I-girders stated as the fatigue cycles induced by the traffic load being less than
due to the concern that a fatigue crack initiating in the flange would the available fatigue life of the bridge details. In this context, the pa-
lead to brittle fracture, leaving therefore a web gap between the stif- vement road-roughness has proven to have a significant effect on the
fener and the tension flange. Afterwards, this detail revealed to be one dynamic response of roadway bridges [4,5]. In some cases, poor road-
of the main causes of distortion-induced cracking, which is caused by roughness levels could lead to even higher amplification factors than
secondary stresses generated by partially restrained deformation. that presumed by the design load models, being as high as 90% of the
Connor & Fisher [3] have estimated that nearly 90% of all fatigue static effects of the moving loads [6]. However, approaches that intend


Corresponding author.
E-mail address: guilherme.alencar@fe.up.pt (G. Alencar).

https://doi.org/10.1016/j.engstruct.2018.02.058
Received 25 September 2017; Received in revised form 16 January 2018; Accepted 19 February 2018
Available online 28 March 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
G. Alencar et al. Engineering Structures 166 (2018) 46–61

bf1
tf1
tw
d
Steel girder
(shell element)
tf2
Concrete deck bf2
(solid element)

a) b)
Fig. 1. Simply supported steel-concrete roadway bridge: (a) overall view, (b) steel girder geometric characteristics (refer to Table 1 for dimensions).

to predict fatigue lives based on the use of a unique road-roughness progressive deterioration of the pavement road-roughness. The local
level for the entire bridge lifecycle, can lead to unrealistic results or stresses are evaluated by the use of the hot-spot stress method [20]
over-conservative lives whether one adopts an excellent or poor through the sub modeling technique in a specific transverse stiffener
roughness level, respectively. Thus, the influence of the progressive welded to the web and subjected to distortion-induced fatigue.
degradation of the road surface roughness in a vehicle-bridge interac-
tion model was considered in the work of Zhang & Cai [7,8], clearly 2. Case study
conducting to more realistic estimations. Zhang et al. [9] have dis-
cussed the effects of adopting a nonlinear fatigue damage accumulation In this study, a typical simply supported steel bridge designed in
model for fatigue assessment of existing bridges considering the pro- accordance with AASHTO LRFD bridge design specification [21] is
gressive road surface deterioration and random dynamic vehicle loads. adopted as an example of illustration of the proposed fatigue assess-
Many studies have been conducted to investigate the effect of truck ment methodology. The selected case study is a good representative of
overloading in bridges, namely to evaluate dynamic impacts in typical real simply-supported steel bridges, corresponding to a composite
highway bridges through long-term traffic measurement [10] and to steel–concrete roadway bridge with a straight axis, and spanning
assess the fatigue design of steel bridges considering a pavement de- 13.0 m by 40.0 m, see Fig. 1 [22]. The structural system consists of four
terioration model [11]. Wang et al. [12] found that the road surface steel girders and a 0.225 m thick concrete slab, as shown in Fig. 2a. The
condition has a great impact on the equivalent number of stress cycles steel sections considered are welded wide flanges made with A588 steel
due to vehicle dynamic loading. Paraskeva et al. [13] have examined with 350 MPa yield strength and 485 MPa ultimate tensile strength. A
the seismic response of a vehicle-bridge interacting system under ver- 2.05 × 105 MPa Young’s modulus with 0.3 Poisson’s ratio and 7850 kg/
tical earthquake excitation, combined with the influence of deteriorated m3 material density was adopted for the steel girders. The concrete slab
road roughness conditions in the acceleration response of vehicles. possesses 2500 kg/m3 density, 25 MPa compression strength and
Deng et al. [14] have studied the effect of the pavement maintenance 3.05 × 104 MPa of Young’s modulus, with 0.2 Poisson’s ratio. Tables 1
cycle on the fatigue reliability index of simply-supported steel I-girder and 2 depict the geometrical characteristics of the used steel plates in
bridges under dynamic vehicle loading. the bridge design. In order to prevent web buckling, steel plate stif-
For standard details usually adopted in bridge design codes, the use feners are welded along the steel girders with 1880 mm spacing at the
of a fatigue strength based on nominal stresses would lead to more span-sections and 1200 mm spacing at the support-sections. Steel plate
reliable life estimations only if the actual detail remains more or less stiffeners were also adopted above the girder supports. Two different
faithful to the original small-scale fatigue tested specimens, following cross sections were adopted along the longitudinal composite beams,
the principle of similitude. On the other hand, for complex welded designated as support cross-section and span cross-section, as presented
joints and for structural details subjected to secondary stresses, where it in Fig. 2b. The structure comprises cross diaphragms in the form of steel
is difficult to define a nominal stress, a more detailed computation of trusses, as shown in Fig. 3. The diaphragms are made of equal angle
the stress field at the connection becomes very relevant. This is leading profiles with 10 mm wall thickness.
to a growing adoption of local stress approaches for civil engineering
structures [15,16], focusing on local stress definitions at the fatigue 2.1. Bridge global finite element model
crack initiation points, in order to mitigate the need of structural stress
concentrations in the design phase, or to better assess the existing The computational model developed for the dynamic analysis of the
structural details of old steel bridges. Eurocode 3 [17] provides the so- composite bridge adopted the usual mesh refinement techniques pre-
called hot-spot stress method in annex B as an available local approach, sent in finite element method simulations implemented in the ANSYS
but giving few instructions, although a review was proposed recently program [19]. The girder top and bottom flanges, the girder web and
[18]. the longitudinal and vertical stiffeners were represented by shell finite
One of the main difficulties in the application of local stress methods elements (SHELL63). The bridge concrete slab was simulated by solid
is the requirement of finite element models with fine meshes, which is finite elements (SOLID45). The transverse steel bracing system was si-
problematic to achieve in the same single model of the bridge due to mulated by beam finite elements (BEAM44). The final computational
high computational costs and the big scale of the involved structure. For model adopted used 28,875 nodes, 21,020 shell elements, 4600 solid
this reason, a shell-to-solid sub modeling technique is generally em- elements and 656 beam elements, which resulted in a numeric model
ployed [19], in order to transfer the dynamic displacements field from with 135,844 degrees of freedom. The strain compatibility between the
the global model to the local one. Therefore, this paper presents a solid elements (concrete slab) and the shell elements (steel plate gir-
framework for fatigue assessment of existing bridges in their lifetime ders) was guaranteed by coupling the corresponding degrees of
serviceability considering the random effects of vehicle speeds and the freedom, simulating the composite bridge decks’ full interaction. The

47
G. Alencar et al. Engineering Structures 166 (2018) 46–61

13000

175
1178

75

225

3178
300
2000

2L 127x127x10
1250 3500 3500 3500 1250
13000
a)
8100 11900 11900 8100

G1
3500 3500 3500

Support cross-section Span cross-section


G2

G3

G4
5629 5629 5628 5628 5628 5629 5629
40000

b)
Fig. 2. Geometry and dimensions of the steel-concrete bridge section: (a) bridge section at support, (b) bridge steel girders top view (units in millimetres).

damping ratio is assumed to be 0.5%, as stated by EN 1991-2 [23] for Table 2


steel and composite steel-concrete bridges. Figs. 4 and 5 illustrate the Geometrical characteristics of the web plate stiffeners (units in millimetres).
bridge finite element model. The composite bridge natural frequencies Location Height Width Thickness
were determined with the aid of the numerical simulations. The asso-
ciated composite bridge main global vibration modes are shown in Support stiffeners 1925 200 22.0
Transverse stiffeners 1845 450 12.5
Fig. 6.
Longitudinal stiffeners 12.5

