Abstract
Reliability analysis has become important to hide human difficulty in predicting
accurately the fate of man-made and natural objects. Reliability-based designs seem
to have gained thrust since late 1970s in civil engineering. The aim of this lecture
note is to give basic understanding of the concept of reliability-based design of steel
structural components. The procedure used to determine the Load and Resistance
Factors has been discussed towards understanding reliability-based design of steel
structural components.
INTRODUCTION
May it be life or business, it is the intention of the human beings to live and work
safely at all times. While, how safe is safe? depend on different parameters, safety
is wanted in all disciplines of engineering. For example, riding a bike at the speed of
40km/hr may be considered to be safe for a person of age 16, while a speed of
60km/hr may be considered to be safe for a person of age 25!... This is purely based
on the assumption that a person of age 25 will have good control over the ride and
would be aware of the risks. Similarly, safety of different elements of a structure is
dependent on the individual mode of failure of particular element. Finally, the safety
of individual elements together dictates the safety of the structure as a whole. Safety
of the individual element, and hence the structure, may be disrupted due to the
application of load more than the load for which the element has been designed to
transfer/carry or due to the over-estimation of the actual load carrying capacity of
the element.
The prediction of the loads and load combinations that would create load effect(s)
that would govern the design, and the safe load carrying capacity of the element
form the basis of design procedure. It should be noted that, the load that will act on
an element/structure with respect to position, magnitude & direction and the load
carrying capacity of the element/structure with respect to magnitude may have
variations. These variations become uncertainties in the reliability analysis. While
the Allowable Stress Design (ASD) or Working Stress Design (WSD) does have
factor of safety, these uncertainties are not taken care rationally. It is the scope of
this lecture note to give an insight into the way the possible uncertainties are taken
care in a rational form in reliability-based design of steel structural components.
PROBABILISTIC ANALYSIS
Probability theory is a way human beings hide their inability to predict accurately
the behavior of nature, and even, man-made objects. Because human beings do not
know what will happen tomorrow certainly, they talk in terms of chance! We
would prefer always to tell tomorrow it may rain to tomorrow it will rain. The
word may gives information that we are not sure about raining tomorrow. If
someone says, there is 70% chance for raining tomorrow, it seems to give a better
quantification of the uncertainty, than simply saying it may rain tomorrow. These
uncertainties are quantified through probabilistic analysis.
Probabilistic analysis aims at quantifying the uncertainties in the dependant variable
due to the uncertainties in the identified independent variables. Reliability analysis is
the same as the probability analysis but by using the term reliability analysis we
mean to fix the probabilities of failure at different limit states of failure and then
estimate partial safety factors to make the structure meet the required reliability.
TYPES OF UNCERTAINTIES
The following are the two types of uncertainties.
1. Aleatory uncertainty
2. Epistemic uncertainty
Aleatory uncertainty, also called as irreducible uncertainty or inherent uncertainty, is
used to represent the unknowns which differ each time when the experiment is run.
Epistemic uncertainty is also called as reducible uncertainty. Epistemic uncertainty
can be reduced considerably once more information is available from past
experience.
PROBABILITY DENSITY FUNCTION
Probability Density Function (PDF) is used to represent the relative frequency of
realizations for a given random variable. In the PDF, the most expected realization is
at the centre (mean) and the realizations for which chances are less to occur are
found near the tail regions. Some characteristics of PDF are given in Appendix.
LIMIT STATES
A limit state is a limit beyond which a member or a structure does not perform to
meet the requirements. When the probability of failure for the structure to cross a
Ellingwood and Galambos9 present the probability based loading and resistance
criteria that are suitable for routine safety checking in design that is based on
statistical on analysis of loads and resistance and examination of reliability levels
implied by the current design standards. Lenz and Galambos 10 have presented a load
factor design criterion for buckling stress of structural steel columns based on FirstOrder probability theory. Load factors and strength factors were evaluated using the
then available information and test results. It has also been shown how different
column theories can be compared on the basis of design resistance obtained by each
theory. The paper concludes by describing how probabilistic problems can be
simplified by using First-Order theory.
ASSESSMENT OF STRUCTURAL RELIABILITY
Classical Reliability Theory
An element/structure is subjected to load(s) that is/are not constant. Also, the load
g R S
g
As long as
is greater than zero, the structure would be safe. So the probability
of failure of the structure is given as,
Pf P g 0
g 0
While
indicates the failure region,
failure surface and safe region, respectively.
