Journal of Constructional Steel Research 139 (2017) 466–472

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

New framework for calibration of partial safety factors for fatigue design
Luca D’Angelo a, * , Alain Nussbaumer b
a
European Organization for Nuclear Research (CERN), Switzerland
b
Resilient Steel Structures Laboratory (RESSLab), Swiss Federal Institute of Technology Lausanne, Switzerland

A R T I C L E I N F O A B S T R A C T

Article history: In Eurocode standards, three verification schemes are proposed for fatigue design under variable amplitude
Received 1 March 2017 loadings: 1) Based on constant amplitude fatigue limit; 2) Based on constant amplitude equivalent stress
Received in revised form 15 September 2017 range at 2 • 106 cycles; 3) Based on accumulated damage. Characteristic values of fatigue resistance and
Accepted 8 October 2017 load effects as well as partial safety factors are introduced in design equations in order to achieve a target
Available online 19 October 2017
reliability level. In this paper a new framework for calibration of fatigue partial safety factors is presented.
Three different fatigue limit state functions are formulated for direct comparison with the three verification
Keywords:
schemes proposed in Eurocodes. The variable amplitude S-N curves used in this framework are defined
Fatigue design
using an original probabilistic approach. The presented framework is then applied to two typical bridge
Partial safety factors
Variable amplitude loadings fatigue sensitive welded joints. The comparison of results with partial safety factors values recommended in
Welded joints Eurocodes shows that the Eurocode-based partial safety factors should be revised by considering different
Fatigue life fatigue sensitive details and by further differentiating between the three verification schemes.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
The safety and serviceability issues in structural design are
addressed in structural codes by defining design equations which
compare load and resistance effects. Due to the inherent uncertainty
in load and resistance terms, these are modeled as random variables.
Characteristic values of random variables and partial safety factors
are introduced in order to ensure a certain level of reliability for the
designed structural component.
The typical design equation for verification of a structural compo-
nent (see Eurocode 1990 [1]) is:
According to the design Eq. (1) a reliability analysis can be made
with the following limit state equation:
g = zR − (E1 + . . . + En ) = 0 (2)
The calibration of partial safety factors is a decision problem, in
which partial safety factors are decision variables which are cali-
brated by minimizing an objective function. Faber et al. [2] proposed
a practical approach for calibration of partial safety factors, in which
the objective function is formulated as follows:


L
 2
zRc W(c) = wj • bj (c) − bt (3)
G= − (cF1 Ec1 + . . . + cFn Ecn ) = 0 (1)
cM j=1

where: L is the number of load cases; wj are the importance factors

where: Rc is the characteristic value of resistance; z is the design
factor; cM is the partial safety factor for resistance; Eci is the char-
acteristic value of the ith action effect; and cFi is the partial safety
factor for the ith action effect.
* Corresponding author.
E-mail addresses: luca.dangelo@cern.ch (L. D’Angelo), alain.nussbaumer@epfl.ch
(A. Nussbaumer).
of different design load cases; and bt is the target reliability index.
Partial safety factors c are computed by minimizing the objective
function in Eq. (3), in which the reliability index is computed by
solving the limit state equation (see Eq. (2)), having determined the
optimal design factor ẑ from the design equation (see Eq. (1)). It is
noted that when partial safety factors are calibrated from Eq. (3) they
are not independent and in the case with one resistance factor and
one loading factor only the product of them can be calculated.
In this paper a new framework for calibration of partial safety
factors for fatigue design is presented. The general limit state Eq. (2)

https://doi.org/10.1016/j.jcsr.2017.10.006
0143-974X/© 2017 Elsevier Ltd. All rights reserved.
 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472 467

and the general design Eq. (1) are adapted to the three verification Table 1
schemes cases proposed in Eurocode 1993-1-9 [3] for fatigue design Recommended values for partial factors for fatigue strength (Table 3.1 of [3]).