3. Fatigue assessment
The stress collective is calculated using the Rainflow cycle counting
3.1. Global SN approach method specified in ASTM E 1049–85 [25] for the traffic simulated with
dynamic numerical analysis. The above fatigue assessment guidelines
The fatigue assessment presented in this paper is based on were implemented into a software [26], following the rules presented in
Palmgren-Miner linear damage summation, Dd [24], as presented in Eq. Eurocode 3 [17].
(1). Current fatigue assessment of steel structures in existing steel
standards are based on the SN curves approach, with typical structural
details organized into different fatigue strength categories. Each detail 3.2. Local approach: the hot-spot stress method
category is represented by the corresponding SN curve, where the fa-
tigue strength Δσ is a function of the number of cycles, NR, both re- Fatigue assessment based on local stresses is often used when no
presented in logarithmic scales. Then, the fatigue resistance is defined specific detail category is available in the nominal approach.
as NR = C.Δσ−m, where NR is the number of cycles to failure for stress Recommendations and procedures are found in IIW [20], both for nu-
range, Δσ; C is the constant representing the influence of the structural merical approaches and field measurements. This paper adopts the
detail and m is the slope of the mean test results line, both constants extrapolated stresses at the plate surface as parameter for the fatigue
being related to material and geometry properties of the detail. life assessment (see Fig. 7). In the literature, such extrapolated stress is
often referred to as structural stress, geometric stress, or structural hot-
k
n E1 n E2 n E3 n Ei spot stress. To avoid any confusion in the discussions in this paper, all
Dd = + + + …= ∑
NR1 NR2 NR3 N Ri (1) surface extrapolation-based stresses are referred to as hot-spot stress
i=1
throughout this paper. The hot-spot stress describes the macrostructural

Table 1
Geometrical characteristics of the girder steel plates (units in millimetres).
Section location Height (d) Top flange width (bf1) Top flange thickness (tf1) Bottom flange width (bf2) Bottom flange thickness (tf2) Web thickness (tw)

Support cross-section 2000 450 25 450 50 9.5


Span cross-section 2000 500 25 670 50 9.5

48
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Longitudinal stiffeners Transverse stiffeners

10
127

127
(a) Cross-bracing B1:
L127x127x10
10
Support
127
stiffners

127
(b) Cross-bracing B2:
2L127x127x10

B2 B1

Fig. 3. Cross-diaphragms sections and illustration of the web plate stiffeners.

26,276 elements
28,875 nodes
y
x
z

a) b)
Fig. 4. Finite element model of the roadway bridge: (a) overall view of the bridge model and (b) a close-up view at the model.

Solid elements
(SOLID45)

Shell elements
(SHEL63)

Beam elements
(BEAM44)

Fig. 5. Finite element modeling of the section at supports.

Fig. 6. Main global vibration modes of the


bridge obtained using the finite element
model.

a) f01 = 2.97 Hz b) f02 = 3.67 Hz c) f03 = 9.69 Hz d) f04 = 10.87 Hz

behaviour excluding local notch effects (e.g. weld toe radius). The hot- where t stands for the plate thickness. Optionally, a linear extrapolation
spot stress is usually evaluated based on extrapolation of stresses at using two points at different distances can also be used [20], but is not
prescribed distances adjacent to the notch, usually the weld toe. In recommended for high bending stress fields.
Fig. 7, the hot-spot stress is obtained at point O based on a quadratic
σHS = 2.52·σ0.4t −2.24·σ0.9t + 0.72·σ1.4t (3)
extrapolation of reference points 0.4t, 0.9t and 1.4t according to Eq. (3),

49
G. Alencar et al. Engineering Structures 166 (2018) 46–61

z
Notch stress modeling technique to reduce computation costs and obtain reliable
Hot-spot stress
estimations of the stress gradient at the weld toe.
Nominal stress

Weld toe
x, y 4. Vehicle-bridge dynamic interaction simulation
O

1.4t
0.4t 0.9t The vehicle-bridge dynamic interaction is performed in a beam-like
2D model in order to obtain the equivalent nodal forces of the vehicle-
t

Reference points
bridge interaction effect. The 2D mathematical model assumes a finite
element representation of the beam-like girder bridge with 40 m span
Fig. 7. Typical representation of the hot-spot stress extrapolation to the weld (refer to Table 3 for detailed characteristics of the model) and the ve-
toe. hicle simulation uses concentrated parameters of mass blocks, springs
and damping devices [6].
3.3. Fatigue prone detail and local finite element model The mathematical model simulates the bridge structure and the
vehicle series as a system, the vehicle-bridge system. The vehicle and
A crucial stage in fatigue assessment of steel bridges is to identify the bridge deck are assembled together to form just one system. The
the most fatigue prone locations, subjected to the major stress ampli- equations of motion for the vehicle and the bridge are derived based on
tudes. The fatigue assessment of the investigated structural detail in this the following matrices, Eqs. (4) and (5):
work, according to steel bridge design codes [17], should use the
[MV ]{d¨v} + [Cv]{d v̇ } + [Kv]{d v} = {FvG} + {Fc } (4)
nominal stresses at the transverse stiffener assessed in the longitudinal
direction and considering a FAT80 class according to global SN ap-
proach. However, there is usually some difficulty to determine the [Mb]{d¨b} + [Cb]{dḃ } + [Kb]{db} = {Fb} (5)
nominal stresses at the web near to the transverse stiffener, due to the
high secondary stress normally involved, leading to a less accurate fa- where [Mv] is the mass matrix, [Cv] is the damping matrix, and [Kv] is
tigue damage assessment. In this situation, the hot-spot stress method the stiffness matrix that are obtained by considering the equilibrium of
can be used, for which the same welded detail possess the FAT100 class, the forces and moments of the system; {FvG} is the self-weight of ve-
type ‘a’ fatigue strength [20]. hicle; {Fc} is the vector of wheel-road contact forces acting on the ve-
Transverse stiffeners are welded with different spacing along the hicle; [Mb], [Cb] and [Kb] are respectively the mass, damping and
web girder, being 1.2 m for the support sections, and 1.88 m for the stiffness matrix of the bridge; and {Fb} is the global nodal loading
span sections. The most critical transverse stiffeners weld ends in the vector due to the interaction between the bridge and the vehicle. The
present bridge case study are clearly located at the midspan in a region effect of the pavement roughness is introduced in the vehicle-bridge
of positive moment. For the slow lane statically loaded by the con- system equations of motion as a load vector analogous to what would
sidered two-axle truck, being 70 kN the 1st axle and 130 kN the 2nd be considered if the vehicle was subjected to a base movement equal to
axle (Table 4.7 of EN 1991-2 [23]), it can be computed a differential the irregular pavement profile, Eq. (6), as stated by Wang & Huang
vertical displacement at the midspan between the external girders of [27]:
almost 12.7 mm (Fig. 8). This eccentric traffic loading is responsible to
the in-plane and out-of-plane bending of the web near to the transverse ̇
{Fb} = −{Fc } = kpi dpyi + cpi dpi (6)
stiffener weld end located in the most external girder, G4 (Fig. 2b).
Therefore, these findings justified the sub modeling of the transverse In which kpi = tire stiffness of the ith axle; cpi = tire damping
stiffener detail at this location (Fig. 9). The finite element mesh of the coefficient of the ith axle; dpyi = relative displacement between the ith
sub model adopts solid elements with quadratic shape functions (the so- axle and the bridge = yai – (– usri) – (– ybi); yai = vertical displacement
called 20-noded SOLID186 element) and it follows the rules indicated of the ith axle; usri = road surface roughness under the ith axle (positive
in IIW [20] for a fine mesh and a quadratic extrapolation with three upward); and ybi = bridge vertical displacements under the ith axle
extrapolation points. In order to take into account the local stiffness at (positive upward). Following this procedure, all the interaction vector
the critical points of the sub model, the fillet welds were completely forces can be generated in the 2D model, which is solved by direct
modelled, and no structural imperfections or misalignments were con- integration of the coupled equations of motions, Eqs. (4) and (5). These
sidered, since all stress concentration due to the former are already equivalent nodal forces are imported to the 3D complex finite element
included in the hot-spot stress determination and any allowance for model in ANSYS as moving loads. Next, the dynamic response of the 3D
misalignment is covered in hot-spot SN curves within a limit of 5% finite element model is assessed using the modal superposition method
stress magnification [20]. As can be seen in Fig. 9d, the sub model of [28,29], for which is generally accepted that it is dependent on the
the investigated detail has much more nodes and elements than the modal forces vector over time presented in Eq. (7):
global FE model of the bridge, which justifies the use of the sub

Slow lane

12,7 mm

a) b)
Fig. 8. (a) Two axles fatigue truck positioned at the slow lane and (b) the corresponding distortion effect due to differential displacement of steel girders due to traffic
static loading.