The mean and the standard deviation for
g R S
g is given as,
g R S 2 RS R S ...................................................(1)
2
g
g
R S
R S 2 RS R S
2
.........................................(2)
g
g
R S
R S
RS
................................................................(3)
As shown in Figure 1, the safety index indicates the distance of the mean margin of
g 0.
safety from
If R and S are assumed to be normally distributed and
uncorrelated, the PDF of limit state function is given by,
fg g
1 g g
exp
2 g
2
.................................................(4)
Pf
f g dg....................................................................................(5)
g
Pf
1 0 g
exp
2 g
2
1
2
exp dg
2
2
Pf
dg..................................................(6)
Pf 1 ...........................................................................(7)
where,
g ( P, L, W , T ) WT
PL
4
g W T
1
P L 40
4
W T P L
4
2
g
1600
g E ( g 2 ) g2
1
P2 L2 L2 P2 L2 P2 16 W2 T2 T2 W2 T2 W2
4
16.2501
2
PL
E ( g ) E WT
4
1
1
( W2 W2 )( T2 T2 ) W T P L ( P2 P2 )( L2 L2)
2
16
2
2.46153
Problem: 2
Consider the same problem: 1 with the same structural and statistical properties.
Now considering two different formulations of the limit-state function as:
PL
4
PL
g 2 ( P, L, W , T ) T
4W
g 1 ( P, L, W , T ) WT
10 8
4
2.48
g2
g2
10 8
4 100 10 6
4
2
(2 10 2) (2.5 10 3 ) 2 (4 10 4 ) 2 (1 10 5 ) 2
3.48
600 103
It can be seen that the safety indexes 1 and 2 determined using FOSM method are
different for different limit states even though the two limit state equations are
equivalent. This lack of invariance was overcome by the Hasofer and Lind12 method.
Hasofer and Lind12 improved the FOSM method by introducing the Hasofer and
Lind (HL) transformation where the design vector X is transformed into the vector
of standardized, independent Gaussian variables, U. Different approximate response
surfaces g(U)=0 correspond to different methods for determining failure probability.
If the response surface is approached by a first-order approximation, the method is
called the First-Order Reliability Method (FORM) and if the response surface is
approached by a second-order approximation at the MPP, the method is called the
Second-Order Reliability Method (SORM)
LRFD FORMAT
The general LRFD format is given by6
R i Si ................................................................(8)
i:
where, R and Si are resistance and load-effect; and i are resistance and load
factors, respectively. It can be noted from the above equation that the resistance and
load effects are considered separately. And, hence any improvements in the
knowledge (viz. development of better models, characterisation of uncertainties) on
resistance or load can be incorporated readily. The other advantages of LRFD format
and
i are relatively insensitive to changes in the
are: (i) the individual factors
design parameters (e.g. the load ratios between various loads) and thus a given set of
load factors can be applied to wide range of design conditions, (ii) with partial load
and resistance factors, any other form of design factors can be readily derived.
Hence, in this investigation partial safety factors are derived for this format. The
basic formulation is described in the next section.
Since, in the present lecture, dead and live load combination is considered, the
corresponding LRFD format is,
R D D L L .........................................................................(9)
,
and
L
D
The determination of
for specified distributions of R, D and L can be
accomplished using AFOSM method, which makes use of Rosenblatt
transformation13 at the design point. The safety of the structural component designed
according to LRFD format can also be expressed in the form of central safety factor.
The concept of central safety factor is presented below.
R
S .........................................................................................(10)
where R and S are the means of resistance and load effect, respectively. To take
into account the COV of variables in the determination of safety factors other
expressions such as characteristic safety factor have been proposed 14,15. Recently,
efforts are being made to include the information about the skewness coefficients of
R and S in the determination of safety factors16.
Determination of Partial Safety Factors
Basic Formulation
The basic variables for the problem under consideration are R, D and L. A linear
performance function assumed in basic variable space is given by,
g(R,D,L) = R(D+L).................................................................(11)
It may be noted that R, D and L should be expressed in consistent units. It is
reasonable to assume that R, D and L are statistically uncorrelated. Random variable
R is a general resistance variable and can represent for example the flexural or
compression or tension resistance of a member. However, the statistical properties
corresponding to the particular type of resistance under consideration should be used
while deriving safety factors. It can be easily seen from Equation 11 that the
combinations of R, D and L which result in g(.) < 0 represent the failure domain.
