You are on page 1of 8

Reliability Engineering and System Safety 73 (2001) 205±212

www.elsevier.com/locate/ress

Cost and safety optimization of structural design speci®cations


Engin Aktas a, Fred Moses a,*, Michel Ghosn b
a
Department of Civil and Environmental Engineering, University of Pittsburgh, Pittsburgh, PA 15261, USA
b
Department of Civil Engineering, The City College of New York, CUNY, New York, NY 10031, USA

Abstract
The design of buildings, bridges, offshore platforms and other civil infrastructure systems is controlled by speci®cations whose purpose is
to provide the engineering principles and procedures required for evaluating the safety of structural systems. The calibration of these codes
and speci®cations is a continuous process necessary to maintain a safe national and global infrastructure system while keeping abreast of new
developments in engineering principles, and data on new materials, and applied loads. The common approach to speci®cation calibration is to
use probabilistic tools to deal with the random behavior of materials and to account for the uncertainties associated with determining
environmental and other load effects. This paper presents a procedure to calibrate load factors for a structural design speci®cation based on
cost and safety optimization. The procedure is illustrated by determining load factors that may be applicable for incorporation in a bridge
design speci®cation. Traditional code calibration procedures require a set of pre-determined safety levels that should be used as target values
that each load combination case should satisfy. The procedure in this paper deduces the failure cost implied in present designs, and provides
consistent safety levels for all load combination cases. For greater accuracy, load effects showing variance in time have been modeled by
separating them into two random variables; time dependent r.v. (wind speed, vehicular loads, etc.) and time independent r.v. (modeling
uncertainties). The total expected lifetime cost is used in the optimization to account for both initial construction cost and future equivalent
failure costs. q 2001 Elsevier Science Ltd. All rights reserved.
Keywords: Cost optimization; Structural reliability; Code calibration; Target reliability; Load combination; Ferry±Borges process; Turkstra's rule

1. Introduction the calibration process should ideally lead to optimized load


and resistance factors that would minimize the total costs
Structural design codes provide the tools and procedures implied in using the new code. The total costs include the
that engineers should follow to produce safe and economic initial cost of construction and future expected risk costs
structures. However, the determination of the target safety that are functions of the probability of failure and the cost
level that design codes should satisfy has been an issue of of failure.
debate since the introduction of the probabilistic approach Previous calibrations of load and resistance factors have
to code calibration in the late 1960s. The pioneering been carried out to satisfy pre-determined target safety
publication by the National Bureau of Standards that devel- indices (b ). The safety index (b or beta) is directly related
oped target safety criteria for the ANSI A-58 document on to the computed risk of failure. In many code developments,
Minimum Design Loads in Buildings [1] has provided the a single target b is extracted from existing design criteria
most widely adopted model for the calibration of structural [2], or else, an actuarial target b is recommended. This
design codes based on probability concepts. target b is then used as the value that the new code should
The calibration of a new structural design code involves uniformly achieve for all applications and load cases. A
®nding load and resistance factors based on updated statis- uniform target beta for all load cases however, does not
tical data on member properties and applied loads; structural distinguish between the marginal costs associated with
reliability modeling; and the relevant performance of increasing the load factors for one load case compared to
structures designed using the current code. It is crucial to another (e.g. gravity compared to seismic); or decreasing the
also consider the economical impact of the new code during resistance factor for one type of element compared to
this calibration process. Because a trade-off exists between another (e.g. main members compared to connections).
the higher costs associated with providing increased safety, Numerous examples exist in structural code develop-
ments where different target betas have been used based
* Corresponding author. Tel.: 11-724-368-8996; fax: 11-724-368-1108. on intuitive economic reasoning and past design experience
E-mail address: moses1@pitt.edu (F. Moses). [3]. For example, use of higher b 's is often observed when
0951-8320/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0951-832 0(01)00046-1
206 E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212

