You are on page 1of 7

JCSS Workshop on Reliability Based Code Calibration

Safety variations in steel designed using Eurocode 3


Mike Byfield David Nethercot
Cranfield University Imperial College
Swindon, SN6 8LA, UK London SW7 2BU, UK

Abstract
It is established that the different design tasks covered by a structural code should conform to
a similar level of reliability. This work demonstrates a wide variation in the reliability of
selected design tasks taken from Eurocode 3. A simple method of correcting this is proposed.
In certain cases this will produce significant gains in economy; in others it will improve
safety.

Keywords: Calibration, Codes, Design, Reliability, Steel Structures, Structural


Safety.

1 Introduction
The structural Eurocodes contain thousands of design expressions for predicting the resistance
of different components in a range of situations for their complete set of structural materials.
These expressions vary in quality, as measured by their ability to correlate accurately against
test data. Some failure modes are highly repeatable and easy to model, such as the bending
failure of laterally restrained beams. Other modes, often involving a high degree of instability,
require complex design expressions to cover all the controlling parameters and physical
effects yet produce only relatively crude correlation with experiments. Such modes include
the shear buckling resistance of plate girders.
The numerical values selected for partial safety factors on resistance, M-factors, have the
potential to significantly affect the economics of one construction material over another. CEN,
the body responsible for drafting the Eurocodes, have adopted what is known as a boxed
values approach to M-factors. The system works by permitting each member state to select
its own M values. These are known as boxed values and are applied to a whole host of
different resistance functions. For example, M1 is applied to the range of design expressions
concerned with buckling in steel construction. This system is attractive since it gives each
member state the freedom to adjust the relative economies achieved by the codes to the levels
already achieved by the existing national standards. However, as this paper will demonstrate,
the numerous expressions governed by M1 include some highly reliable design tasks, and
others for which the boxed value produces reliability levels far lower than desired for safe
design.
The probability of the resistance of a structural member falling below the design resistance
(not the probability of failure) is influenced by 4 factors:
Reliability of material properties
Reliability of geometric properties
Design expression accuracy

1
JCSS Workshop on Reliability Based Code Calibration

The value of partial safety factor, M


A survey of the reliability of basic material and geometric properties for hot rolled steel
sections has recently been carried out, based on over 7000 mill test results from two leading
EU producers [1]. The survey found that steel is produced to a higher quality than was
assumed during the calibration of the safety factors for Eurocode 3-1 [2]. Variability of
material and geometric properties was found to have little impact on reliability. Indeed, the
work showed that the assumptions concerning variability used during the EC3 calibration are
conservative, being based on surveys carried out in the early 1970s [3]. Subsequent work
[6,7] has indicated that accuracy of the design expression has by far the greatest impact on
design reliability.
This paper examines whether a link exists between design reliability and the complexity of
the structural phenomenon being considered. Thus, the probability of resistance falling below
design resistance is established for three radically different resistance functions:
1. The tensile load capacity of ordinary bolts
2. Bending moment capacity of laterally restrained beams
3. Shear buckling resistance of plate girders
The analysis shows that reliability levels vary considerably between these three different
resistance functions. This lack of uniformity is due to the inability of the present system to
accommodate the large variations that exist in the quality of the design expressions. The
importance to each member state of being able to adjust the relative economics of the codes is
recognised and an adjustment to the Eurocodes is proposed that enable reliability levels to
become more uniform. The new method relies on an additional safety factor embedded within
the resistance functions, supplementing the existing boxed-value partial safety factors. In
situations such as restrained beam design, the approach will lead to substantial design
economies. Conversely, safety factors may have to be increased where reliability would
otherwise fall short of the target reliability.

2 Basis of M calibration
The objective of calibration is to provide a scientific basis for selecting values for the -
factors that ensure a given (or target) level of confidence in achieving safe design; i.e. the
probability of resistance (R) minus load (S) < 0 is suitably small. If the statistical distributions
of R and S are known as illustrated in Fig. 1, then the probability that (R-S) will fall below
zero may be represented in terms of the safety index . Where is the number of standard
deviations (R-S) between the mean of R-S (R-S) and the origin, as illustrated in Fig. 1.
The logarithmic normal probability distribution function is used to model both R and S. Basic
geometric and material properties are also assumed to be lognormally distributed [4]. A
lognormal distribution has the advantage that it does not produce negative values.