50
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Fig. 9. Distortion-induced fatigue and local sub


modeling: (a) web gaps in transverse stiffener;
(b) investigated web stiffener and its corre-
sponding FAT class for nominal stress approach
[17]; (c) structural detail in the global finite
element model (shell elements) and (d) struc-
tural detail in the local finite element model
(solid elements).

Table 3 nodes of the local finite element model. These displacements are ex-
Characteristics of the simply-supported beam-like 2D model. changed and interpolated between the global and the local model using
No. Nodes No. Elements Concentrated mass Equivalent bridge f01 (Hz)
the so-called cut boundary interpolation files [19] (also known as CBDO
values (kg) inertia (m4) files), being required a different CBDO file for each vibration mode used
in the dynamic solution. Next, static analyses are performed with the
Support Span Support Span local finite element model for each group of imposed boundary di-
section section section section
mensionless modal displacements j, in order to obtain the corre-
nodes nodes
sponding three nodal modal stress components ([ σx], [σy] and [σxy]) of
41 40 8960 9340 0.336 0.397 2.85 the three nodes at the plate surface used to compute the extrapolated
hot-spot stress at the weld toe. It is highlighted that the nodal stresses
are actually obtained from an interpolation of the element stresses
⎡ Φ1,1 ⋯ Φ1,k ⋯ Φ1,K ⎤
Fb1 (t ) ⎤ evaluated at the element integration points [19]. The modal stress
⎡ F1…(t ) ⎤ ⎢ ⋮ ⋱ ⋮ ⋱ ⋮ ⎥⎡ …
⎢ ⎥ ⎢ ⎥ component [σz] perpendicular to the plane was neglected, since four
F (t ) = ⎢ Fj (t ) ⎥ = ⎢ Φj,1 ⋯ Φj,k ⎥
⋯ Φj,K ·⎢ Fbk (t ) ⎥ elements were discretized through the plate thickness and it was con-
⎢ ⎥
⎢ … ⎥ ⎢ ⋮ ⋱ ⋮ ⋱ ⋮ ⎥⎢ … ⎥ sidered that at the free surface there is a plane stress state.

⎣ F (t ) ⎥
⎦ ⎢ ΦJ ,1 ⎢
⎣ FbK (t ) ⎥
J
⎣ ⋯ ΦJ ,k ⋯ ΦJ ,K ⎥

⎦ (7) The stresses time-histories components of each traffic event, that is,
σx(t), σy(t) and σxy(t) were computed as the sum of the corresponding
where Φj,k is the vertical modal displacement exported from ANSYS of
nodal modal stress components multiplied by the generalized modal
the degree of freedom k, for the mode of vibration j, and Fb(t) is the
coordinates of every truck passages obtained from the mode super-
loading vector imported from the beam-like 2D model, which includes
position solution of the 3D finite element model. Different road-
the vehicle self-weight and the vehicle road-roughness interaction
roughness levels and vehicle speeds are taken into account through the
forces. It is important to highlight that the implemented approach only
vector interaction forces imported from the beam-like 2D model. These
needs to use the vertical modal displacements of the degrees of free-
time-histories are extrapolated by the hot-spot method to the weld toe,
doms of the nodes with applied loads [29]. For the present case, these
and then the stress components are used to obtain the principal stress
nodes correspond with those in contact with the tires during the vehicle
time-history by finding the eigenvalues of the symmetric tensor.
crossing. Provided the roadway traffic lanes are well defined, the nodes
Finally, the principal stresses time-histories are used to compute the
in contact with the vehicle tires can be easily pre-determined in the 3D
fatigue damage, as recommended by the IIW [20], since its direction
finite element model.
remains almost within ± 60° to the weld line, which is ensured by
calculating the direction cosines of the stress tensor components of the
4.1. Determination of local stress time-histories by the finite element method extrapolation points. This overall process, including the importation of
the interaction forces from the beam-like 2D model, the dynamic ana-
Starting from the modal dynamic response of the bridge 3D finite lysis and the sub modeling procedure required the development of a
element model it is possible to obtain the modal displacements at the computer routine thus obtaining the local stress time-histories in an
boundaries of the local detail for each mode of vibration j and then to automated mode by integrating ANSYS and Matlab [30].
apply them as imposed displacements on the corresponding boundary

51
G. Alencar et al. Engineering Structures 166 (2018) 46–61

4.2. Determination of nominal stress time-histories Table 4


Dynamic properties of the assumed fatigue truck.
Nominal stress time-histories were obtained from two different Parameter 1st axle 2nd axle Units Description
sources: (a) the 3D FE model of the bridge, more specifically the global
σx(t) stresses located near the investigated detail (Fig. 9c), determined kv 864 2340 kN/m Suspension spring stiffness
kp 1620 6720 kN/m Tire spring stiffness
in the same manner as the local stresses, and (b) the beam-like 2D
cv 6 12 kNs/m Suspension damping coefficient
model, for which the dynamic analysis of the vehicle-bridge system is cp 2 2 kNs/m Tire damping coefficient
solved by full numerical integration of the coupled equations of motion mp 635 1066 kg Suspension mass
through the Newmark algorithm [4]. Subsequently, one can compute Total mass (m) 20.3 t
the global bending moment ranges over time for the beam-like 2D Truck body mass (ms) 18,599 kg
Natural frequencies (f) [1.32;2.41;11.39;14.74] Hz
model, M(t). These bending moment ranges are used to obtain the
longitudinal stresses over time using the section properties of the
composite steel-concrete girder, which were also validated with the 5. Modeling of progressive deterioration for road surface
global tridimensional finite element model in Alencar [5]. In order to
correctly obtain the stresses at one steel girder, one must compute the 5.1. Generation of the road surface roughness spectra
applied moment in a specific girder. This is performed by multiplying M
(t) by the AASHTO distribution factor for steel girder bridges with re- Road surface roughness is generally defined as an expression of ir-
inforced concrete slabs. In this paper, a 0.547 value was computed for regularities of the road surface, and it is the primary factor affecting the
the girder distribution factor considering one design lane loaded. Fur- dynamic response of both vehicles and bridges, as shown by Silva and
thermore, for the fatigue limit state, the distribution factor should be Roehl [32]. Based on the studies carried out by Dodds & Robson [33]
divided by 1.2, since multiple presence factors are not to be applied for the road surface roughness was assumed as a zero-mean stationary
fatigue if one design truck is used, as specified in Article 3.6.1.1.2 of Gaussian random process, and it can be generated through an inverse
AASHTO [21]. The nominal stress at the bottom flange is obtained by Fourier transformation as shown in Eq. (8):
dividing the applied moment in the girder to the composite bending
modulus, WSG, equal to 81,389 cm3. Then, the stress at the bottom N