calibrating load factors for a case when gravity loads are be independent between one recurrence period and another,
applied alone as compared to the b 's used when environ- but the modeling uncertainties are assumed constant over
mental loads such as wind, earthquake or wave loadings are the design life.
applied [4]. Similarly, higher b 's are often used for connec- For those design checks that contain more than one time
tions compared to the b 's used for main members or foun- dependent random variable, a load combination process
dations [3]. Despite such experiences, it appears that must be used to ®nd the maximum combined effect of the
structural codes are moving to introduce criteria for a different loads. In this paper, the Ferry±Borges method is
uniform target design safety index applicable to all load used to effect the load combination analysis.
combination cases, especially where system reliability In general, codes are formulated to check the design of
models can be carried out. The basis for accepting a uniform individual components although in reality these components
b criterion is that it leads to uniform risk of failure for all interact to form one structural system. Further work is still
load cases [1]. Optimization in the context of uniform risk is needed to extend the proposed minimization process to
based on a formulation in which load and resistance factors account for system safety as well as component safety.
are found to minimize the deviation of computed betas from The application of the proposed method is illustrated with
the target beta for a large population of representative an example for the AASHTO bridge speci®cations. The
design examples. design check for the combination of the dead load and
The current approach to optimization that uses a single live load (vehicular load) that have been calibrated in the
target b is very limited in its scope since it ignores past current code to a target safety index bT ˆ 3:5 is used as the
design experiences which did recognize the economic base case for the proposed calibration procedure. The safety
implications of providing particular components with provided by the current code for this base case is assumed to
additional safety levels taking into considerations the risk be at a satisfactory level. Hence, the deduced parameters of
consequences of a component's failure; and the marginal this base case including the cost of failure are applied to
costs associated with increasing each respective load and calibrate load factors for other load combinations. The
resistance factor. This paper presents a comprehensive optimization procedure is performed for a range of design
approach to structural code optimization, which considers examples to cover the possible design space. The load
relative cost margins for each load case and load combina- factors obtained from the optimization analyses are
tion. Load and resistance factors are selected with a criterion tabulated for the given example.
to minimize the total cost including the associated risk cost.
Minimization may be carried out over a sample of represen-
tative designs similar to the sample that is usually used 2. Total cost function
when optimizing with a ®xed target beta. In the proposed
optimization model, the betas are not ®xed but target betas The cost function used in this example contains only the
are determined as a part of the minimization process and initial cost and the cost of failure. Maintenance costs or
will vary with each load combination case and component future rehabilitation costs are not considered herein. The
type. total cost function can be represented by the sum of the
A major obstacle in the proposed cost optimization initial cost and the failure costs all brought to the same
approach is to quantify the failure cost. Failure costs are reference time as expressed in Eq. (1):
dif®cult to be assessed a value especially when human Xn
lives are involved. Such direct cost estimates are avoided Pfi
CT ˆ CI 1 Cf …1†
in the proposed approach, and instead the cost of failure is iˆ1
eij
deduced as an implied or `nominal' cost based on current
code practice. where CT is the total cost, CI the initial cost, e ij the dis-
In most current reliability applications, the data for count factor, `j' the discount percentage rate …real rate ˆ
environmental loads are represented as time dependent nominal 2 inflation†; Pfi the annual failure probability for
random variables. The probability distributions of maxi- year `i', Cf is the failure cost, and n is the total number of
mum load effects are projected based on the exposure period years in the design life of the structure. The discount rates in
or design life considered. This approach normally ignores developed countries are between 2 and 4%, and assuming
the difference between the time dependent variables and the that rate equal to 3% throughout the design life is suf®cient
time invariant variables. However, the proposed cost to represent the trend. Eq. (1) assumes that the failure costs
optimization model requires the monitoring of yearly in real terms are constant throughout the design life.
variations in the probability of failure to have an accurate The design of structural components is normally
evaluation of the equivalent failure cost as all costs must be controlled by an equation of the form:
brought to their present worth values for comparison. Thus, fRn $ gD Dn 1 gL Ln 1 gW Wn …2†
a more accurate separation between time dependent and
time invariant variables is necessary. In particular, during where Rn is the nominal component capacity, f the resis-
the reliability calculations, load repetitions are assumed to tance factor, Dn, Ln and Wn stand for nominal values of dead,
E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212 207

