You are on page 1of 14

Int. J. Forensic Engineering, Vol. x, No.

x, xxxx

Stochastic failure analysis of the gusset plates in the


Mississippi River Bridge
Mojtaba Mahmoodian, Amir Alani and
Kong Fah Tee*
Department of Civil Engineering,
University of Greenwich,
Central Avenue, Chatham Maritime,
Kent ME4 4TB, UK
E-mail: M.Mahmoodian@gre.ac.uk
E-mail: M.Alani@gre.ac.uk
E-mail: K.F.Tee@gre.ac.uk
*Corresponding author
Abstract: The I-35W Mississippi River Bridge over Minneapolis, Minnesota,
collapsed suddenly on August 1, 2007. Previous studies showed that the
demand-to-capacity ratio for one of the gusset plates had become extremely
high after about 40 years of service of the bridge and therefore, the failure of the
gusset plate caused the failure of the whole bridge. A forensic assessment using
stochastic reliability analysis is carried out in this research to check whether the
collapse of the bridge could have been predicted at the design stage. For this
purpose, the probabilities of failure for different types of stresses in the gusset
plate are estimated. To consider the uncertainties involved in dead load and live
load increments with time, the gamma process concept is employed to model
stress increments. It is shown that the probability of failure in the year 2007 was
higher than the recommended value. Therefore, it can be concluded that if the
results of this study had been available at the design stage, the lack of reliability
in 2007 could have been predicted and the collapse of the bridge and its disastrous
consequences could have been prevented. It is also concluded that stochastic
reliability analysis can be used as a rational tool for failure analysis and reliability
assessment of bridges to prevent the risk of collapse.
Keywords: probabilistic failure analysis; Mississippi river bridge; gusset plates;
reliability index; bridge collapses.
Reference to this paper should be made as follows: Mahmoodian, M., Alani,
A. and Tee, K.F. (xxxx) Stochastic failure analysis of the gusset plates in the
Mississippi River Bridge, Int. J. Forensic Engineering, Vol. x, No. x, pp.xxxxxx.
Biographical notes: Mojtaba Mahmoodian is a PhD candidate working on
reliability analysis and service life prediction of infrastructures, including
pipelines and bridges. His working experience includes working for about eight
years for construction companies and with consulting engineers, mainly in quality
control and assessment of structures and infrastructure such as bridges, dams,
tunnels and high rise buildings. His publications include journal and conference
SDSHUV LQ WKH HOG RI VWRFKDVWLF UHOLDELOLW\ DQDO\VLV FRUURVLRQ RI FRQVWUXFWLRQDO
materials (concrete, steel and cast iron), concrete technology and quality control
and optimisation methods in civil engineering.
Copyright 20xx Inderscience Enterprises Ltd.

M. Mahmoodian et al.
Amir Alani is Head of the Department of Civil Engineering at the University of
Greenwich and the Bridge Wardens Chair in Tunnel and Bridge Engineering, both
since 2009. He previously was Head of the Department of Mechanical and Design
Engineering at the University of Portsmouth. He completed his MSc and PhD in
Saturated Soil Media with applications in Petroleum Engineering at the University
of Science and Technology of Montpellier in France. His research interests are
bridge and tunnel engineering (assessment and monitoring), applications of nonGHVWUXFWLYHWHVWLQJPHWKRGVLQWKHKHDOWKPRQLWRULQJRIVWUXFWXUHVEUHUHLQIRUFHG
concrete, the reliability of cementitious concrete sewer pipes and repair and
maintenance of buildings and structures.
Kong Fah Tee is a Senior Lecturer in infrastructure engineering at the University
of Greenwich. He did his BEng (Hons) in Civil Engineering from the University
Putra Malaysia and his PhD from the National University of Singapore. He also
FRPSOHWHG D SRVWJUDGXDWH FHUWLFDWH LQ +LJKHU (GXFDWLRQ DW WKH 8QLYHUVLW\ RI
Greenwich and is a Fellow of the Higher Education Academy. His research effort
KDVEHHQIRFXVHGRQVWUXFWXUDOV\VWHPLGHQWLFDWLRQVWUXFWXUDOKHDOWKPRQLWRULQJ
experimental stress analysis, fatigue, fracture mechanics, structural dynamics and
control and reliability. He was an invited visiting foreign expert at the Nanjing
University of Aeronautics and Astronautics.

