Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

IABSE CONFERENCE

GENEVA, 2015

Structural Engineering: Providing Solutions


to Global Challenges

REPORT

International Association for Bridge and Structural Engineering IABSE


iv Structural Engineering: Providing Solutions to Global Challenges

Table of Contents

Theme 1: Climate Change and the Energy Challenge


CE-01 T: Sustainability and the Structural Engineer
Andrew Martin
The DIOGEN and CIOGEN works, a specific approach in environmental assessment of civil engineering structures
Tardivel, Yannick; Feraille, Adélaïde; Montens, Serge; Tessier, Christian. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Sustainable Bridge Design in Germany
Retzepis, Ioannis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Major sustainability drives on a large project, the Queen Elizabeth Park.
Kirk, Martin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Bridges, Sustainability and the Structural Engineer
Martin, Andrew. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
A Sustainable Structural Design method to analyse structural and environmental performances of a building
Tsimplokoukou, Kassiani; Lamperti Tornaghi, Marco; Negro, Paolo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Sustainable Bridge Design Using Precast Concrete Joints with Monolithic Behaviour
Saad, Fathy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

CE-02 T: Ultra-long service-life of structures


Christian Cremona
Designing Bridges for a Long Service Life
Gabler, Markus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Structure Birth Certificate for Service Life Design and Through-Life Management
Bartholomew, Michael . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Durability Features of the Vimy Memorial Bridge
Ajrab, Jack; Anderson, William; Montminy, Sylvain; O’Connell, Ryan; Delph, Marcel . . . . . . . . . . . . . . . . . . . . . . . 66
Corrosion Protection Plan Requirements in APD: Defining Requirements
Elza, Peter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Increased Service Life of Fatigue Effected Structures through High Frequency Mechanical Impact Treatments
Breunig, Stephanie; Kuhlmann, Ulrike. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Verification of Infinite Life Fatigue Performance of a Cost-Effective Steel Orthotropic Deck Design by Full Scale
Laboratory Testing
Mukherjee, Soham; Roy, Sougata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

CE-03 S: Sustainable construction materials


Daia Zwicky
Material quantities and embodied carbon of footbridges
De Wolf, Catherine; Bianquis, Rosalie; Verbeeck, Kenny; Ochsendorf, John . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Wood-lightweight concrete - Development of energy and resource resource-efficient construction materials
Fadai, Alireza; Winter, Wolfgang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Hybrid Timber Construction – combining material properties for energy efficiency and sustainability
Hein, Carsten; Lawrence, Andrew; Campbell, Adrian; Snelson, Timothy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Research of Performance and Durability of long span prestress concrete box girder bridge of gravel concrete
Weng, Fang-Wen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
IABSE Conference, Geneva, 2015 v

Application of Prestressed Concrete Composite Box-girder Bridges with Corrugated Steel Webs in Bridge
Engineering in China
Li, Ming; Song, Jianyong; Li, Wanheng; Zhang, Jinquan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Build with weathering steel : environmental and economical impacts–T7 tramway bridge in Paris area
Michel, Loïc; Gogny, Eric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

CE-04 T: Sustainable buildings


Daia Zwicky
Developing a Basis for Design – Embodied Carbon in Structures
Sarkisian, Mark; Shook, David; Desai, Devki; Wang, Nicole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Carbon Emission Assessment for Super Tall Buildings with Viscoelastic Coupling Dampers
Zhao, Xin; Qin, Lang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Structural Optimization for a Sustainable Architecture
Beghini, Alessandro; Mathias, Neville; Sarkisian, Mark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Innovative construction system for sustainable buildings
Loss, Cristiano; Piazza, Maurizio; Zandonini, Riccardo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Integrated Embodied Carbon Optimal Design of Super Tall Building Structures with Viscous Dampers
Zhao, Xin; Shi, Tao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Indicators for Serviceability for Low-Carbon Building Slab Types
Molina-Moreno, Francisca; Yepes-Piqueras, Víctor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