flange must be interpolated 130 mm above, since the investigated


r (x ) = ∑ 2·ΔΩ·Gd (Ωi) ·cos(2π·Ωi x + θi )
i=1 (8)
welded detail is at the end of the transverse stiffener. This is made
though the calculation of the theoretical constant neutral line at the where θi = random phase-angle uniformly distributed from 0 to 2π,
investigated composite section; in this case approximately equal to Gd(Ω) = power spectral density (PSD) function (cm3/cycle) for the road
1.49 m (adopting the bottom flange as horizontal axis). Finally, fol- surface elevation, and Ωi = wave number (cycles/m). The PSD function
lowing this procedure it is possible to obtain the nominal stress time- for road surface roughness were developed by Dodds & Robson [33], as
histories at the referred detail using the beam-like 2D model subjected presented in Eq. (9):
to vehicle dynamic loads for different speeds and pavement roughness −2
Ω
levels. Gd (Ωi) = Gd (Ω0) ⎛ i ⎞
⎜ ⎟

Ω
⎝ 0⎠ (9)

where Ω = spatial frequency of the pavement harmonic i (cycles/m),


4.3. The assumed fatigue vehicle
Ω0 = discontinuity frequency of 1/2π (equal to 1 rad/m), and
Gd(Ω0) = road roughness coefficient (m3/cycle), also called Road-
The fatigue truck used in the present work is presented in Fig. 10a,
Roughness Coefficient (RRC), whose value is chosen depending on the
being one of the most common vehicles in local roads in Brazil (see
road class shown in Table 5, EN 1991-2, annex B [23].
Fig. 10a). The two-axle truck structural-mechanical model used is
shown in Fig. 10b, with their dynamic properties based on experimental
measurements [31]. The geometry, mass distribution, damping, and 5.2. Progressive deterioration model for the road-roughness
stiffness of the tires and suspension systems of the truck are listed in
Table 4. In the present study, only one vehicle in one lane, traveling In order to consider the road surface damages due to loads or cor-
along the bridge at different speeds was considered for fatigue analysis rosions, a progressive deterioration model for the road-roughness is
due to bridge relatively short span length. necessary when generating the random road profiles. Paterson & Attoh-
Okine [34] have developed such a model considering the International
Roughness Index (IRI), with the values at any time after the service of
road surface being calculated using Eq. (10):

ms șv
CM

k v2 cv2
uv k v1 cv1
mp2
2.0 m mp1
up1
k p1 cp1 k p2 cp2 up2
Road-Roughness
4.5 m

a) b)
Fig. 10. Model of prototype two axles truck: (a) 2C truck geometry scheme, (b) modeling of the rigid body, springs and dampers.

52
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Table 5
Average values of Gd(Ω0) for different levels of road roughness quality (in cm3)
according to EN 1991-2 [23].
Road class Road Gd(Ω0) (cm3) Analogous
quality description
level Lower limit Geometric Upper limit
Mean

A Excellent – 1 2 New roadways


B Good 2 4 8 layers made of
concrete or
asphalt
C Average 8 16 32 Old roadway
layers without
maintenance
D Poor 32 64 128 Roadway
E Very 128 256 512 layers Fig. 11. Deterioration of road roughness pavement in a 15-years period in
poor consisting of terms of ln(RRC × 106).
cobblestones
or similar
material
Others ESAL values were obtained in the mentioned works, since they
assumed different fatigue vehicles, but leading also to similar required
periods of surface renovation.
IRIt = 1.04e η·t ·[IRI0 + 263(1 + SNC )−5 (CESAL)t ] (10)

where IRIt is the IRI value at time t, IRI0 is the initial roughness value 6. Results and discussions
directly after completing the construction and before opening to traffic,
t is the time in years, and η is the environmental coefficient varying 6.1. Static analysis: Local stresses at the hot-spot
from 0.01 to 0.7 that depends on dry or wet, freezing or nonfreezing
conditions, usually adopted equal to 0.10 for bridges exposed in general First, the local stresses at the vertical stiffener weld end due to static
environment conditions. Structural number (SNC) is a parameter that is loading were investigated. The obtained local stresses for the fatigue
calculated from data on the strength and thickness of each layer in the truck positioned in the slow lane at the midspan, 4.5 m ahead and be-
pavement, herein adopted equal to 4, and (CESAL)t is the estimated hind, are shown in Figs. 12–14, with the directions referring to the same
number of traffic in terms of AASHTO 80-kN (18-kip) cumulative Cartesian coordinate system in the global model (Fig. 4a). The direct
equivalent single axle load at time t in millions. Following strictly the stresses σx (Fig. 12a) are typical due to the global bending of the girder,
rules of AASHTO Guide for Design of Pavement Structures [35], the 80- while the direct stresses σy (Fig. 12b) are mainly due to local out-of-
kN Equivalent Single Axle Load (ESAL), which is required to obtain plane bending of the local web panel. The relative location of the
(CESAL)t, was computed equal to 5.693 for the assumed fatigue vehicle maximum direct stresses σx did not change with truck position (Fig. 13),
(Fig. 10). The total number of vehicles at first year is considered equal while the location of the maximum direct stresses σy moved within the
to 50,000, due to the localization of the bridge within a local road with vertical stiffener thickness (12.5 mm) from the beginning (Fig. 14a) to
a low traffic of trucks [23]. Therefore, the (CESAL)t at first year is the end (Fig. 14b) of the longitudinal weld seam, with a global max-
computed as equal to (5.693 × 50,000)/106 = 0.28465, and this value imum in the middle (Fig. 12b). The location of the maximum shear
changes if the number of trucks per year increases. Finally, the RRC can stresses σxy did not move from the middle of the weld corners, and their
be correlated with the IRIt using Eq. (11), as shown in [7]: magnitudes were within the limits of Fig. 12c. Thus, since Fig. 12c
shows that shear stresses are not located in the same point of maximum
RRCt = 6.1972 × 10−9 × exp{[1.04ηt ·[IRI0 + 263(1 + SNC )−5 (CESAL)t ]] normal stresses and have a relatively lower magnitude, one can in-
vestigate where a fatigue crack can be expected to initiate based on
/0.42808} + 2 × 10−6 (11)
normal stresses σx, either in the right or in the left of the transverse
The initial IRI0 varies from one region to another. It may range from weld seams (Fig. 12a), and σy in the middle of the longitudinal weld
0.90 to 2.0 m/km depending on the specifications adopted in each seam (Fig. 12b). Moreover, according to EN 1991-2 [23], a normal
country for roads construction. The initial standard value for the IRI0 distribution of the vehicle center line in the transverse direction within
adopted in this work is 0.90 m/km. The terminal value for the rough- a prescribed range of ± 0.2 m can be adopted when local effects are
ness which indicates that the road is no more serviceable has been es- being assessed. However, a change in the centerline of the vehicle
tablished as being equal to IRI = 3.0 [36]. Substituting the values of of ± 0.5 m resulted in less than 4% of change in the calculated hot-spot
SNC, (CESAL)t and η into Eq. (11), the adopted progressive pavement stress. Therefore, the probabilistic transverse normal distribution of the
deterioration model used in the numerical simulations of the present truck centerline in the slow lane was neglected.
work can be obtained (Fig. 11). In this figure, a logarithmic function Since the maximum stresses presented in Figs. 12–14 were com-
was applied to the y-axis in order to obtain a better representation of puted at singularity points, it must be more accurate to compare the
the different road-roughness conditions. extrapolated stresses instead of the maximum stresses, deciding be-
It is expected that pavement will serve a period less than the time tween the most appropriate stress components to be extrapolated based
span for which it was originally designed, which is usually selected as on the principal stress directions. Therefore, as can be seen from
more or less 10-years. After a 10-years period, a surface renovation is Fig. 16, which illustrates the adopted extrapolation rule for the hot spot
performed, and the RRC value recovers to the initial value. Three traffic stresses, since the principal stresses directions are almost perpendicular
scenario growths are considered through a traffic growth rate α, equal to the transverse welds, in spite of a small deviation angle, stress ex-
to 0%, 3% and 5%. The use of the Brazilian truck 2C, which has a trapolation must use principal stresses σ1 for the transverse weld (HS1
computed 80-kN Equivalent Single Axle Load (ESAL) equal to 5.693, region) and normal stresses σy for the longitudinal weld (HS2 region),
and the adoption of the same characteristics for the pavement found in as recommended by the IIW [20]. Thus, for the horizontal weld the
[7–9], namely the environmental coefficient (η = 0.10) and the Struc- computed hot-spot stress was slightly lower (21.4 MPa) than for the
tural number (SNC = 4), lead to a computed period of surface re- transverse weld (25.1 MPa) (see Fig. 17), which confirmed that the
novation of more or less 10 years, which is reasonable with the practice. extrapolated stresses in the transverse welds are always higher than the

