You are on page 1of 9

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283936919

Comparison of load combination models for probabilistic calibrations

Chapter · July 2011


DOI: 10.1201/b11332-147

CITATIONS READS

8 389

2 authors, including:

Miroslav Sýkora
Czech Technical University in Prague
146 PUBLICATIONS   757 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Standardization activities related to the COST Action TU1402: Quantifying the Value of Structural Health Monitoring View project

All content following this page was uploaded by Miroslav Sýkora on 28 January 2016.

The user has requested enhancement of the downloaded file.


Comparison of load combination models for probabilistic calibrations
M. Sykora & M. Holicky
Czech Technical University in Prague, Klokner Institute, Prague, Czech Republic

ABSTRACT: Civil engineering structures are, as a rule, exposed to time-variant actions. Selection of an ap-
propriate model for the load combination is one of the key issues of probabilistic calibrations. The submitted
paper attempts to compare Turkstra’s rule with analyses based on Ferry Borges-Castanheta (FBC) models and
rectangular wave renewal processes with intermittencies. Load combinations of two independent variable ac-
tions are numerically analysed for different load ratios and various types of loads. It appears that Turkstra’s
rule and the FBC processes provide sufficiently accurate results for a wide range of reliability problems while
the analysis based on an upper bound on the failure probability for the intermittent renewal processes yields in
some cases rather conservative estimates. For probabilistic calibrations application of Turkstra’s rule seems to
be particularly convenient as available reliability software products can be directly used and the partial factors
of basic variables can be readily derived.

1 INTRODUCTION 2 BASIC ASSUMPTIONS

Civil engineering structures are, as a rule, exposed to Basic variables X(t) used in the analysis include:
combinations of time-variant loads. Selection of an  Time-invariant variables R  X(t) - resistance
appropriate model for the load combination is one of and geometry variables, permanent actions and
the key issues of probabilistic calibrations of reli- model uncertainties,
ability elements in codes.  Time-variant actions Q1(t) and Q2(t)  X(t) - cli-
Models for load combinations are often based on matic actions, imposed loads etc., that can be de-
transformation of the time-variant case into a time- scribed by stationary, ergodic and regular proc-
invariant one using the rule proposed by Turkstra esses.
(1970) that is considered to be particularly useful for The time-variant actions are assumed to be mutually
probabilistic calibrations. Another widely used independent. Neither cumulative phenomenon such
model is based on rectangular wave renewal proc- as fatigue or creep nor deterioration effects is taken
esses with fixed durations of load pulses proposed into account. Stationary and ergodic conditions are
by Ferry Borges & Castanheta (1971). thus assumed. Note that load effects are referred to
Alternatively random load fluctuations in time as loads or actions for convenience of notation.
can often be described by rectangular wave renewal The limit state function is denoted as g[X(t)] =
processes with random durations between renewals g[R,Q1(t),Q2(t)]; g[X(t)] < 0 indicating failure. A
and random durations of load pulses. Such processes reference period tref (e.g. working life) is considered
seem to be applicable for a wide range of loads on as a deterministic variable, t  [0,tref]. The instanta-
structures as recognised e.g. by Wen (1990), neous failure probability is:
Ditlevsen & Madsen (1996) and JCSS (2006).
Pf(t) = P{g[X(t) < 0]} (1)
The paper is aimed at comparison of these tech-
niques. Differences are briefly discussed and re- The failure probability over the reference period
marks on applicability of the models in probabilistic is the probability of the limit state being exceeded at
calibration studies are provided. Combination of two least once:
time-variant actions is considered throughout the
paper to clarify general concepts. Reliability of steel Pf(tref) = P{t  [0,tref], g[X(t)] < 0} (2)
frames of different configurations, exposed to snow
and wind actions, is then analysed to compare the
three approaches numerically.
t* 3.2 Partial factors
Q1 Q1,tref q1 Partial factors for the basic variables can be readily
Q1,tref
derived from characteristic values xk, actual sensitiv-
ity factors X obtained from a FORM analysis of the
failure probability given in Equation 3 and a target
Q2
Q2,t*
q2 reliability level t:
Q2,t*
Favourable effect: Xi = xk,i / FXi-1[(-Xi × t)],
Unfavourable effect: Xi = FXi-1[(-Xi × t)] / xk,i (4)
t* t* tref t fQ(q)
where FX-1(·) = inverse cumulative distribution func-
Figure 1. Turkstra’s rule. tion of X. Note that in Equations 4, the leading ac-
tion is described by the maxima Q1,tref while Q2,t* is
Q1 q1 Q1,tren1 used for an accompanying action.
Q1,t
ren1

