You are on page 1of 8

PROBABILISTIC ASSESSMENT OF WELDED JOINTS VERSUS

FATIGUE AND FRACTURE


By Mladen Lukic1 and Christian Cremona2

ABSTRACT: This paper presents a probabilistic reliability assessment procedure for steel components damaged
by fatigue. The crack growth model is based on the principles of fracture mechanics theory. It is compared to
experimental results and gives a good prediction. The fatigue safety margin includes the crack growth from an
initial crack depth to a final crack depth determined according to the fracture mechanics theory: brittle and
ductile fractures. The reliability calculus is performed using a first-order reliability method. The sensitivity
analysis of different parameters shows that some variables can be taken as deterministic. Applications are made
on a transverse-stiffener-to-bottom-flange welded joint of a typical steel bridge.

INTRODUCTION The traffic load data must be related to component stresses


at locations expected to be vulnerable to fatigue damage. Ex-
The phenomenon of fatigue is a slow process that introduces
amples of such sections for a steel girder include girders with
numerous uncertainties. In welded joints, cracks are often lo-
cover plates, girder joints and splices, and section disconti-
calized at the weld. Welding induces some defects that help
nuities.
small cracks to appear. They can grow under loading and lead
The proposed approach consists of assessing the reliability
to joint failure. The conditions governing crack growth are
of welded joints based on the principles of fracture mechanics.
structural geometry, initiation site, material characteristics, and
The originality of this paper is the combination of the crack
loadings. In general, these conditions are random. Therefore,
growth model expressed by the Paris law with the criteria of
an appropriate analysis of fatigue phenomena consists of treat-
fracture toughness and plastic yield, which allows one to eval-
ing the problem in a probabilistic manner. Although fatigue
uate the risk of fatigue damage on a better estimation of the
analysis of bridges is well established [e.g., Ponts (1996)],
critical crack depth. This paper also attempts to highlight the
the use of probabilistic methods has only been considered in
effect of the different variables introduced by the model on
recent years (Byers et al. 1997). The standard probabilistic
the reliability assessment. The total number of variables can
method of fatigue analysis for highway bridges consists of the
be large enough to try to reduce this number by a sensitivity
development of stress-range distributions from field data or
study. This analysis is performed on a transverse-stiffener-to-
simulation, use of the Palmgren-Miner damage rule for fatigue
bottom-flange welded joint of a typical steel bridge.
damage analysis along with an appropriate S-N relationship
for the critical structural details, and use of a probability func-
tion to describe the reliability of a critical component and its CRACK GROWTH
corresponding fatigue life. Methods based on crack growth and
The model used in this paper is the commonly adopted Paris
fracture mechanics have also been used but to a more limited
law (Paris and Erdogan 1963) corresponding to the opening
extent [e.g., Zhao and Halder (1996)]. For highway bridges,
of a semielliptical crack in the flange of a transverse-stiffener-
the techniques of fatigue and fracture reliability have been
to-bottom-flange welded joint (Fig. 1)
applied mainly to
da
Condition assessment and estimation of the remaining = CK mI (1)
dN
lifetime of bridges (Cremona 1996), where probabilistic
methods can be used to obtain estimates of the adequacy where a = crack depth; N = number of loading cycles; KI =
of the existing structure, need for increased inspection in stress intensity factor range [Mode I (tension) crack growth];
the future to prevent failure, and approximate remaining and C and m = two material parameters. The Paris law is based
fatigue lifetime based on projections of the future loads on the elasticity theory and omits plasticity inside the crack.
Development of probability-based design stress ranges for In logarithmic coordinates, the Paris law is represented by a
fatigue-critical bridge components (Kretz and Jacob line, which is not merely the reality. Eq. (1) overestimates the
1991), where accurate traffic load data can be acquired crack growth in the vicinity of KIth and underestimates it in
through weight-in-motion systems (an extensive amount
of such data is available showing distribution of load by
its time of appearance, transversal position, speed, number
of axles, gross weight of axles, and distance between
axles)
1
Res. Engr., Ctr. Technique Industriel de la Constr. Metallique, Do-
maine de Saint-Paul, 78470 Saint-Remy le`s Chevreuse, France; formerly,
Laboratoire Central des Ponts et Chaussees, Sect. Durabilite des Ouvrages
dArt, 58, Blvd. Lefebvre, 75732 Paris CEDEX 15, France.
2
Res. Engr., Laboratoire Central des Ponts et Chaussees, Section
Durabilite des Ouvrages dArt, 58, Blvd. Lefebvre, 75732 Paris CEDEX
15, France.
Note. Associate Editor: Jamshid Mohammadi. Discussion open until
July 1, 2001. To extend the closing date one month, a written request
must be filed with the ASCE Manager of Journals. The manuscript for
this paper was submitted for review and possible publication on May 21,
1999. This paper is part of the Journal of Structural Engineering, Vol.
127, No. 2, February, 2001. ASCE, ISSN 0733-9445/01/0002-0211
0218/$8.00 $.50 per page. Paper No. 21024. FIG. 1. Transverse-Stiffener-to-Bottom-Flange Welded Joint

JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001 / 211

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
the vicinity of KIf . Therefore, this equation is only valid for Mk (a) = stress concentration factor, depending not only
stress intensity factor ranges greater than the threshold stress on the crack depth a but also on other geometric param-
intensity range KIth and smaller than the fracture stress in- eters, such as flange thickness b, weld height h, and weld
tensity range KIf . The model keeps its linearity for KI far angle . The stress concentration factor represents the
from KIth. This threshold value can be taken into account for magnification factor to take account of stress concentra-
discriminating between damaging and nondamaging cycles. In tions due to specific structural detail. The classical solu-
the present study, the stress intensity factor range threshold tion (Sedlacek et al. 1997) is adopted for this factor


KIth is taken equal to zero (all cycles are damaging). Details w
regarding the release of that hypothesis can be found, for ex- a
Mk = v
ample, in Lukic (1999). b
The stress intensity factor range can be expressed by
where the parameters v and w are calculated according to
KI = Y(a)Mk (a)S a Hobbacher (1993)


2
where h h h
v = 0.8068 0.1554 0.0429 0.0784 tg
b b b


Y(a) = stress intensity correction factor, a function of the 2
crack depth a as well as of other geometric parameters h h
concerning the welded joint, such as the crack half-length w = 0.1993 0.1839 0.0495
b b


c, flange thickness b, and flange width d. The solution
adopted in this paper is from Newman and Raju (1983), h
0.0815 tg
where the cracked bottom flange is represented by a plate b
with a semielliptical surface crack


2 4
S = stress range at the hot spot for the uncracked section
1 a a (effective stress range).
Y(a) = M1 M2 M3 fw
1 1.464(a/c)1.65 b b
With these developments, the initial form of the Paris law
with [(1)] is transformed into the following:
M1 = 1.13 0.09a/c
da
= C(Y(a)Mk (a) a)mS m (2)
0.89 dN
M2 = 0.54
0.2 a/c
MODEL VALIDATION
1
M3 = 0.5 14(1 a/c)24
0.65 a/c To validate the crack growth model adopted, a set of fatigue
tests performed at the Rheinische-Westfael Technische Hochs-
1 chule, Aachen, Germany, was used (Bleck et al. 1998). In these


fw =
1 a 3 tests, the design concept to avoid brittle fracture (Eurocode 3)
cos was checked by means of experimental investigations of thick
d/b a/c b
plates from fine-grained thermomechanically rolled steels. Two
steel grades, S355M and S460M (EN 10025), at 80-mm thick-
TABLE 1. Material and Geometrical Characteristics of Spec-
imens ness, were chosen for small-scale tests and component-like
large-scale tests as well. Test datainputs and outputsare
Variable S35D21 S35D23 S46D22 S46D24 given in Tables 1 and 2. The effect of the structural details
(1) (2) (3) (4) (5) was covered by a multiplication factor from a finite-element
d (mm) 198.0 198.5 198.0 198.2 analysis carried out at the Institute of Steel Construction of
h (mm) 28.28 28.28 28.28 28.28 the Rheinische-Westfael Technische Hochschule.
(degrees)


45.00 45.00 45.00 45.00 0.076
m 3.250 1.730 2.460 2.430 a
C[1013 mm/cycles(N/mm3/2)m ] 0.650 3,820 45.10 63.60 Mk = 0.7919
a/c 0.715 0.675 0.515 0.602
b
b (mm) 81.46 81.42 81.50 81.60
a0 (mm) 3.430 4.320 4.600 3.300
The results are presented in Fig. 2. A fair agreement between
tests results and the crack growth model can be noticed.