CE-05 S: Multifunctional energy-efficient building envelopes


Georgios Tsionis
Energy-Efficient Building Envelopes: use of Phase Change Materials in Cement-Based Composites
Lamperti Tornaghi, Marco; Caverzan, Alessio. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Numerical modelling of UHPC and TRC sandwich elements for building envelopes
Miccoli, Lorenzo; Johansson, Gabriel; Zandi, Kamyab; Williams Portal, Natalie; Fontana, Patrick;
Müller, Urs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

CE-06 S: Extreme environmental actions on structures


Yaojun Ge and Yongxin Yang
Wind Resistant Performance and Countermeasures for Cable-stayed Bridges with Open-section Girder
Yang, Yongxin; Zhou, Rui; Ge, Yaojun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Reevaluation of the Design Loads for an Existing Avalanche Protection Gallery - A Probabilistic Approach
Fischer, Katharina; Schubert, Matthias; Schaer, Mark; Margreth, Stefan; Schellenberg, Kristian . . . . . . . . . . . . . . 212
Effect of Statistical Uncertainties in Ground Snow Load on Structural Reliability
Rózsás, Árpád; Sýkora, Miroslav . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Climate change effects on design thermal actions for concrete structures
Larsson, Oskar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Truss Structure Capacity Monitoring Considering Accelerated Steel Corrosion under Climate Change
Feng, Yansheng; Sun, Zhi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

CE-07 S: Environmental hazards on structures


Inamullah Khan, Qiao Li and Deshan Shan
Effects of Tsunami Measures for Bridges
Hasegawa, Akira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

Effect of Statistical Uncertainties in Ground Snow Load on Structural


Reliability
Árpád Rózsás
Budapest University of Technology and Economics, Budapest, Hungary

Miroslav Sýkora
Czech Technical University in Prague, Prague, Czech Republic

Contact: rozsas.arpad@epito.bme.hu

Abstract
This paper studies the effect of commonly neglected statistical uncertainties on structural
reliability. The failure probability of a generic structural member, subjected to snow load is
analysed using frequentist and Bayesian techniques to quantify parameter estimation and model
selection uncertainties in ground snow load. Various variable to dead load ratios are considered to
cover a wide range of real structures. The analysis reveals that statistical uncertainties may have a
substantial effect on reliability. By accounting for parameter estimation uncertainty, the failure
probability can increase by more than an order of magnitude. Bayesian posterior predictive
distribution is recommended to incorporate parameter estimation uncertainty in reliability
studies.
Keywords: structural reliability; statistical uncertainty; Bayesian statistics; posterior predictive;
model averaging; maximum likelihood; snow load.

and provide practical recommendations regarding


1 Introduction their treatment. For this purpose, the reliability
One source of uncertainty in structural reliability is level of a generic structural member, subjected to
statistical uncertainty which stems from the snow load is investigated. The effect of parameter
scarcity of available information to identify estimation and model selection uncertainty in
probabilistic models. This uncertainty is often ground snow maxima is considered using
neglected or inadequately addressed in frequentist and Bayesian techniques. In contrast
probabilistic engineering analysis and design even with the typically used fixed model and ‘best’
if its effect on reliability can be significant [1-3]. point estimates, model averaging and uncertainty
Albeit this uncertainty is acknowledged and intervals are used respectively.
recognized in the literature [4-6], only relatively The adopted techniques are not restricted to the
few studies have been focused on its practical considered problem and can be utilized for other
consequences [2, 7-9]. random variables, phenomena as well, e.g. floods
The aim of this paper is to investigate effects of and wind loads.
statistical uncertainties on structural reliability

Back to
Back to Table
Contents
of Contents 220
1
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

2 Conceptual framework shape factor;


, ground snow, 50-year maxima;
To study the effect of statistical uncertainties, two  permanent load;
sets of mechanical-probabilistic models with  beam span, 10 m;
different representation of the ground snow load  bay distance, 2 m.
are compared. These two model sets are referred
hereinafter as analysed and reference models, and Different load ratios are achieved by varying the
they differ solely in the ground snow distribution. mean value of the permanent load while keeping
For example, to illustrate the effect of parameter its coefficient of variation fixed. The load ratio is
estimation uncertainty, the reference model defined as:
contains the posterior mean distribution while the  ∙ !,"
analysed model uses the posterior predictive = (2)
 ∙ !," #$
distribution of ground snow. The reference model
is used to find a mean value of the resistance to where the m and k subscripts refer to mean and
reach the target reliability level, which is selected characteristic values respectively.
as 3,8 for a considered 50-year reference period.
The analysed model has the same mean resistance
and thus results based on this model show the s=μ sg•