under variable amplitude loadings: 1) Based on constant amplitude Design method Consequence of failure
fatigue limit; 2) Based on constant amplitude equivalent stress range
Low consequence High consequence
at 2 • 106 cycles; 3) Based on accumulated damage.
Fatigue resistance terms in design and limit state equations Damage tolerant 1.00 1.15
Safe life 1.15 1.35
are characterized by using ML-MCS method [4], which allows to
increase the confidence in characteristic and design values of fatigue
resistance with respect to current standards.1 The target reliability
indexes for fatigue limit state functions are based on the recommen- 2.2. Verification scheme 2: based on CA equivalent stress range
dations of JCSS [5].
The paper is structured as follows: The fatigue design based on CA equivalent stress range has to
meet the following criterion:
• In Section 2 the three design verification schemes proposed
in Eurocode 1993-1-9 for fatigue design under VA loadings Sc
cFf • SE,2 ≤ (5)
are recalled and shortcomings of Eurocode-based partial safety cMf
factors are highlighted.
• In Section 3 the new framework for calibration of fatigue partial where: SE,2 is the equivalent stress range, at 2 • 106 cycles, which is
safety factors using ML-MCS S-N model is presented. computed by using Fatigue Load Model (FLM) 3 and k damage equiv-
• In Section 4 an application of the framework to two typical alent factors; and Sc is the characteristic fatigue strength at 2 • 106
bridge fatigue sensitive welded joints is considered. cycles (FAT).
• In Section 5 results of analyses of the two considered study
cases are presented.
• In Section 6 results are discussed and comparison with par- 2.3. Verification scheme 3: based on damage accumulation
tial safety factor values recommended in Eurocode 1993-1-9 is
The fatigue design based on damage accumulation has to meet
made.
the following criterion:

The developed framework has general applicability and can be

used with different refinement degrees: differentiating between
detail categories and design methodologies, to set for example differ-
ent partial safety factors in function of the methodology, or serve as
basis to re-calibrate fatigue partial safety factors valid for all details
and methodologies for a given target reliability level.
2. Fatigue design under VA loadings in Eurocode standards
In Eurocode standards, three verification schemes are proposed
for fatigue design under VA loadings: 1) Verification scheme based
on CAFL (see EN-1991-2 [6]); 2) Verification scheme based on CA
equivalent stress range at 2 • 106 cycles (see A.6 of EN-1993-1-9 [3]);
and 3) Verification scheme based on accumulated damage (see A.6 of
EN-1993-1-9 [3]).
2.1. Verification scheme 1: based on CAFL
The fatigue design based on CAFL exceedance has to meet the
following criterion:
CAFL
cFf • Smax ≤ (4)
c cMf
where: CAFL is the characteristic value of the fatigue limit; Smaxc
is the characteristic value of maximum loading stress range; cFf is
the loading partial safety factor, which is set to 1.0; and cMf is the
resistance partial safety factor.

Ntot
ni
Dd = ≤ 1.0 (6)
Ni
i
where: ni is the number of cycles corresponding to the design loading
stress range cFf • Si ; Ni is the endurance to failure obtained from the
factored cSc − N curve.
Mf
The definition of the partial resistance factor cMf in Eurocode
1993-1-9 (Section 1.4, pp. 9, [3]) is ambiguous because cMf is strictly
defined for fatigue strength at 2 • 106 cycles (therefore relevant only
for verification scheme 2), but it is applied by extension for fatigue
strengths at any number of cycles (verification schemes 1 and 3).
Recommended values of partial factor cMf are presented in
Table 1. Sedlacek et al. address the issue of calibration of fatigue
design partial safety factors in [7]. Nevertheless, a proof of rigorous
calibration of the values recommended in Table 1 is not available and
the real reliability level, b, corresponding to these values is still under
debate.
3. New framework for partial safety factor calibration
In this section a new framework for calibration of fatigue par-
tial safety factors using ML-MCS approach-based S-N curves is
presented. The framework includes the three verification schemes
which have been presented in Section 2. One different limit state
function is formulated for each of verification schemes. Partial safety
factors are calibrated by using the following general objective func-
tion, which is valid for all three verifications:

  
L
   2
1
W cFf , cMf = wj • bj cFf , cMf − bt (7)
In [4] the authors presented new probabilistic method which allows to improve
j=1
accuracy in the estimation of CA and VA fatigue S-N curves of welded steel connec-
tion, using combination of Maximum Likelihood Method and Monte-Carlo Simulations
Method. The new probabilistic method is referred to as ML-MCS approach. Improve- where: cFf is the partial safety factor for fatigue loading; cMf is the
ment of accuracy in the estimation of the high cycle fatigue region of S-N curves is
of primary importance when assessing remaining fatigue life of existing structures.
partial safety factor for fatigue resistance; L is the number of load
Nevertheless, use of accurate S-N curves is a primary requirement also when fatigue cases; wj are the load case importance factors; bj are the computed
design partial safety factors have to be calibrated reliability indexes, for j = 1, . . . , L; bt = −V −1 ([V(4.2)]100 ) =
468 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472

(char.) (char.)