Introduction

The I-35W bridge over the Mississippi River in Minneapolis, Minnesota, collapsed suddenly
on 1 August, 2007. Approximately 300 m of the 570 m long bridge collapsed onto the water
and ground, resulting in 13 fatalities and 145 injuries, with 111 vehicles involved in the
collapse. The collapse of the bridge in a major US downtown area was unprecedented. This
event was peculiar because the bridge was a typical steel truss structure and the collapse
occurred under normal operating conditions. In general, steel truss bridges such as the I-35W
Mississippi River bridge are a very common form for long-span bridges in the world. Before
this event occurred, steel truss bridges had earned the reputation of being economical and
reliable. Although the small redundancy of the trusses may be of concern, it is believed that
mandated maintenance procedures assure that this structural system is as safe and reliable as
any other (Liao et al., 2011).
A failure analysis of the I-35W bridge (Astaneh-Asl, 2008; Hill et al., 2008) showed that
the bridge failure was initiated at the gusset plates that connected the top chord members to a
compression diagonal and tension diagonal. In a comprehensive structural analysis, Holt and
Hartmann (2008) concluded that the demand-to-capacity ratio of one gusset plate (U10) was
high enough to cause the failure of the plate and eventually, the failure of the whole bridge.
Gusset plates are designed to bear up stresses at the Whitmore section (Whitmore, 1952),
to block shear from tension and buckling under compression; the stresses are calculated
based on simple beam equations. The stress distribution in gusset plates for bridge trusses
differs substantially from that in simple tension or compression connections and varies with
member arrangements (Liao et al., 2011).
To resolve uncertainties about loads and stresses at the design stage, the use of stochastic
models for reliability analyses of bridges can be considered. In a stochastic model, temporal

Stochastic failure analysis of the gusset plates in the Mississippi

variability associated with the stresses can be taken into account. In this study, a stochastic
reliability analysis is conducted on the critical gusset plate (U10) in the collapsed I-35W
bridge to check the predictability of the collapse based on the probability of failure.
To model the monotonic progression of stresses, this paper proposes a gamma process model
for stress variations due to loads increments.
A gamma process is a stochastic process with independent, non-negative increments
having a gamma distribution with an identical scale parameter and a time-dependent
shape parameter. A random variable X that is gamma-distributed with shape and scale
is denoted: X ~ (, *DPPD , ). A stochastic process model, such as the gamma
process, incorporates the temporal uncertainty associated with the evolution of stress. The
gamma process allows modelling of gradual stress increment monotonically accumulating
over time. The mathematical aspects of gamma processes can be found in Dufresne et al.
(1991), Ferguson and Klass (1972), Singpurwalla (1997) and van der Weide (1997).
In this study, based on the available information, the loading condition of the bridge at
the design stage (year 1967) is determined and stresses are calculated. For two consequential
20-year periods (ending in the years 1987 and 2007), stresses are then predicted. The gamma
model parameters (scale and shape parameters) are estimated using the maximum likelihood
method or the method of moments. The probability of failure of the bridge due to different
stresses (shear, tension and compression) is calculated and the probabilities of failure for the
gusset plate are checked with the maximum acceptable probability of failure suggested by
design codes such as LRFD AASHTO (1998).

Description of I-35W Mississippi River bridge failure

According to the reports from the Minnesota Department of Transportation (Mn/DOT, 2008)
and the National Transportation Safety Board (NTSB, 2008b and 2008d), construction of the
I-35W Mississippi bridge started in 1964 and after four years, in 1967, the bridge was opened
to the public. The Mississippi River bridge comprised 14 spans and 13 piers consisting of the
south approach spans, the north approach spans and the central deck-truss spans. The length
between the two end supports was 581 m. The spans from the south and the north approaches
were rested respectively at the main trusses at panel points U0 and U7KHPDLQWUXVVZDVRQ
roller supports at Piers 5, 6 and 8 and a hinge support at Pier 7. The main truss was 11.6 m long
DQGLQWRWDOSDQHOVDQGPDLQWUXVVHVZHUHMRLQHGE\RRUWUXVVHVWKDWVSDQQHGWUDQVYHUVHO\
between the panels (Figure 1). The bottom trusses were supported on longitudinal stringers,
ZKLFKVXSSRUWHGWKHUHLQIRUFHGFRQFUHWHGHFNDQGWUDIFORDG%HORZWKHRRUWUXVVHVVZD\
IUDPHVLQDFKHYURQFRQJXUDWLRQZHUHMRLQHGWRWKHWZRPDLQWUXVVHV7KHPDLQWUXVVHVZHUH
also connected by lateral bracing spanning among the upper chords and lower chords.
7KHGHFNZDVVHSDUDWHGIRUVRXWKERXQGDQGQRUWKERXQGWUDIF(DFKKDOIRIWKHGHFNZDV
meant to carry four 3.7 PWUDIFODQHVDQGWZR m shoulders. The total concrete thickness
was 215 mm, which, according to the original design, should have been 165 mm thick, but
the thickness was increased by 50 mm because of concrete overlay in later years. External
barriers and middle railings were also added to address maintenance and operational issues.
At the time of the collapse, a bridge patching and overlay project had been underway since
June, 2007. Two northbound outside lanes and two southbound inside lanes were closed to
KHDY\WUDIF(TXLSPHQWDQGPDWHULDOZHUHVHWXSRQWKHVHFXUHVRXWKERXQGWUDIFODQHVQHDU
the location of the failure.

