You are on page 1of 11

International Journal of Steel Structures

September 2013, Vol 13, No 3, 1-11


DOI 10.1007/s13296-

www.springer.com/journal/13296

Probabilistic Distributions of Plate Buckling Strength for


Normal and Bridge High-performance Steels
Dang Viet Duc1,*, Yoshiaki Okui2, Koichi Hagiwara3, and Masatsugu Nagai4
1

Department of Civil and Environmental Engineering, Saitama University,


Shimo Okubo 255, Sakura-ku, Saitama 338-8570, Japan
2
Department of Civil and Environmental Engineering, Nagaoka University of Technology,
1603-1, Kamitomioka, Nagaoka, Niigata 940-2188, Japan

Abstract
The probabilistic distributions of buckling strengths for compressive plates of normal and bridge high-performance steels
were obtained through numerical analyses in order to develop a nominal design strength and a corresponding safety factor. In
the numerical analyses, Monte Carlo simulation was used in combination with the response surface method to reduce the effort
associated with the finite element analyses. For each value of the slenderness parameter R, a response surface of the normalized
local bucking strength was determined based on the results of 114 finite element analyses using different residual stresses and
initial defections. The response surface is approximated as a simple algebraic function of the residual stress and the initial
deflection. Monte Carlo simulation is then carried out in order to evaluate the probabilistic distribution of the local bucking
strength. The mean values obtained in the present study approach those of a mean curve proposed based on experiments. The
standard deviation of the present study was approximately half that obtained based on experimental results in the range of 0.6
<R<1.2.
Keywords: bridge high-performance steels, local bucking strength, residual stress, initial deflection, local buckling

1. Introduction
Structural parts constructed from unstiffened plates,
such as box columns, box chord members in trusses, and
flanges of box girders, are widely used in steel structures.
In the design process of these structural parts, which are
under compression, load carrying capacity is frequently
governed by the local buckling strength (LBS) of the
constitutive unstiffened steel plates.
The current design equation of the Japanese Specifications
for Highway Bridges (JSHB, 2002) regarding the LBS of
compressive steel plates was originally proposed in the
1980 version of the specifications (Japan Road Association,
1980). Usami and Fukumoto (1989), Usami (1993), and
Kitada et al. (2002) demonstrated that the LBS design
equation in the JSHB is not conservative for 0.5<R<0.75
(intermediate range) and is overly conservative for R>
Note.-Discussion open until February 1, 2014. This manuscript for
this paper was submitted for review and possible publication on May
14, 2013; approved on August 22, 2013.
KSSC and Springer 2013
*Corresponding author
Tel: +81-48-858-3849; Fax: +81-48-858-9419
E-mail: email dangviet.duc@gmail.com

0.8 (slender range), where R is the width-thickness ratio


parameter defined by
2
b 12(1 )
R = --- -----y ---------------------2
t E
k

(1)

where b, t, sy , E, m, and k represent the plate width, plate


thickness, yield strength, elastic modulus, Poisson ratio,
and buckling coefficient, respectively. However, these
studies were based on a deterministic method and did not
yield any probabilistic information, such as the mean value
or standard deviation of the LBS, which is indispensable
in the determination of the safety factor in reliabilitybased design specifications, such as AASHTO LRFD
bridge design specifications (AASHTO, 2007); see also
ISO 2394 (ISO, 1998) and Bijlaard et al. (2003).
Few studies have attempted to obtain the probabilistic
characteristics of the LBS. Komatsu and Nara (1983)
performed a number of initial deflection measurements of
compressive members in steel bridge structures and
carried out FE analyses in order to obtain the probabilistic
characteristics of the LBS. However, they considered
only the initial displacement of the plates as a random
variable, and the residual stress was treated as a
deterministic parameter. In fact, the magnitude of residual