4 FERRY BORGES-CASTANHETA MODELS


Q2 tren1 q2
Q2,tren2
Q2,t
ren1 Q2,tren1 4.1 Failure probability
Q2,t
t
ren2 In this model the reference period is divided into a
tren2 tref ft (q)
ren number of intervals having fixed durations tren. Load
Figure 2. Intermittent FBC models. pulses of a constant magnitude during tren are de-
scribed by identically distributed independent ran-
Ton1 Ton1 dom variables. Such processes are often referred to
as Ferry Borges-Castanheta (FBC) models after J.
Q1
Ferry Borges and M. Castanheta, who suggested
Q1
their application in standardisation, Ferry Borges &
Castanheta (1971). These processes are hereafter re-
Ton2 Ton2 ferred to as “FBC models”.
Tren1 Tren1 Tren1
Q2 For intermittent loads, it can be assumed that the
Q2
load may be present in the interval with the occur-
rence probability pon. The process is fully character-
tref Tren2 Tren2 t ized by the deterministic duration tren, distribution of
Figure 3. Intermittent renewal processes.
the load intensity within a pulse (given the load is
present), and by the probability pon. Intermittent
FBC models are indicated in Figure 2.
3 TURKSTRA’S RULE The maximum of load effect can be estimated by
approximate analytical (FORM/SORM) or simula-
tion methods. The failure probability is then ob-
3.1 Failure probability tained using the following relationship:
The failure probability given in Equation 2 may be Pf(tref) ≈ P[g(R, Emax) < 0] (5)
assessed by a transformation of the time-variant
problem into a time-invariant one. The time- More details on the analysis using the FBC models
invariant maximum effect of simultaneously acting are provided by Melchers (2001).
load processes over the reference period may be es-
timated using the rule proposed by Turkstra (1970). 4.2 Partial factors
For the combination of two independent time-
variant loads, the maximum effect of the combina- The partial factors can be assessed using Equa-
tion Emax occurs when one of the processes, the lead- tions 4, however, the partial factor of Emax is ob-
ing action, takes its maximum Q1,tref during the ref- tained only. To simplify the analysis it is assumed
erence period and the other, an accompanying that the ratio Q1 / Q2 is similar to that obtained by
action, is at an “arbitrary point-in-time value” Q2,t*: an additional reliability analysis based on the limit
state function g(R,Q1,tren1,Q2,tren1) < 0 in which
Emax ≈ E(Q1,tref, Q2,t*), maxima of the actions during tren1 are applied (as-
Pf(tref) ≈ P[g(R, Q1,tref, Q2,t*) < 0] (3) suming tren1 ≥ tren2, see Figure 2). The partial factors
where E = load combination effect; and t = period * can then be derived from the partial factor Emax as:
selected taking into account duration of the maxi- EmaxE(q1k, q2k) = E(Q1 q1k, Q2 q2k) (6)
mum of the leading action as shown in Figure 1.
Apparently, disadvantage of this approach is that re- 5.3 Failure probability
sults of the additional analysis need to be available.
Failure probabilities for all the load cases are then
evaluated, e.g. Pf,1i = P[g(R,Qi) < 0]. For the combi-
nation of two processes, the upper bound proposed
5 RECTANGULAR WAVE RENEWAL by Sykora (2005) reads:
PROCESSES WITH INTERMITTENCIES
Pf 0, t ref   p0 Pf , 0  p11 Pf1,1  p12 Pf2,1  p12, 2 Pf1,,22 