a
effect of different ground snow representation. g

Throughout the study, different analysed- L


reference model pairs are considered to illustrate
various effects. Figure 1. Illustration of the analysed structural
member
3 Location and snow data
4.2 Probabilistic model
The ground snow data are obtained from the
Carpatclim database, covering a 50-year The properties of the random variables are
observation period (1962-2011) [10]. The used summarized in Table 1.
daily snow water equivalent data are from Table 1: Properties of random variables
Budapest, this location represents low-land areas
in the Carpathian region. Annual maxima from Variable Distr. Mean CV Ref.
winter seasons are extracted to infer the
 [-] LN 1,00 0,10 [12]
distribution of ground snow load. This approach is
referred to as the block method [11]. [kNm] LN * 0,15 [12]

 [-] LN 1,00 0,10 [13]


4 Numerical example
[-] LN 0,80 0,17 [14]
4.1 Mechanical model ,% [kN/m ]
2
† 0,4‡ 0,7‡ [10]
The mechanical model and loads of the analysed
& [kN/m ]
2
N § 0,10 [13]
structural member are presented in Figure 1. The
flexural failure of a mid-span cross-section is * Based on the FORM analysis to reach ()*+,) =
selected as a representative ultimate limit state: 3,8 with a fixed coefficient of variation.
† Gumbel, GEV, LN2, LN3 [15].
 ‡ Using unbiased moment estimates of the sample.
 =  ∙ −  ∙ ∙ , +  ∙  ∙ (1) § Varied to get different load ratios ().

For the annual ground snow load (,%) the sample
where:
 resistance model uncertainty; mean and coefficient of variation (CV) are
resistance; provided since these are also dependent on the
 load effect model uncertainty; fitted distribution type and technique applied to

Back to
Back to Table
Contents
of Contents 2
221
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

infer parameters of a distribution. The skewness 5.1.2 Bayesian - credible interval


of the analysed sample is 1,6, indicating Fréchet-
Bayesian uncertainty intervals are called credible
type distribution family [16]. The commonly used
intervals and can be constructed from the
Gumbel, generalised extreme value (GEV), two-
posterior distribution of the parameters.
parameter (LN2) and three-parameter lognormal
Approximate credible intervals can be constructed
(LN3) distributions are applied to represent the
using an approach similar to the delta method [6]
annual ground snow load [15]. The connection
for derived parameters where the evaluation of
between the 1-year and 50-year maxima
the likelihood function is expensive. The point
distributions is established by assuming that the
estimate is obtained by using the point estimate
annual maxima are mutually independent.
of the parameters (mean), and the covariance
matrix is derived from the joint distribution
5 Statistical analysis function of the parameters.
Frequentist and Bayesian statistical techniques are The parameter estimation uncertainty can be
applied to fit models to the snow maxima, and to incorporated into the distribution by averaging
quantify statistical uncertainties. Frequentist over the posterior distribution of parameters:
statistics bases the inference on comparing the
relative likelihood of datasets given a parameter
37 = 8: 37|9 ∙ 39|01 ∙ 09 (3)
value, /01|3456164, while Bayesian
statistics bases inference on the relative evidence
37 is the posterior predictive distribution [18].
of the parameter values given a dataset,
/3456164|01 [17]. For the current
5.2 Model selection uncertainty
analysis the most important difference is that the
Bayesian approach treats parameters as random Model selection uncertainty is the uncertainty in
variables, thus makes it possible to integrate the type of the probabilistic model, and it is also
uncertainties in their estimation into the failure stems from the scarcity of information. Herein it is
probability. treated considering different models, and
averaging them based on their ‘goodness’.
Maximum likelihood (ML) and posterior mean
(B.PM) are used as point estimates per frequentist 5.2.1 Frequentist model averaging (FMA)
and Bayesian approaches, respectively.
Accordingly, the distributions specified by these For  number of models (;), the model-averaged
point estimates are referred to as maximum point estimate (<=) can be obtained as the
likelihood (ML) and posterior mean (B.PM) weighted average of the model point estimates
distributions. (<=> ) [19]:

5.1 Parameter estimation uncertainty <= = ∑A =


>B% @> ∙ <> (4)

5.1.1 Frequentist - confidence interval where @> is the Akaike weight corresponding to
model C that is calculated using the sample size
In the frequentist paradigm the uncertainty
corrected Akaike information criterion. The
interval – to express parameter estimation
variance of the averaged parameter can be
uncertainty – is termed confidence interval and
estimated as follows [19]:
the delta method is used here to establish these
intervals [11]. This approximate method always I
produces symmetric confidence intervals; vaFr <= = ∑A F r <=>  + <=> − <= J
>B% @> ∙ Hva (5)
although the real intervals are often skewed. Yet it
is deemed sufficient here since confidence The averaging can be conducted for a selected
intervals are used only for illustrative purposes. study parameter such as a fractile or reliability
index.

Back to
Back to Table
Contents
of Contents 3
222
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

5.2.2 Bayesian model averaging (BMA) is too narrow. This is also supported by the
evidence that maxima of the sample are outside
Bayesian model averaging is based on density
of the 90% confidence interval (Figure 2). This can
functions [20]:
be observed also for other locations from the
3KLM 7 = ∑A
>B% N> ∙ 37|;>  (6) Carpatclim database with skewness exceeding 1,2.

Ground snow, annual maxima [kN/m2]


2
Strictly speaking 37|;>  is the posterior 1,5
Gumbel
FMA
LN2
FMA
predictive distribution under model C ; however, 1
for illustration the averaging is performed with the 0,5
posterior mean distributions as well. The
0
corresponding model averaged distributions are
2
denoted as BMA.PM and BMA.PP respectively. GEV LN3
1,5 FMA FMA
N> is the Bayes weight, given as:
1
PQRSR|LT ∙PLT  0,5
N> = 3;> |01 = ∑V (7)
UWX P QRSR|LU ∙P LU  0
1,1 10 100 1000 1,1 10 100 1000
Return period [year]
where 301|;>  is calculated from the
posterior predictive distribution under each Figure 2: Annual maxima return value-return
model, and 3;>  is the prior of the model. period plots with 90% confidence bands for the
selected (green + solid white) and FMA (blue +
In both model averaging approaches the outcome
dashed white) distributions in Gumbel space
depends on the considered model set. Herein the
GEV, Gumbel, LN3 and LN2 models are The Bayesian analysis yields to similar results as
considered, hence the results and conclusions are the frequentist, although the averaging weights
conditioned on this choice. are slightly favouring the three-parameter models
over the two-parameter ones (Table 2). Both the
For the Bayesian calculations vague priors are
methods indicate that there is a strong evidence
applied for both to the parameters and to the
against the LN2 model for this location.
models as well, and numerical integration is used
to calculate the posterior and posterior predictive Table 2: Summary of model averaging weights
distributions. Practically infinite ranges are
selected for the integrations and uniform priors Model Y Z
are adopted on these intervals. GEV 0,29 0,37

5.3 Results of distribution fitting Gumbel 0,43 0,25

The statistical uncertainties of ground snow LN3 0,28 0,38


models are illustrated in return value-return LN2 0,004 0,002
period plots with confidence intervals (Figure 2).
The solid white lines and green regions show the Table 3 shows the characteristic values for the
ML point estimates and 90% confidence intervals considered models and statistical approaches.
respectively; the latter illustrates the parameter With the exception of LN2, which is not supported
estimation uncertainty. The dashed white line and by the data, the models yield to comparable
blue region correspond to the FMA point estimate values. Due to parameter estimation uncertainty
and 90% confidence intervals respectively. The the largest increase is observed for LN3, it is about
extent of model selection uncertainty can be 8%. The Gumbel model underestimates the model
judged by comparing the blue regions to the green averaged characteristic value by about 10%. The
ones. For example for the Gumbel distribution the results are in good agreement with the 1,25
blue region is considerably wider than the green, kN/m2 characteristic value specified in the
which implies that the Gumbel confidence interval Hungarian National Annex of EN 1991-1-3.