TABLE 2. Loading Characteristics of Specimens


S35D21 S35D23 S46D22 S46D24
Si Ni Si Ni Si Ni Si Ni
(MPa) (cycles) (MPa) (cycles) (MPa) (cycles) (MPa) (cycles)
(1) (2) (3) (4) (5) (6) (7) (8)
71.00 512,000 72.00 508,600 70.00 503,000 68.00 500,600
48.00 634,000 50.00 658,200 52.00 648,800 50.00 722,500
71.00 1,074,000 72.00 1,084,800 70.00 1,100,400 68.00 1,798,700
48.00 1,125,000 50.00 1,209,800 52.00 1,244,900 50.00 1,895,600
71.00 1,490,400 72.00 1,649,600 70.00 1,691,400 68.00 2,496,000
48.00 1,622,400 50.00 1,792,300 52.00 1,839,400 50.00 2,598,100
71.00 1,890,400 72.00 2,317,100 70.00 2,266,000 68.00 3,119,600
48.00 1,989,900 50.00 3,198,600
71.00 2,258,400 68.00 3,427,100
48.00 2,435,400
71.00 2,526,400

212 / JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
adopted in this paper is therefore to perform an analysis based
on the R6 rule (Milne et al. 1986) to obtain the critical crack
depth.
Two criteriafracture and plastic collapselimit the load
capacity of a cracked structure. The elastic and plastic com-
ponents of the analysis are separated in a way that aids cal-
culation, facilitates a sensitivity analysis, and provides an in-
sight into the way in which a structure will perform. First, the
stress intensity factor should not exceed the fracture toughness.
Second, the applied load should not exceed the plastic yield
load. These criteria are defined as follows.
The value of Kr represents the measure of the proximity to
linear elastic fracture mechanics failure
KI
Kr = 1 (4)
KIC
where KI = Mode I (tension) stress intensity factor; and KIC =
fracture toughness.
The value of Lr represents the measure of the proximity to
plastic yield
FIG. 2. Comparison between Test Results and Crack Growth P
Model Lr = L max
r (5)
PL
where P and PL = applied and plastic yield loads, respectively.
SAFETY MARGIN Furthermore, certain interactions exist between these two
A limit state defines the frontier between damage and non- modes of failure and the R6 rule treats their safety margin as
damage. On the other hand, a safety margin expresses a function g (Lr , Kr ) (Sedlacek et al. 1997)
the distance between the actual performance of the structure 1
and the corresponding limit state. For fatigue reliability g (Lr , Kr ) = Kr , Lr 1 (6a)
assessment, a straightforward safety margin can be defined 1 0.5L r2
by g (Lr , Kr ) = 0, Lr > 1 (6b)
M = af a(t)
Measure of Proximity to Plastic Yield Lr
where af = final or failure crack depth, which can be chosen
This is a measure of how close the structure containing the
equal to some conventional value or calculated according to
flaw is to the plastic yield. The applied loads to be used in
fracture mechanics criteria; and a(t) = crack depth in time t
evaluating Lr [(5)] are those contributing to plastic collapse.
from the beginning of crack growth.
The yield load for part-through cracks is the load needed
The crack depth a(t) is difficult to obtain, and another form
to cause plasticity to spread across the remaining ligament,
of the safety margin can be obtained from (2), where the var-
calculated for an elastic perfectly plastic material. The effect
iables a and N are separated and the equation integrated from
of the flaw must be included in evaluating the plastic yield
the beginning of crack growth to time t. Welded joints such
load
as those in steel bridges, are generally submitted to variable
amplitude stresses. It is therefore important to define the sta- sG sQ bd
tistics of the stochastic process S. Two possibilities are given Lr =
fy a 2
for studying the fatigue phenomenon under variable amplitude bd
2a/c
loading: range counting or equivalent stress range approach.
If each cycle is assumed not to depend on its preceding one, where sG , sQ , and fy represent the dead load stress, peak traffic
the equivalent stress range approach can be used (Madsen et load stress, and yield strength, respectively.
al. 1988)

