Professional Documents
Culture Documents
Pf =
I (r , q ) f
R (r ) f Q (q)
dr dq
(1)
FO = ln (Rm Qm )
V R2 + VQ2
(3)
Pf =
I ( x, y ) f R ( x ) f Q ( y )
hXY ( x, y )
hXY ( x, y ) dx dy
(5)
P f =
(6)
N i =1
h XY ( x i , y i )
where the samples (xi,yi) are drawn from the importance sampling
density hXY. For this application we perform importance sampling over
the load dimension only, hXY=fQ*. A normal distribution is selected for
Q* with a mean equal to the average of the means of R and Q, and a
COV equal to that of Q. Figure 1 compares estimates of the probability
of failure by direct Monte Carlo sampling and importance sampling for
an increasing number of samples.
1000
SP50LA
Direct Monte Carlo
Importance Sampling
Pf x106
800
600
400
200
0
2000
4000
6000
Number of samples
8000
10000
FRAME STRUCTURES
The steel frames analyzed are based on those of Ziemian (1990) typical
of low-rise industrial structures. Figure 2 shows the geometry, support
conditions and loads for the 16 frames discussed in this paper. The
frames are labeled with the following nomenclature:
S, U: symmetric or unsymmetric geometry;
P, F: pinned or fixed base;
50: 50 ksi (345 MPa) nominal yield strength steel;
H, L: heavy or light gravity load;
A, E: member sizes determined by advanced analysis or elastic LRFD.
Member sizes for frames UP50HA and UP50HE are given in Table 1;
the member sizes of all frames are listed in Ziemian (1990). The yield
strengths of beams and columns, and gravity loads are modeled as
random variables. Other potential random effects, such as residual
4.57 m
B3
C5
L: 32.84 kN/m
6.10 m
B1
C1
B4
C6
H: 109.45 kN/m
B2
C2
C3
P
U: 6.10 m
U: 14.63 m
S: 10.36 m
S: 10.36 m
Figure 2. Dimensions and loads of frames.
Frames with correlated Fy have a slightly larger mean strength but also
a significantly larger COV than the frames with uncorrelated Fy. For a
given frame, the failure mode of each sample with uncorrelated Fy
depends on the spatial distribution of the random Fy, and the effect of a
single weak member may be offset by a strong member elsewhere in
the frame. In contrast, for a given frame with correlated Fy, all samples
exhibit the same failure mode.
5
4
mean=1.927
cov= 0.062
3
2
1
0
0.5
1
1.5
2
Normalized Frame Strength
2.5
5
4
mean=1.521
cov= 0.079
3
2
1
0
0.5
1
1.5
2
Normalized Strength at 1st Plastic Hinge
2.5
5
4
mean=1.969
cov= 0.093
3
2
1
0
0.5
1
1.5
2
Normalized Frame Strength
2.5
5
4
mean=1.618
cov= 0.100
3
2
1
0
0.5
1
1.5
2
Normalized Strength at 1st Plastic Hinge
2.5
Figure 5 plots frame strength against first plastic hinge strength for two
of the frames. For UP50HA no correlation exists. For UP50LE, a band
of correlated samples is apparent at the upper left of the plot; it is likely
that this subset of samples has a common failure mode. Other frames
exhibited various correlation structures between the two extremes of
Figure 5. This lack of consistent correlation between these two
performance measures suggests that there is no simple means of
relating the member-based failure criterion to the system-based
criterion.
1.4
2
1.8
1.2
1.6
1.4
1.2
0.8
UP50HA
1.4
1.6
1.8
Frame Strength
1 UP50LE
1.6 1.8 2 2.2 2.4
Frame Strength
Reliability results
Tables 4 and 5 list the reliabilities of all 16 frames for both frame
strength and first plastic hinge strength for uncorrelated and correlated
Fy. The reliability is expressed as a probability of failure and a
reliability index, =-1(Pf) where is the standard normal
cumulative distribution function. The reliability by sampling is also
compared to the first order reliability, FO, computed from Eq. 3. The
first-order approximation generally overestimates compared to the
sampling estimate, since the lower tail of the strength distributions have
a higher probability mass than the normal distribution assumed for FO.