M. Mahmoodian et al.

Figure 1 Plan and elevation of I-35 Mississippi River Bridge

Source:

Liao et al., 2011

2QFHWKHFROODSVHRIWKHEULGJHWRRNSODFHHOGLQYHVWLJDWLRQVZHUHSHUIRUPHGE\WKH176%
and the Mn/DOT. Observations (Astaneh-Asl, 2008; Hill et al., 2008; NTSB, 2008d) showed
that the collapse was initiated at panel point U10. Figure 2(a) is a picture taken before the
FROODSVH 176%F VKRZLQJWZRJXVVHWSODWHVOLQNLQJYHWUXVVPHPEHUVDWSDQHOSRLQW
8RIWKHZHVWPDLQWUXVV 8: )LJXUH E VKRZVWKHYHWUXVVPHPEHUVXSSHUFKRUGV
U9/U10 and U10/U11, diagonals members L9/U10 and U10/L11 and a vertical member
8/ 7KH YH WUXVV PHPEHUV ZHUH FRQQHFWHG WKURXJK D SDLU RI PP WKLFN JXVVHW
plates of ASTM A441 grade 50 steel, using river bolts of 25 mm diameter. Upper chords U9/
U10 and U10/U11 and compression diagonal member L9/U10 were precast box sections,
while tension diagonal member U10/L11 and vertical member U10/L10 were W sections.
All the truss members were welded prefabricated sections. Because U10 was located near an
LQHFWLRQSRLQWRIWKHFRQWLQXRXVWUXVVWKHIRUFHLQWKHXSSHUFKRUGXFWXDWHGIURPWHQVLRQ
on one side to compression on the other side, as can be seen in Figure 2(b); the diagonal
members exerted considerable compression and tension on the joining plate. Consequently,
DKXJHVKHDUIRUFHZDVIRUPHGRQDFULWLFDOODWHUDOVHFWLRQDVVKRZQLQWKHJXUH
Figure 2 Panel point U10 before collapse: (a) Photo U10W (NTSB, 2008c); (b) member description and truss forces (units in kN)

Source:

Liao et al., 2011

Based on commonly used design checks, Holt and Hartmann (2008) suggested that if the
thickness of the gusset plate was twice that of the original one at panel point U10, it would

Stochastic failure analysis of the gusset plates in the Mississippi

have been much safer to transfer the shear forces and design loads. Figure 3(a) shows a
picture of panel point U10W taken after the collapse (NTSB, 2008c). Figure 3(b) shows the
three primary cracks seen in Figure 3(a); these were commonly observed in all four U10W
and U10E gusset plates. Unbelievably, fracture of gusset plates was detected in the U10
panel points. This observation, in addition to the study by Astaneh-Asl (2008) and by Holt
DQG+DUWPDQ  ZKLFKLVEULH\SUHVHQWHGLQWKHQH[WVHFWLRQSRZHUIXOO\VXJJHVWWKDW
the collapse of the I-35W Bridge started in the U10 gusset plate.
Figure 3 Panel point U10 after collapse: (a) Photo U10W (NTSB, 2008c); (b) Reported locations
of fracture (see online version for colours)

Source:

Liao et al., 2011

Previous results of the structural analysis of the gusset plates

The study by Holt and Hartman (2008) on the truss gusset plates indicates that the
Demand-to-Capacity (D/C) ratios of some of the gusset plates (especially gusset plate
U10) were more than the allowable values. The D/C ratio is a comparative measure of
WKHHIFLHQF\RIWKHGHVLJQ$'&YDOXHRIOHVVWKDQLQGLFDWHVDFRQVHUYDWLYHGHVLJQ
D'&UDWLRRILQGLFDWHVDQHIFLHQWGHVLJQDQGD'&UDWLRJUHDWHUWKDQLQGLFDWHVD
liberal design with a reduction in the intended factor of safety. They found out that in
designing the gusset plates, as usually is the case, the width and length of many of them
were dictated by the number of connectors (rivets) needed to fasten the truss members
to the gusset plate. The study indicated that the thicknesses of the gusset plates supplied
at L5, L7, L9, U12 and L13 were the result of the demands of the applied loading. All
of these gusset plates have one or more D/C ratios equal to or greater than 0.80. They
showed that all the gusset plates at U2, L3, U4, U10 and L11 failed the D/C ratio check
for shear along Section A-A and the gusset plates at U10 and L11 on Section B-B also
IDLOHG WKH VKHDU '& UDWLR FKHFN E\ D VLJQLFDQW PDUJLQ ,Q DGGLWLRQ WKHVH WZR SODWHV
also had principal tension and compression overstresses that were again a result of the
dominating shear on the section.
Holt and Hartman (2008) also concluded that only the U10 gusset plate violated the
unsupported edge limitations. Reviewing the entire set of D/C ratios in total reveals that the

