Structural Safety: Fatemeh Jalayer, Iunio Iervolino, Gaetano Manfredi

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Structural Safety 32 (2010) 220228

Contents lists available at ScienceDirect

Structural Safety
journal homepage: www.elsevier.com/locate/strusafe

Structural modeling uncertainties and their inuence on seismic assessment


of existing RC structures
Fatemeh Jalayer *, Iunio Iervolino, Gaetano Manfredi
Department of Structural Engineering, University of Naples Federico II, Via Claudio 21, Naples 80125, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The structural modeling uncertainty may be comparable to that of the ground motion representation in
Received 21 April 2008 the seismic assessment of existing buildings. This work aims to characterize the uncertainties in material
Received in revised form 19 February 2010 properties and in construction details and to propagate them to estimate the structural performance con-
Accepted 19 February 2010
ditional on code-based seismic demand and capacity denitions. A prior probability distribution for the
Available online 12 March 2010
uncertain parameters is constructed based on the state of knowledge about the building before any in situ
test is conducted. The uncertainties in the parameters are then propagated using simulation in order to
Keywords:
obtain a probability distribution for structural resistance. The simulation-based methods allow for suit-
Structural reliability
Modeling uncertainty
able grouping of uncertain parameters in order to build a simplied model of correlation across structural
Bayesian updating parameters. The results of tests and inspections are used both to update the prior probability distribu-
Condence factors tions for parameters and also to update the structural reliability in a Bayesian framework. The updated
Existing structures probability distributions take into account the increased level of information eventually gathered as a
Markov Chain Monte Carlo simulation results of destructive and/or non-destructive tests and inspections on the structure.
2010 Elsevier Ltd. All rights reserved.

1. Introduction in the so-called condence factors (FC) which have to be applied to


mean material properties when performing seismic assessment of
A signicant portion of the total seismic risk in Italy comes from the structure. The primary scope for introduction of these con-
the various types of losses induced by inadequate response of dence factors is to allow for a certain level of conservatism,
existing buildings to ground shaking. This is the case for several through the use of material strength values smaller than that of
other European countries in which the average service life for the corresponding mean values determined based on available
buildings is larger than that of countries like the United States. knowledge, in the seismic assessment of existing structures. It is
Therefore, management of existing building stock is a major con- worthwhile mentioning that after applying the code-based con-
cern in such regions. The majority of reinforced concrete (RC) exist- dence factors, the assessment of an existing building will be per-
ing structures have been designed mainly for gravity loading only formed almost the same as the design of new construction and
and the seismic provisions considered in the design process are may be subjected to partial factors employed by the code.1
either very poor or non-existent. As a result, more recent European Evaluation of these condence factors seems to be rather qual-
seismic guidelines (e.g., Eurocode 8 [8], OPCM [2022], NTC[19]) itative and based on three increasing levels of knowledge about the
pay particular attention to seismic assessment of existing struc- structure in question. These levels of knowledge are classied
tures, which is distinguished from that of the new construction based on a prescribed set of (destructive and non-destructive) ver-
by lack of information about both the original features and the cur- ifying tests and inspections. To each level of knowledge a specic
rent state of building in consideration. In other words, determining value of condence factor is assigned; for example, the Italian
the structural modeling parameters such as, material properties and European codes, recommend condence factors equal to
and structural detailing parameters is considered a major chal- 1.35, 1.20 and 1.0 for knowledge levels, poor, sufcient and accu-
lenge in the assessment of existing buildings. rate, respectively. Although the condence factors are applied to
These European and Italian seismic guidelines synthesize the ef-
fect of structural modeling uncertainties specic to existing build- 1
The uncertainties in the component capacity models used for the assessment of
ings due to lack of knowledge about buildings characteristics (e.g., existing buildings in both Eurocode 8 and the Italian seismic code are taken into
uncertainty in the material properties and in construction details) account separately by an overall security factor denoted by cel in order to take the
mean minus standard deviation value for component capacity instead of the mean
value. As far as it concerns the uncertainty in the seismic action, the above-mentioned
* Corresponding author. Tel.: +39 7683672; fax: +39 7683491. codes do not seem to make a clear distinction between the new construction and the
E-mail address: [email protected] (F. Jalayer). existing buildings.