live and wind load, respectively, and g D, g L, and g W are the quantifying Cf becomes more of a political issue rather than
load factors for dead, live and wind loads. an engineering one. Ditlevsen [6] suggests using some ®xed
The total initial and failure costs associated with using reference from the current code to determine the cost of
Eq. (2) should be applicable to the full range of design failure. In the proposed approach, instead of estimating
applications covered by the particular speci®cation includ- the values of Cf by using real cost values, previous codes
ing different structural types, sizes and geometries. Nominal and designs and performance histories are used to deduce an
load ratios such as Ln/Dn and Wn/Dn, etc. may be used to implied failure cost value. For example, one load combina-
represent the design space. Cost terms, such as initial cost tion case such as the application of the dead load alone using
and failure cost show variance from one structure to another. the AISC building design code, or the combination of dead
Therefore, these cost terms will be normalized by a base and live loads acting together using the AASHTO bridge
cost factor to generalize the results of the optimization speci®cations, can be used to deduce the Cf value.
process. Therefore, a current code with its prede®ned target safety
level for a particular combination is assumed to present a
2.1. Initial cost balanced cost with a suf®cient level of safety. The implied
failure cost can then be deduced using the total cost function
Given a structural con®guration and size, the initial cost and assuming it is minimized at the current code's load and
of a structure can be expressed as a function of its safety resistance factors. Deducing the Cf value from the current
level. The initial cost, CI can be represented by the follow- code allows code writers to reach a consistent code for all
ing formula [5]: other load combinations and avoids the issue of assigning
! !
gD gL monetary values to the cost of failure. The latter can be a
CI ˆ C0 1 1 KD 2 1 1 KL 21 politically intractable problem, which would create grounds
gD0 gL0
for many discussions and would not necessarily lead to an
! ! !! acceptable solution.
gW X g
1 KW 2 1 1 ¼ ˆ C0 1 1 Ka a
21 In the proposed approach, the implied failure cost value is
gW0 ga0
a introduced in the optimization process of the rest of the load
…3† combination cases to be included in the design code. Thus,
C0 is the base cost which is the construction cost of the the proposed approach maintains consistency in the design
structure associated with the use of the reference load code by considering the respective marginal initial cost
factors designated by the subscript 0, g a0 …a ˆ slopes (see Eq. (3)) even though the safety index values
{D; L; W; ¼}†; where D, L, W are dead, live and wind will not generally be found to be uniform for all load combi-
loads), and Ka is the normalized cost factor for the respec- nation cases. The cost of the consequences of a failure
tive load effects. The load factors are represented by ga …a ˆ however would be the same for all load combinations.
D; L; W; ¼}†: Eq. (3) is a general form of the initial cost It is important to realize that the procedure outlined in this
function and assumes a linear relationship between the load paper deals only with the component safety, as it is
factors and the initial cost. commonly done in current code speci®cations so failure
Different types of load such as gravity and environmental is de®ned as the component failure. When system safety is
loads may affect the initial cost differently. For example, in considered, other factors such as redundancy and collapse
most structures, the marginal cost slope for increased lateral mechanisms must be considered. The effect of system safety
loading such as wind or earthquake can signi®cantly exceed is being investigated in an ongoing research project and will
that of gravity loads; therefore different cost factors are used be the subject of another paper.
to represent these effects. It is reasonable to use either unity
or the current code's load factors as the reference load 2.3. Normalization of the terms in total cost function
factors, because these affect only the base cost value. As mentioned above a design code must cover a full range
Normalization makes the ®nal conclusions independent of of design applications. Therefore, the terms used in Eq. (1)
the actual value of C0, therefore the optimization is not have to be generalized. The total cost function (Eq. (1)) is
dependent on the reference load factors. normalized by the base cost C0, and a total cost factor, TCF
is expressed as
2.2. Failure cost      
C gD gL
Determining the cost of failure, Cf is a key issue in the TCF ˆ T ˆ 1 1 KD 2 1 1 KL 21 1¼
C0 gD0 gL0
cost minimization process. This term has to include the
economic cost of failure, including intangible costs and Xn
Pfai
1g
consequential costs. The importance of a structure and the eij
iˆ1
costs of keeping that structure in service are contained
…4†
within the Cf value. It is clearly not easy to quantify Cf
especially when human loss is considered and the issue of where the failure cost ratio, g is the ratio of the failure cost,
208 E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212