5.1 Description of the process 1  pon, 2  11  pon , 2 11,2   (9)
t ref  
1  pon,1  1  pon ,1 1, 2 
In this model renewal process of load pulse initia- 2 2
tions follows a Poisson process obtained as the se-
quence of mutually independent exponential dura- where + = outcrossing rate. For instance the out-
tions between renewals Tren with the rate . crossing rate  11,2 due to load intensity change of
Durations of “on”-states Ton, independent of the du- Q1(t) given “on”-state of Q2(t), is defined as follows:
rations Tren, are also mutually independent exponen-
tial variables with the rate . Load pulses are as-  11,2 = 1P[g(R,Q1-,Q2) ≥ 0  g(R,Q1+,Q2) < 0] (10)
sumed to be non-overlapping.
In accordance with Shinozuka (1981), probability where Q1- = (random) value of the process Q1(t) be-
of “on”-state is obtained for stationary conditions: fore the jump when the process may be “on” or
“off”; and Q1+ = random value after the jump when
pon =  / ( + ) =  / (1 + ) (7) the process is “on”. Evaluation of the outcrossing
where  = interarrival duration intensity. rate {k}+ given the set {k} of loads is “on” is de-
Given an “on”-state, load intensities Q are identi- scribed by Rackwitz (1998) and Sykora (2007).
cally distributed independent variables. Two inter- The upper bound in Equation 9 is similar, but not
mittent renewal processes are shown in Figure 3. identical with that provided by Rackwitz (1998).
Such models are in the following denoted as “inter- The bounds were compared by Sykora (2009).
mittent renewal processes”.
It is noted that compared to the processes ac- 5.4 Partial factors
cepted here, Rackwitz (1998) considered more gen-
eral processes with multiple load pulses during an It is assumed that a dominant load case when one of
“on”-state using an additional parameter . How- the processes, or both the processes are “on” - can be
ever, in practical applications it is often sufficient to identified from Equation 9. For instance when the
assume that each “on”-state is associated with one combination of both the processes is dominant,
rectangular wave and thus  =  as considered e.g. Equation 9 can be rewritten as follows:
by Holicky & Rackwitz (1998).  
Pf 0, tref   p12, 2 Pf1,,22  tref pon , 2 11,2  pon ,1 12, 2 
(11)
Pf1,,22 1  tref 1 pon , 2  tref  2 pon ,1   N 21, 2 Pf1,,22
5.2 Load combination
The load coincidence method was proposed by where N = expected number of events when Q1(t)
Shinozuka (1981) to describe combinations of in- and Q2(t) are “on”.
termittent renewal processes. For the combination of Following a procedure proposed by Rackwitz
two processes, the probabilities pni,j of the exhaus- (1997), the partial factors can be derived from Equa-
tive, mutually exclusive set of load cases that: tions 4 using the sensitivity factors obtained for the
 Both the processes are “off” (denoted as p0), analysis of the dominant load case and a reduced re-
liability index given approximately by:
 The only process Qi(t) is “on” while the other is
“off” ( p1i ), red = --1{(-t) / N} (12)
 Both the processes are “on” ( p12, 2 )
are determined for independent processes as follows: 6 COMPARISON OF DIFFERENT
1 i APPROACHES
p0  , p1i  ,
1  1 1   2  1  1 1   2 
(8) 6.1 Applicability of reliability methods
1  2
p 1, 2