Dang Viet Duc et al. / International Journal of Steel Structures, 13(3), 000-000, 2013

stresses is considered to be a stochastic variable, which


significantly affects the LBS. Fukumoto and Itoh (1984)
reported the results for a large number of single plate and
box tests along with measurements of the initial deflection
and residual stress. Based on these experimental results,
they reported the mean value and standard deviation of
the LBS. However, their experimental data included test
results for specimens with an initial displacement that
exceeded the upper limit of the initial deflection for steel
bridge fabrication specified by the JSHB. Accordingly,
their results on the variability of the LBS based on the
initial displacement appear to be overestimated.
Furthermore, there are three more important reasons to
reexamine the LBS design equation. First, the thickness
of steel plates has tended to increase in recent steel bridge
construction. In fact, the regulation for the maximum
plate thickness in the JSHB has increased from 50 to 100
mm since 1996. Hence, the experimental data used to
establish the current LBS design equation in the JSHB as
well as in previous studies (Komatsu and Nara, 198;
Fukumoto and Itoh, 1984) were obtained using thinner
plate specimens. Furthermore, the LBS design equation in
the JSHB has not been revised to account for the use of
thicker plates.
Second, the Steels for Bridge High-performance Structures
(SBHS) specifications have been standardized in the
Japanese Industrial Standards (JIS) in 2008 (JIS, 2008),
and SBHS steels possess the advantages of high yield
strength and good weldability, although their inelastic
behavior differs from that of ordinary steels. For example,
SBHS steels have almost no yield plateau and a greater
yield-to-tensile strength ratio compared to ordinary steels
(Shin et al., 2013 with name HPS). Therefore, the
applicability of the current LBS design equation in JSHB
to SBHS steel plates should be investigated.
Third, recent design specifications, such as AASHTO
LRFD specifications and Eurocode, use the partial factor
format, in which safety factors differentiate between the
influences of uncertainties and variabilities originating
from individual sources. For example, the material factor
represents the variation of yield strengths, whereas the
member factor represents the uncertainty of the capacity
estimation equation. In order to establish an LBS design
equation in the partial factor format, it is necessary to
remove the effect of material strength variations on the
LBS, which is very difficult by means of an experimental
approach because the material strength variation is
inevitably taken into account in the experiments.
The present paper attempts to obtain a probabilistic
distribution of the LBS for compressive steel plates
considering thick steel plates of up to 100 mm in
thickness, conventional steel, and SBHS steels. In order
to propose a nominal LBS and corresponding safety
factor in the partial factor format, a combination of Monte
Carlo simulation and the response surface method is used,
in which only the initial displacement and residual stress

Figure 1. Procedure of the Monte Carlo-based method.

are considered as random variables.

2. Outline of the Present Study


In the present study, the probabilistic distribution of
buckling strengths for compressive plates of normal and
bridge high-performance steels is determined by a random
simulation, which is based on Monte Carlo simulation.
This Monte Carlo-based simulation involves the steps
illustrated by the flowchart in Fig. 1.
Step 1
Initial deflection and residual stress are assumed to be
the only two statistically independent random variables,
which affect the variability of the LBS. The probabilistic
density functions and the parameters of these random
variables are assigned in this step based on previous
research. The variation in yield strength is not considered
in this simulation in order to consider probabilistic
distribution of the LBS and the associated safety factor
excluding the effects of the yield strength variation. The
variation of the steel plate thickness is neglected as well,
because the effect of thickness variation on the LBS is
considered to be less than 0.2%, based on a statistical
report on steel plate thickness (Murakoshi et al., 2008).
Step 2
In this step, a response surface is identified for each
steel grade and several specific values of the width-

Probabilistic Distributions of Plate Buckling Strength for Normal and Bridge High-performance Steels

thickness ratio parameter. The response surface is an


approximate function of the normalized LBS with respect
to two variables: the residual stress and the initial
deflection. In standard Monte Carlo simulation, a large
number of deterministic FEM analyses are required,
which takes a significant amount of time for input data
preparation and computation. In order to avoid this
problem, the response surface method is used to evaluate
the LBS for corresponding residual stress and initial
deflection input data. The response surface is identified
through a sufficient number of FEM analysis results
based on the least squares method. The FEM simulation
model and the identification of the response surface will
be presented in more detail in Subsections 4.1 and 4.5,
respectively.
Step 3
A pair of values for the residual stress and initial
deflection is generated in accordance with the probabilistic
characteristic assigned in Step 1.

Figure 2. Histogram of the initial deflection reported by


Fukumoto and Itoh (1984) and Weibull distribution fit,
where W0/b=1/150 is the maximum allowable initial defection
in the Specifications of Highway Bridges (Japan Road
Association, 2012).