CF, to the reference initial cost, C0 ; …g ˆ CF =C0 †: Eq. (4) is other variables such as resistance, dead load or modeling
used herein to deduce the implied failure cost ratio, g for the uncertainty remains invariant over time. It is assumed that
reference case of the design code. This value of g is then the time dependent loads are independently repeated from
used consistently to optimize load factors for all other load one recurrence period to the other, but that the modeling
combination cases. uncertainties make their effects on the structure correlated.
Therefore, if the modeling uncertainties are separated from
the load intensities, one could take advantage of the proper-
3. Reliability considerations
ties of independent variables in the FORM algorithm. The
separation of the time-dependent from the time invariant
Current code calibration procedures use the ®rst order
variables is important because the presence of the modeling
reliability method (FORM) to calculate the safety index b
uncertainty and other non-time dependent variables affect
that is used as the primary measure of structural safety.
the annual probability of failure Pfak that decreases over time
3.1. First order reliability method (FORM) as k increases.
The reliability analysis of combined time dependent loads
FORM is based on the ®rst order approximation of the such as wind plus live load requires a combination tech-
failure function, G, at the design point. Since the design nique to handle the different time dependency effects.
point is not known in advance, an iterative procedure is Load combination techniques, including Turkstra's rule
required. The safety index b represents the shortest distance and Ferry±Borges method are available to combine the
from the failure surface to the origin of the normalized effects from several time dependent loads.
design space and the closest point on the failure surface is
de®ned as the design point. Different versions of the FORM 3.2. Load combination processes
algorithm are provided in many sources [7,8].
Time dependency can be handled with a minor modi®ca- As the intensity of wind, live and earthquake loads vary
tion of the traditional FORM if the limit function, G, during the design life of a structure, it is important to
contains only one time dependent random variable. In the account for the probable maximum response when analyz-
traditional FORM algorithms, a transformation is carried ing the structure for the combined effect of two or more of
out for every non-normal random variable into an equivalent these extreme events.
normal variable at a trial design point. The calculation of b Simultaneous occurrences of these extreme events at their
is executed for that trial point and the process is repeated maximum intensities are rare, so the combined maximum
until convergence. An identical procedure is followed for effect is not the sum of the expected maximum of each load
the time dependent variable by observing that the probabil- effect. In the early design codes the combination effect was
ity distribution of the variable after k recurrence periods is accounted for by using a reduction factor. So, when more
the probability distribution for one event raised to the kth than one time dependent load was involved in a combination
power. As an example, assume that an environmental load case, the sum of the expected maximum of each load effect
effect, Y1 can be represented as follows: was multiplied by a factor less than one. This was an
intuitive approach to reach affordable designs. Currently
Y 1 ˆ Xm Q 1 …5† three methods namely Turkstra's rule, Ferry±Borges
where Xm is the structural modeling factor including process and Wen's load coincidence method are available
modeling uncertainty, and Q1 is the annual maximum load to handle load combination processes. The ®rst two methods
intensity. Although the Q1 are independent from year to are simpler to use when compared to the last one, and suita-
year, Xm does not vary with time. The maximum environ- ble for implementation in a FORM analysis, therefore this
mental load effect, Yk in k years can be expressed as study is focused on Turkstra's rule and the Ferry±Borges
process that are discussed next.
Yk ˆ Xm Qk …6†
p
The probability distribution of Qk is FQk …Q † ˆ 3.2.1. Turkstra's rule
‰FQ1 …Qp †Šk which is used to calculate the distribution in k Turkstra's rule has frequently been used during code
years during the calculation of the reliability index b . The developments for formulating load combination problems.
annual probabilities of failure are calculated for each year of Although easy to apply, the accuracy of the rule is dif®cult
the design life as to estimate. It may be conservative or unconservative
Pfak ˆ Pf…k11† 2 Pfk …7† depending on the number of repetitions of each individual
load case and the relative uncertainty of the magnitude of
where Pfak is the probability of failure for year k, Pf(k11) the each load. For that reason more accurate load combination
probability of failure for a period of k 1 1 years, and Pfk is models such as the Ferry±Borges model and Wen's
the probability of failure for a period of k years. coincidence model have been developed. A major limitation
As mentioned above, random variables such as wind in all load combination work has been the incorporation of
speed or vehicular live loading are time-dependent while the modeling uncertainties. Such model random variables
E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212 209

