Professional Documents
Culture Documents
Summary
The subject of reliability concepts in Load Resistance Factor Design (LRFD) would seem to be a pretty dry subject –
perhaps the domain of academics and code writers – and it is. But it is also a subject which has very real and practical
implications for all foundation projects. There are many projects in which achieving the specified factor of safety can
cause major difficulties, and allowing some rational evaluation of the design requirements and some flexibility in the
application of acceptance criteria might be very helpful to the progress of the contract. This paper presents some of
the author’s current thoughts on the application of reliability principles to foundation acceptance. The paper is
developed around the LRFD approach, however as will be shown, the conversion between LRFD load and resistance
factors and global factor of safety is simple and direct.
Why do we need a factor of safety, and what is the So far we are still above ground, dealing with
appropriate value? Simply, we need to ensure that there uncertainty in applied loads. Below ground there is
will be a sufficient excess of capacity above the applied even greater uncertainty. First, we must deal with the
loads, so that there is no risk of failure. Furthermore, method of capacity estimation. This may be by
requiring a minimum factor of safety probably has a geotechnical design using one of the many design
secondary consequence of achieving satisfactory methods presented in text books or technical papers.
deflection performance at working load. These methods rely on input of appropriate strength
parameters determined from the site investigation. As
To be honest, we should refer to the factor of safety as geotechnical design is predictive and generally empirical
the factor of ignorance - in dealing at the interface or semi-empirical, it is associated with a large degree of
between structures and nature there is much of which we uncertainty. The results of a prediction exercise for a
are ignorant. As practitioners in the exciting art of recent pile load test in Singapore (Wong, 2002), show
foundation engineering, this is both our privilege, and the wide distribution of capacity predictions typical for
the cross we must bear. such predictions events (see Figure 1). Most predictions
in this case were conservative, however, this cannot be
Of what then are we ignorant? First, the designer has assumed as a general rule.
provided us with design loads based on an estimation of
1
Fulcrum The Newsletter of the Deep Foundations Institute Summer 2002
250 0
200 0
Fa ilure Lo ad = 18 06 kN
More recent foundation design code practice has been to
150 0 adopt a limit state approach, in which uncertainty
100 0 associated with the loads is explicitly included in a load
50 0
0
factor (ψ) greater than 1, and uncertainties associated
1 3 5 7 9 11 1 3 1 5 1 7 19 21 23 25 27 2 9 31 33 with determination of capacity are incorporated into a
P a rti ci pa nts resistance factor (φ) less than 1. The limit state or
LRFD approach then requires that the factored up loads
Figure 1 – individual capacity predictions from
pile prediction exercise must always be less than the factored-down resistance.
Put mathematically, this is :
Table 1
Factors affecting choice of resistance factor (extracted and summarized from the Australian Piling Code)
Aspect of Design or Verification φ min φ max Case for φ min Case for φ max
Site investigation - - limited comprehensive
Static design calculation (CPT based) 0.45 0.65 simple sophisticated
Static design calculation (SPT based) 0.40 0.55 simple sophisticated
Properties chosen - - average conservative
Correlations used - - published site-specific
Construction control - - limited careful
Dynamic pile testing with signal match 0.65 0.85 <3% tested >15% tested
Dynamic pile testing no signal match 0.50 0.70 <3% tested >15% tested
Static testing 0.70 0.90 <1% tested >3% tested
the capacity reduction factor should be lower if the implemented in a spreadsheet in order to establish a
consequences of failure are higher. statistically-based capacity reduction factor.
In statistics, the student t-distribution determines the This approach has been adopted on a number of projects
confidence that a given limited sample of n pile capacity to establish the relationship between the number of tests
tests with sample mean m and standard deviation s is and capacity requirement.
representative of all N piles on a site with mean capacity
µ and standard deviation σ. This concept of “rewarding” the project by allowing a
reduction in capacity requirements for increased or
tn-1,α = m - µ (4) higher order testing makes good technical and economic
s /√ n sense. The cost reductions available to the contract
where for instance α=0.005 for a 99.5% confidence through increased testing can often far outweigh the
limit. A normal distribution is assumed. For this costs of testing, especially for cost-effective techniques
analysis, an assumption is made that the standard such as pile driving analyzer (PDA) testing. The use of
deviation of the sample is a reliable estimate of the a limited number of static load tests on a project may
standard deviation of all the piles. give the engineer a good feeling about the construction
technique, but it can be seen from the graphs shown in
An intrinsic reduction factor, φi is defined for each the next section, that it does not provide a high reliability
method of estimating pile capacity. Paikowsky (2001) is in a statistical sense. PDA testing really offers engineers
undertaking a study of a wide range of pile capacity for the first time a statistically significant method of
estimation methods in order to propose intrinsic establishing the reliability of the whole foundation.
reliability factors on the basis of statistical analysis.
This intrinsic reduction factor is the maximum reduction The coefficient of variation is a measure of the
factor that could be adopted for that method when every variability of soils. Typical values quoted in the
pile is tested. (n = N). literature vary from 0.10 (low variability) to 0.35 (high
variaibility). The coefficient of variability governs the
When only a sample of piles are tested, the net reduction confidence with which individual results can be
factor, φr will be less than φi. The value of φr can be extrapolated to the whole site. Although the coefficient
computed as follows: of variation can be ‘guessed’ from typical values, or
estimated based on the results of the site investigation, it
k.tN-1,α is preferable to get site-specific estimates which can be
1+ based on test results (e.g. SPT-N, blow counts or pile test
√ N (1 - k.z1- α ) (5)
φr = φi results).
k.tn-1,α
1+
√ n (1 - k.z1- α ) 3. RESULTS OF STATISTICAL ANALYSIS.
where, k is the coefficient of variation σ/µ. Despite the As noted previously, this method can easily be
apparent complexity of this expression, it can easily be incorporated in a spreadsheet analysis. Interested
persons can download a spreadsheet which implements
3
Fulcrum The Newsletter of the Deep Foundations Institute Summer 2002
between 25 and 500. The curves show that in all cases, 0.1
4
Fulcrum The Newsletter of the Deep Foundations Institute Summer 2002