The frames designed by advanced analysis have a lower reliability
index than those designed by LRFD.
FO
0.77
1.17
1.06
1.10
1.42
1.04
1.43
0.83
3.06
3.68
3.06
3.35
3.61
4.90
3.40
3.48
FO
1.09
1.56
1.54
1.69
1.91
1.49
1.81
1.51
3.23
4.09
3.49
3.69
3.66
4.85
3.85
3.90
For the LRFD designed frames, FO for first plastic hinge is greater
than 3 for all frames. The target for steel beam-columns in LRFD is
approximately 3, and thus these analyses demonstrate that the LRFD
code is effective in achieving this goal. For these frames, the sampling
estimate of is also generally near 3 as well. Examining the effect of
correlation on reliability, reveals that based on frame strength and
uncorrelated Fy is greater than the for correlated Fy. Whereas for first
plastic hinge strength, for correlated Fy is greater than that for
uncorrelated Fy. Correlation among member yield strengths tends to
decrease the probability of occurrence of the first plastic hinge but to
increase the probability of overall frame failure.
For the frames designed by advanced analysis, the reliability indices
based on frame strength are above 3 and thus could be considered
acceptable designs. The values of for first plastic hinge are generally
between 1 and 2, corresponding to probabilities of occurrence of up to
14%. This result suggests that the occurrence of a plastic hinge in a
frame designed by advanced analysis and subjected to nominal gravity
loads may not be an infrequent event. In the selection of member sizes
for these frames, Ziemian et al. (1992a) prohibited the formation of
plastic hinges under service loads. However, this deterministic design
requirement does not necessarily result in a zero probability of
occurrence of plastic hinging when random loads and strengths are
included in the analysis. The occurrence of a plastic hinge may affect
the serviceability performance, even though it does not compromise
overall strength and safety. This observation highlights the importance
of serviceability issues in frames designed by advanced analyses.
Elastic LRFD design controls the occurrence of the first plastic hinge,
while advanced analysis may allow plastic hinging at service load
levels in cases where it does not compromise frame strength. The
existence of plastic hinges may also impact structural behavior for
extreme load events, such as seismic, since the structure cannot be
assumed to begin in an elastic, undamaged state.
Resistance factors for advanced analysis
The general expression for the resistance factor has been given in Eq. 4.
The distribution of true resistance is ideally based on experimental
data, and related to nominal resistance by
R = Rn PMF
(7)
with mean and COV given by
Rm = Rn Pm M m Fm ,
V R = V P2 + VM2 + V F2
(8)
these values and t=3.00 in Eqs. 4 and 8, results in =0.84. The value
of =0.90 used in LRFD for beams corresponds to t=2.10.
The goal of advanced analysis methods is to provide predictions of
strength which are closer to the true strength than existing elastic-based
code methods. Since experimental data are not available for the frames
analyzed here, we assume the limiting case that the distribution of true
strength is exactly predicted by the advanced analysis with random
material properties. Practical design with advanced analysis will be
based on a single analysis with nominal yield strengths, resulting in the
nominal strength RnAA . We relate the deterministic nominal strength to
the probability distribution of strength by
R AA = RnAA B AA F
(9)
AA
where B is the random variable bias factor and F is the LRFD
fabrication factor, retained since our analyses do not consider random
geometric properties. This equation is analogous to Eq. 7. Without test
data for frame strength it is not possible determine individual bias
factors equivalent to P, M and F.
Table 6 gives values of nominal strengths and mean bias factors for
frame strength and first plastic hinge for both uncorrelated and
correlated Fy. The mean bias factor, B mAA , is the mean strength from
Table 1 or 2 divided by RnAA . The bias factors fall within the range of
0.93 to 1.08, and the mean of the bias factors is close to 1.0 indicating
that RnAA is a good predictor of RmAA with little bias. For comparison,
the combined bias factor assumed in LRFD for beam-columns is 1.07.