2
JCSS Workshop on Reliability Based Code Calibration

R-S
.R-S Resistance (R)

Loading (S)

R-S

Fig. 1: Assumed distributions of R, S and R-S

3 The target reliability


The Structural Eurocodes aim to provide a probability of the resistance of structural
components falling below the design resistance (P(r<rd)) of 10-3. This is known as the target
reliability and is not the probability of failure. The design resistance is given by:
rn
rd = (1)
M
Where M is the partial safety factor on resistance and the nominal resistance rn is calculated
using manufacturers values of geometric and material properties. This is sometimes called
the characteristic resistance, which is misleading. Whilst manufacturers specify characteristic
values (95% confidence limit) for material strength, they specify mean values for geometric
properties such as web thickness. For this and other reasons resistances calculated using
nominal material and geometrical properties are not characteristic values, i.e. they do not
represent the 95% confidence limit.
Ln[rd] r
ud.r
frequency

log resistance (ln[r])

Fig. 2: The statistical basis of the reliability calculations

4 The calculation of P(r<rd)


Fig. 2 shows the basis of the reliability calculations, with the log design resistance, Ln(rd),
located ud number of standard deviations ( r ) from the mean of log resistance ( r ) i.e:

3
JCSS Workshop on Reliability Based Code Calibration

ln rd = r u d r (2)
r ln rd
Rearranging, u d = (3)

r
The target reliability is achieved when ud = 3.04 with a sample size n . This correlates to
a P(r<rd) of 10-3. If ud 3.04 and the sample size is small then from Students t-distribution
we can determine the probability of resistance falling below design resistance, i.e:
ln rd
P(r < rd ) = n 1 r (4)
r

Where:
Standard deviation of log resistance, r = ln 1 + v 2r ( ) (5)
2
And the mean value of log resistance, r = ln (br )
m
r
(6)
2

V is the coefficient of variation of resistance


r

rm is the resistance calculated using mean values of basic variables.


b is a measure of any difference between experimental and predicted resistances; i.e, a b
of 1.10 represents a resistance function that on average underestimates resistance by
10%.
Alternatively the safety factor required to achieve the target reliability (known as *R ) can be
determined directly from:
rn
*R =
[
exp ((n + 1) / n ) t
r
0.5
n 1 r ] (7)

Where the tn-1 factor is derived from students t-distribution for a probability of 10-3. This
produces essentially the same result as the method used for calibrating the Eurocodes [4].

5 The reliability analysis


Using the method outlined above three different resistance functions have been calibrated.
Each of these utilises readily available test data to illustrate the main contention of the paper;
it is not claimed that this illustrative treatment is comprehensive in terms of calibrating
against all suitable test data. These include:
1. Tensile resistance of ordinary bolts. This was based on 135 direct tensile tests on 20mm
diameter grade 8.8 ordinary bolts [5].
2. Bending resistance of restrained beams. This was based on 20 tests carried out specifically
for the purpose of calibrating the plastic moment of resistance design function [6]. Lateral
restraints were positioned and section sizes selected to produce what may be considered as
a worst case scenario.
3. Shear buckling resistance of plate girders. This was based on a survey of 35 different tests
on plate girders [7]. Resistance calculations were carried out using the simple post-critical
design method, using formulae relating to girders with web slenderness > 1.2.
In all cases the design expressions were taken from Eurocode 3-1 [2]. The results from the
analysis are listed in Table 1.

4
JCSS Workshop on Reliability Based Code Calibration

Design task EC3 Boxed value Pr (r<rd) M value required to achieve


the target reliability ( *R )
Tensile resistance of ordinary Mb = 1.25 <10-8 0.95
bolts
Bending resistance of M0 = 1.10 4.6x10-6 0.95
restrained beams
Shear buckling resistance of M1 = 1.10 1.0x10-2 1.33
plate girders

Table 1: results from reliability analysis of 3 separate design tasks


It is clear that a significant variation in reliability levels exists between these three design
tasks. Excessive reliability is present for the tensile capacity of bolts and the bending strength
of laterally restrained beams. In both cases it would actually be possible to achieve the target
reliability with a M value of less than unity. Conversely, the shear buckling resistance of plate
girders is a design situation where the probability of design resistance not being achieved is of
the order of 1 in 100. To meet the basic target reliability, 1 should be increased from 1.10 to
1.33.
It seems no coincidence that the most complex design task requires the highest safety factor.
Moreover, it is likely that this variation in design reliability is not restricted to the 3 examples
considered herein. Were the present study to be extended then the variability in reliability
levels might be expected to be of the form shown in Fig. 3. Clearly, dramatic variations in
reliability is both uneconomic and potentially unsafe.
1.0E-08 1.0E-08
1.0E-07 1.0E-07
1.0E-06 1.0E-06
1.0E-05 1.0E-05
P(r<rd)