2
1  1 1   2  In probabilistic calibration studies efficient reliabil-
ity methods should be preferably applied since nu-
merous analyses are commonly required to assess in-
fluence of various input parameters on the predicted
reliability level. For the three different approaches,
the following remarks may be made with respect to terministic intervals may be rather unrealistic for
the applicability of different reliability methods: spike-like processes such as wind storms, earth-
 Turkstra’s rule: reliability can be assessed by any quakes, loads due to crowds of people etc.
of well-established methods for the time-invariant  Intermittent renewal process: this model is gener-
analysis such as the approximate analytical or ally applicable for many types of intermittent and
simulation methods implemented in commercial non-intermittent actions on structures. However,
software products, the upper bound given in Equation 9 has been
 FBC models: the exact analytical or numerical in- found to yield somewhat conservative estimates
tegration methods may be used to calculate the when time-invariant variables are governing
load combination maximum. However, computa- structural reliability and load combination aspects
tions become rather cumbersome for a larger become less important.
number of time-variant actions, say n > 3. Appli-
cation of the FORM/SORM methods is usually
based on the well-known algorithm proposed by 6.3 Estimation of partial factors
Rackwitz & Fiessler (1978), available in few Partial factors, derived for a given target reliability
commercial software products. Direct simulations level, are commonly the main result of probabilistic
of processes mostly make computations demand- calibration studies. Concerning this aspect, it follows
ing. from the above information that:
 Intermittent renewal processes: failure probabili-  Turkstra’s rule: estimation of partial factors of all
ties in the upper bound in Equation 9 can be the basic variables is straightforward.
evaluated by any of reliability methods for time-  FBC models: estimation of partial factors of time-
invariant analysis; the outcrossing rates can be es- invariant variables is straightforward. Assessment
timated by the system reliability technique pro- of the partial factors of time-variant actions may
posed by Hagen & Tvedt (1991) using the effi- be based on results of an additional analysis as
cient FORM/SORM methods. In an described in Section 4.2 which, however, may
approximation the outcrossing rates can be re- complicate calibration studies.
placed by failure probabilities (see Equation 11)  Intermittent renewal processes: estimation of par-
without loosing accuracy for the cases when time- tial factors of all the basic variables is straight-
variant loads are governing reliability. The upper forward given that a dominant load case can be
bound is, however, not available in commercial identified (see Section 5.2). Otherwise, the proce-
software products and its applicability for calibra- dure proposed by Rackwitz (1997) can be applied
tion studies seems to be disputable. Direct simu- and the analysis becomes complicated.
lation of processes is mostly time-demanding.
6.4 Remarks on non-stationary cases
6.2 Accuracy The present analysis is based on the assumption of
Several studies investigating accuracy of the three stationary properties of basic variables. The com-
approaches are available, see for instance Wen parison of the three approaches reveals that Turk-
(1990), Holicky & Vrouwenvelder (1997), Sykora stra’s rule or FBC models are more convenient for
(2007), Sykora (2008) and Sykora (2009). These calibration studies than the intermittent renewal
studies reveal that: processes.
 Turkstra’s rule: Turkstra (1970) indicated that In the case of non-stationary conditions (degrad-
“since the maximum effect of the load combina- ing structures, fatigue, creep effects, non-stationary
tion may occur at a time when all loads are less actions), analysis based on the intermittent renewal
than their life-cycle maxima, the described ap- processes may become superior to the other ap-
proximation will underestimate the failure prob- proaches as reliability can be efficiently analysed us-
ability”. However, accuracy of the rule was found ing the Laplace transform, Rackwitz (1997). Turk-
to be good particularly for “low” failure prob- stra’s rule or FBC models can be used to estimate an
abilities (approximately lower than 10-3), consid- upper bound (combining the maximum effect of load
ering various combinations of intermittent and combination with the minimum resistance), which
non-intermittent loads. Moreover, even for “high” may be overly conservative.
failure probabilities (serviceability or durability
limit states), the rule seems to yield sufficiently
accurate estimates for calibration studies. The ru- 7 NUMERICAL EXAMPLE
le may lead to misleading results when a leading
action is not identified correctly.
 FBC models: given that time-variant loads can be 7.1 Basis of analysis
well described by FBC models, the exact solution The approaches to load combinations are applied in
is found. However, the model based on the de- reliability analysis of low-rise industrial buildings.
For convenience of notation, the ratio of the total
characteristic variable action to the total characteris-