M. Mahmoodian et al.

gusset plates at U10 were the most under-designed plates. Coupling the lack of capacity of
these gusset plates with a violation of the detailing requirements for the unsupported edge,
makes the U10 gusset plates the most vulnerable plates in the truss. Figure 4 illustrates the
demand to capacity ratios for gusset plate U10 obtained by Holt and Hartman (2008). In
almost all cases, the ratio is more than one.
Figure 4

Demand to capacity ratio, results from Holt and Hartman, 2008 (see online version for
colours)

$QRWKHUVWXG\E\+DR  ZKLFKXVHGDQRQOLQHDU'QLWHHOHPHQWFRPSXWDWLRQEDVHG


load rating, also indicated that some of the gusset plates had almost reached their yield
limit when the bridge experienced the design load condition. Both previous studies (Holt
and Hartman, 2008; Hao, 2010) showed that the capacity inadequacies for the U10 gussets
were considerable for all conditions investigated with the plate, providing approximately
50% of the resistance required by the design loadings. In the next section, a reliability-based
methodology is introduced to analyse the reliability of the gusset plate.

Problem formulation for reliability analysis

In the theory of structural reliability, the criterion that should be checked for reliability
analysis can be expressed in the form of a limit state function (or safety margin) as follows
(Melchers 1999):
Z ( R , S , t ) = R (t ) S (t ) ,

(1)

where:
R(t): structural resistance or capacity at time t
S(t): load effect or stress at time t.
For an in-service bridge, the stresses SLQFUHDVHZLWKWLPHGXHWRWKHLQFUHDVHLQWUDIFQHZ
pavement overlays and so on. The capacity of the structure or the resistance of the material
also may change due to deterioration processes and aging. In practical examples, there are

Stochastic failure analysis of the gusset plates in the Mississippi

always uncertainties when predicting stresses in the future because the prediction of changes
LQ WUDIF DQG UHSDLU SODQV LV XQFHUWDLQ 7KLV LVVXH OHDGV LQIUDVWUXFWXUH PDQDJHUV WR DSSO\
stochastic analyses for the reliability assessment of bridges.
Because no corrosion in gusset plates had been reported in the Mississippi River bridge,
in this study the resistance (R) is assumed to be constant with time (i.e., there is no material
deterioration and resistance reduction within the bridges lifetime). However, because of
the addition of more dead load from overlaying new pavements and remarkable changes
LQ WUDIF YROXPHV RQ WKH EULGJH WKH VWUHVVHV RQ WKH EULGJH DUH FRQVLGHUHG DV VWRFKDVWLF
processes of the loads effect.

Stochastic process of stresses

For an in-service bridge, dead loads increase due to repair and maintenance actions and
OLYHORDGVLQFUHDVHGXHWRKLJKHUWUDIFGHPDQG:LWKWKHVHDVVXPSWLRQVDVWUXFWXUDOORDG
effect (stress) that varies in time can be modelled as a gamma process. The occurrence of a
VWUHVVPD\EHGHVFULEHGE\DJDPPDGLVWULEXWLRQIXQFWLRQ7KHPDWKHPDWLFDOGHQLWLRQRI
the gamma process is given in Equation (2). Recall that a random quantity S has a gamma
distribution with shape parameter > 0 and scale parameter > 0 if its probability density
function is as given by Papoulis (1965):
Ga ( S | , ) =

1 S
S e .
( )

(2)

Let (t) be a non-decreasing, right continuous, real-valued function for t 0. The gamma
process is a continuous-time stochastic process {S(t), t 0} with S(t) denoting the stress at
time t, t 0; the probability density function is S(t LQDFFRUGDQFHZLWKWKHGHQLWLRQRIWKH
gamma process, given by
(3)
f S (t ) ( S ) = Ga ( S | (t ), ),
with mean and variance as follows:
E ( S (t )) =

(t )
,

Var ( S (t )) =

(4)

(t )
.
2

(5)

A structure is considered to fail when its stress, denoted by S(t), becomes more than its
resistance (which is to say that the limit state function, Z, in Equation (1), is negative).
Assuming that resistance R is deterministic, the time at which failure occurs is denoted by
the lifetime T. By considering the probability density function (Equation (3)), the probability
of failure can then be written as:

Pr(T t ) = Pr( S (t ) R ) = f S (t ) ( S )d S =
R

where (v, x ) =

t=x

( (t ), R )
,
( (t ))

(6)

t v 1e t dt is the incomplete gamma function for x 0 and v > 0.