(design) (design)

S
c
S [MPa]

S [MPa]
Sc/γMf
CAFL CAFL
γMf
CAFL/γMf CAFL/γMf
−m γMf
γMf 1
−m
γMf 2

6
2⋅ 10 N N

Fig. 1. Resistance partial safety factor: verification schemes 1 and 2. Fig. 2. Resistance partial safety factor: verification scheme 3.

3.00, is the target reliability index for a 100 year reference period2 . The optimal design, ẑj , for the load case j, is computed by solving
The reliability indexes bj are computed by solving the limit state the design Eq. (8), Gj = 0, for assigned values of partial safety factors.
equation for different load cases ( j = 1, . . . , L), after having deter- The reliability index bj is then computed by solving the following
mined the optimal design, ẑ, as solution of the design equation. For- limit state equation3 :
mulation of limit state and design equations for the three considered
verification cases is discussed in the following three subsections. gj = ẑj • exp(V) − Smax,j (12)

3.1. Verification scheme 1: based on CAFL Fatigue partial safety factors are calibrated by minimizing the follow-
ing objective function:
The design equation, for the generic load case j, is formulated as
follows:

20
 2
W(c) = wj • bj (c) − 3.00 (13)
CAFL
Gj = zj • − cFf • Smax, j (8) j=1
cMf c

where wj = 1 for j = 1, . . . , 20, since all load cases have the same
where zj is the design factor and Smaxc , j is the characteristic value importance.
of the maximum stress range, Smax, j . The geometrical interpreta- It is noted that when cFf and cMf are determined with Eqs. (8),
tion of the resistance partial safety factor cMf is illustrated in Fig. 1. (12), and (13), they are not independent and only the product cFf • cMf
 
Smax is modeled as a Gumbel random variable, G1 l G 1 , sG1 , having is determined. Moreover it is possible to set the load partial factor
coefficient of variation equal to: equal to 1.0 by proper choice of the return period of the characteristic
value of the load effect: hence the value of cMf |cFf =1 can be com-
 
1.28 sG1 /l G 1 puted, allowing for direct comparison with recommended values of
cvG =   (9)
1 0.58 sG1 /l G 1 + 1 cMf in Eurocode standards.
For the verification scheme 1 this can be done by using the follow-
ing three step-scheme: 1) The c-characteristic return period plot (see
In order to take into account both the effects of position and dis-
Fig. 3), which relates the product cMf • cFf to the return period of the
persion of loading term, 20 load cases are built by considering all
characteristic value of the maximum stress range, qSmax , is produced
possible combinations of the following values of l G 1 and cvG :
1 by minimizing the objective Eq. (13), where the reliability indexes
bj are computed by considering different values of qSmax in Eq. (12);
(exp(l V ) − exp(v.05 ) 2) The resistance partial safety factor ĉMf = cMf |cFf = 1 is computed
l G 1 = (exp(v.05 ), • , D, • , exp(lV )) with D =
4 by minimizing the objective Eq. (13), where the reliability indexes
(10) bj are computed by setting Smax,j = Smaxc ,j in Eq. (12); 3) The return
period q̂Smax is computed by intersecting the c-characteristic return
period plot with the straight line cFf = 1 (see again Fig. 3).
cvG = (0.20, 0.27, 0.33, 0.40) (11) The characteristic period is represented in terms of years (and not
1
in number of cycles) to have consistent and homogeneous definition
where V = N (l V , sV ) is the natural logarithm of the CAFL and v.05 of the characteristic load effect for all three verification schemes.
represents the 0.05 quantile of its distribution.