20 m

20 m
tic action is introduced:

10 m
 = (mSsk + mWwk) / (mGgk + mSsk + mWwk) (14)
The load ratio may vary within the interval from
25 m 22 m 15 m nearly 0 (underground structures, foundations) up to
Figure 4. Configuration of the representative portal frames. nearly 1 (local effects on bridges, crane girders). For
a given ratio, the characteristic value of the perma-
Table 1. Probabilistic models of basic variables.
__________________________________________________ nent action can be determined.
Variable Dist. () xk X /xk VX pon,X
__________________________________________________ It is assumed that the frames are designed in an
Resistance R LN 1.0 Eq. 15 1.18 0.08 - economical way so that the following condition,
Perm. load G N 1.35 Eq. 14 1 0.10 - based on the fundamental relationship (6.10) pro-
Snow on roof S* GU 1.5(0.5) vided in EN 1990 (2002), is fulfilled:
Münster (altitude 60 m) 0.31 0.26 1.17 0.23
Bremen (12 m) 0.29 0.26 1.14 0.26 rk / M0 = mGG gk +
Aachen (266 m) 0.38 0.27 1.08 0.27 Q max[mS sk + mW 0,W wk; mS 0,S sk + mW wk] (15)
Stuttgart (245 m) 0.38 0.25 1.11 0.32
Berlin (75 m) 0.44 0.25 1.13 0.31 Values of the factors  and  are given in Table 1.
Braunlage (560 m) 2.87 0.24 1.08 0.54
Wind action W** GU 1.5(0.6)
Münster 0.37 0.17 0.67 1 7.2 Turkstra’s rule
Bremen 0.43 0.19 0.61 1
Aachen 0.45 0.19 0.60 1 Two alternatives when the snow load or wind action
Stuttgart 0.33 0.20 0.59 1 is the leading action are considered. Firstly, the
Berlin 0.38 0.20 0.56 1 snow load maximum S50 is combined with the arbi-
Braunlage 0.34 0.21 0.54 1 trary point-in-time value of the wind action Wt* ob-
Resist. unc. KR LN - 1 1.15 0.05 -
Load ef. unc. KE LN - 1 1.0 0.10 -
tained as follows:
Unit mom. of G mG det - frame A: 43.4, B: 27.7, C: 11.3  Considering the continental climate, the 50-year
Unit mom. of S mS det - frame A: 34.7, B: 22.2, C: 9.0 maximum of the snow load may be assumed to be
Unit mom. of W mW det - frame A: 7.5, B: 38.2, C: 78.0
__________________________________________________ “on” one week, see Holicky & Sykora (2011),
LN = lognormal, N = normal, GU = Gumbel distribution of  Weekly maxima of the wind thus should be taken
maximum values, det = deterministic value. into account. As these are not available, monthly
*Monthly maxima given snow is present, **monthly maxima. maxima are conservatively applied.
Secondly, the wind maximum W50 is combined with
the arbitrary point-in-time value of the snow load as-
The example is accepted from previous studies by sessed as follows:
Schleich et al. (2002) and Sadovsky & Pales (2008)  Duration of a 50-year wind storm is estimated to
where more details on input data and analysis are be one day,
provided. Three representative steel portal frames  Daily maxima of the snow should thus be taken
exposed to snow and wind loads are shown in Fig- into account. As these are not available, monthly
ure 4. Models for the climatic loads and material maxima of the snow load (including months with-
properties are based on meteorological data for six out snow) are applied, using the mixed distribu-
locations in Germany and data collected from pro- tion with a cumulative distribution function:
ducers of rolled profiles, respectively. The snow
load is assumed to be present with the probability pon FSmix(x) = (1 - pon)10 ≤ x + ponFS(x) (16)
while the wind action is always present. Reference where 10 ≤ x is 0 if the condition 0 ≤ x is fulfilled and
period is 50 years. 1 otherwise. More unfavourable of the two alterna-
The limit state function reads: tives is then considered in the reliability analysis.
g[X(t)] = KRR - KE[mGG + mSS(t) + mWW(t)] (13)
Notation and probabilistic models of the basic vari- 7.3 FBC models
ables are given in Table 1. Fixed duration of load pulses is assumed to be one
Note that the uncertainty factors KR and KE are month. Within the period of a month, snow load is
taken into account to cover imprecision of models described by the mixed distribution described by
for resistance and load effects. Compared to the pre- Equation 16. Cumulative distribution function of
vious studies, probabilistic characteristics of snow Emax is obtained as follows:
and wind loads are modified to include uncertainties
related to the shape and exposure coefficients (snow) Emonth = Smix + W; FEmax(x) = FEmonth(x)50 × 12 (17)
and gust and roughness factors (wind), see also
Holicky & Sykora (2011).
7.4 Intermittent renewal processes 6
Considering the data in Table 1, the expected num- 
ber of snow events per year is S = 12pon [1/year]
and the rate S is 12(1 - pon) [1/year]; for wind pa- 5
rameters W ≈ 12 [1/year] and W → 0 [1/year] are
used. Turkstra - wind
4
3.8
7.5 Reliability analysis
Reliability index  obtained by the probabilistic
analysis for the six locations is indicated in Figures 3 renewal processes
5a to 5c for the frames A to C. Considering light- FBC
Turkstra - snow
weight roofing and low self-weight of a structure,
the load ratio  = 0.8 is assumed. 2
It follows from Figures 5a to 5c that all the ap-