To model stress change due to load increment, in terms of a gamma process, the question
which remains to be answered is: how does the expected stress increase over time (i.e., the

M. Mahmoodian et al.

shape function for the gamma process model)? It is very common to model progressive
processes such as structural deterioration by using a power law formulation (Ellingwood and
Mori, 1993; van Noortwijk and Klatter, 1999). In this study, the same formulation is used for
modelling stress increments within the service life of the structure:
(7)

(t ) = ct b ,

where c and b are constants. Most of the time, by considering a few predicted values for
stress, there is engineering knowledge available about the shape of the expected stress and
therefore, parameter b in Equation (7) can be assumed.
To model the stress as a gamma process with shape function (t) = ctb and scale parameter
, the parameters c and should also be estimated. For this purpose, statistical methods are
suggested. The two most common methods that can be used for parameter estimation are
maximum likelihood and the method of moments. Both methods for deriving the estimators
of c and were initially presented by Cinlar et al. (1977) and were developed by van
Noortwijk and Pandey (2003).

5.1 Maximum Likelihood Estimation


In statistics, Maximum Likelihood Estimation (MLE) is a method of estimating the
parameters of a statistical model. When applied to a data set and given a statistical model,
0/(SURYLGHVHVWLPDWHVIRUWKHPRGHOVSDUDPHWHUV,QJHQHUDOIRUD[HGVHWRIGDWDDQGDQ
underlying statistical model, the method of maximum likelihood selects values of the model
parameters that produce a distribution giving the predicted data the greatest probability (i.e.,
parameters that maximise the likelihood function). Let us assume n predicted stresses are
denoted by S1, S2, , Sn. The principle of maximum likelihood assumes that the sample data
set is representative of the population, with a probability density function of fS (S1, S2, ,
Sn  DQG FKRRVHV WKDW YDOXH IRU WKH  XQNQRZQ SDUDPHWHU  WKDW PRVW OLNHO\ FDXVHG WKH
predicted data to occur, i.e., once predictions S1, S2, , Sn are given. fS (S1, S2, , Sn LVD
IXQFWLRQRIDORQHDQGWKHYDOXHRIWKDWPD[LPLVHVWKHDERYHSUREDELOLW\GHQVLW\IXQFWLRQ
LVWKHPRVWOLNHO\YDOXHIRU
In the current study a typical data set consists of prediction times ti, i = 1, , n where
0 = t0 < t1 < t2 < ... < tn and the corresponding predictions of the cumulative amounts of
stresses Si, i = 1, , n are assumed to be given as inputs of the model. The maximum
likelihood estimators of c and can be determined by maximising the logarithm of the
likelihood function of the increments. The likelihood function of the predicted stress
increments i = Si Si 1 , i = 1,! , n is a product of independent gamma densities:
b

b
b
c[ti ti 1 ]
ic[ ti ti 1 ]1e i .
l (( 1 ,! , n | c, ) = f S (t ) S (t ) ( i ) =
b
b

1
i
i
i =1
i =1 (c[t t
i
i 1 ])

(8)

To maximise the logarithm of the likelihood function, its derivatives are set to zero.

b b

ctnb
b
b
c
[
t

t
]log
i
i 1
S log (c[ti ti 1 ])
n

n
log l ( 1 ,! , n | c) = c

c
ctn
i =1

b
b
[
]
1
log

+
c
t

i
i 1
i
i

S
n

ct b n
= tnb log n + [tib tib1 ] log i c[tib tib1 ] = 0
S n i =1

)}

(9)

Stochastic failure analysis of the gusset plates in the Mississippi

where the function (x) is the derivative of the logarithm of the gamma function:
'

( x) log ( x)
=
.
( x)
x

( x) =

(10)

It follows that the maximum likelihood estimator of is (van Noortwijk and Pandey, 2003):
b
ct
= n
Sn

(11)

where c must be computed iteratively from the following equation:

ct
nb
b
b
b
b
b

t
t
c
t
t
t
[
]
[
]
log
log

i
i
i
i
i
n

1
S .
i =1
n

{ (

(12)

5.2 Method of moments


In statistics, the method of moments is a method of estimation of population parameters such
as mean and variance by equating sample moments with unpredictable population moments
and then solving those equations for the quantities to be estimated. The method of moments
estimates c and can be found from van Noortwijk and Pandey (2003):
n

c
=

i =1

t
i =1

b
i

tib1

Sn
=
tnb

(13)

2
n

n
t b tib1
Sn

i =1 i
1

=
i tib tib1
n t b t b 2
i =1
i =1 i
i 1

)) .
2

(14)

7KHUVWHTXDWLRQIURPERWKPHWKRGV PD[LPXPOLNHOLKRRGDQGPHWKRGRIPRPHQWV LVWKH


same and the second equation in the method of moments is simpler, because this method
PDNHVLWXQQHFHVVDU\WRGRLWHUDWLRQIRUQGLQJWKHXQNQRZQSDUDPHWHU (c) .

Resistance capacity

$FFRUGLQJWRRIFLDOGRFXPHQWV 176%DQGD WKHEULGJHVVWHHOPHPEHUVZHUH


PDGHRIJUDGHPLOGVWHHOZLWK<RXQJVPRGXOXVRI*3DD3RLVVRQVUDWLRRID
\LHOGVWUHQJWKRI03DDQGDQHQJLQHHULQJXOWLPDWHVWUHQJWKRI03DDWQRUPDO
strain. Assuming constant material properties during the service life of the bridge (i.e., no
corrosion is expected), the resistance capacity (RLQ(TXDWLRQ  LVVHWWR03DIRUWKH
calculation of the probability of failure.