53
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Fig. 12. Local stresses due to static fatigue truck loading at mid-span location (units: MPa).

extrapolated stresses in the longitudinal welds for the considered static reasons, a modified SeN curve (FAT80) for nominal stresses without
load cases. Therefore, based only on static analysis results, one probably cut-off limit and a unique slope (m = 3) was adopted for the fatigue
could consider the ends of transverse welds (HS1) as the most critical damage calculation.
locations. However, dynamic analyses have shown that stresses normal Fig. 17 shows that the dynamic responses computed using the im-
to the longitudinal weld are much more prone to dynamic amplification plemented approach for the 3D finite element model in general match
effects. well with those from the coupled dynamic approach for the beam-like
2D model. In spite of the higher difference observed during the free
vibration phase between both models only for the worst case
7. Dynamic analysis (RRC = 32 cm3) in the 10th year, mainly due to the much less-damped
response of the beam-like 2D model, the proposed approach for dy-
7.1. Comparison of numerical response in terms of nominal stresses namic analysis can be considered to provide an acceptable level of
computational accuracy for fatigue damage evaluation, and can be
A verification of the engineering approach which consists in export extended to evaluate local stresses for the hot-spot stress method.
the vehicle-bridge dynamic interaction forces from the beam-like 2D Complex dynamic loads which occur with the simultaneous crossing of
model to the 3D FE model is presented for the crossing of a single two- multiple vehicles are not expected since for the fatigue life estimations
axle vehicle at the speed of 20 m/s (72 km/h) considering three dif- of short span bridges the common practice is to use solely one type of
ferent road-roughness conditions and a time-span of 20 s. Fig. 17 shows vehicle crossing alone [13]. Therefore, expected problems that could
the obtained nominal stresses at the investigated detail. A preliminary occur due to different behavior of both dynamic systems, namely the 3D
fatigue damage analysis covering the first 10 years of the bridge’s life- finite element model solved in the framework of the dynamic mode
span period is also presented in Fig. 18 for both models and the same superposition method with imported interaction forces and the coupled
vehicle speeds distribution as Zhang & Cai [7] (see Table 6). For safety

15.5 m 24.5 m 24.5 m 15.5 m

MAX
(22 MPa)

MAX
(25 MPa)

a) b)
Fig. 13. Location of maximum local direct stresses σx due to static fatigue truck loading before (a) and after (b) mid-span location (units: MPa).

54
G. Alencar et al. Engineering Structures 166 (2018) 46–61

15.5 m 24.5 m 24.5 m 15.5 m

MAX
(14 MPa)
MAX
(17 MPa)

a) b)
Fig. 14. Change in the location of maximum local direct stresses σy due to static fatigue truck loading before (a) and after (b) mid-span location (units: MPa).

Fig. 15. Hot-spot stress definition: (a) and (b) extrapolation points used to obtain the hot-spot stress at the weld; (b) principal stress directions at element centers.

55
G. Alencar et al. Engineering Structures 166 (2018) 46–61

40 3.5
3D FEM
35 1
3.0 2D beam
x
30
y 2.5
Stress (MPa)

25

Dcalc (×10-3)
HS1: x or 1 2.0
20 HS2: y
15 1.5

10 1.0
5 0.5
0
0 5 10 15 20 25 30 0.0
0 1 2 3 4 5 6 7 8 9 10
Distance from weld toe (mm)
Time (years)
Fig. 16. Hot-spot stresses computed for static loading considering the assumed
fatigue vehicle at the midspan. Fig. 18. Accumulated fatigue damage comparison between the nominal stress
methods for the 3D FE bridge global model and the beam-like 2D model.

Excellent road-roughness (RRC = 2 cm3)


8 Table 6
7 3D FEM Adopted vehicles speed distribution per year [7].
Nominal stress (MPa)

6 2D BEAM Speed 10 m/s 20 m/s 30 m/s 40 m/s 50 m/s 60 m/s


5
4 Distribution 29.57% 48.42% 20.12% 1.83% 0.03% 0.03%
3
2
1 vehicle-bridge 2D model solved with a full-integration method can be
0 diminished.
-1
-2 7.2. Fatigue damage sensitivity to the number of vibration modes
0 2 4 6 8 10 12 14 16 18 20
Time (s) Since a structure comprises several vibration modes, however some
a) of them could have a negligible or even a null contribution to the in-
Poor road-roughness (RRC = 8 cm3) vestigated dynamic load effect. Thus, usually one must perform a sen-
10 sitivity analysis concerning the number of vibration modes used in the
analysis in order to optimize the numerical computation. Herein the
8 3D FEM
Nominal stress (MPa)

best parameter to evaluate through this sensitivity analysis is the fa-


2D BEAM
6 tigue damage, since it comprises individually the contribution of all
4
stress ranges from a single traffic event. Fig. 19a shows the fatigue
damage in the longitudinal weld in a logarithm scale due to an in-
2 dividual crossing of the fatigue truck at different speeds and poor road-
0 roughness condition considering the fatigue strength class FAT100. It
can be concluded that a convergent response was achieved between 50
-2
and 60 Hz. Consequently, the frequency limit, which needs to be con-
-4 sidered in an analysis, is 60 Hz, that is equivalent to a total of 361 vi-
0 2 4 6 8 10 12 14 16 18 20
bration modes.
Time (s) On the contrary, Fig. 19b shows a comparison between the same
b) sensitivity analysis for the fatigue damage in the transverse weld and in
the longitudinal weld, with an envelope of maximum and minimum
Poor road-roughness (RRC = 32 cm3)
12 damages indexes for different vehicle speeds. It can be observed a
10 higher contribution to the out-of-plane bending stresses σy and hence to
3D FEM
Nominal stress (MPa)

8 the fatigue damage in the longitudinal weld given by the 12th local
2D BEAM
6 vibration mode, presented in Fig. 20a. It consists of a local transverse
4
bending mode of the concrete deck and the corresponding in-phase
2
local bending of the related steel web panel. It seems clear that the
0
vehicle acting on the slow lane can transfer high mechanical vibration
-2
energy mainly through this vibration mode, which leads to an ampli-
fication of the stresses in the steel web panel with maximum modal
-4
displacements (Fig. 20b). Finally, since this mode contribution is higher
-6
0 2 4 6 8 10 12 14 16 18 20 than the global bending modes contributions, the highest damage is
Time (s) expected to occur in the longitudinal weld seam due to normal stresses,
σy .
c)
Fig. 17. A comparison of nominal stress time-histories between the 3D FE 7.3. Stress time-histories for the fatigue prone detail
model and the beam-like 2D model for different road-roughness conditions.
After the selection of all relevant vibration modes, dynamic analyses
were performed covering all the design lifespan period of the bridge
(100 years), considering different speeds and different road-roughness

56
G. Alencar et al. Engineering Structures 166 (2018) 46–61

a)

Higher contribution of
first vibration mode
(f01=2.97Hz) on in-plane
bending stresses.