may be applicable to all load occurrences; thus treating a results. Therefore, the time dependent r.v.'s are separated
modeling uncertainty r.v. as if it occurs independently with into time independent and dependent variables, and only the
each load repetition may be excessively conservative. The latter are used in the evaluation of the convolution integral.
separation of the modeling uncertainties from the indepen- The effect of the time independent variables is taken into
dent random variables, which occur randomly with each consideration through numerical integration. This is done by
load repetition, is a major aim in this paper. ®nding b in the FORM algorithm conditional on the values
Although based on probabilistic concepts, Turkstra's rule of the time independent modeling variables and then
reduces to a practical quasi-deterministic method for integrating on these modeling variables to obtain the overall
solving the load combination process. Since the probability b value. The accuracy of the integration was veri®ed by
of the simultaneous occurrence of the loads at their comparison with b values found by Monte Carlo
maximum is negligibly small, Turkstra proposed to approx- Simulation.
imate the combined effect by assuming that one of the loads
is at its maximum value while the others are acting at their
point in time values [9]. By taking the maximum value for 4. Cost and reliability optimization
one extreme event at a time, a set of load effect cases is
created. The maximum load effect from this set of sub- Providing a balance between cost and safety in a design
combinations (as shown in Eq. (9)) will give an approxima- code is one of the aims of code calibration. Reliability-based
tion the max effect of the combination of all the extreme cost optimization is the necessary tool to attain that goal.
events. This represented by the following equations Reliability techniques must be included in the optimization
assuming that the total effect Z(t) at time t is process to evaluate the random behavior of member resis-
tances and applied loads [12]. The main advantage of the
Z…t† ˆ X1 …t† 1 X2 …t† 1 ¼ 1 Xn …t† …8† proposed code optimization procedure is the continuity it
then the maximum value of Z(t) ;t can be approximated by provides with current code practice through the use of the
8 cost of failure implied in the current code while accounting
>
> X1max 1 X2 1 ¼Xn for the uncertainties associated with the design process.
>
>
>
< X1 1 X2max 1 ¼Xn
> The optimal solution for a design is attained when the
max Z ˆ max …9† total cost represented by Eq. (1) is minimized. During the
>
> ..
> . code optimization process, designs representing a family of
>
>
>
: typical con®gurations and spanning a range of parameters,
X1 1 X2 1 ¼Xnmax de®ned herein as the design space, have to be optimized and
where Xi represents the instantaneous values of load variable the average factors of these typical designs introduced as the
i. optimal solution for the code. In this study, instead of ®nd-
ing the individual optimal solutions for each typical design
3.2.2. Ferry±Borges process and averaging them, the total cost factor is summed over this
The Ferry±Borges method models the load processes as design space and then this summation is minimized to
rectangular pulses that change after prescribed, equal and obtain the optimal load factors. The objective function
deterministic intervals [10]. It is assumed that the loads are which is related to the load factors, g 's, the probability of
constant during these elementary intervals. The lifetime of a failure, Pf, the normalized cost factors, K's, and the failure
structure is an integer multiple of the load intervals. Also, cost ratio, g, is presented as shown in Eq. (10)
the interval time of the load with the longer interval is an X
min ‰TCF…g; Pf ; K; g†Ša …10†
integer multiple of the interval time of the loads with shorter a[Des:Sp:
intervals.
The cumulative distribution of the maximum response It is important to realize that the optimization process will
can be calculated using a convolution integral [8]. The not give an optimum solution to every single design, but that
FORM algorithm developed by Rackwitz and Fiessler the whole design space covered by the code is optimized as
[11], which is capable of evaluating convolution integrals, a set.
is used in the present reliability analysis. The algorithm A design space is de®ned herein by the range of applic-
evaluates the integral during the process of converting able values for different types of load effects (W, L, etc.).
r.v.'s to normally distributed equivalent variables. The Nominal values of these load effects are normalized by the
assumption of independence from one event to another nominal dead load, Dn. As a result, the nominal load effects
makes the Ferry±Borges method easy to apply. For these are represented via dimensionless factors of the nominal
reasons, the Ferry±Borges method has been selected for use dead load. It is only required to vary these factors over a
in this study in implementation of the optimization process reasonable range to cover the family of typical designs. For
in Section 5. example, for typical short to medium span bridges the ratio
As mentioned earlier, the correlation between the events of the live to dead load, Ln/Dn, varies between 0.25 and 2.0.
due to modeling uncertainties may produce conservative The proposed code calibration procedure is illustrated in
210 E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212