P(r<rd)

1.0E-04 1.0E-04
1.0E-03 1.0E-03
1.0E-02 1.0E-02
1.0E-01 1.0E-01
1.0E+00
1.0E+00
Individual resistance functions Individual resistance functions

Fig. 3: An idealised view of the variations in Fig. 4: An idealised view of the variations in
reliability between different design tasks reliability, using the system of supplementary
safety factors proposed herein

A more rational method for applying the resistance functions contained in the codes would be
to determine a M factor for each resistance function. The factor would take the form of a
numerical constant incorporated into the design expression, with the designer being largely
unaware of the origin of the factor. No other safety factors on resistance would be applied.

5
JCSS Workshop on Reliability Based Code Calibration

It is appreciated that nation states may be unwilling to give up the freedom provided by the
boxed value system of safety factors. Given this, it would be possible to embed a
supplementary safety factor into each resistance function, whilst retaining the boxed value
system of M factors. The supplementary safety factor would take the form of a numerical
constant, whose purpose is to unify reliability levels. The boxed values selected by nation
states would merely adjust design economy and target reliability, with the variations in
reliability between design tasks being closer to the pattern shown in Fig. 4. The concept is
illustrated as follows:
M
Supplementary factor, SF =
*R
Where M is the boxed value
And *R is the safety factor output from reliability analysis, such as from equation 7.
Thus the design resistance, rd = SF.rn / M
1.10
In the case of the plastic moment capacity of restrained beams, SF = = 1.17
0.94
Therefore, M pl.Rd = 1.17 Wpl f y / M 0 , this would represent a 17% increase in the design
moment, whilst still achieving the target reliability.

6 Conclusions
The role of calibration in seeking to ensure consistency of reliability in design predictions across
a range of different structural phenomena has been discussed. Using 3 topics from steel design
and the expressions from EC3, it has been shown that the processes adopted for calibrating the
Eurocodes lead to widely different levels of reliability. A simple correction, including the
introduction of an additional (hidden) factor whose value depends directly on the closeness with
which the associated design expression fits the supporting test data has been proposed.

7 Acknowledgements
The paper was originally published in the proceedings of the IABSE conference on safety, risk
and reliability, held in Malta in 2001 [8].

8 References
[1] Byfield, M.P. and Nethercot, D.A., Material and geometric properties of structural steel
for use in design, The Structural Engineer, vol. 75/No. 21, Nov. 1997, pp. 1-5.
[2] CEN, Eurocode 3: Design of steel structures - Part 1.1: General rules and rules for
buildings, British Standards Institution, London, 1993.
[3] Alpsten, G., Variations in mechanical and cross-sectional properties of steel, E.C.C.S.,
Second International Colloquium on stability, Introductory Report, Leige, 1977.
[4] CEN, ENV 1993-1-1: Eurocode 3 Tiel 1-1: Annex Z - Determination of design
resistance from tests, European Committee for Standardisation, Brussels, 1993.
[5] Tizani, W., Assessment of the Quality of Imported Bolts for Use in Tensile
Applications, The University of Nottingham, UK, reference no. SC 2000 004
(RG0629), 1999.
[6] Byfield, M.P. and Nethercot, D.A., An analysis of the true bending strength of steel
beams, Proceedings of the Institution of Civil Engineers, ISSN 0965-0911, May 1998,
pp. 188-197.

6
JCSS Workshop on Reliability Based Code Calibration

[7] Nethercot, D.A. and Byfield, M.P () Calibration of design procedures for steel plate
girder design, Advances in Structural Engineering, Vol. 1 No. 2, 1997, pp. 111-126.
[8] Byfield, M.P. and Nethercot, D.A. (2001). Reliability of Steelwork Designed to
Eurocode 3. Proceedings of IABSE Conference on Safety, Risk and Reliability in
Engineering Malta, 203-208.

You might also like