Back to
Back to Table
Contents
of Contents 4
223
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

Table 3: Summary of ground snow load largely underestimated by the Gumbel


characteristic values [kN/m2] model.
5
Model ML B.PM B.PP
4
GEV 1,23 1,34 1,38 3

Reliability index, β [-]


Gumbel 1,07 1,11 1,12 2 Gumbel LN2
1 FMA FMA
LN3 1,20 1,18 1,28 5
LN2 1,78 1,90 1,98 4
3
FMA 1,16 N/A N/A
2 GEV LN3
BMA N/A 1,21 1,27 1 FMA FMA

0 0,2 0,4 0,6 0,8 0 0,2 0,4 0,6 0,8


N/A - not applicable
Load ra!o, χ [-]

6 Reliability analysis Figure 3: Load ratio– reliability index plots with


90% confidence bands for the selected (green +
solid white) and FMA (blue + dashed white)
6.1 Reliability index
ground snow distributions, using Gumbel ML as
The reliability of the selected structural member is reference model
analysed using the First order reliability method
As the LN2 model is clearly far from the FMA, and
(FORM) considering the various probabilistic
not supported by the data (Table 2), it is discarded
representation of the ground snow maxima.
from the following analyses. The same analysis
Initially a commonly applied Gumbel distribution
with B.PM models yields to similar results as those
is investigated. The reference model is the Gumbel
of the maximum likelihood based. Since the Bayes
maximum likelihood distribution while the
weights prefer the GEV and LN3 models, the
analysed models are the four selected maximum
underestimation of Gumbel model is even larger
likelihood and the frequentist model averaged
than in the case of FMA.
distributions. The reliability indices with
approximate 90% confidence intervals are Figure 4 shows the effect of statistical
presented in Figure 3. These intervals express uncertainties (here the reference model is the
solely the statistical uncertainties in the ground B.PM for each selected model, hence the solid
snow distributions. green line is horizontal, indicating the 3,8 value).
The analysed models are the B.PP and BMA
This analysis represents the following scenario:
models. The main observations from Figure 4 are
Gumbel ML distribution is adopted for snow
as follows:
maxima and a structural member is designed to
achieve the target reliability. The failure • The effect of parameter estimation
probability is then calculated assuming that the uncertainty within the Gumbel model is
snow maxima follows other distributions. Figure 3 small; however the model uncertainty has
shows the followings: substantial effect, significantly exceeding an
order of magnitude in failure probability.
• The distribution type has substantial effect on
the reliability, e.g. for large load ratios the • For GEV and LN3 the consideration of
GEV model yields to about 50 times larger parameter estimation uncertainty gives about
failure probability than the Gumbel. 20 and 5 times larger failure probability,
respectively.
• All models yield to smaller reliability index
than the reference Gumbel. • These effects are rapidly increasing with
increasing load ratio and reaching a relatively
• Compared to the FMA model, the failure
stable level after ratio of 0,3.
probability and uncertainty intervals are

Back to
Back to Table
Contents
of Contents 5
224
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

4
B.PM
The presented uncertainty interval construction
3,5 B.PP techniques require no or several additional limit
Reliability index, β [-]