Results and discussions

7.1 Estimation of stresses and gamma distribution parameters


%\ XVLQJ DYDLODEOH LQIRUPDWLRQ +ROW DQG +DUWPDQ   GHDG ORDGV OLYH ORDGV DQG
impacts can be considered for loading on the gusset plates at the beginning of their service

10

M. Mahmoodian et al.

LHWKH\HDU 7KHORDGVIRUWKHWZRVXEVHTXHQW\HDUSHULRGV HQGLQJLQWKH\HDU


DQGLQWKH\HDU FDQEHSUHGLFWHGEDVHGRQHVWLPDWLRQRIWKHLQFUHDVHVLQGHDG
ORDG DQG OLYH ORDG GXH WR UHSDLU DQG PDLQWHQDQFH DFWLRQV DQG JURZLQJ WUDIF GHPDQG
UHVSHFWLYHO\6KHDUSULQFLSDOWHQVLRQDQGSULQFLSDOFRPSUHVVLRQVWUHVVHVDWWKHJXVVHWSODWH
sections then can be determined accordingly. The calculated and predicted stress values
are presented in Table 1.
Table 1

6WUHVVHVDWJXVVHWSODWH8IRU6HFWLRQ$$DQG6HFWLRQ%%LQWKH\HDUV
DQG

6HFWLRQ%%

6HFWLRQ$$

Stress type
(S)
6KHDU
Principal
Tension
Principal
Compression
6KHDU
Principal
Tension
Principal
Compression

Calculated stresses at
year 
t0 = 0 03D

Predicted stresses at
year 1987,
t1 = 20 yr 03D

Predicted stresses at
year 2007,
t2 = 40 yr 03D

103


155


212
212

151



212

103








151





To model the stress increments as a gamma process, the scale and shape parameters (i.e., (t)
and ) should be estimated. For this purpose, the two statistical methods mentioned above
PD[LPXPOLNHOLKRRGDQGPHWKRGRIPRPHQWV DUHXVHG7KHUHVXOWVIRUVHFWLRQV$$DQG
%%RIWKHJXVVHWSODWH8 )LJXUH DUHSUHVHQWHGLQ7DEOH
Figure 5 )UHHGLDJUDPRIJXVVHWSODWH8 VHHRQOLQHYHUVLRQIRUFRORXUV

Source +ROWDQG+DUWPDQ

11

Stochastic failure analysis of the gusset plates in the Mississippi


Table 2

*DPPDPRGHOSDUDPHWHUVHVWLPDWHGE\PD[LPXPOLNHOLKRRGPHWKRGDQGPHWKRGRI
moments

6HFWLRQ%%

6HFWLRQ$$

Stress type
(S)
6KHDU
Principal
Tension
Principal
Compression
6KHDU
Principal
Tension
Principal
Compression

Parameter

Maximum likelihood
method

Method of moments

(t)

(t)

(t)

(t)

(t)

(t)