Measure of Proximity to Linear Elastic Fracture


af
dx N0
M= Ct 1 (E [S] (E [S])) m Mechanics Failure, Kr
a0 (Y(x)Mk (x) x)m t
(3) A common input for any evaluation of this measure [(4)] is
the linear elastic stress intensity factor KI , which is evaluated
where a0 = initial crack depth; = annual number of cycles;
from the elastically calculated stress field in the uncracked
t = time elapsed from the joint commissioning; N0 = number
body at the location of the crack. Code methods are quick and
of cycles related to the initiation phase; and (E [S]) = uncer-
easy and are applicable o semielliptical flaws. They may, how-
tainty factor for the expected value E [S] of the stress range S.
ever, be overly pessimistic for very steep stress gradients, but
that is not the case. Therefore, for the crack tip (Sedlacek et
CRITICAL CRACK DEPTH DETERMINATION al. 1997)
Up to now different possibilities have been proposed, of KI = Y(a) (Mk (a)(sG sQ ) sS ) a
which the two most frequently used have been to choose the
critical crack depth equal to the half-thickness b/2 (Cremona where sS represents the residual stress.
1996) or the full-thickness b (Zhao and Haldar 1996) of the The fracture toughness KIC of the material containing the
bottom flange (Fig. 1). Such choices can provide optimistic flaw should preferably be obtained by direct testing. If the
values (Lukic 1999) in comparison to the critical crack depth fracture toughness cannot be measured directly, indirect meth-
determined using fracture mechanics criteria. The approach ods of estimating fracture toughness may be used provided
JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001 / 213

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
these have been validated for the material in question. Such TABLE 3. Variables Used from Existing Joint
methods include, among others, the use of appropriate data- Distribution VX
bases and material specifications. The fracture toughness can Variable X type X X (%)
thus be expressed as a function of the Charpy-V transition (1) (2) (3) (4) (5)
temperature (Wallin 1995)
a0 (mm) Lognormal 0.125 0.045 36.0
KIC [MPAm] = 20 (KIC ) a/c Lognormal 0.390 0.160 41.0


b (mm) Normal 30.00 3.000 10.0
1/4
25 d (mm) Normal 800.0 8.000 1.00
{11 77 exp[0.019(Te TK 28 18)]} ma Normal 3.000 0.030 1.00
2a C a [1013 mm/cycles(N/mm3/ 2)m ]
2 Lognormal 2.503 0.923 36.9
a/c h (mm) Normal 8.400 0.700 8.33
(degrees) Normal 35.00 2.000 5.71
where (KIC ) represents the fracture toughness uncertainty fac- Te [C] Gumbel 9.658 3.995 41.4
tor (Nussbaumer 1997); Te = material temperature; and TK28 = minimum
test temperature for 28-J minimum average Charpy V-notch TK 28 [C] Normal 20.0 2.000 10.0
impact energy. The value of af , such that the fracture safety (KIC) Weibull 0.906 0.254 28.0
fy (MPa) Lognormal 345.0 34.50 10.0
margin [(6)] becomes the fracture limit state, g(Lr , Kr ) = 0, is sG (MPa) Lognormal 172.5 17.25 10.0
introduced into the fatigue safety margin [(3)]. sQ (MPa) Gumbel 40.75 1.221 3.00
maximum
PROBABILITY OF FAILURE sS (MPa) Lognormal 355.0 35.50 10.0
E [S] (MPa) Deterministic 7.800 0.000 0.00
The probability of failure of the structural element is iden- (E [S]) Normal 1.000 0.100 10.0
tical to the probability of the limit-state violation and can be (106 cycles/year) Normal 3.882 0.388 10.0
stated N0 (cycles) Deterministic 0.000 0.000 0.00

Pf = P(a(t)) > af ) = P(M < 0)


a
Correlation: (m, ln C) = 0.99.