Also important is the observation that the bias factors appear
independent of the method by which the frame was designed.
Tables 7 and 8 present values of the variance of the advanced analysis
strength prediction (including VF =0.05) and values of for both
t =3.00 and 2.10. The COVs of the strengths are in the range of 0.07 to
0.10, somewhat less than the LRFD value of 0.15. For a target
reliability of 3.00 on frame strength, the values of range from 0.86 to
0.91 for both correlated and uncorrelated Fy. For first plastic hinge
strength, the values of for =3.00 range from 0.80 to 0.90. For a
for
for
for
VRAA
VRAA
t=3.00 t=2.10
t=3.00
UP50HA 0.071
0.89
0.93
0.094
0.85
UP50LA 0.077
0.91
0.94
0.094
0.86
UF50HA 0.074
0.90
0.94
0.098
0.85
UF50LA 0.070
0.90
0.93
0.092
0.84
SP50HA 0.079
0.89
0.93
0.101
0.84
SP50LA 0.085
0.87
0.91
0.099
0.83
SF50HA 0.082
0.86
0.90
0.098
0.83
SF50LA 0.081
0.87
0.90
0.092
0.80
UP50HE 0.080
0.90
0.93
0.094
0.84
UP50LE 0.083
0.90
0.94
0.090
0.82
UF50HE 0.073
0.89
0.93
0.090
0.83
UF50LE 0.082
0.89
0.92
0.092
0.84
SP50HE 0.076
0.91
0.95
0.092
0.84
SP50LE 0.086
0.87
0.90
0.093
0.84
SF50HE 0.094
0.89
0.94
0.088
0.81
SF50LE 0.080
0.87
0.90
0.090
0.80
min
0.86
0.90
0.80
mean
0.89
0.92
0.83
max
0.91
0.95
0.86
for
t=2.10
0.89
0.90
0.90
0.87
0.88
0.87
0.87
0.84
0.88
0.85
0.87
0.88
0.88
0.88
0.84
0.84
0.84
0.87
0.90
for
for
for
VRAA
VRAA
t=3.00 t=2.10
t=3.00
UP50HA 0.106
0.88
0.93
0.112
0.87
UP50LA 0.107
0.88
0.93
0.107
0.90
UF50HA 0.108
0.88
0.92
0.109
0.90
UF50LA 0.108
0.88
0.93
0.109
0.89
SP50HA 0.092
0.89
0.93
0.105
0.90
SP50LA 0.109
0.88
0.92
0.109
0.88
SF50HA 0.109
0.88
0.92
0.110
0.87
SF50LA 0.108
0.88
0.92
0.112
0.87
UP50HE 0.105
0.88
0.93
0.112
0.87
UP50LE 0.108
0.88
0.92
0.109
0.89
UF50HE 0.110
0.87
0.92
0.110
0.89
UF50LE 0.107
0.88
0.92
0.109
0.89
SP50HE 0.082
0.90
0.93
0.112
0.86
SP50LE 0.109
0.88
0.92
0.110
0.87
SF50HE 0.112
0.87
0.92
0.110
0.88
SF50LE 0.108
0.88
0.93
0.110
0.87
min
0.87
0.92
0.86
mean
0.88
0.93
0.88
max
0.90
0.93
0.90
for
t=2.10
0.92
0.95
0.95
0.94
0.95
0.92
0.92
0.91
0.92
0.93
0.94
0.94
0.91
0.92
0.93
0.92
0.91
0.93
0.95
with random yield strengths and random applied gravity loads, nonlinear structural analysis simulations were performed to calculate
distributions of frame and first plastic hinge strengths.
Frames designed by advanced analysis had a smaller mean strength
than those designed by LRFD, since they contain smaller member
sizes. However, the calculated reliability indices of the frames designed
by advanced analysis were still above 3.0 based on the failure condition
of frame strength. Because advanced analysis primarily controls frame
strength, some frames exhibit non-negligible probabilities of plastic