0.064t
0.012
0.039t
0.007
0.052t
0.009
0.059t
0.013
0.022t
0.005
0.027t
0.006

0.105t
0.019
0.032t
0.006
0.049t
0.009
0.073t
0.016
0.025t
0.005
0.030t
0.007

7.2 Calculation of the probability of failure


The estimated gamma parameters are used in Equation (6) for calculation of the
SUREDELOLW\RIIDLOXUH7KHUHVXOWVDUHLOOXVWUDWHGLQ)LJXUHVDQG7KHUHVXOWVVKRZWKDW
in all cases, there is a good agreement between the results obtained from the maximum
likelihood method and the method of moments. For instance, in Figure 6a, considering
the acceptable probability of failure of 0.1, the service life of the gusset plate from the
PD[LPXPOLNHOLKRRGPHWKRGZRXOGEH\HDUVZKLOHIRUWKHPHWKRGRIPRPHQWVWKH
VHUYLFHOLIHLV\HDUV
'HVLJQFRGHVVXJJHVWDUHOLDELOLW\LQGH[RIIRUEULGJHGHVLJQ /5)'$$6+72 
This value of reliability index corresponds to a probability of failure Pf = 0.232 10 3 . An
RYHUDOO UHYLHZ RI WKH UHVXOWV SUHVHQWHG LQ )LJXUH D VKRZV WKDW WKH JXVVHW SODWH 8 ZLOO
UHDFKWKLVSUREDELOLW\RIIDLOXUHDIWHU\HDUVIRUFURVVVHFWLRQ$$7KHOLIHWLPHRIWKH
JXVVHWSODWHFRQFOXGHGIURPRWKHUJUDSKVIRUFURVVVHFWLRQ$$ )LJXUHVEDQGF DUHHYHQ
shorter. This means that if the current study had been carried out at the design stage or before
WKHEULGJHFROODSVHGLWZDVFOHDUO\SUHGLFWDEOHWKDWDIWHU\HDUVRIVHUYLFHWKHEULGJHV
stability would be at high risk.
7KHREWDLQHGUHVXOWVSUHVHQWHGLQ)LJXUHVDQGDUHDOVRLQDJRRGDJUHHPHQWZLWK
WKHUHVXOWVRIWKHSUHYLRXVVWXG\E\+ROWDQG+DUWPDQ  SUHVHQWHGLQ)LJXUHRIWKH
FXUUHQWVWXG\7KLVKDVEHHQPRUHFOHDUO\LOOXVWUDWHGLQ)LJXUH7KHSUHVHQWHGUHVXOWVLQ
)LJXUHVKRZWKDWWKHSUREDELOLW\RIIDLOXUHRIWKHJXVVHWSODWH8LQLVPXFKKLJKHU
than the permissible value ( Pf = 0.232 10 3 ).
Although it is expected to have a deterministic conclusion in a forensic analysis, the
QDWXUHRIDVWRFKDVWLFVWXG\ZLOOHQGXSWRDSUREDELOLVWLFIRUPRIWKHUHVXOWV+RZHYHULQ
the current stochastic study, the criterion of Pf < 0.232 10 3 is checked to reach certain
conclusions about the failure or safety of the structure and therefore, the consequence of the
study is certain and reliable.

12

M. Mahmoodian et al.

Figure 6 3UREDELOLW\RIIDLOXUHRIWKHEULGJHGXHWRGLIIHUHQWW\SHRIVWUHVVHVRQJXVVHWSODWH8
VHFWLRQ$$

Figure 7 3UREDELOLW\RIIDLOXUHRIWKHEULGJHGXHWRGLIIHUHQWW\SHRIVWUHVVHVRQJXVVHWSODWH8
VHFWLRQ%%

Stochastic failure analysis of the gusset plates in the Mississippi

13

Figure 8 3UREDELOLW\RIIDLOXUHLQWKH\HDUGXHWRGLIIHUHQWW\SHRIVWUHVVHVRQJXVVHWSODWH8
(see online version for colours)

Conclusion

7KHFROODSVHRIWKH,:0LVVLVVLSSL5LYHUEULGJHLQZDVUHYLHZHGZLWKDIRFXV
RQWKHIDLOXUHRIWKHJXVVHWSODWHV8VLQJDVWRFKDVWLFIDLOXUHDVVHVVPHQWPHWKRGVKRZHG
how a reliability analysis could have helped the designers to predict the increase in
WKH SUREDELOLW\ RI IDLOXUH ZLWKLQ WKH EULGJHV VHUYLFH OLIH 7KH UHVXOW REWDLQHG LQ WKLV
study showed that if a proper stochastic reliability analysis had been carried out at the
GHVLJQVWDJHRUEHIRUHWKHEULGJHFROODSVHGWKHLQVXIFLHQF\RIUHOLDELOLW\RIWKHZHDN
JXVVHW SODWH 8  ZDV SUHGLFWDEOH DQG SUHYHQWDWLYH DFWLRQV FRXOG KDYH EHHQ FDUULHG
RXW WR SURKLELW WKH FDWDVWURSKLF FROODSVH RI WKH 0LVVLVVLSSL 5LYHU ULGJH7KH SURSRVHG
stochastic method can be used for failure analysis of different sections of a structure
at the design stage or during the service of the structure. It can be suggested that the
proposed study should be carried out on all the designed sections in a project or just
for some critical sections, the failure of which may cause a catastrophic collapse of the
whole structure.

References
$$6+72   /5)' %ULGJH 'HVLJQ 6SHFLFDWLRQV QG HG $PHULFDQ $VVRFLDWLRQ RI 6WDWH
+LJKZD\DQG7UDQVSRUWDWLRQ2IFLDOV:DVKLQJWRQ'&
$VWDQHK$VO$  3URJUHVVLYHFROODSVHRIVWHHOWUXVVEULGJHVWKHFDVHRI,ZFROODSVH7th
International Conference on Steel Bridges*XLPDUHV3RUWXJDO
&LQODU(%D]DQW=3DQG2VPDQ(  6WRFKDVWLFSURFHVVIRUH[WUDSRODWLQJFRQFUHWHFUHHS
Journal of the Engineering Mechanics Division9RO1R(0SS
'XIUHVQH ) *HUEHU +8 DQG 6KLX (6:   5LVN WKHRU\ ZLWK WKH JDPPD SURFHVV ASTIN
Bulletin9RO1RSS
(OOLQJZRRG %5 DQG 0RUL <   3UREDELOLVWLF PHWKRGV IRU FRQGLWLRQ DVVHVVPHQW DQG
OLIHSUHGLFWLRQRIFRQFUHWHVWUXFWXUHVLQQXFOHDUSRZHUSODQWVNuclear Engineering and Design,
9ROSS