The technique that has been considered is the first-order


reliability method. This method easily expresses the failure studies can serve. A European project (Carracilli et al. 1995)
probability in terms of a reliability index to determine the fatigue resistance of welded joints was used.
The dimensions and type of the welded joint tested in the study
Pf () were similar to the one analyzed in this paper.
where = cumulative probability function of a standard nor- Initiation Number of Cycles N0 . The number of cycles
mal distribution. to crack initiation typically relates to the formation of small
detectable cracks. It includes processes of nucleation as well
STUDY OF PARTICULAR WELDED JOINT as the propagation of small cracks. In the design for damage
tolerance, the presence of an initial defect is assumed a priori
A particular welded joint, Transverse Stiffener to Bottom and inspection intervals are set according to the application of
Flange, from an existing motorway composite steel and con- propagation mechanics. Crack initiation mechanics is of little
crete bridge has been considered (Lukic 1999). use for this purpose (McDowell 1996).
Yield Strength fy . Because the exact material properties
Basic Variables are not available, the European standards are used (Eurocode
Table 3 synthesizes characteristics of all variables used. 3 1992). Yield strength is taken as the S355-quality steel least-
Note that the threshold value of the growth of the stress in- nominal value for the nominal plate thickness between 16 and
tensity factor KIth is not considered in this example. Basic 40 mm (EN 10025 1993).
variables to be taken into account are obtained from three dif- Dead-Load Stress sG . The hypothesis is adopted that the
ferent sources: literature, bridge drawings, and measures and dead-load stress in the bottom flange of the joint remains con-
subsequent calculi. stant. It is supposed to be equal to half the yield strength,
which is considered to be the uppermost limit of the real dead-
Variables from Literature load stress and is therefore kept as conservative (Sedlacek et
Initial Crack Depth a 0 . Fatigue crack nucleation and al. 1997).
growth occurs along a progression of length scales ranging Residual Stress sS . The propagation of a fatigue crack is
from the order of 1 m to the scale of individual grains (Mc- influenced by local strains in the surroundings of the crack
Dowell 1996). The crack growth can be calculated starting resulting from two effects, applied loads and residual stresses.
from two different types of initial crack depth: smallest mea- The latter are generally divided into microscopical and mac-
sured or virtual calculated. In many cases measures of existing roscopical. From the linear elastic fracture mechanics point of
flaws have been made and the initial crack depth is considered view, only the macroscopical ones can be taken into account.
as the smallest crack depth measured (Engesvik and Torgeir They are supposed to be equal, not to the least thickness-de-
1983). pendent nominal-yield stress, but to the real one (Nussbaumer
Crack Shape a/c. In the case of welded joints, it is pos- 1997).
sible to say that the crack shape is found between 0 and 1. Test Temperature for 28-J Minimum Average Charpy
The crack shape depends not only on the structural detail but V-Notch Impact Energy TK28 . Because the exact material
also on local geometry, crack depth, coalescence phenomena, properties are not available, the European standards are used
and type and range of applied stress. For the sake of simplicity, (Eurocode 3 1992). The test temperature for 28-J minimum
it is preferable to choose the distribution laws in the function average Charpy V-notch impact energy is taken as the S355J2-
of the structural detail type only. For fillet welds, a lognormal quality steel nominal value for the nominal plate thickness
distribution can be adopted (Yamada et al. 1989). Furthermore, between 10 and 150 mm (EN 10025 1993).
this shape is not supposed to change during crack growth. Material Temperature Te . It is necessary to know statis-
Weld Height h and Weld Angle . This variable should tically the extreme values that can be reached during the joint
be either directly measured on the site or taken from the cor- lifetime. For every random variable described by a distribu-
responding drawings. If neither of the two is possible, similar tion, its extreme values also correspond to a random variable,
214 / JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
so the air shadow minimal temperature follows one of these Variables from Measures and Subsequent Calculi
distributions. Its parameters are adjusted on the English data
as the comparison of minimal temperatures shows that the The relationship between traffic loads and member stresses
French and the English air temperatures are similar. For the can be obtained by direct stress measurements using strain
composite bridges, the material temperature may be considered gauges or by traffic measures and subsequent structural anal-
equal to the air shade temperature (Nussbaumer 1997). ysis. Both may need advanced data acquisition systems allow-
Fracture Toughness Uncertainty Factor (KIC ). In the ing on-site data reduction and processing. The duration of field
brittle fracture or transition zone, the material toughness fol- data collection for highway bridges is often limited to a few
lows one of the extreme value distributions. That is taken into days. For most bridges this is adequate to obtain a reasonable
account in the resilience-toughness equation with the intro- estimate of stress or load ranges experienced by a bridge. Such
duction of the fracture toughness uncertainty factor. This ex- field data would probably not detect the high stress levels that
pression is applicable to quasi-static loads only. In the case of occur when overloaded vehicles are allowed to use the bridge
road bridges, the loads may be considered as quasi-static and by special permit. The field data described can be represented
the fracture toughness is supposed to follow a Weibull distri- with theoretical stress-range distribution models.
bution (Wallin 1995). The calculated variables are issued from traffic measure-
Material Parameter m. Different theories exist in quali- ments and influence lines of the bridge. This procedure gives
fying the two material parameters. Some consider them as ran- histograms of certain phenomena. The histograms used in this
dom variables, with or without correlation between them. Oth- work are level crossings and rain-flow histograms.
ers consider the parameter m as a material dependant constant. Peak Traffic Load Stress sQ. To verify the reliability of
The approach adopted in this paper is to consider it as a nor- a bridge, it is necessary to know the extreme values of the
mally distributed random variable, the value of which is de- effects due to traffic load that can be reached during the struc-
termined from the tests on structural steel (Bremen 1989) and tural lifetime. If the traffic recordingsone week in general
to correlate it with the other material parameter C. are available, the target consists of extrapolating the ex-
Material Parameter C. As can be shown, this variable is treme values from the recording period to the structural
the most influential variable in the model adopted. Therefore, lifetime. When the extreme values are statistically studied, it
special attention must be paid in the determination of its value. is often the case that the particular, extreme value distributions
Contrary to the parameter m, this parameter is always consid- are used. The Gumbel distribution is the most appropriate for
ered as random. To obtain the most precise values of its dis- traffic load effects and is used for this variable in this study.
tribution, the same test that served in the determination of the The peak traffic load stress is obtained in two phases (Cre-
material parameter m served for this one. The natural loga- mona and Carracilli 1998):
rithm ln C was supposed normally distributed between its ex-
treme values in the tests, namely, 29.78 and 28.38, with Fitting the Rices formula on the queue of the level
95% confidence level (Bremen 1989), which gives it the mean crossing histogram is based on the hypothesis that the
of 29.01 and standard deviation of 0.3571. The values of effect is a stationary Gaussian process. The Rices formula
this material parameter are shown in Table 3. permits the expression of the density function for level
Correlation between Material Parameters (m, ln C). crossings