Table 1 Table 3
Statistical data Dead and wind load combined

r.v. Time dependent a Time independent KW/KG g ˆ 58 g ˆ 116

Distr. type Bias COV Distr. type Bias COV b gD gW b gD gW

R ± ± ± Lognormal 1.12 0.10 0.5 3.46 1.28 2.20 3.65 1.30 2.34
D ± ± ± Normal 1.03 0.08 1.0 3.41 1.32 2.02 3.60 1.34 2.16
W Extreme I 0.87 0.24 Normal 1.00 0.23 2.0 3.44 1.39 1.87 3.63 1.41 2.01
L Lognormal 1.00 0.13 Normal 1.00 0.12 3.0 3.57 1.49 1.76 3.65 1.50 1.91
a
75 year data.

calculate a normalized cost factor for gravitational load, KG,


Section 5 for the development of load factors for a bridge which can be calculated from Eq. (3). By looking at the
design code. relationship between the differences in the initial cost
when compared to the differences in the live load factors,
Eq. (3) leads to Eq. (12) which may be used for different
5. Illustrative example values of a . Eq. (12) shows that a change in the live load
factor by Dg L, has a corresponding initial cost change repre-
A highway bridge code is used to illustrate the proposed sented by DCI. Averaged along the range of a , the factor KG
procedure for ®nding optimized load factors for different calculated from Eq. (12) is found to be equal to 0.14.
load combinations [13]. AASHTO's basic load combination  
for dead and live loads represented by the equation: Rn . g 1 gL0 …1 1 …DgL =gL0 †a DCI
KG D0 21 ˆ …12†
gD Dn 1 gL Ln ; is used as the reference base case. The load gD0 1 gL0 a C0
factors in the AASHTO code are 1.25 and 1.75 for dead and
live loads, respectively. This reference load combination It is assumed that the total cost is optimized by the present
has been calibrated to satisfy a target safety index, bT ˆ code for the base load combination (dead plus live). Thus,
3:5 [2]. The load and resistance statistical data were differentiating TCF in Eq. (11) with respect to the load
assembled during the AASHTO LRFD calibration for a factors about the current target safety index (3.5) should
design life of 75 years [2]. The statistical data used in this be equal to zero due to the assumption that the current
example are given in Table 1. Wind and live loads are load factors and the target safety level are optimal. There-
separated into time dependent and independent random fore, by setting the differentiation of TCF equal to zero, the
variables as provided in Table 1 for illustration. For the optimum failure cost ratio, g is calculated as
combination of dead and live loads, the total cost factor,      
P 2 Pfi 21 P 2 Pfi 21
TCF, can be expressed as
  2gD eij 2gL eij
X Pfi gˆ2 KG ˆ 2a KG
gD 1 agL gD0 1 agL0 gD0 1 agL0
TCF ˆ 1 1 KG 21 1g …11†
gD0 1 agL0 eij …13†
where KG is the normalized cost factor for gravity load (dead The failure cost ratio, g values deduced using Eq. (13)
plus live), and a is the nominal live load to nominal dead along with the code speci®ed gD0 ˆ 1:25 and gL0 ˆ 1:75
load ratio …a ˆ Ln =Dn † taken herein in the range of 0.5±2.0. are averaged for all the a values and found to be equal to 58.
The probability of failure Pf is calculated from the limit state Other load combination cases speci®ed in the AASHTO
function G ˆ R 2 D 2 L: Eq. (11) is used to deduce the bridge speci®cation; namely dead plus wind represented by
implied failure cost ratio, g value. the check Rn . gD Dn 1 gW Wn and dead plus live plus wind
In a study conducted on bridge costs and loadings, Moses represented by the check Rn . gD Dn 1 gW Wn 1 gL Ln are
[14] showed that a 25% change in g L produces a change of next studied using the deduced value of g. In addition to
2% in the initial cost of bridges. This information is used to these two load combinations speci®ed by the code, Rn .
gD Dn and Rn . gW Wn cases are also studied to show the
Table 2
Wind load alone
effectiveness of the proposed method.
Optimized load factors are calculated for different ratios
KW/KG g ˆ 58 g ˆ 116 of normalized initial cost factors of wind and gravitational
load. The results can then be used by code bodies using
b gW b gW
examples of bridge cost data to select optimized load factors
0.5 3.20 2.23 3.39 2.38 for the code. This process is also repeated with the implied
1.0 3.00 2.08 3.20 2.23 failure cost ratio, g increased by two, namely g ˆ 116 to
2.0 2.79 1.93 3.00 2.08
illustrate the effect of the failure cost factor on the optimized
3.0 2.66 1.85 2.88 1.99
load factors. The analyses were run using an algorithm that
E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212 211