14

M. Mahmoodian et al.

)HUJXVRQ76DQG.ODVV0-  $UHSUHVHQWDWLRQRILQGHSHQGHQWLQFUHPHQWSURFHVVHVZLWKRXW


*DXVVLDQFRPSRQHQWVAnn. Math. Stat.9RO1RSS
+DR6  ,:%ULGJH&ROODSVHJournal of Bridge Engineering9RO1RSS
+LOO+-0F*RUPOH\-&.RRE0-DQG1XJHQW:-  L:%ULGJHRYHUWKHPLVVLVVLSSL
ULYHUFROODSVHLQYHVWLJDWLRQFinal Report Prepared for Mn/DOT%ULGJH1R0LQQHDSROLV
0LQQHVRWD
+ROW5DQG+DUWPDQQ-  $GHTXDF\RIWKH8JXVVHWSODWHGHVLJQIRUWKH0LQQHVRWDEULGJH
1R,:RYHUWKH0LVVLVVLSSLULYHUFinal Report. Technical Report Prepared for Federal
Highway Administration7XUQHU)DLUEDQN+LJKZD\5HVHDUFK&HQWHU:DVKLQJWRQ'&
/LDR 0 2ND]DNL 7 %DOODULQL 5 6FKXOW] $( DQG *DODPERV 79   1RQOLQHDU QLWH
HOHPHQWDQDO\VLVRIFULWLFDOJXVVHWSODWHVLQWKH,:EULGJHLQ0LQQHVRWDJournal of Structural
Engineering9RO1RSS
0HOFKHUV5(  Structural Reliability Analysis and PredictionQGHG-RKQ:LOH\DQG6RQV
Chichester.
1DWLRQDO7UDQVSRUWDWLRQ6DIHW\%RDUG176%  /RDGVRQWKHEULGJHDWWKHWLPHRIWKHDFFLGHQW
Modelling Group Study Report No. 07-115.
1DWLRQDO7UDQVSRUWDWLRQ6DIHW\%RDUG176% D 0DWHULDOODERUDWRU\IDFWXDOUHSRUWReport No.
07-119.
1DWLRQDO7UDQVSRUWDWLRQ6DIHW\%RDUG176% E 6WUXFWXUDOLQYHVWLJDWLRQJURXSFKDLUPDQIDFWXDO
UHSRUWReport No. 08-015.
1DWLRQDO7UDQVSRUWDWLRQ6DIHW\%RDUG176% F 'DWDUHSRUWVWDWHE\VWDWHEULGJHFRXQWVCase
Number HWY07MH024.
1DWLRQDO7UDQVSRUWDWLRQ6DIHW\%RDUG176% G &ROODSVHRI,:KLJKZD\EULGJHAccident
Report NTSB/HAR-08/03VW$XJXVW0LQQHDSROLV0LQQHVRWD
3DSRXOLV$  Probability, Random Variables And Stochastic Processes0F*UDZ+LOO1HZ<RUN
6LQJSXUZDOOD1  *DPPDSURFHVVHVDQGWKHLUJHQHUDOL]DWLRQVDQRYHUYLHZ,Q&RRNH5
0HQGHO 0 DQG 9ULMOLQJ + Engineering Probabilistic Design And Maintenance For Flood
Protection'RUGUHFKW.OXZHU$FDGHPLF3XEOLVKHUVSS
YDQGHU:HLGH+  *DPPDSURFHVVHV,Q&RRNH50HQGHO0DQG9ULMOLQJ+Engineering
Probabilistic Design And Maintenance for Flood Protection 'RUGUHFKW .OXZHU $FDGHPLF
3XEOLVKHUVSS
YDQ1RRUWZLMN-0DQG.ODWWHU+(  2SWLPDOLQVSHFWLRQGHFLVLRQVIRUWKHEORFNPDWVRIWKH
(DVWHUQ6FKHOGWEDUULHUReliability Engineering and System Safety9RO1RSS
YDQ1RRUWZLMN-0DQG3DQGH\0'  $VWRFKDVWLFGHWHULRUDWLRQSURFHVVIRUWLPHGHSHQGHQW
UHOLDELOLW\DQDO\VLVProceedings of the Eleventh IFIP WG 7.5 Working Conference on Reliability
and Optimization of Structural Systems1RYHPEHU%DQII&DQDGDSS
:KLWPRUH 5(   ([SHULPHQWDO LQYHVWLJDWLRQ RI VWUHVVHV LQ JXVVHW SODWHV %XOOHWLQ 1R 
8QLYHUVLW\RI7HQQHVVHH(QJLQHHULQJ([SHULPHQW6WDWLRQ.QR[YLOOH7HQQ

You might also like