Step 4
The LBS is evaluated by means of the response surface
for the generated values of initial deflection and residual
stress.
Step 5
The termination of the random simulation process is
decided based on the verification of the convergence of
the LBS probabilistic characteristics, such as the mean
value and standard deviation. The convergence verification
process is described in detail in Subsection 5.1.

3. Probabilistic Characteristics of Residual


Stress and Initial Deflection
Fukumoto and Itoh (1984) reported statistical data on
the residual stresses and initial out-of-plane deflections of
single plates and square boxes. Figure 2 shows the
histogram they obtained for the normalized initial defection
W0=W0/b for 220 cases, where W0 and b are the
maximum deflection and plate width, respectively. The
mean value and standard distribution of the normalized
initial defection are 1/400 and 1/520, respectively. In this
figure, the solid curve represents a Weibull distribution
fitting their histogram. Since the maximum initial defection
W0 in box girders is restricted to within b/150, in
accordance with a provision for member precision in the
Specifications of Highway Bridges (Japan Road Association,
2012), the Weibull distribution curve is truncated at W0/
b=1/150. The Weibull distribution is used at generation
of random variables for W0/b in Monte Carlo simulation.
Figure 3 shows a histogram of the normalized residual
stress rc=rc/y, which is the ratio of the measured
residual compressive stress rc to the yield stress y . The

Figure 3. Histogram of the residual stress reported by


Fukumoto and Itoh (1984) and the lognormal distribution
fit.

mean value and standard deviation of the reported


normalized residual stress obtained from the reported data
are 0.232 and 0.145, respectively. The normalized residual
stress is assumed to follow a lognormal probability
distribution. The curve shown in Fig. 3 is a lognormal
distribution fit, which is used to realize the random
variable of the normalized residual stress in the following
Monte Carlo simulation. Note that, in the present study,
the normalized initial deflection and residual stress are
assumed to be independent random variables.

4. Deterministic Analysis
Before conducting the probabilistic simulation, a series
of deterministic finite element analyses of four-edge simply
supported compressive plates is carried out for the purpose
of comparison with reported experimental data. Furthermore,
these numerical results are used to determine the response
surfaces.

Dang Viet Duc et al. / International Journal of Steel Structures, 13(3), 000-000, 2013

Figure 4. Idealization of the stress-strain relation based on


actual test data.

Figure 6. Idealized residual stress distribution and sinusoidal


initial deflection surface.

four edges of the plate model. The idealized residual


stress distribution and sinusoidal initial deflection (Kim
K. D., 2006) applied to an FE plate model are shown in
Fig. 6.
The compression is applied to the plate model by the
displacement control method through one loading edge.
The plate simulation is modeled by ABAQUS S4R
(Dassault, 2008) shell elements with a mesh size of 30
30 elements.

Figure 5. Idealized stress-strain relations of the steel


grades considered in the present study.

4.1. Nonlinear elasto-plastic FE model


Both material and geometric nonlinearity are considered
in the following finite element analyses. The von Mises
yield surface with the isotropic strain hardening hypothesis
is applied to a steel material model. Uni-axial stress-strain
relationships of different steel grades are idealized from
test data, as shown in Fig. 4 for SBHS500 as an example,
and all idealized stress-strain relationships of the six steel
grades considered in the current study are illustrated in
Fig. 5. The inelastic characteristics of these steel grades
are listed in Table 1.
The simply supported boundary condition is applied to

4.2. Effect of aspect ratio


Nara et al. (1987) pointed out that plates with an aspect
ratio (=a/b)=0.5 yield the minimum LBS. This is true
under the condition of the same normalized residual
stress rc and normalized initial deflection W0. However,
according to measurement data on compression plates of
actual steel bridges (Komatsu and Nara, 1983), the mean
value of the normalized initial deflection of plates for =
0.5 is much lower than that for =1, specifically 1/2,069
for =0.5 and 1/591 for =1.0.
Figure 7 shows the LBSs obtained through FE analyses
for =0.5 and 1.0 along with the respective average
normalized initial defections. In these analyses, the
residual stress is set to the mean value of 0.23 for all
cases. The LBSs for =1 are more critical than those for
=0.5. Hence, the aspect ratio is assigned to =1 for all
subsequent numerical simulations.