Table 4 cost of failure is used. This `implied' cost of failure is


Dead, wind and live load combined deduced from experience with the existing currently used
KW/KG g ˆ 58 g ˆ 116 code. Instead of a pre-determined target reliability index,
the cost of failure is adopted as a target for optimizing
b gD gW gL b gD gW gL load factors for different load combination cases. The
0.5 3.95 1.25 1.20 1.00 4.18 1.25 1.30 1.10
input to the optimization includes the marginal initial load
1.0 3.93 1.25 1.10 1.05 4.10 1.25 1.20 1.10 slope for each pertinent load type. The output includes the
2.0 3.84 1.25 1.00 1.05 4.08 1.25 1.10 1.15 optimized load factors and the corresponding safety indices.
3.0 3.87 1.25 0.95 1.10 4.04 1.25 1.05 1.15 The results show, for example, in Table 2 that at the
optimum, b decreases, along with the increasing wind
load initial cost factor. This observation is consistent with
optimizes the sum of the total cost factors along the design experience in codes such as those for buildings and offshore
space. The case for dead load alone gives an optimum safety structures where computed safety indices for gravity loads
index of 3.70 and a dead load factor of 1.46. These change to are higher than for environmental loads such as wind. Such
3.88 and 1.50, respectively, when g is increased by a factor results are expected as the natural outcome of the trade-off
of 2 …from g ˆ 58 to g ˆ 116†: between cost and safety: as the marginal cost required to
The results for the cases of: (a) wind load alone; (b) dead upgrade the design of a structure increases the optimum
and wind load; and (c) dead, wind and live loads are safety level decreases.
tabulated in Tables 2±4, respectively. These tables show The proposed code calibration approach using cost
the optimized load factors along with the corresponding optimization successfully ®nds the optimum levels of safety
safety indices obtained for the optimized load factors. In index and corresponding optimum load factors. These
Table 3, the safety indices are the average values over the methods require accurate models for cost and reliability.
Wn/Dn range of 0.1±2. In Table 4, the safety indices are the In the reliability formulation, time dependent and time
average values over the Wn/Dn range of 0.1±2 and Ln/Dn independent variables including modeling uncertainties
range of 0.25±2.0. The dead load factor, g D for the combi- are separated so that accurate year to year failure probabil-
nation of dead, live and wind loads, is kept constant at the ities may be calculated. Future failure probabilities and the
current code speci®ed value of 1.25 to keep the optimization costs of such failures are brought to present worth costs
problem two dimensional in function of g L and g W. using discounted interest value. In order for the procedure
Tables 2±4 clearly indicate that increasing the failure cost to be applied to make ®nal code recommendations on load
factor, g, which implies that the consequences of failure factors, better initial cost data still need to be gathered.
increase, affects both the safety index, b , and the load The calibration procedure studied in this paper is formu-
factors requiring higher values for both in order to produce lated for the individual components and structural failure is
an optimal code. This ®nding is consistent with the goals of de®ned as the failure of one component. Further work still
the proposed optimization procedure because it demon- needs to be done to extend the proposed minimization
strates that as the cost of failure increases there is need to process, to include system safety indices as well as compo-