Many studies have shown not only that the material parameters 2 2
follow the normal distributions but also that there exists a 1
1 x 1 x
(x) = exp = 0 exp
strong negative correlation between them. That correlation is 2 2 2
not linked to a physical property of the material but comes
rather from the mathematical expression of the Paris law. where = mean of the Gaussian stationary process for
high values of x; = process standard deviation; and
Variables from Drawings = standard deviation derivative. The fitting procedure
consists, therefore, in identifying these three parameters.
Bottom flange thickness b and bottom flange width d are When this identification is brought to its end, it is possible
the only variables taken from the appropriate drawings. to extrapolate the extreme values for any return period tr .

FIG. 3. Reliability Index and Failure Probability versus Time

JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001 / 215

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
TABLE 4. Proposal about Whether Variables from Existing
Joint SHould Be Probabilistic or Deterministic

Mean Standard deviation


Variable influence influence Proposal
(1) (2) (3) (4)
b Deterministic
d Deterministic
h Deterministic
Deterministic
fy Deterministic
sG Deterministic
sQ Deterministic
sS Deterministic
Te Deterministic
TK 28 Deterministic
a0 Probabilistic
a/c Probabilistic
Probabilistic
m Probabilistic
(KIC ) Probabilistic
(E[S]) Probabilistic
ln C Probabilistic
Note: = growing mean or standard deviation implies longer life;
= growing mean or standard deviation implies no significant change; and
= growing mean or standard deviation implies shorter life.

FIG. 5. Lifetime in Function of Initial Crack Depth a 0

where ts = reference period, usually 50 or 100 years.


Fitting the Gumbel distribution parameters on the extrap-
olated values effects for different return periods is ob-
tained for different values of from corresponding values
of x .

Effective Stress Range Mean E[S]. This variable repre-


sents the mean of the corresponding rain-flow histogram.
Annual Number of Cycles . This variable is the number
of effective stress ranges obtained from the corresponding
rain-flow histogram.

Analysis
The results of the procedure are shown in Fig. 3, where the
evolution of the reliability index as well as the evolution of
the failure probability with time are given. It is obvious that
the use of a conventional crack depth at failure (b or b/2) gives
overestimated reliability indexes in comparison with the ap-
proach including fracture mechanics criteria. In this latter case,
note that a 3.8 reliability index (minimal value usually re-
quired for design in bridge engineering in Eurocodes) is ob-
tained for a 35-year crack propagation period. According to
FIG. 4. Lifetime in Function of Weld Height h this example, the reliability index after 100 years is about 1.7,
which corresponds to the failure probability of about 5%.
The probability of failure also corresponds to the lifetime
The extrapolated value x , being a probability to be probability function. The important values can be deduced: the
exceeded, is given by mean lifetime is 280 years, median lifetime is 235 years, and