3
It is noted that both the epistemic uncertainty of the model parameters and the
aleatory randomness of the CAFL are taken into account in the definition of the fatigue
2 1y.
Based on bt = 4.2 [5], assuming independency. limit state function
 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472 469

The reliability index bj is then computed by solving the following

limit state equation5 :

gj = ẑj • S − SE,2, j (18)

Fatigue partial safety factors are calibrated by minimizing the follow-

γ >1 ing objective function:
Ff
γMf ⋅ γFf


20
 2
γFf<1 W(c) = wj • bj (c) − 3.00 (19)
j=1

where wj = 1 for j = 1, . . . , 20, since all load cases have the same
importance.
It is noted again that when cFf and cMf are determined with
(γ =1) Eqs. (14), (18), and (19), they are not independent and only the
Ff
product cFf • cMf is determined. Moreover, as for the verification
ρ, characteristic return period scheme 1, a three step-scheme allows for computing the resistance
partial safety factor ĉMf and the return period q̂SE,2 , corresponding to
Fig. 3. Illustration of c– characteristic return period plot. cFf = 1: Eqs. (14) and (18) are used in place of respectively Eqs. (8)
and (12).

3.2. Verification scheme 2: based on CA equivalent stress range 3.3. Verification scheme 3: based on damage accumulation

The design equation, for the generic load case j, is formulated as The design equation, for the generic load case j, is formulated as
follows: follows:

⎧ ⎫
Sc ⎪
⎪  
CAFL
  ⎪
⎪
Gj = zj • − cFf • SE,2,c,j (14) ⎪
⎨ ∞ ntot • fR cFf • s; kR 
cMf
⎪
⎬
cMf c,j
ntotc,j • fR cFf • s; kR
Gj = dc,0.5 − ds + ds
⎪
⎪ zj • Nc (s) zj • Nc (s) ⎪
⎪
⎪
⎩ CAFL (cMf )−m1 0 (cMf )−m2 ⎪
⎭
where zj is the design factor, Sc is the characteristic value of fatigue cMf

strength at 2 • 106 cycles, and SE,2,c,j is the characteristic value of (20)

equivalent stress range at 2 • 106 cycles, SE,2,j . The experimental vari-
ability in life of the design curves is included in the definition of the where zj is the design factor, dc,0.5 is the characteristic value of the
  m1
m
equivalent stress range, SE,2 = 2 • 1106 ki=1 Si 1 ni 1 , since the term critical damage, ntotc,j is the characteristic value of the total number of
 cycles, fR (cFf • s; kR ) is the pdf of the Rayleigh loading spectrum with
m −m
Si 1 expresses the variability in life Ni = C • Si 1 .
scale parameter equal to kR and Nc (s) is the characteristic value of
The geometrical interpretation of the resistance partial safety the number of cycles to failure. It is noted that the natural logarithm
factor is illustrated in Fig. 1. of the number of cycles to failure, Y, is modeled as follows [4]:
SE,2 is modeled as a log-Normal random variable4 ,
 
log N2 lN2 , sN2 , having coefficient of variation equal to: 
m0 + m1 X + 4 (0, exp(s)) , for X > V

Y=
 m0 + VDm + (m1 − Dm) X + 4 (0, exp(s)) , for X ≤ V
2
cvN = exp sN −1 (15)
2 2 (21)

In order to take into account both the effects of position and dis-
persion of loading term, 20 load cases are built by considering all
possible combinations of the following values of lN2 and cvN :
2
lN2 = (3.40, 3.57, 3.75, 3.92, 4.10)
cvN = (0.10, 0.13, 0.17, 0.20)
2 
The optimal design, ẑj , for the load case j, is computed by solving the
design Eq. (14), Gj = 0, for assigned values of partial safety factors.
where m0 and m1 are the intercept and the slope of the S-N curve
above the CAFL, 4(0, exp(s)) is the error term and Dm = m1 − m2 ,
with m2 representing the slope of the S-N curve below the CAFL. The
higher slope below the CAFL expresses the fact that stress ranges
below the CAFL do not contribute to crack propagation until the
(16)
crack reaches a certain size (decreasing fatigue limit concept, first
introduced by Haibach [8] and then used in the Eurocodes).
The geometrical interpretation of the resistance partial safety
(17) factor is illustrated in Fig. 2.