3 BMA.PM state function evaluations as compared to the


Gumbel BMA.PP standard reliability methods such as FORM and
2,5
thus are applicable to complex problems as well.
4
It should be noted that the results are based on
3,5 the analysis of a single location with fixed number
3 of observations. For other phenomenon, such as
GEV LN3
2,5 wind extremes, which are typically following
Weibull distribution, the effects are anticipated to
0 0,2 0,4 0,6 0,8 0 0,2 0,4 0,6 0,8
Load ra"o, χ [-]
be smaller. However, smaller number of
observations can considerably increase statistical
Figure 4: Load ratio – reliability index plots for uncertainties. Moreover, it is expected that these
models with (dash-dot) and without parameter uncertainties can be reduced by taking into
estimation uncertainty (solid). account more informative prior information that is
Table 2indicates minor differences in the often available.
characteristic values for the selected probabilistic
models. Contrary to this the reliability indices in 8 Conclusions
Figure 4suggest the model selection is a key issue The following conclusions are drawn from the
for reliability analysis. Consequently, comparison analysis of the effect of statistical uncertainties on
of characteristic values seems to be insufficient to structural reliability:
support decision about appropriate probabilistic
model. • The effect of parameter estimation
uncertainty within the Gumbel model is
7 Discussion small; however the model uncertainty has
substantial effect, considerable exceeding an
The main focus of this paper is the quantification order of magnitude in failure probability.
and propagation of statistical uncertainties. From • The distribution choice may lead to 50-times
a broader engineering perspective that views increase in failure probability.
reliability analysis as a tool for decision making, • For the generalised extreme value and three-
the model selection uncertainty problem is parameter lognormal distributions the
recommended to be resolved by agreeing on a consideration of parameter estimation
conventional distribution type [5, 21]. The authors uncertainty can yield to about 20- and 5-
support this approach. However, the decision on times increase in failure probability,
the selected distribution should be backed by respectively.
appropriate techniques for model uncertainty • The use of ‘best’ point estimates such as
analysis. For safety-critical facilities and actions it maximum likelihood estimator are not
still might be desirable to analyse this source of conservative and can have practically
uncertainty in detail. significant effect.
Regarding parameter estimation uncertainty it is The following practical recommendations and
argued that it should be considered in reliability observations are made based on the presented
studies by using the posterior predictive and previous studies by the authors [22, 24, 25]:
distribution [22, 23]. The main arguments are that
reliability analysis is inherently predictive, the 1) For common structures and standardization:
failure probability is governed by the very • Consensus on the conventional distribution
uncertain distribution tails, and after the selection type is needed.
of distribution function it is unambiguous • The commonly applied Gumbel model for
construct the posterior predictive estimates. annual maxima is often inadequate, i.e.
underestimates failure probability and has

Back to
Back to Table
Contents
of Contents 6
225
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

deceptively narrow uncertainty interval, this on slope reliability. Reliability Engineering


model assumption should be revisited. & System Safety. 2015; 140(0): 99-106.
• Bayesian approach is recommended to [3] Coles S., Pericchi L. Anticipating
handle parameter estimation uncertainties. catastrophes through extreme value
However, further studies should clarify which modelling. Journal of the Royal Statistical
probabilistic model, e.g. posterior mean or
Society Series C (Applied Statistics). 2003;
posterior predictive, should be applied in the 52(4): 405-16.
framework of economic optimization, and
derivation of safety factors. [4] JCSS. JCSS Probabilistic Model Code. Part I
2) For structures which require fully probabilistic - Basis of Design: Joint Committee on
risk assessment framework: Structural Safety; 2000.
• Bayesian statistics is promoted to incorporate [5] Melchers R.E. Structural Reliability
model selection and parameter estimation Analysis and Prediction. Chichester, UK:
uncertainty. John Wiley and Sons Ltd.; 1999.
• Generalised extreme value distribution is
recommended for probabilistic modelling of [6] Der Kiureghian A. Analysis of structural
extremes. reliability under parameter uncertainties.
• Inclusion of all available data and prior Probabilistic Engineering Mechanics. 2008;
information is recommended, however 23(4): 351-8.
guidance on priors is needed. [7] Gardoni P., Der Kiureghian A. Probabilistic
The consideration of statistical uncertainties can Models and Fragility Estimates for Bridge
be especially important for predicting extremes Components and Systems. PEER Report
under changing climate for which one cannot rely 2002/13. Berkeley, USA: Pacific
on historical observations. Earthquake Engineering Research Center,
College of Engineering, University of
California, Berkeley, 2002.
9 Acknowledgements
[8] Van Gelder P.H.A.J.M. Statistical Methods
This work was partly supported by the
for the Risk-Based Design of Civil
International Visegrad Fund Intra-Visegrad
Structures: Delft University of Technology;
Scholarship (contract no. 51401089) and by the
2000.
Ministry of the Interior of the Czech Republic
(project no. VG20122015089). The calculations are [9] Most T., Knabe T. Reliability analysis of the
completed using Matlab [26] and FERUM [27], the bearing failure problem considering
work and commitment of the developers of these uncertain stochastic parameters.
applications are highly appreciated. All the related Computers and Geotechnics. 2010; 37(3):
codes/scripts, processed data, and results are 299-310.
available from the corresponding author. [10] Szalai S., Auer I., Hiebl J., Milkovich J.,
Radim T., Stepanek P., et al. Climate of the
10 References Greater Carpathian Region.
[1] Sisson S.A., Pericchi L.R., Coles S.G. A case www.carpatclim-eu.org: 2013.
for a reassessment of the risks of extreme [11] Coles S. An Introduction to Statistical
hydrological hazards in the Caribbean. Modeling of Extreme Values. London:
Stochastic Environmental Research and Springer-Verlag; 2001.
Risk Assessment. 2006; 20(4): 296-306.
[12] JCSS. JCSS Probabilistic Model Code. Part
[2] Li D.-Q., Tang X.-S., Phoon K.-K. Bootstrap III - Resistance Models: Joint Committee
method for characterizing the effect of on Structural Safety; 2000.
uncertainty in shear strength parameters