Münster

Bremen

Aachen

Braunlage
Stuttgart

Berlin
proaches lead to similar reliability levels in most
cases. Two basic cases are distinguished:
1 Snow and wind loads are of comparable effects Figure 5a. Reliability index for frame A ( = 0.8).
(frame B except for Braunlage): Turkstra’s rule
rather overestimates reliability level (particularly 6
for Bremen) as maximum of the load combination
may not occur when one of the loads takes its 
Turkstra - wind
maximum with respect to a reference period. FBC
5
However, the differences between Turkstra’s rule
and FBC models are small (up to 0.3 in terms of Turkstra - snow
). The upper bound for intermittent renewal
processes is in some cases rather conservative (re- 4
liability index lower by about 0.1-0.4 compared 3.8
to FBC models). renewal processes
2 Snow or wind load is dominant (snow: frame A 3
for all locations, frames B and C for Braunlage;
wind: frame C except for Braunlage): differences
among the approaches are negligible (up to 0.2)
since load combination aspects are of a minor 2
Münster

Bremen

Aachen

Braunlage
Stuttgart

Berlin
importance.
Frame B and Berlin are hereafter considered as a
representative frame configuration and location, re-
spectively. Reliability index as a function of the load Figure 5b. Reliability index for frame B ( = 0.8).
ratio is shown in Figure 6. It follows that the differ-
ence between Turkstra’s rule and FBC models is in- 6
Turkstra - snow
significant, for the whole range of the load ratio up 
to 0.1. The upper bound for intermittent renewal
processes seems to be conservative; for  > 0.4 it FBC Turkstra - wind
5
yields reliability indices lower by about 0.2 com-
pared to FBC models, for  < 0.4 the upper bound
becomes overly conservative as the difference in- renewal processes
4
creases up to 0.8. 3.8
Considering the target reliability level 3.8 as indi-
cated in EN 1990 (2002) for structures with medium
failure consequences and a 50-year reference period, 3
the partial factors are derived for all the three ap-
proaches. Partial factors for all the basic variables
are initially determined and the partial factors for re-
2
sistance of a cross-section (M0) and effects of the
Münster

Bremen

Aachen

Braunlage
Stuttgart

Berlin

permanent load (G), snow load (S) and accompany-


ing wind action (W × W) are obtained as follows:
M0 = KR × R; G = KE × g; Figure 5c. Reliability index for frame C ( = 0.8).
S = KE × s; W × W = KE × w (18)
6 pears that for the time-invariant variables (Figure
7a), Turkstra’s rule and FBC models lead to nearly