Table 1. Inelastic characteristic of six steel grades


SHBS700
1
2
3
4
5

SHBS500

SM570

SM490Y

SM490

SM400

/y

/y

/y

/y

/y

/y

/y

/y

e/ey

s/sy

/y

/y

1
11
25

1
1.06
1.06

1
3
15
32
45

1
1
1.1
1.13
1.13

1
3
15
32
50

1
1
1.12
1.17
1.17

1
10
23
50
100

1
1
1.14
1.28
1.28

1
12
25
70
130

1
1
1.16
1.36
1.36

1
12
32
90
200

1
1
1.26
1.47
1.47

Probabilistic Distributions of Plate Buckling Strength for Normal and Bridge High-performance Steels

Figure 7. LBSs for a = 0.5 with W0= 1/2,069 and = 1.0


with W0= 1/591.The normalized initial defections W0 are
assigned to the mean values corresponding to the aspect
ratio .

Figure 9. Experimental points used to identify a response


surface. The FEM results at the experimental points are
used to identify a response surface.
Table 2. Mean values and standard deviations of input
random variables
Mean
Std. Dev.

rc/y

W0/b

0.23
0.145

0.0025
0.0019

5. Response surface
The response surface method (Guan and Melchers,
2001; Gasper et al., 2012) is used to reduce the excessive
computational time for Monte Carlo simulation of the
LBS. A cubic response function is used to approximate
the dependency of the normalized LBS on the normalized
initial defection and residual stress:
Figure 8. FE analysis results (r= 0.4 and W0=1/150) for
SM400 and SBHS700 along with previously reported
experimental results.

cr = cr(x1, x2)

i j

pijx1x2

(2)

i, j = 0

i+j3

4.3. Comparison of FE results with experimental


results
Before showing the stochastic simulation, the results of
deterministic FE analyses are compared with the experimental
data reported in this section. The normalized residual
stress and initial deflection are assigned to r=0.4 and
W0=1/150, respectively. The normalized initial defection
of 1/150 is the maximum tolerance for steel bridges in the
Specifications of Highway Bridges (Japan Road Association,
2012). On the other hand, r=0.4 corresponds approximately
to the 90th percentile value fdor the lognormal distribution
of the normalized residual stress shown in Fig. 3, because
the exceedance probability of r=0.4 is 10.8%.
Figure 8 shows the results of FE analyses for SM 400
and SBHS700 along with previously reported experimental
data (Dwight and Moxham, 1969; Rasmussen and Hancock,
1992). The FE analysis results fall around the minimum
LBS given by the experimental data.

where x1 r and x2 W0 are the input random variables,


and pij (i, j=0 to 3) are coefficients of the polynomial to
be determined through regression analysis using FE
analyses results.
In order to obtain the unknown coefficients, the
experimental points for the input variables should be
chosen. The experimental points are selected as shown in
Fig. 9 according to the mean values and standard deviations
of the input random variables listed in Table 2. In Fig. 9,
m and s represent the mean value and standard deviation,
respectively, for the corresponding input random variables.
A set of coefficients pij in Eq. (2) for a response surface
were determined from a set of 114 deterministic FE
analysis results at the experimental points regarding all
six steel grades for each R value using the least squares
method. Figure 10 shows R values ranging from 0.4 to
1.4 and the yield strengths used in the FE analyses to
identify the response surfaces. The minimum R value

Dang Viet Duc et al. / International Journal of Steel Structures, 13(3), 000-000, 2013

Figure 10. Slenderness parameter and yield strength in FE


analyses used to identify response surfaces.
Figure 11. Response surface and FEM results for all steel
grades and R=0.8.

corresponds to a plate thickness of 10 mm and a


maximum plate thickness of 100 mm. Table 3 shows a list
of the identified coefficients in Eq. (3), and Fig. 11 shows
the response surface for R=0.8, as an example.
The fit of the obtained response surfaces is evaluated
based on values of R2, which is referred to as the
coefficient of determination, presented in Table 4. For
R>0.7, the response surfaces exhibit very good fitting
with R2 >95%. For R0.5, the goodness of fit appears to
be poor, which can be explained by the fact that the
difference in normalized LBSs among different steel
grades is significantly affected by the inelastic behavior
of individual steel grades.

the random simulation. Figures 12(a) and 12(b) shows the


mean value and standard deviation of the LBS as a
function of the number of random input couples N. The
mean value and standard deviation converge as the
number of simulation cycles is increased. The variation of
mean values for the entire range of R values is less than
0.2%, whereas the variation of the standard deviation is
less than approximately 2%. In the next section, the
simulation results for N=100,000 are considered.