Back to
Back to Table
Contents
of Contents 7
226
IABSE Conference – Structural Engineering: Providing Solutions to Global Challenges
September 23-25 2015, Geneva, Switzerland

[13] JCSS. JCSS Probabilistic Model Code. Part II Reliability. Modelling in Mechanics; 2015;
- Load Models: Joint Committee on Ostrava, Czech Republic.
Structural Safety; 2001.
[25] Rózsás Á., Sýkora M., Vigh L.G. Long-term
[14] Ellingwood B., O'Rourke M. Probabilistic Trends in Annual Ground Snow Maxima
models of snow loads on structures. for the Carpathian Region. Engineering
Structural Safety. 1985; 2(4): 291-9. Mechanics 2015; 2015 May 11-15, 2015;
Svratka, Czech Republic.
[15] STRUREL. COMREL & SYSREL: Users
Manual. Componental and Systel [26] MATLAB. The Language of Technical
Reliability Anlysis. Using Built-in Symbolic Computing. R2014a (8.3.0.532) ed. Natick,
Processor. München, Germany: RCP Massachusetts: MathWorks Inc.; 2014.
GmbH; 2003.
[27] Der Kiureghian A., Haukaas T., Fujimura K.
[16] Reiss R.-D., Thomas M. Statistical Analysis Structural reliability software at the
of Extreme Values with Applications to University of California, Berkeley.
Insurance, Finance, Hydrology and Other Structural Safety. 2006; 28(1–2): 44-67.
Fields. 3 ed: Birkhäuser Basel; 2007.
[17] Spiegelhalter D., Rice K. Bayesian
statistics. Scholarpedia, 4(8):52302009.
[18] Aitchison J., Dunsmore I.R. Statistical
Prediction Analysis. New York, USA:
Cambridge University Press; 1980.
[19] Burnham K.P., Anderson D.R. Model
Selection and Multimodel Inference. A
Practical Information-Theoretic Approach.
Second Edition ed. New York: Springer-
Verlag; 2002.
[20] Hoeting J.A., Madigan D., Raftery A.E.,
Volinsky C.T. Bayesian Model Averaging: A
Tutorial. Statistical Science. 1999; 14(4):
382-417.
[21] Ditlevsen O. Distribution arbitrariness in
structural reliability. ICOSSAR'93,
Structural Safety and Reliability; 1993;
Rotterdam, The Netherlands1994.
[22] Rózsás Á., Sýkora M. Model comparison
and quantification of statistical
uncertainties for annual maxima of
ground snow loads (accepted for
publication). European Safety and
Reliability Conference, ESREL; 7-10,
September; Zürich, Switzerland 2015.
[23] McRobie A. The Bayesian View of Extreme
Events. Henderson Colloquium; July 5;
Magdalene College, Cambridge 2004.
[24] Rózsás Á., Sýkora M. Effect of Parameter
Estimation Uncertainty on Structural

Back to
Back to Table
Contents
of Contents 8
227

You might also like