Higher contribution of vibration modes


f10=19.6 Hz and f12=21.4 Hz on out-of-
plane bending stresses.

b)
Fig. 19. Fatigue damage due to one passage of the fatigue truck at different speeds and poor road-roughness conditions for: (a) out-of-plane bending stresses (σy) and
(b) in-plane bending stresses (σx) vs. out-of-plane bending stresses (σy).

levels of degradation. A time increment of 0.005 s and a time interval of between the different stress definition methods, e.g. the nominal stress
2 s in free vibration was adopted for the dynamic analyses. At first, in methods obtained from both the 3D FE model and the beam-like 2D
order to measure the severity of the computed stress signals at different model, and the principal hot-spot stresses (σ1) obtained at the trans-
periods of time, the equivalent stress range, defined as the constant verse weld of the 3D FE solid sub model (Fig. 15). The obtained nu-
stress amplitude equivalent to a variable stress history for the same merical signal for the hot-spot stress method have been submitted to a
number of applied cycles [20], was computed for the most re- low pass filter for a better representation in comparison to the nominal
presentative signals. Thus, concerning the out-of-plane hot-spot stresses stress definitions, with the filtered signal represented in red and ori-
obtained at the longitudinal weld (σyHS) due to one fatigue truck ginal signal represented in grey. In all cases two road-roughness con-
crossing, computed results have shown that after a 10-years period the ditions are compared: (i) in the 1st year, when an excellent road-
equivalent stress range Δσye increased more than twice, from 7.33 MPa roughness quality level is expected, and (ii) after 10-years of intense
to 16.70 MPa (see Fig. 21b). This leads to a severe magnification of the use, when the degradation of the pavement already contributed to an
individual damage due to one fatigue truck crossing, by the order of 14- increase of undesirable dynamic effects.
fold, which highlights the importance of consider a progressive dete- Computed equivalent hot-spot shear stress ranges after a 10-years
rioration model for the pavement in this kind of analysis. period was equal to 2.46 MPa (Fig. 22), hence less than 15% of the
A comparison between different hot-spot stress components and maximum computed equivalent normal stress range
stress definitions methods was also performed. Fig. 22 shows different (16.70 × 15% = 2.51 MPa). This means that a multiaxial fatigue as-
stress components analyzed, i.e. σx and σ1 at the transverse weld (HS1) sessment based on normal stresses and shear stresses do not need to be
and σxy at the weld corners, which is the region of maximum shear performed for the investigated fatigue prone detail [20], since low
stresses (Fig. 12c). On the other hand, Fig. 23 shows a comparison shear stresses were usually found. In respect of the transverse weld, the

Fig. 20. 12th vibration mode (f12 = 21.36 Hz): (a) transverse local bending of the concrete deck; (b) local bending of the web panel adjacent to the stiffener detail.

57
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Fig. 21. (a) Hot-spot stress σy over time for v = 10 m/s and (b) Rainflow cycle counting for one fatigue truck crossing for the initial road-roughness level and after a
10-years period.

use of σx or σ1 would not lead to significant differences in terms of global and local 3D finite element models and the beam-like 2D model.
fatigue damage, as can be inferred from Fig. 22. Concerning the stress The influence of the traffic growth in the estimated life was also con-
definitions methods, despite a less damped signal for the beam-like 2D sidered, through the adoption of three traffic growth rates α, 0%, 3%
model in general a good agreement was usually found between both and 5%. The original Miner's rule assumes that the design life is con-
nominal stress methods (3D finite element model vs. beam-like 2D sumed when the cumulative damage DL = 1. However, according to
model), which is also a good indication of the accuracy of the computed Niemi et al. [37] many tests have shown that failure may occur at a
vehicle load distribution factors for highway girder bridges according to computed damage of DL = 0.5. Therefore, in this work a value of
AASHTO LRFD Bridge Design Specifications [21]. It can be observed DL = 0.5 is adopted for safety purposes. Moreover, for a safe life as-
that the 3D finite element model is much more damped in terms of sessment with high consequences of failure, Eurocode 3 [17] re-
computed stresses, mainly due to the simple element definition adopted commends that the fatigue strength reduction factor, γMf, should be
for the beam-like 2D model, which uses only Bernoulli beams and have taken as 1.35. For the sake of higher conservatism, the cut-off limit ΔσL
only planar beam vibration modes. On the contrary, the 3D bridge finite was not considered, being adopted a modified SN curve with a unique
element model has much more parts that can contribute to dissipation slope (m = 3). The fatigue damage is then computed for the entire
of the vibration and can be less susceptible to dynamic effects than the bridge design lifespan period (100-years), considering surface renova-
beam-like 2D model. Higher local stresses were obtained in relation tions of the pavement being executed over time. Table 7 and
with nominal stresses as expected since it also has a higher FAT class Figs. 24–26 show the obtained fatigue lives in years and the fatigue
(FAT100) than the nominal approaches (FAT80), and of course the fi- damage evolution, respectively, for different stress definitions and for
nite element size used is much refined. Therefore, an accurate com- both analyzed hot-spot points. In these graphs, the points indicate the
parison between the three stress definition methods can be only per- total accumulated fatigue damage (Dcalc) for each year since the bridge
formed in terms of estimated fatigue life, as presented in the next traffic opening, while the lines are regular data fit functions, being used
section. a linear fit for a traffic increase rate of 0% and an exponential fit for
increase rates of 3% and 5%.
The adopted criteria to define when a surface renovation is required
7.4. Fatigue life estimation is whether the RRC exceeds 32 cm3, which is classified as a “poor”
pavement. When a surface renovation is executed, one considers that
Fatigue life estimations based on Palmgren-Miner’s rule were per- the road-roughness becomes equal to as when the road was released to
formed for both points, the transverse weld end location (HS1) and the traffic, that is RRC = 2 cm3, which is classified as an “excellent” pa-
longitudinal weld middle location (HS2), considering the three different vement. The highest damages are obtained in the last years of each
stress definition methods based on the developed numerical models, the

40 50
35
Hot-spot: σ1 Hot-spot: σ1
Hot-spot: σx 40 Hot-spot: σx
30
Hot-spot: σxy Hot-spot: σxy
25
Stress (MPa)

30
Stress (MPa)

20
20
15
10 10
5
0
0
-5 -10
0 1 2 3 4 5 0 1 2 3 4 5
Time (s) Time (s)
a) b)
Fig. 22. Different hot-spot stress time-histories for v = 10 m/s considering different stress definitions for initial road-roughness level (a) and poor road roughness
level (after 10-year) (b).

58
G. Alencar et al. Engineering Structures 166 (2018) 46–61

Table 7
Fatigue life estimation in years considering different stress definitions (γMf = 1.35).
Stress definition method Fatigue strength class (FAT) Traffic increase

α = 0% α = 3% α = 5%

Beam-like 2D model 80 »100 61 47


3D finite element model 80 »100 88 60
σ1 hot-spot stress (Transverse weld) 100 106 38 34
σy hot-spot stress (Longitudinal weld) 100 41 20 19

40 50
35 2D beam 3D FEM
3D FEM 40 2D beam
30
Hot-spot: σ1 Hot-spot: σ1
25 30
Stress (MPa)

Stress (MPa)
20
20
15
10 10
5
0
0
-5 -10
0 1 2 3 4 5 0 1 2 3 4 5
Time (s) Time (s)
a) b)
Fig. 23. Comparison of nominal stresses versus hot-spot stresses for v = 10 m/s for initial road-roughness level (a) and poor road roughness level (after 10-year) (b).