6.2. Statistical information of local buckling strength


The probabilistic distribution of the normalized LBS is
obtained for each slenderness parameter value, as shown
in Fig. 13 for R=0.8 as an example. The mean and
standard deviation of the LBS are calculated based on
the random simulation results and are listed in Table 5.
The mean and the mean minus twice the standard
deviation (-2) for different slenderness parameters are

6. Monte Carlo Simulation of Local Buckling


Strength
6.1. Convergence of random simulation
Random simulations are implemented with 100; 1,000;
10,000; and 100,000 random input couples of residual
stress and initial deflection to check the convergence of

Table 3. Response surface coefficients considering all steel grades for individual R values
R value

p00

p01

p02

p03

p10

p11

p12

p20

p21

p30

0.40
0.50
0.60
0.70
0.80
0.92
1.04
1.16
1.28
1.40

1.098
1.034
1.012
1.037
1.047
0.963
0.850
0.757
0.697
0.650

-40.22
-15.15
-1.72
-34.54
-65.98
-40.44
-23.31
-12.66
-10.08
-7.44

5442.0
1683.0
-1894.0
3169.0
8382.0
3081.0
1188.0
377.6
367.4
229.3

-248100
-90660
100500
-135000
-389100
-112100
-34700
-11430
-12430
-8387

0.007
-0.012
-0.087
-0.234
-0.584
-0.937
-0.788
-0.567
-0.435
-0.343

-2.320
-8.400
-21.150
-32.100
-0.520
57.210
47.270
25.720
23.330
18.580

-48.4
-42.4
507.9
1465.0
1309.0
-891.1
-769.2
46.8
-269.2
-208.5

-0.007
0.069
0.284
0.568
1.006
1.368
1.169
0.856
0.617
0.462

2.25
6.72
11.01
9.74
-15.13
-35.91
-29.12
-20.27
-15.11
-11.63

0.000
-0.057
-0.198
-0.342
-0.492
-0.632
-0.556
-0.419
-0.288
-0.211

Table 4. R-square values of response surfaces in the range of 0.4R1.4


R

0.4

0.5

0.6

0.7

0.8

0.92

1.04

1.16

1.28

1.4

R-squared

0.702

0.837

0.920

0.945

0.964

0.964

0.959

0.956

0.961

0.964

Probabilistic Distributions of Plate Buckling Strength for Normal and Bridge High-performance Steels

Figure 13. Probabilistic distribution of LBS for R=0.8.

Figure 12. Convergence of random simulations. N and


N are the mean values and standard deviations of the
LBSs obtained from N random simulations.

Figure 14. Results of the present study, JSHB (2002), and


Fukumoto and Itoh (1984).

shown in Fig. 14 for comparison with the current JSHB


design strength (JSHB, 2002) and the corresponding
and -2 curves proposed by Fukumoto and Itoh (1984)
based on experimental data.
As shown in Fig. 14, the mean values of the current
study are close to the mean curve reported by Fukumoto
and Itoh (1984). Within the range of 0.65<R<0.85, the
mean values of the current study are slightly greater than
those proposed by Fukumoto and Itoh (1984). This
greater mean value of the LBS may be attributed to
inclusion of SBHS steel and the exclusion of simulation
results obtained with the normalized initial deflection W0
>1/150 in the present study, because the influence of
normalized initial deflection and steel grade on the LBS
is more significant than that in the range R>0.85 (slender
range).
Figure 15 shows the standard deviations of the LBSs

obtained in the present study and those reported by


Fukumoto and Itoh (1984). Within the practical range of
0.6<R<1.2, the results of the present study are
approximately half those reported by Fukumoto and Itoh
(1984). For R<0.6, the difference between the present
results and those obtained experimentally is more
significant.
The differences between the current simulation condition
and Fukumoto and Itohs experimental data are as
follows:
(a) Boundary condition
Fukumoto and Itoh reported experimental data for the
LBSs of welded boxes and single plates with weld beads,
whereas the present simulation considers only simply
supported plates. The flanges of welded boxes are assumed
to be simply supported plates in design practice. However,

Table 5. Means and standard deviations obtained from Monte Carlo simulations for different R values
R value