lifetime with 5% probability of failure is 100 years. Sensitivity
0 ts analysis in relation to the lifetime of the joint was performed
x = 2 ln(0 tr) 2 ln
ln(1 ) for the whole set of variables, initially taken as random. This
216 / JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
CONCLUSIONS
A probabilistic model for assessing fatigue damage has been
proposed in this paper. The crack growth model is the classical
Paris law including appropriate stress concentration and stress
intensity correction factors. Compared to test results, it pro-
vides fair predictions of the crack growth. The model pre-
sented in the paper does not include the threshold value
Kth . A more general approach can nevertheless be developed
allowing one to distinguish between damaging and nondamag-
ing cycles [e.g., Lukic (1999)]. The model used in the relia-
bility analysis is not the kinetic expression of the Paris law
but an integral form more amenable to computations. That
integral model avoids the step-by-step calculation of crack
growth. It requires one to determine a critical size that can be
assessed in different ways. To be physically realistic, the crit-
ical crack size is calculated from a model (called the R6 rule)
that takes into account the risk of fracture and plastic collapse.
The integral Paris law and the R6 rule then constitute a com-
pound model for assessing failure by fatigue and fracture col-
lapse. That model has been applied to a typical transverse-
stiffener-to-bottom-flange welded joint used in steel bridges.
The results show that its use leads to improved safety, as it is
more pessimistic than models classically used. Sensitivity
analysis has shown that only a limited number of variables
must be taken into account in the probabilistic manner. Any
other variable may be treated deterministically. That provides
a gain in computation as well as in precision. From 17 vari-
ables, only 7 have to be taken as random because of the strong
influence of their coefficient of variation on the reliability anal-
ysis. Note that a probabilistic model as presented in this paper
helps to easily calculate updated failure probabilities by means
of conditional probabilities (Bayesian analysis) as soon as new
inspection results on data are available.

FIG. 6. Lifetime in Function of Material Parameter m


APPENDIX I. REFERENCES
Bleck, W., et al. (1998). Composite bridge design improvement for high
speed railways. ECCS Proj. 7210/SA/128, Draft, Final Rep., Rhein-
analysis showed that some variables can be taken as deter- ische-Westfael Technische Hochschule, Aachen, Germany.
ministic, if necessary. An omission factor i for the i th variable Bremen, U. (1989). Amelioration du comportement a` la fatigue

dassemblages soudes: Etude et modelisation de leffet de contraintes
Xi is defined as the ratio between the reliability index calcu-
lated with all random variables minus the variable Xi taken as residuelles [Improvement of the fatigue behavior of welded joints:
Study and modeling of residual stresses]. PhD thesis, EPFL, Lau-
deterministic and equal to a specific value and the reliability sanne, France (in French).
index calculated with all variables Byers, W. G., Marley, M. J., Mohammadi, J., Nielsen, R. J., and Sarkani,
i (t) xi S. (1997). Fatigue reliability reassessment applications: State-of-the-
i (t) = art paper. J. Struct. Engrg., ASCE, 123(3), 277285.
i (t) Carracilli, J., Le Pautremat, E., Jacob, B., and Galtier, A. (1995). Com-
portement en fatigue des poutres metalliques de ponts [Fatigue behav-
An omission factor in the vicinity of 1 leads one to consider ior of bridge steel girders]. ECCS Proj. 7210/SA/311, Draft, Final
the variable Xi as deterministic and equal to the specific value. Rep., LCPC, Paris (in French).
It is nevertheless essential to dispose of precise statistical in- Cremona, C. (1996). Reliability updating of welded joints damaged by
formation for all other variables. The proposal to distinguish fatigue. Int. J. Fatigue, 18(8), 567575.
those variables that can be taken as deterministic from the Cremona, C., and Carracilli, J. (1998). Evaluation of extreme traffic
others that must be taken as probabilistic is given in Table 4. loads effects in cable stayed and suspension bridges by use of WIM
records. Proc., 2nd Eur. Conf. on Weigh-in-Motion of Road Vehicles,
The analysis has been performed on the basis of an omission 243251.
factor study for each variable (specific value chosen equal to EN 10025. (1993). Hot-rolled products of non alloy structural steels
the mean value). Some examples of this analysis are shown in Technical delivery conditions. European standard, CEN, Bruxelles.
Figs. 46 for a variable h that can be taken as deterministic, Engesvik, K. M., and Torgeir, M. (1983). Probabilistic analysis of the
variable a0 that leads to slightly sensitive results, and variable uncertainty in the fatigue capacity of welded joints. Engrg. Fracture
m that gives results highly sensitive to uncertainties. It is ob- Mech., 18(4), 743762.
Eurocode 3. (1992). Design of steel structures: General rules and rules
vious that the most important variables are those connected to for buildings. European prestandard, CEN, Bruxelles.
the fatigue crack growth only. Four of themboth material Hobbacher, A. (1993). Stress intensity factors of welded joints. Engrg.
parameters, effective stress range, and annual number of cy- Fracture Mech., 46(2), 173182.
clesmust be taken as probabilistic. Concerning the variables Kretz, T., and Jacob, B. (1991). Convoi de fatigue pour les ponts-route
connected to the fracture, only the fracture toughness uncer- mixtes [Fatigue train for composite road bridges]. Constr. Metallique,
tainty factor must be treated as probabilistic, whereas others 1, 4151 (in French).