Turkstra - wind
same partial factors while somewhat higher values
5
are obtained for intermittent renewal processes (the
Turkstra - snow
difference up to 0.2 for G).
Different partial factors are, however, obtained
for time-variant loads. For  > 0.4 (time-variant ac-
4 tions are significant), the differences among the ap-
proaches are small (up to 0.3). For  < 0.4 consid-
FBC erably different partial factors are obtained.
3 However, influence on design is insignificant in
renewal processes such cases - for  > 0.2 estimates of the characteris-
tic resistance using the derived partial factors in
Equation 15 differ up to 5 % for all the approaches,
2 for  < 0.2 estimates based on the intermittent re-
0 0.2 0.4 0.6 0.8 1
 newal processes may differ up to 10 %.
Figure 6. Reliability index as a function of the load ratio  Two general trends are observed in Figures 7a
(frame B, Berlin). and 7b - Turkstra’s rule rather overestimates signifi-
cance of the leading action while intermittent re-
3.5 newal processes overestimates the influence of time-
M0, G invariant variables for low .
3

2.5
8 CONCLUSIONS
2 renewal processes Selection of an appropriate model for the load com-
bination may be one of the key issues of probabilis-
1.5 G tic calibrations. Comparison of three approaches to
load combination indicates that:
1
1 Turkstra’s rule:
M0  Reliability can be assessed by any method for the
0.5 Turkstra – snow ≈ FBC
time-invariant analysis.
 When applied strictly as proposed, the rule may
0
0 0.2 0.40.8  0.6
1 lead to overestimation of an actual reliability, but
Figure 7a. Variation of the partial factors M0 and G with the
the error seems to be insignificant; misleading re-
load ratio  (frame B, Berlin). sults may be obtained when a leading action is
not identified correctly.
3.5  Estimation of partial factors is straightforward.
S, W × W 2 Ferry Borges-Castanheta (FBC) models:
3 S  Application of FORM/SORM methods is usually
based on the Rackwitz-Fiessler algorithm that
2.5 may not be available in software products.
 The exact solution is found if time-variant loads
2 are well described by FBC models.
 Estimation of partial factors requires an addi-
renewal processes
1.5 FBC tional analysis and may be complicated.
Turkstra – snow 3 Intermittent renewal processes:
1  An upper bound on failure probability can be
W × W evaluated by any system reliability method; how-
0.5 ever, it is not available in software products.
 Estimation of partial factors is straightforward if
0 a dominant load case is identified.
0 0.2 0.4
0.8  1 0.6 Numerical example reveals that:
Figure 7b. Variation of the partial factors S and W × W with  Differences between Turkstra’s rule and FBC
the load ratio  (frame B, Berlin). models are insignificant,
 When time-invariant variables are dominant, the
upper bound for intermittent renewal processes
Figures 7a and 7b show variation of the partial fac- becomes conservative.
tors with the load ratio  (frame B, Berlin). It ap-
 Turkstra’s rule rather overestimates significance REFERENCES
of the leading action while intermittent renewal
processes overestimate the influence of time- Ditlevsen, O. & Madsen, H.O. 1996. Structural Reliability
invariant variables when time-invariant variables Methods. Chichester (England): John Wiley & Sons.
EN 1990: 2002. Eurocode - Basis of structural design.
are dominant. Ferry Borges, J. & Castanheta, M. 1971. Structural Safety,
For common calibration studies, Turkstra’s rule is Course 101 (2nd ed.). Lisbon: Laboratorio National de En-
recommended. Results can be verified by the analy- genharia Civil.
sis based on FBC models. Application of intermit- Hagen, I. & Tvedt, L. 1991. Vector Process Out-Crossing as
tent renewal processes is recommended for non- Parallel System Sensitivity Measure. Journal of Engineer-
stationary conditions. ing Mechanics 117(10): 2201-2220.
Holicky, M. & Rackwitz, R. 1998. Time variant reliability of a
column under variable loads with intermittencies. In S.
Lydersen, G.K. Hansen & H.A. Sandtorv (eds.), Proc. ES-