0.4

0.5

0.6

0.7

0.8

0.92

1.04

1.16

1.28

1.4

value
value

1.039
0.0277

1.006
0.0157

0.982
0.0212

0.938
0.0413

0.862
0.0515

0.766
0.0399

0.701
0.0294

0.653
0.0216

0.617
0.0171

0.586
0.0145

Dang Viet Duc et al. / International Journal of Steel Structures, 13(3), 000-000, 2013

Figure 15. Standard deviations obtained in the present


study and those reported by Fukumoto and Itoh (1984).

there is a degree of rotational fixity at flange-web junctures,


which affects the variation of the LBS.
(b) Inelastic material property
In the random simulation, the variation of yield stresses
is not considered, and only six deterministic stress-strain
curves for different steel grades are used to account for
material property differences in the inelastic regime. The
effect of inelastic behavior on the LBS is considered to be
more significant for plates with smaller R vales, which is
consistent with the results shown in Fig. 15.
(c) Initial deflection
In the present study, as explained previously, any initial
deflection that exceeded the maximum limit specified in
the design specifications was omitted from the random
simulation, whereas the experimental data reported by
Fukumoto and Itoh include the LBSs of plates with initial
deflections that exceed the maximum limit specified in
the specifications, which results in larger variation in the
experimental LBS.
The tendency of the standard deviation will be explained
further in Subsection 6.3.

6.3. Approximation of mean and variance


In order to obtain the probabilistic distribution of the
LBS, it is necessary to use a numerical simulation method
such as Monte Carlo, as shown in the previous section,
because the explicit functional relationship between the
LBS and the input variables is not known in the problem
examined herein. However, an approximate mean and
variance of the LBS can be extracted from the mean and
variance of the input variables with the correct deterministic
analysis. In this section, an approximate mean and variance
of the LBS will be calculated using a Taylor series finite
difference (TSFD) estimation procedure (Haldarand and
Mahadevan, 1999).
The first-order approximation of the mean E(cr) and
variance Var(cr) of the LBS can be obtained as follows:
E(cr) cr(x1, x2)

(3)

Figure 16. Mean values of LBSs obtained through random


simulation and Taylor series finite difference (TSFD)
estimation.

Figure 17. Standard deviations of LBSs obtained through


random simulation and Taylor series finite difference
(TSFD) estimation.
2

cr
cr, i cr, i
Var(cr) --------- Var(xi) --------------------- xi

2
2

i=1

(4)

i=1

where x1 (i = 1,2) is the mean of two input random


variables: x1 r and x2 ; and
W0

cr, 1 cr(x1x1, x2)

(5)

cr, 2 cr(x1, x2x2)


xi (i = 1,2) is the standard deviation of two input
variables: x1 r and x2 . In Eqs. (3) and (4),
W0

cr(x1, x2) and cr, 1 can be evaluated through


deterministic FEM analyses.
Figure 16 shows the mean values of the LBS obtained
from the random simulation and those obtained from the
TSFD. The mean values obtained from the random
simulation are slightly greater than those obtained from
the TSFD.
In Fig. 17, the standard deviations obtained from both
methods are presented together. The TSFD results and the
random simulation results have the same tendency. At

Probabilistic Distributions of Plate Buckling Strength for Normal and Bridge High-performance Steels

Figure 18. Variance of LBS due to individual input


random variables: residual stress r and initial deflection
W0 obtained from Taylor series finite difference (TSFD)
and random simulation (RS).

Figure 19. Explanation of Eq. (7).

R0.8, the standard deviations from both methods


become the maximum values. For R>0.9, the results
obtained from both methods are similar. For R<0.9, the
TSFD results are significantly greater than those obtained
by the random simulation.
The variances of the LBS obtained from the two
methods are plotted together in Fig. 18. In this figure, the
LBS variances due to the variation of the residual stress
and the initial deflection are plotted separately. The LBS
variances obtained from both methods have a similar
tendency to reach maximums at R=0.8 and 0.9 for the
initial deflection and residual stress, respectively. However,
the results obtained from the TSFD estimation are greater
than that from the random simulation. This difference
results from the curvature of the response surface, which
is not considered in the first-order approximation of the
TSFD.
For R<0.7, the residual stress has almost no effect on
the LBS variance, because even though the residual stress
is introduced from the initial stage of loading, the
ultimate state is the same full plastic state, regardless of
the presence or absence of the residual stress. On the
other hand, for R>1.1, the initial deflection has almost no
effect, because the post buckling behavior governs the
LBS. Within the range of 0.8 R 0.9, both the initial
deflection and the residual stress have a significant effect
on the variance of the LBS, and the effect of the initial
defection on the LBS variance is more significant than
the effect of the residual stress.