The plane represented by the equation g(.) = 0 is known as the limit state surface.
Using the linear optimization technique it can be shown that the formal reliability
index (which is a measure of reliability of a structural component) for linear
performance function considered is given by:
R D L
R2 D2 L2
................................................................(12)
Pf ( )....................................................................................(13)
The probability of failure computed using Eq. (13) is exact in the sense of numerical
integration, only when basic variables follow normal distribution and when the
number of variables are small. However, when any one (or all) of the variables
follow non-normal distribution, the value of should be determined from14,17,18.
RN DN LN
( RN ) 2 ( DN ) 2 ( LN ) 2
..........................................................................(14)
D*
W*
RN
( RN ) 2 ( DN ) 2 ( LN ) 2
...........................................................(15)
DN
( RN ) 2 ( DN ) 2 ( LN ) 2
LN
( RN ) 2 ( DN ) 2 ( LN ) 2
.........................................................(16)
............................................................(17)
D 1 D* t D ..............................................................(18)
L 1 L* t L
The central safety factor corresponding to the LRFD format considered is given by:
1
1
c
D L L
D
R ........................................ ................................................(19)
D
P ()
where fi
= failure probability of ith structural component designed using set of
partial safety factors ; Pft is the target failure probability; = (1, 2, ..., m) = set
of weighting factors indicating the relative importance of each of m structural
components included in the partial factor evaluation j = 1.0 (j=1,...,m).
The overall procedure of code calibration is illustrated in Figure 2. The
recommended values of target failure probabilities and target reliability indices by
Nordic committee on building regulation are presented in Table 1. It is important to
note that to obtain consistent safety factors to be used in the design requires large
amount of data and hence, normally resistance factors are determined for structural
members for specified limit states.
Failure Type
I
II
Not serious
10-3 (3.09)
10-4 (3.71)
-4
Serious
10 (3.71)
10-5 (4.26)
-5
Very serious
10 (4.26)
10-6 (4.75)
Note: Values within brackets are target reliability indices
III
10-5 (4.26)
10-6 (4.75)
10-7 (5.20)
Design the components to the limits of specified existing old code with which a new code for which partial safety factors are sug
Design components according recent code for which are suggested and which contains more rational design clauses reflecting n
Modify
NO
Is S a minimum
YES
STOP
Compute Si
depth of web, thickness of flanges and web, Youngs modulus of steel, yield strength
of the material, effective length of the column are treated as random variables in the
simulation. The Probability Density Function (PDF) of P/P y and the fractiles of the
strength estimated by IS: 800-2007 with respect to the strength estimated by SSRC
curves are reported. The scope of the work is to determine the reliability associated
with the IS:800-2007 column strengths as estimated by the column curves a and
b of IS: 800-2007 with respect to SSRC-1 and SSRC-2, respectively. It was noted
from this study table that the deterministic values of P/P y estimated by the IS:8002007 are less than the values estimated by the SSRC curves. This suggests that the
deterministic column strength estimated using IS:800-2007 is conservative by 3-14
% in the case of major axis buckling and around 1-6 % for minor axis buckling.
Recommendations made:
1. The non-dimensional slenderness ratio defined in IS:800-2007 should be
redefined in-line with Eurocode 3, because the column curves suggested in this code
are based on Eurocode.
2. The partial safety factors should be made a function of to achieve uniform
reliability in the design of compression members.
Proposed characteristic non-dimensional slenderness ratio equations are,
For major axis buckling,
* = 1.184, Correlation coefficient, R2=1.0
For minor axis buckling,
* = 1.197+0.003, (>0), Correlation coefficient, R2=0.99
Probabilistic analysis of a steel beam-column designed as per IS: 800-200720
The probabilistic analysis of steel beam-column designed as per the
recommendations of IS: 800-2007 has been carried out. By using the test results of
the steel beam-column collected from the literature, it was observed that the codal
interaction equation under-estimated the strength of the beam-column. Hence,
modeling error was used to multiply the interaction equation to reflect the failure
criteria (i.e, interaction ratio = 1.0). Then the probabilistic analysis was carried out
using the modified interaction equation and the probability distribution function for
the interaction ratio was determined, from which, the probability of failure of the
beam-column for all the load combination considered was determined. The
probabilistic analysis was carried with all the load combinations considered and the
interaction ratios were determined using the modified interaction equation. The
statistics of the interaction ratios were studied and it was found that, for all the
combinations of load, lognormal distribution cannot be rejected to represent the
interaction values at 20% significance level.