Lukic , M. (1999). Evaluation et maintenance probabilistes des assem-
have a rather small influence and may be considered as deter- `
blages soudes vis-a-vis de la fatigue et de la rupture. Application aux
ministic. Finally, among the variables that influence both the ponts mixtes [Probabilistic assessment and maintenance of welded
fatigue crack growth and the fracture, only crack shape should joints versus fatigue and fracture. Application to composite bridges].
be taken as probabilistic. PhD thesis, ENPC, Paris (in French).

JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001 / 217

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
McDowell, D. L. (1996). Basic issues in the mechanics of high cycle C = material parameter;
metal fatigue. Int. J. Fracture, 80, 103145. c = crack half-length;
Madsen, H. O., Krenk, S., and Lind, N. C. (1988). Methods of structural d = flange width;
safety, Prentice-Hall, Englewood Cliffs, N.J.
E[] = expected value;
Milne, I., Ainsworth, R. A., Dowling, A. R., and Stewart, A. T. (1986).
Assessment of the integrity of structures containing defects. Rep. fy = yield strength;
No. R/H/R6-Revision 3, CEGB, Berkeley, Calif. h = weld height;
Newman, J. C., Jr., and Raju, I. S. (1983). Stress-intensity factor equa- KI = Mode I (opening) stress intensity factor;
tions for cracks in three-dimensional finite bodies. Proc., Fracture KI C = fracture toughness;
Mech.: 14th Symp., J. C. Lewis and G. Sines, eds., ASTM, West Con- Kr = measure of proximity to linear elastic fracture me-
shohocken, Pa., I-238I-265. chanic failure;
Nussbaumer, A. (1997). Developpement dun modele ` de fiabilite pour Lr = measure of proximity to plastic yield;
les ponts metalliques combinant fatigue et rupture [Development of M = safety margin;
reliability model for steel bridges combining fatigue and fracture]. Mk = stress concentration factor;
`
Internal Rep., CTICM, Saint-Remy-les-Chevreuse, France (in French).
Paris, P. C., and Erdogan, F. (1963). A critical analysis of crack prop-
m = material parameter;
agation laws. J. Basic Engrg., 85, 528534. N = number of cycles;

Ponts metalliques
et mixtes: Resitance a` la fatigue [Steel and composite N0 = initiation number of cycles;
bridges: Fatigue resistance]. (1996). Guide de conception et de jus- Pf = failure probability;
tifications, SETRA, Bagneux, France (in French). S = effective stress range;
Sedlacek, G., et al. (1997). Design of steel structures, Part 2Bridges, s = stress;
for chapter 3Materials, choice of steel material to avoid brittle frac- Te = material temperature;
ture. Background documentation to Eurocode 3, Draft, Rheinische- TK 28 = test temperature for 28-J minimum energy Charpy
Westfael Technische Hochschule, Aachen, Germany.
V-notch impact energy;
Wallin, K. (1995). Validation of methodology for selecting Charpy
toughness criteria for old thin low strength steels. Publ. 216, VTT, t = time;
Espoo. tr = return period;
Yamada, K., Nagatsu, S., and Mitsugi, Y. (1989). Evaluation of scatter ts = reference period;
of fatigue life of welded details using fracture mechanics. First draft, VX = coefficient of variation of random variable X;
Draft, Dept. of Civ. Engrg., Nagoya University, Nagoya, Japan. Y = stress intensity correction factor;
Zhao, Z., and Haldar, A. (1996). Bridge fatigue damage evaluation and = reliability index;
updating using non-destructive inspections. Engrg. Fracture Mech., KI = growth of stress intensity factor caused by stress
53(5), 775788. range S;
KI th = threshold value of KI ;
APPENDIX II. NOTATION () = uncertainty factor;
The following symbols are used in this paper: = angle between weld and vertical axis;
X = mean of random variable X;
a = crack depth; = annual number of cycles;
af = crack depth at failure; () = density function for level crossing;
a0 = initial crack depth; (X, Y) = correlation between random variables X and Y; and
b = flange thickness; X = standard deviation of random variable X.

218 / JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 2001

Downloaded 07 Jan 2010 to 141.213.232.87. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright

You might also like