6.4. Proposal of a partial safety factor


As mentioned in the introduction, the probabilistic
characteristics of the LBS are an important basis for the
proposal of a partial safety factor and the associated
nominal strength. Determining the safety factor and the
nominal strength based on probability of failure as
calculated from the probability density of the LBS is a
rigorous method. However, the probability of failure
depends largely on the shape of the foot of the probability
density function, and, accordingly, high accuracy is not
expected. Therefore, the reliability index is used to
specify a safety margin
T = 1
---fN

(7)

where s and m are the mean and the standard deviation,


respectively, of the normalized LBS, T is the target
reliability index, and and fN are the safety factor and the
corresponding nominal strength (See Fig. 19 for the
probability density function and the approximated normal
distribution at R=0.8). Once the target reliability index
and the nominal strength have been specified, the
corresponding safety factor is calculated from Eq. (7) and
the mean and standard deviation obtained from the Monte
Carlo simulation.
Assuming that the probability density of the LBS is
described by a normal distribution and assigning nonexceedance probabilities of the LBS of 5.0, 3.0, and
1.0%, the corresponding target reliability indices become

Table 6. Partial safety factors in the range of 0.4R1.4


R
fN =
(5%)
(3%)
(1%)

0.4
1.039
1.05
1.05
1.07

0.5
1.006
1.03
1.03
1.04

0.6
0.982
1.04
1.04
1.05

0.7
0.938
1.08
1.09
1.11

0.8
0.862
1.11
1.13
1.16

0.92
0.766
1.09
1.11
1.14

1.04
0.701
1.07
1.09
1.11

1.16
0.653
1.06
1.07
1.08

1.28
0.617
1.05
1.06
1.07

1.4
0.586
1.04
1.05
1.06

10

Dang Viet Duc et al. / International Journal of Steel Structures, 13(3), 000-000, 2013

Figure 20. Partial safety factors obtained in the range of


0.4R1.4

Figure 21. Design LBS for three levels of exceedance


probability: 5, 3, and 1%, obtained in the present study
and corresponding design values as specified by AASHTO
(2007).

1.64, 1.88, and 2.33. Furthermore, setting the nominal


LBS to be equal to the mean of the LBS as an example,
the partial safety factor can be obtained as shown in Table
6.
As shown in Fig. 20, the maximum partial safety factor
values for exceedance probabilities of 5, 3, and 1% are
1.11, 1.13, and 1.16, respectively.
Calculating the partial safety factors for exceedance
probabilities of 5, 3, and 1% and assuming that the mean
LBS values are the nominal resistances, the obtained
design LBSs are compared to the corresponding design
values as specified by AASHTO (2007) and presented in
Fig. 21. As shown in the figure, the mean strengths
obtained in the present study are significantly lower than
the nominal strengths specified by AASHTO (2007)
within the range of 0.65<R<1.20.

7. Conclusion
The mean results of the LBS obtained from approximate
random simulation are similar to the mean curve reported

by Fukumoto and Itoh (1984) based on experimental data.


Compared with the standard deviation obtained from
the experimental results of Fukumoto and Itoh (1984), the
standard deviation based on the approximate random
simulation of the present study is approximately half for
0.6<R<1.2 and is significantly lower for R<0.6. The
causes of this significant difference are summarized as
follows: (a) the difference of the boundary condition
between experiments and numerical simulation; (b) the
variation of stress-strain relationship; and (c) the consideration
of the maximum initial deflection in the random simulation
in accordance with the design code.
The results of M-2S curve of the present study indicate
that the proposed JSHB design equation for steel plates in
compression is not conservative for 0.5<R<0.8 and is
overly conservative for R>0.85.
The M-2S curve proposed by Fukumoto and Itoh (1984)
is overly conservative compared to the corresponding
results of the present study.
Based on the LBS variance results obtained from the
three methods discussed herein, we have the following
conclusions regarding the influences of residual stress
and initial deflection on the LBS of steel plates:
-In the range of R<0.8, the influence of initial deflection
on the LBS of steel plates is more sensitive than that of
residual stress.
-In the range of R>0.9, the influence of residual stress
on the LBS of steel plates is more sensitive than that of
initial deflection.
-Within the range of 0.8<R<0.9, the influence of initial
deflection on the LBS of steel plates is dominant.
For R<0.5, the hardening behavior of steels significantly
affects the LBS of steel plates.
The approximation of response surface shows a very
good fit with the FEM results.