ntot is modeled as a Gumbel random variable, G3 108 , sG3 , having
a coefficient of variation cvG .
3
In order to take into account both the effects of uncertainty on
loading spectrum stress ranges and uncertainty on total number of

4 5
Modeling the logarithm of the equivalent stress range as a Normal random vari- It is noted that both the epistemic uncertainty of the model parameters and the
able allows for consideration of the variability on the maximum stress range obtained aleatory randomness of the fatigue life are taken into account in the definition of the
from fatigue load model as well as of the variabilities on different k factors. fatigue limit state function.
470 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472

cycles, 32 load cases are built by considering all possible combina- Table 2
tions of the following values of kR and cvG : VA S-N model parameters.
3

kR : r = (0.05, 0.08, 0.11, . . . , 0.25) (22)

Y|V N (lY|V , exp(s)) N (lY|V , exp(s))

V N (lV , exp(sV )) N (lV , exp(sV ))
cvG = (0.20, 0.27, 0.33, 0.40) (23)
3 l Y |V f(m0 , m1 , s) f(m0 , m1 , Dm, s)
m0 N (28.66, 2.85) N (25.77, 0.95)
where r is the percentage of stress range cycles exceeding the m1 N (−3.42, 0.79) N (−2.67, 0.21)
characteristic value of the CAFL. Rayleigh distribution was chosen s N (−0.54, 0.20) N (−1.05, 0.14)
lV N (3.45, 0.06) N (3.86, 0.13)
to model the loading spectrum since this probability distribution sV N (−1.62, 0.41) N (−1.67, 0.50)
provides a quite accurate characterization of the upper tail of the Dm 6 5
spectrum, with only one distribution parameter. Dc log N (0.27, 0.58) log N (−0.31, 0.60)
The optimal design, ẑj , for the load case j, is computed by solving dc,0.5 1.31 0.74
CAFL 21 MPa 30 MPa
the design Eq. (20), Gj = 0, for assigned values of partial safety
a05 27.58 25.14
factors. The reliability index bj is then computed by solving the
following limit state equation6 :
⎧ ⎫
⎪
⎨ ∞ 
exp(V) ⎪
⎬ Table 3
ntotj • fR (s; kR ) ntotj • fR (s; kR ) Resistance partial safety factors and related characteristic return period (cFf = 1.0).
gj = Dc − ds + ds (24)
⎪
⎩ ẑj • N(s) ẑj • Nc (s) ⎪
⎭
exp(V) 0

Fatigue partial safety factors are calibrated by minimizing the follow-

(1) ĉMf = 1.80 ĉMf = 2.27
ing objective function: Smax ≤ CAFL
q̂Smax = 750 years q̂Smax = 500 years
cMf
(2) ĉMf = 1.33 ĉMf = 1.26

32
 2 SE,2 ≤ Sc
cMf q̂SE,2 = 750 years q̂SE,2 = 1000 years
W(c) = wj • bj (c) − 3.00 (25) (3) ĉMf = 1.37 ĉMf = 1.45
n
tot
j=1 ni
N(cMf )
≤ dc,0.5 q̂ntot = 500 years q̂ntot = 500 years
i

where wj = 1 for j = 1, . . . , 32, since all load cases have the same
importance.
It is noted again that when cFf and cMf are determined with Table 4
Eqs. (20), (24), and (25), they are not independent and only a com- Comparison of design values for verification schemes 1 and 2.
bination of them is determined. Moreover, as for the verification
scheme 1, a three step-scheme allows for computing the resistance
partial safety factor ĉMf and the return period q̂SE,2 , corresponding to
cFf = 1: Eqs. (20) and (24) are used in place of respectively Eqs. (8) ML-MCS Eurocode ML-MCS Eurocode
and (12).
1 CAFL[MPa] 21 33 31 29
It is recalled here that in order to increase the confidence in
CAFLd [MPa] 12 24 14 21
characteristic and design values of fatigue resistance with respect 2 Sc [MPa] 46 45 53 40
to current standards, ML-MCS methods [4] is used to characterize Sc,d [MPa] 35 33 42 30
fatigue resistance terms for all three verifications schemes.