ACKNOWLEDGEMENTS REL'98. Rotterdam: Balkema.
Holicky, M. & Sykora, M. 2011. Conventional probabilistic
The study has been conducted at the Klokner Insti- models for calibration of codes. Proc. ICASP11, ETH Zu-
rich, 1-4 August 2011. Leiden: CRC Press/Balkema.
tute CTU within the project GACR 103/09/0693. Holicky, M. & Vrouwenvelder, A.C.W.M. 1997. Time variant
Results of the project LG11043 have been utilised. reliability of a reinforced concrete column. In C. Guedes
Soares (ed.), Proc. ESREL ’97, Lisbon, Pergamon.
JCSS 2006. JCSS Probabilistic Model Code. Zurich: Joint
SYMBOLS Committee on Structural Safety.
Melchers, R.E. 2001. Structural Reliability Analysis and Pre-
diction. Chichester, England: John Wiley & Sons Ltd.
f probability density function Rackwitz, R. 1997. A Concept for Deriving Partial Safety Fac-
g limit state function tors for Time-variant Reliability. In C. Guedes Soares (ed.),
n number of time-variant actions (processes) Proc. ESREL’ 97, Lisbon, 17-20 June 1997. Pergamon.
pmi , j ,..., k
probability that m processes Qi(t), Qj(t),…, Rackwitz, R. 1998. Computational techniques in stationary and
non-stationary load combination - A re-view and some ex-
Qk(t) out of n processes are “on” tensions. Journal of Structural Engineering 25(1): 1-20.
pon probability of “on”-state Rackwitz, R. & Fiessler, B. 1978. Structural reliability under
t time combined random load sequences. Computers & Structures
9(5): 489-494.
t* reference period for an accompanying action Sadovsky, Z. & Pales, D. 2008. Probabilistic optimization of
tref reference period partial safety factors for the design of industrial buildings.
tren basic (deterministic) interval of FBC model International Journal of Reliability, Quality and Safety En-
xk characteristic value of basic variable gineering 15(5(2008)): 411-424.
E(max) (maximum) load combination effect Schleich, J.B., Sedlacek, G. & Kraus, O. 2002. Realistic Safety
F cumulative distribution function Approach for Steel Structures. Eurosteel 2002, Coimbra.
Shinozuka, M. 1981. Notes on the Combinations of Random
G permanent action Loads (I), BLN Report No. NUREG/CR 1979. Brookhaven
KE, KR uncertainty of load and resistance models National Laboratory.
N expected number of events when processes Sykora, M. 2005. Load combination model based on intermit-
are “on” tent rectangular wave renewal processes. In G. Augusti,
Pf failure probability G.I. Schuëller & M. Ciampoli (eds.), Proc. ICOSSAR 2005,
Rome, 19-23 June 2005. Rotterdam: Millpress.
Q(t) time-variant action that can be described by Sykora, M. 2007. On accuracy of different approaches to time-
stationary, ergodic and regular processes variant structural reliability. In T. Aven & J.E. Vinnem
Qi,t maximum of Qi(t) related to a period t (eds.), Proc. ESREL 2007, Stavanger. London: Balkema.
R time-invariant variable; resistance in example Sykora, M. 2008. Accuracy of selected approaches to time-
S snow load variant reliability analysis of serviceability limit states. In
Ton random duration of “on”-states C.A. Graubner, H. Schmidt & D. Proske (eds.), Proc. 6th
Int. Probab. Workshop, Darmstadt. TU Darmstadt.
Tren random duration between renewals Sykora, M. 2009. Time variant analysis of structural reliability
VX coefficient of variation of basic variable - numerical comparison of various approaches. In H. Fu-
W wind action ruta, D.M. Frangopol & M. Shinozuka (eds.), Proc. ICOS-
X basic variable SAR 2009. Leiden: CRC Press/Balkema.
 FORM sensitivity factor Turkstra, C.J. 1970. Theory of Structural Design Decisions,
SM Studies Series No. 2. Ontario, Canada: Solid Mechanics
(t) (target) reliability index Division, University of Waterloo.
 partial factor Wen, Y.K. 1990. Structural load modeling and combination
 reciprocal of the mean duration Tren for performance and safety evaluation. Elsevier.
 reciprocal of the mean duration Ton
X mean of basic variable
+ outcrossing rate
 interarrival duration intensity
 combination factor of variable action

View publication stats

You might also like