develop codes that produce higher reliability levels. nent safety indices. When system safety is investigated,
The KW/KG ratio also plays an important role in de®ning other factors such as redundancy and collapse mechanisms
the optimal safety level and load factors. Increasing values must be considered. The effect of system safety is being
of KW/KG mean that the load factor for wind becomes more investigated as part of an ongoing research study.
costly to increase. Therefore, instead of increasing the wind
load factor, which also increases the initial cost proportional
to KW, it may be more cost ef®cient to increase the gravita-
References
tional load factors (dead and live loads). Such information
would lead to selecting optimum load factors for all load [1] Ellingwod B, Galambos TV, MacGregor JG, Cornell CA. Probability
combinations provided that the relevant cost factors are based load criteria for american national standard A58, NBS Special
available from existing practice. Publication No. 577. Washington: National Bureau of Standards, US
Department of Commerce, 1980.
[2] Nowak AS. Calibration of LRFD bridge design code, Final Report,
NCHRP 12-33, Transportation Research Board, Washington, DC,
6. Conclusions 1993.
[3] AISC, Manual of steel construction Ð load and resistance factor
It is important that structural design codes be calibrated to design, American Institute of Steel Construction, Chicago, 1996.
produce a balance between safety and economy. A cost [4] API RP 2A-LRFD. Recommended practice for planning, designing
optimization procedure including the initial cost of design and constructing ®xed offshore platforms Ð load and resistance
factor design, American Petroleum Institute, 1993.
and future failure costs based on probability of failure is [5] Ellingwood B, Kanda J. Formulation of load factors based on opti-
formulated in this paper and applied for the AASHTO mum reliability. J Struct Safety 1991;9:197±210.
bridge speci®cations. A single consistent value for the [6] Ditlevsen O. Structural reliability codes for probabilistic design Ð a
212 E. Aktas et al. / Reliability Engineering and System Safety 73 (2001) 205±212

debate paper based on elementary reliability and decision analysis [12] Frangopol DM. Reliability-based optimum structural design. In:
concepts. J Struct Safety 1997;19(3):253±70. Sundarajan CJ, editor. Probabilistic structural mechanics handbook,
[7] Thoft-Christensen P, Baker MJ. Structural reliability and its applica- theory and applications. London: Chapman & Hall, 1995. p. 352±87.
tions. New York: Springer, 1982. [13] Aktas E, Moses F, Ghosn M. Calibration of load factors using relia-
[8] Melchers RE. Structural reliability analysis and prediction. 2nd ed. bility based cost optimization. In: Kareem, Haldar, Spencer, Johnson,
New York: Wiley, 1999. editors. Proceedings of the Eighth Specialty Conference on Probabil-
[9] Turkstra CJ. Theory of structural design decisions, Solid Mechanics istic Mechanics and Structural Reliability, CD-ROM Proceedings.
Study No. 2. Waterloo, Ontario, Canada: University of Waterloo, Notre Dame, Indiana: University of Notre Dame, July 2000 (paper
1972. PMC2000-042, 6p.).
[10] Turkstra CJ, Madsen HO. Load Combination in codi®ed structural [14] Moses F. Effects on bridges of alternative truck con®gurations and
design. J Struct Div ASCE 1980;106(ST12):2527±43. weights, Final Report, TRB, National Research Council, Washington,
[11] Rackwitz R, Fiessler B. Structural reliability under combined random DC, 1989.
load sequences. Comput Struct 1978;9:489±94.

You might also like