4. Case studies
The framework built in Section 3 was applied to two typical
bridge fatigue sensitive joints: 1) Welded cover plate; and 2) Welded
in-plane gusset. The VA S-N stochastic model of these two details was
estimated in [4] using ML-MCS approach. The parameters of VA S-N
curves are resumed in Table 2 for the convenience of the reader. It is
recalled that the cover plate detail and the in-plane gusset detail are
classified respectively as FAT 45 and FAT 50 in Eurocode 1993-1-9.
5. Results
The resistance partial safety factor, ĉMf = cMf |cFf =1 , and the char-
acteristic return period, q̂ = q|cFf =1 are presented in Table 3, for the
three verification schemes7 .
6
It is noted that both the epistemic uncertainty of the model parameters and the
aleatory randomness of the CAFL, of the fatigue life and of the critical damage sum, are
taken into account in the definition of the fatigue limit state function.
7
It is noted that the characteristic return period, q̂, represents the number of years
needed to exceed the characteristic value and not a life factor.
For both details the formation of a fatigue crack could rapidly lead
to structural failure, therefore the case Safe life-High consequences has
to be chosen as reference in Eurocode 1993-1-9. For this case, Table
3.1 in Eurocode 1993-1-9 recommends cMf = 1.35.
It has to be noted that values of cMf |cFf =1 presented in Table 3
cannot be directly compared with recommended values of cM f |cFf =1
in Eurocode standards, due to the following reasons: 1) For the veri-
fication schemes 1 and 2, cM f |cFf =1 values in Table 3 refer to ML-MCS
based CA S-N curves, which are different from Eurocode standards
characteristic S-N curves; 2) For the verification scheme 2, cM f |cFf =1
values in Table 3 refer to ML-MCS based VA S-N curves and ML-
MCS approach-based critical value of accumulated damage, dc,0.5
(see Table 2), which are different from respectively the Eurocode-
based S-N curves and Eurocode-based critical value of accumulated
damage8 .
But results for verification schemes 1 and 2 can be compared to
Eurocode standards in terms of design value of CAFL and design value
of fatigue strength at 2 • 106 cycles (see Table 4).
8
dc,0.5 = 1.0 for all details
 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472
As for results for verification scheme 3, they can be compared to
Eurocode standards in terms of design VA S-N curves (see Figs. 4 and
5). In order to make direct comparison with Eurocode design S-N
curves, the ML-MCS design S-N curves were re-scaled at dc,0.5 = 1.0
(see re-scaling method in [4]).
6. Conclusions and discussion
This paper presented a new framework for calibration of partial
safety factors used in fatigue design under variable amplitude
loadings.
The presented framework improves consistency in achieving
target levels of safety by:
• Defining one different design equation for each of three
design verification criteria proposed in Eurocode standards;
this allows to distinguish three different resistance partial
safety factors, cMf , to be used in: 1) Fatigue design based on
CAFL; 2) Fatigue design based on CA equivalent stress range
at 2 • 106 cycles; and 3) Fatigue design based on accumulated
damage;
• Defining a method to assess the characteristic return period
of the load effect which allows to set cFf = 1.0 in the design
equation;
• Using ML-MCS-based S-N curves which results in a more
realistic consideration of the CAFL and of the fatigue strength
in HCF region (i.e. in computation of damage sum).
The application of the presented framework to two typical bridge
fatigue sensitive welded joints shows that:
• Calibration of partial safety factors for verification scheme 1
(based on CAFL) gives higher values of ĉMf with respect to
the verification scheme 2 (based on equivalent stress range at
2 • 106 cycles), for both details (see Table 3) . This is probably
due to the higher fatigue life scatter in the HCF region with
respect to the scatter in the finite life region.
• The ĉMf for the verification scheme 1 is considerably higher
for the in-plane gusset (ĉMf = 2.27) with respect to the cover
plate (ĉMf = 1.80). This is probably due to the more dispersed
CAFL distribution of the in-plane gusset detail with respect to
the cover plate detail (see l V distributions in Table 2).
60 ML−MCS characteristic
Eurocode characteristic
ML−MCS design (d =1.0)
c,.5
Eurocode design (d
c,.5
S [MPa]
30
20
15 6 7
10 10 10
N
Fig. 4. Comparison of VA design S-N curves for verification scheme 3, cover plate.
471
60 ML−MCS characteristic
Eurocode characteristic
ML−MCS design (d =1.0)
c,.5
Eurocode design (d
c,.5
=1.0)
S [MPa]
30
20
15 6 7 8
10 10 10
N
Fig. 5. Comparison of VA design S-N curves for verification scheme 3, in-plane gusset.
• For all verification schemes the return period of characteristic
load effect, q̂, ranges between 500 and 1000 years (see Table 3).
• Eurocode standards give unsafe estimate of the design value
of the CAFL, for both considered details (see Table 4). For
the in-plane gusset detail this is due to the unsafe Eurocode-
based calibration of resistance partial factor (see Tables 1 and
3), while for the cover-plate gusset this is both due to the
unsafe Eurocode-based calibration of resistance partial factor
(see again Tables 1 and 3) as well as to the unsafe Eurocode-
based estimate of the characteristic value of the CAFL (see first
line in Table 4).
• Eurocode standards and ML-MCS approach give similar values
of ĉMf for the verification scheme 2 (see Tables 1 and 3).
The considerable difference on the design fatigue strength at
2 • 106 cycles, for the in-plane gusset detail, is due to the over-
conservative Eurocode-based estimate of the characteristic
fatigue strength (see fourth line in Table 4).
• Fig. 4 shows that the Eurocode-based VA design S-N curve
of the cover-plate detail is over-conservative with respect to
the ML-MCS-based S-N curve for stress ranges higher than 19
MPa, while it becomes slightly under-conservative for stress
ranges lower than 19 MPa. Fig. 5 shows that the Eurocode-
based VA design S-N curve of the in-plane gusset detail is over-
conservative with respect to the ML-MCS based S-N curve, at
all stress ranges.
=1.0)
The reliability framework for calibration of fatigue partial safety
factors, set up in this paper and applied to two fatigue-sensitive
details, constitutes a powerful tool that can be used to revise the
Eurocode basis for fatigue design of structures. The three Eurocode
formats for fatigue design and associated partial safety factors can
be revised by considering different fatigue sensitive details and
by further differentiating between: 1) Verification based on CAFL
exceedance; 2) Verification based on equivalent stress range at 2 • 106
cycles, using lambda factors; and 3) Verification based on damage
accumulation.
8
Acknowledgments
This research was supported by the Swiss Federal Roads Office
(FEDRO Project AGB 2010/003).
472 L. D’Angelo, A. Nussbaumer / Journal of Constructional Steel Research 139 (2017) 466–472