References
Bijlaard, F., Sedlacek, G., Muller, C., and Trumpf, H. (2003).
Unified European design rules for steel and composite
structures. International Journal of Steel Structures, 3,
pp. 117-125.
Dassault (2008). Abaqus Theory Manual v.6.8-EF. Dassault
Systemes Simulia Corp., the place of publication.
AASHTO (2007). AASHTO LRFD bridge design specifications.
4th Ed., American Association of State Highway and
Transportation Officials.
Eurocode 3 (2004). Design of steel structures. Parts 1-5.
Plated structural elements. European Committee for
Standardization (CEN), Brussels, Belgium.
Eurocode 4 (1994). Design of composite steel and concrete
structures. Part 2. General rules and rules for bridges.
European Committee for Standardization (CEN), Brussels,
Belgium.
Dwight, J. B. and Moxham, K. E. (1969). Welded steel
plates in compression. The Structural Engineer, 47(2),
pp. 49-66.

Probabilistic Distributions of Plate Buckling Strength for Normal and Bridge High-performance Steels

Fukumoto, Y. and Itoh, Y. (1984). Basic compressive


strength of steel plates from test data. Proc. of JSCE, No.
334/I-1, pp. 129-139.
Gaspar, B., Naess, A., Leira, B. J., and Guedes Soares, C.
(2012). System reliability analysis of a stiffened panel
under combined uniaxial compression and lateral
pressure loads. Structural Safety, 39, pp. 30-43.
Guan, X. L. and Melchers R. E. (2001). Effect of response
parameter variation on structural reliability estimates.
Structural Safety, 23, pp. 429-444.
Haldar, A. and Mahadevan, S. (1999). Probability,
Reliability, and Statistical Methods in Engineering
Design. 1st Ed., John Wiley and Sons.
ISO 2394 (1998). General principles on reliability for
structures. 3rd Ed., International Standard.
JRA (2002). Specifications for highway bridges. Part II.
Steel bridges. Japan Road Association (in Language).
JRA (1980). Specifications for highway bridges. Part II.
Steel bridges. Japan Road Association (in Language).
JRA (1996). Specifications for highway bridges. Part II.
Steel bridges. Japan Road Association (in Language).
JIS (2008). Higher yield strength steel plates for bridges. JIS
G 3140, Japanese Industrial Standard (in Language).
Kim, K. D. (2006). Large Displacement of Elasto-plastic
Analysis of Stiffened Plate and Shell Structures

11

International Journal of Steel Structures, 6, pp. 65-68.


Kitada, T., Yamaguchi, T., Matsumura, M., Okada, J., Ono,
K., and Ochi, N. (2002). New technology of steel bridge
in Japan. Journal of Constructional Steel Research, 58,
pp. 21-70.
Komatsu, S. and Nara, S. (1983). Statistical study on steel
plate members. Journal of Structural Engineering,
ASCE, 109(4), pp. 977-992.
Murakoshi, J., Yanatory, N., Arima, K., Shimizu, H., and
Komori, D. (2008). Report on statistical data of steel
members and strength. Technical Note of Public Work
Research Institute, No. 4090, the place of publication.
Rasmussen, K. J. R. and Hancock, J. G. (1992). Plate
slenderness limits for high strength steel sections. J.
Construct. Steel Research. 23, pp. 73-96.
Shin, D. K., Cho E. Y., Kim, K. S. (2013). Ultimate flexural
strengths of plate girders subjected to web local
buckling. International Journal of Steel Structures, 13,
pp. 291-303.
Usami, T. and Fukumoto, Y. (1989). Deformation analysis
of locally buckled steel compression members. Journal
of Constructional Steel Research, 13, pp. 111-135.
Usami, T. (1993). Effective width of locally buckled plates
in compression and bending. Journal of Structural
Engineering, 119(5), pp. 1358-1373.