Fig. 24. Accumulated damage considering different stress definitions (α = 0%). Fig. 25. Accumulated damage considering different stress definitions (α = 3%).

period of 10 years, mainly between the 9th and the 10th year, for which damage evolution is the reason because there is almost no difference for
the surface renovation is not performed until the beginning of the 11th the estimated lives between traffic increases rates of 3% and 5%, when
year. For this reason, the graphs of Figs. 24–26 present almost a linear considering the highest local stresses for dynamic loading, i.e. the σy
damage growth between the 1st and the 8th year and big leaps between hot-spot stresses at the longitudinal weld of the investigated detail
the 8th and the 11th year with an exponential growth of the fatigue (Table 7).
damage within this period. The reason is that in the first stage (1st – Life estimations have shown that the approaches based on the
8th) the pavement quality is between excellent-average, and during the nominal stress method yields to very non-conservative results. For null
period between 9th – 10th years the pavement quality is between traffic increase, the fatigue lives based on nominal stress leads to in-
average-poor (see Fig. 11). This leap effect repeats for every period of finite fatigue life (much higher than 100 years), as shown in Fig. 24. In
10 years for a traffic increase rate of α = 0%. However, it is highlighted general, when compared to the 3D finite element model, the beam-like
that, although the pavement is considered totally renewed after each 2D model yields to less conservative results, mainly due to less damped
pavement restoration, the number of trucks and hence the (CESAL)t response of the simplified model. The less damped response clearly
increase every year for traffic increase rates of α = 3% or 5%. The re- leads to more stress cycles in the free vibration phase, contributing
sult of that is that the initial required renovation surface period will more to the damage.
decrease (10-years) over time. Moreover, the decrease in the required For the worst case presented in Table 7, i.e. the σy hot-spot stress
renovation surface period and the leap effect observed in the fatigue component and a traffic increase rate of α = 5%, a less conservative

59
G. Alencar et al. Engineering Structures 166 (2018) 46–61

• The nominal stress approach yields to very non-conservative results


for both the beam-like 2D model and the 3D finite element model.
For the hot-spot stress, the longitudinal weld, which is much more
prone to dynamic effects, revealed to be most prone to fatigue da-
mage than the transverse weld, which is the crack initiation point
for the nominal stress approach when the FAT80 structural detail
fatigue strength is considered. Therefore, a fatigue crack is much
more likely to initiate in the middle of the longitudinal weld sub-
jected to distortion-induced stresses (σy) instead of the point ana-
lyzed in the transverse weld subjected to longitudinal stresses due to
global bending of the girder bridge (σx or σ1). A small weld defect
with dimensions of the order of 1 mm and with a crack-like shape at
this position could be a point subjected to fatigue crack initiation
and high stress concentration due to vehicles dynamic loads;
• The nominal stress method applied for both numerical models
clearly could not determine the fatigue life for out-of-plane stresses,
such as that demonstrated to exist in the present case study, which
Fig. 26. Accumulated damage considering different stress definitions (α = 5%). can be a serious limitation for global approaches used in the design
phase. In this situation, the more non-conservative results are ap-
fatigue life estimation was made, considering a cut-off limit equal to proached by the 3D finite element model using nominal stresses in
40 MPa (ΔσL = 0.404 × 100 = 40 MPa) and γMf = 1.0. Under these the shell elements of the web girder near the examined detail. This
conditions, the obtained fatigue life was 35 years, when the number of could be against sense, since a great time and efforts can be con-
trucks reached 275,800 per year per slow lane. This life can be seen as a sumed developing a complex 3D numerical model to obtain very
measure of the estimated fatigue life of the roadway bridge since its over-estimate fatigue lives, which is against the structural safety. In
opening without any safety factor. If one considers the fatigue damage this case, the hot-spot stress method could lead to more realistic
limited to 1.0, this fatigue life increases to 41 years, when the number results of welded bridges, since it captures the stress concentration
of trucks reached 369,600 per year per slow lane, which can be con- due to the overall geometry of the welded details.
sidered a heavy traffic condition.
Some challenges remain for further studies: (a) the consideration of
8. Conclusions more complex vehicle numerical models with more axles and car-
bodies, accompanied by the implementation of an integrated 3D cou-
In this study, a fatigue assessment is carried out for a fatigue detail of pled dynamic system of the vehicles and the bridge; (b) the computa-
a composite steel-concrete roadway girder bridge by utilizing two con- tion of a reliability index using a fatigue limit state function based on
ventional methods according to the EN 1993-1-9 [17] and the IIW re- SN curves obtained for hot-spot stresses for the local detail and con-
commendation [20] caused by vehicle passages at varied vehicle speeds sidering the randomness of all variables (vehicle speeds, vehicle types,
under various road conditions. The effects of long-term asphalt pavement etc); (c) the consideration of different road-roughness profiles for the
deterioration and various vehicle parameters, such as vehicle speeds and pavement in the transverse direction; d) the change in the position of
types, can be included in the developed numerical methodology. A nu- the transverse stiffener end, in order to respect the good design practice
merical simulation toward solving a coupled vehicle-bridge system in- of keeping it in the maximum of 60 mm or 4 times the web thickness,
cluding a suspension-vehicle model and a 3D dynamic bridge model is the lesser of the two, above the bottom flange girder [38], and hence
used to obtain the stress time-histories due to dynamic loading. The the impact on the expected fatigue life of this structural detail. These
calculated fatigue lives from the three different approaches, namely shortcomings will be studied in a future work.
using the beam-like 2D model for nominal stresses and the 3D finite
element model for both nominal and hot-spot stresses are compared with Acknowledgements
each other to acquire a reasonable fatigue-life estimation.
Since the estimated fatigue life is affected by the accuracy of the The authors gratefully acknowledge the financial support for this
numerical model including modeling options near the fatigue detail, research work provided by the Brazilian Science Foundation’s CNPq
three types of numerical models are utilized for the computation of (doctoral scholarship process 203662/2014-8), CAPES and FAPERJ.
fatigue truck-induced stress ranges at the fatigue detail. The simplified Authors acknowledge the Portuguese Foundation for Science and
global beam-like model is developed for the determination of section Technology for the funding, particularly through the Project POCI-01-
bending moments at the region of the fatigue detail, and a refined local 0145-FEDER-007457 - CONSTRUCT - Institute of R&D in Structures and
finite element model is also utilized for the evaluation of local stress for Construction funded by FEDER funds through COMPETE2020 -
the fatigue assessment. A third approach based on the global stresses Programa Operacional Competitividade e Internacionalização (POCI).
obtained in the shell elements for the global 3D bridge element is also
considered. From this study, the following conclusions are drawn: References