References [5] Joint Committee Structural Safety, JCSS Probabilistic Model Code: Resistance
Models, 2013.
[6] European Committee for Standardization, Eurocode 1 : Actions on Structures.
[1] European Committee for Standardization, Eurocode EN 1990 - Basis of
Part 2: Traffic Loads on Bridges, 2002.
Structural Design, 2001.
[7] G. Sedlacek, H. Eisel, W. Hensen, B. Kühn, M. Paschen, Leitfaden Zum DIN-Fach-
[2] M.H. Faber, J.D. Sørensen, Applications of statistics and probability in Civil
bericht 103: Stahlbrücken, Ernst & Sohn. 2003.
Engineering, in: A. Der Kiureghian, S. Madanat, J.M. Pestana (Eds.), Proceedings
[8] E. Haibach, Modifizierte Lineare Schadensakkumulations-Hypothese
of the 9th International Conference on Applications of Statistics and Probability
Zur Berücksichtigung Des Dauerfestigkeitsabfalls Mit Fortschreitender
in Civil Engineering, Millpress, San Francisco, California, 2003, pp. 927–935.
Schädigung.TM Nr. 50/70., Laboratorium für Betriebsfestigkeit. 1970.
[3] European Committee for Standardization, Eurocode 3: Design of Steel
Structures - Part 1–9: Fatigue, 2005.
[4] L. D’Angelo, A. Nussbaumer, Estimation of fatigue S-N curves of welded joints
using advanced probabilistic approach, Int. J. Fatigue 97 (2017) 98–113. https://
doi.org/10.1016/j.ijfatigue.2016.12.032.