Reliability-Based Design of Steel Imperfect Columns against Buckling 21
Probabilistic analysis of load carrying capacity of geometrically imperfect columns,
with slenderness parameters covering the range of practical interest, is carried out
using Monte Carlo simulation technique. In simulation, the cross-sectional
dimensions, effective length of column, modulus of elasticity of steel and yield
strength of steel are treated as random variables. The effects of changes in mean
central imperfection, coefficient of variation of effective length parameter and the
yield strength of steel on statistical properties of load carrying capacity are studied.
Using the results of probabilistic analysis an equation for characteristic load carrying
capacity is proposed. Depending on the value of slenderness parameter, two values
for characteristic resistance factor have been suggested, which will help in achieving
uniform reliability in the column design.
REFERENCES
1. Yao, J.T.P, Kawamura, H. (2001), On Structural Reliability, Jl of
Temporal Design in Architecture and the Environment, 1, pp.1-5.
2. Moses, F. (1977), Structural system reliability and optimization,
Computers & Structures, 7, No.2, pp.283-290.
3. Moses, F. (19821983), System reliability developments in structural
engineering, Structural Safety, 1, No.1, pp. 3-13.
4. Hohenbichler, M., Rackwitz, R. (19821983), First-order concepts in
system reliability, Structural Safety, 1, No.3, pp.177-188.
5. Dolinski, K. (1982-1983), First-order second-moment approximation in
reliability of structural systems: Critical review and alternative approach,
Structural Safety, 1, No.3, pp.211-231.
6. Ellingwood, B., Galambos T.V., MacGregor, J. G., and Cornell C.A.
(1980), Development of a Probability Based Load Criterion for American
National Standard A58: Building Code Requirements for Minimum Design
Loads in Buildings and Other Structures, National Bureau of Standards
Special Publication, U.S, 577.
7. Galambos, T.V., Ellingwood, B., MacGregor, J. G., and Cornell C.A.
(1982), Probability-Based Load Criteria: Assessment of Current Design
Practice, J. Struct. Eng., ASCE 108, pp.959-977.
8. Ellingwood, B., MacGregor, J. G., Galambos T.V., and Cornell C.A.
(1982), Probability-Based Load Criteria: Load Factors and Load
Combinations, J. Struct. Eng., ASCE, 108, pp.978-997.
9. Ellingwood, B., Galambos, T.V. (1982-1983), Probability-based criteria
for structural design, Structural Safety, 1, No.1, pp.15-26.
10. Lenz, J., Ravindra, M.K., Galambos, T.V. (1973), Reliability based design
rules for column buckling, Computers & Structures, 3, No.3, , pp.573
588.
11. Choi, S. K., Grandhi, R., Canfield, R. A. (2007), Reliability-based
Structural Design, Springer-Verlag London limited.
12. Hasofer, A. M, and Lind, N. C. (1974), Exact and Invariant SecondMoment Code Format, Jl of the Engineering Mechanics Division, ASCE
100, EM1, pp.111-121.
13. Rosenblatt, M., (1969), Remarks on a multivariate transformation,
Annals of Mathematical studies, Vol.23, pp. 470-472.
14. Madsen, H.O., Krenk, S. and Lind, N.C., (1986), Methods of structural
safety, Prentice Hall, Inc., Englewood Cliffs, NJ.
APPENDIX
Some Characteristics of PDF
Random variable
A random variable, X, can take any values in the range - < x < . While a random
variable is denoted by a capital letter, say X, its realization is denoted by a small
letter, x.
Fx x f x ( x)dx
x
x E ( X ) xf x dx
average or mean, .
Dispersion Measures
How the realized values of the random variable are distributed about the mean is
represented by Second Central Moment of X, called variance. It is given by,
V ( X ) E ( X x) 2
Standard deviation is defined as the square root of variance.
x V (X )
x
and is denoted by .
x
x
Correlation Measures
If any two random variables X and Y are correlated, the value of X can be affected
by the value of Y. The linear association between these two variables can be
represented by covariance and is given by,
XY Cov( X , Y ) E X X Y Y
and, the correlation coefficient, which is a non-dimensional measure of correlation,
is given by,
XY
XY
XY
If the two variables X and Y are independent, correlation coefficient is zero and
therefore, correlation coefficient of 1 indicates perfect correlation.
List of common probability distributions
1.
2.
3.