• Static analyses have shown that the extrapolated stress to the [1] Nishikawa K, Murakoshi J, Matsuki T. Study on the fatigue of steel highway bridges
in Japan. Constr Build Mater 1998;12:133–41. http://dx.doi.org/10.1016/S0950-
transverse weld are always higher than the extrapolated stress to the 0618(97)00015-9.
longitudinal weld. Based only on the above static analyses results, [2] Fisher JW, Roy S. Fatigue damage in steel bridges and extending their life. Adv Steel
one probably could consider HS1 at transverse welds as the most Constr 2015;11:250–68. http://dx.doi.org/10.18057/IJASC.2015.11.3.1.
[3] Connor RJ, Fisher JW. Identifying effective and ineffective retrofits for distortion
critical location. However, dynamic analyses have shown that HS2 fatigue cracking in steel bridges using field instrumentation. J Bridg Eng
at longitudinal weld is much more prone to dynamic effects and 2006;11:745–52.
hence to fatigue issues. The reason is mainly due to a higher am- [4] Silva JGS. Non-deterministic dynamic analysis of highway bridge decks with surface
irregularities (in Portuguese). Dsc Thesis. PUC-Rio, 1996.
plification effect due to local vibration modes for the examined
[5] Alencar G. Dynamic analysis and fatigue assessment of composite (steel-concrete)
structural detail; highway bridges subjected to the traffic of vehicles over the irregular pavement

60
G. Alencar et al. Engineering Structures 166 (2018) 46–61

surface (in Portuguese). Master Thesis. University of State of Rio de Janeiro, 2015. docume. International Institute of Welding (IIW); 2016.
[6] da Silva JGS. Dynamical performance of highway bridge decks with irregular pa- [21] AASHTO. AASHTO LRFD Bridge design specifications. 6th Ed. Washington, DC:
vement surface. Comput Struct 2004;82:871–81. http://dx.doi.org/10.1016/j. 2012.
compstruc.2004.02.016. [22] Leitão FN, Da Silva JGS, Da Vellasco PCGS, De Andrade SaL, De Lima LRO.
[7] Zhang W, Cai CS. Fatigue reliability assessment for existing bridges considering Composite (steel-concrete) highway bridge fatigue assessment. J Constr Steel Res
vehicle speed and road surface conditions. J Bridg Eng 2012;17:443–53. http://dx. 2011;67:14–24. http://dx.doi.org/10.1016/j.jcsr.2010.07.013.
doi.org/10.1061/(ASCE)BE.1943-5592.0000272. [23] EN 1991–2. Eurocode 1: Actions on structures - Part 2: Traffic loads on bridges
[8] Zhang W, Cai CS. Reliability-based dynamic amplification factor on stress ranges for 2003.
fatigue design of existing. Bridges 2013;18:538–52. http://dx.doi.org/10.1061/ [24] Miner MA. Cumulative damage in fatigue. J Appl Mech 1945;12:A159–64.
(ASCE)BE.1943-5592.0000387. [25] Astm E. Standard practices for cycle counting in fatigue. Analysis 2005:1–10.
[9] Zhang W, Cai CS, Pan F. Nonlinear fatigue damage assessment of existing bridges [26] Alencar G, Correia JAFO. A user-friendly tool for fatigue assessment of steel
considering progressively deteriorated road conditions. Eng Struct structures according to Eurocode 3. New Trends Integrity, Reliab. Fail., Porto,
2013;56:1922–32. http://dx.doi.org/10.1016/j.engstruct.2013.06.027. Portugal: IRF 2016–5th International Conference on Integrity, Reliability and
[10] Han W, Wu J, Cai CS, Asce F, Chen S, Asce M. Characteristics and dynamic impact Failure; 2016, p. 1–9.
of overloaded extra heavy trucks on typical highway bridges. J Bridg Eng [27] Wang T-L, Huang D. Cable-stayed bridge vibration due to road surface roughness. J
2015;20:1–11. http://dx.doi.org/10.1061/(ASCE)BE.1943-5592.0000666. Struct Eng 1992;118:1354–74.
[11] Wang W, Deng L, Ph D, Asce M, Shao X. Fatigue Design of Steel Bridges Considering [28] Liu K, Zhou H, Shi G, Wang YQ, Shi YJ, De Roeck G. Fatigue assessment of a
the Effect of Dynamic Vehicle Loading and Overloaded Trucks 2016;21:1–12. composite railway bridge for high speed trains. Part II: conditions for which a dy-
doi:10.1061/(ASCE)BE.1943-5592.0000914. namic analysis is needed. J Constr Steel Res 2013;82:246–54. http://dx.doi.org/10.
[12] Wang W, Deng L, Shao X. Number of stress cycles for fatigue design of simply- 1016/j.jcsr.2012.11.014.
supported steel I-girder bridges considering the dynamic effect of vehicle loading. [29] Albuquerque C, Silva ALL, de Jesus AMP, Calçada R. An efficient methodology for
Eng Struct 2016;110:70–8. http://dx.doi.org/10.1016/j.engstruct.2015.11.054. fatigue damage assessment of bridge details using modal superposition of stress
[13] Paraskeva TS, Dimitrakopoulos EG, Zeng Q. Dynamic vehicle – bridge interaction intensity factors. Int J Fatigue 2015;81:61–77. http://dx.doi.org/10.1016/j.
under simultaneous vertical earthquake excitation. Bull Earthq Eng 2017;15:71–95. ijfatigue.2015.07.002.
http://dx.doi.org/10.1007/s10518-016-9954-z. [30] MATLAB. MATLAB 2015a Reference Manual 2015.
[14] Deng L, Wang W, Cai CS. Effect of pavement maintenance cycle on the fatigue [31] Battista RC, Pfeil MS, Carvalho EML. Fatigue life estimates for a slender orthotropic
reliability of simply-supported steel I-girder bridges under dynamic vehicle loading. steel deck. J Constr Steel Res 2008;64:134–43. http://dx.doi.org/10.1016/j.jcsr.
Eng Struct 2017;133:124–32. http://dx.doi.org/10.1016/j.engstruct.2016.12.022. 2007.03.002.
[15] Radaj D, Sonsino CM, Fricke W. Recent developments in local concepts of fatigue [32] da Silva JGS, Roehl JLP. Probabilistic formulation for the analysis of highway
assessment of welded joints. Int J Fatigue 2009;31:2–11. http://dx.doi.org/10. bridge decks with irregular pavement surface. J Braz Soc Mech Sci 1999;21:1–10.
1016/j.ijfatigue.2008.05.019. [33] Dodds CJ, Robson JD. The description of road surface roughness. J Sound Vib
[16] Alencar G, Ferreira G, de Jesus A, Calçada R. Fatigue assessment of a high-speed 1973;31:175–83.
railway composite steel-concrete bridge by the hot-spot stress method. Int J Struct [34] Paterson WD, Attoh-Okine B. Simplified models of paved road deterioration based
Integr 2018. on HDM-III. Annu. Meet. Transporatation Res. Board, 1992, p. 1–30.
[17] EN 1993-1-9. Eurocode 3: Design of steel structures - Part 1–9: Fatigue 2005. [35] AASHTO. AASHTO Guide for Design of Pavement Structures 1993.
[18] Ladinek M, Lang R, Lener G. Ermüdungsfestigkeit nach EN 1993–1-9 Anhang B : [36] Shiyab AMSH. Optimum Use of the Flexible Pavement Condition Indicators in
Strukturspannungen – Gedanken zur Neufassung. Stahlbau 2016;85:274–80. http:// Pavement Management System (PhD Thesis). Curtin University of Technology,
dx.doi.org/10.1002/stab.201610373. 2007.
[19] ANSYS. Advanced Analysis Techniques Guide. Release 12.0 Documentation for [37] Niemi E, Fricke W, Maddox SJ. Fatigue Analysis of Welded Components: Designer’s
ANSYS. ANSYS, Inc. 2009. guide to the structural hot-spot stress approach. IIW-1430-00; 2006.
[20] IIW. Recommendations for Fatigue Design of Welded Joints and Components. IIW [38] ECCS. ECCS - Technical Committee 6 - Fatigue - Good design practice 2000.

61

You might also like