0167-4730/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.strusafe.2010.02.004
F. Jalayer et al. / Structural Safety 32 (2010) 220228 221

the properties of materials, the uncertainties in structural model- [1]. Nevertheless, model updating is not an end by itself, and it is
ing specic to existing buildings are not limited to them and in- normally desirable to also improve the predictions of structural
clude also other structural detailing parameters entering into the reliability. The Bayesian framework used for updating the struc-
seismic assessment problem (e.g., reinforcement detailing, cover tural model and its reliability is described in detail in this section.
thickness, etc.). However, the extent to which the uncertainties
in structural detailing parameters could affect the seismic assess- 2.1. Evaluation of robust reliability
ment of existing structures seems not yet to be studied in depth.
This work aims to outline a theoretical basis for quantifying and Let the vector h denote the set of uncertain model parameters.2
updating the uncertainties in the material properties and construc- Let D denote some test data and consider that the set of possible
tion details specic to existing buildings which are the basis for the structural models can be dened by M to specify (both the structural
designation of condence factors in Eurocode 8. This methodology and the probabilistic) modeling assumptions used in the analysis.
is presented in the context of an existing case-study building The Bayesian framework used herein provides a rigorous method
where the available in situ test results are used in order to update for updating the plausibility of each of the models in representing
the modeling uncertainties and the structural reliability, given the structure. The plausibility of a model may be quantied by the
code-specic seismic actions and limit state capacity threshold, in- probability distribution over the vector of model parameters
side a Bayesian probabilistic framework. The focus of the study is h h1 ; . . . ; hn  that dene a model within the set of possible models.
on the uncertain parameters that are specic to an existing build- In other words, each model is uniquely identied by a realization of
ing as opposed to a building of new construction. Thus, the uncer- the vector h. Therefore, the plausibility of the model can be quanti-
tainties in the seismic action (common between new and existing ed by the probability of such realization among the set of possible
construction) and the modeling uncertainties in the component values of the vector.
capacities such as the modeling uncertainty in determining the The updated probability distribution can be dened using the
ultimate rotation in a section (addressed in the assessment of Bayes Theorem [2]:
existing buildings by applying a separate security factor) are not
taken into account. This is done particularly in order to single- pDjh; M
pD h phjD; M phjM 1
out those uncertainties specic to seismic assessment of existing pDjM
construction related to the estimation of code-based condence
where phjM is the prior probability distribution for h specied by
factors. The characterization of uncertainties in this framework is
M, p(D|M) is the probability distribution for data D specied by M,
preformed in two levels. In the rst level, prior probability distri-
and pDjh; M is the (updated) probability distribution for observed
butions for the uncertain modeling parameters are constructed
data D given the vector of parameters h specied by M.
based on information available from original design documents
Updated response predictions can be made implementing data
and (qualitative) professional judgment. In the second level, the re-
D through pD h given by Eq. (1). For example, if the probability
sults of in situ tests and inspections are implemented in the Bayes-
of a failure event F based on modeling parameters h is denoted
ian framework in order to both update the prior distributions for
by P(F|D, M), the robust failure probability3 can be calculated from
the modeling parameters and also to update the distribution for
the following integral dened over the whole domain of h:
structural performance variable and structural reliability using
Z
simulation-based reliability methods. The Bayesian updating pro-
PFjD; M PFjh; MphjD; M dh 2
cedure employed allows for updating the probability distributions
for both structural modeling parameters and the structural global
response within the simulation routine. Moreover, it is general en- where PFjh; M is the failure probability for the structural model
ough to allow for both consideration of various types of inspections dened by h. In particular, given a specic representation of ground
ranging from drilled cores tests, ultrasonic tests, and pacometric motion, PFjh; M reduces to a deterministic index function IF h; M.
tests to pseudo-dynamic health-monitoring tests and the consider- This index function is equal to one in the event of failure and equal
ation of the corresponding measurement errors. to zero otherwise:
Updating of structural reliability across increasing amount of test Z
pDjh; M
results makes it possible to (i) introduce a performance-based prob- PFjD; M IF h; M phjM dh 3
PDjM
abilistic denition of the condence factor as the value that, once ap-
plied to the mean material properties, leads to a value for structural This paper utilizes a Markov Chain Monte Carlo (MCMC) simu-
performance measure with a specied probability of being exceeded lation method to evaluate the robust reliability in Eq. (3) [2]. The
(e.g., 5%) and (ii) evaluate the code-based recommendations regard- MCMC method employs the MetropolisHastings (MH) algorithm
ing condence factors and the corresponding knowledge levels. The [17,13] in order to generate samples as a Markov Chain sequence
methodology presented in this study, in the context of an existing which are used later to estimate the robust reliability by statistical
case-study, lays out the fundamentals for a comprehensive work averaging. The MetropolisHastings algorithm is normally used to
on the characterization of uncertainties specic to existing build- generate samples according to an arbitrary PDF when the target
ings, implementation of in situ test results, and verication of PDF is known only up to a scaling constant.
code-based requirements regarding the condence factors.
2
It should be noted that the vector h includes all the uncertain parameters in the
problem, such as those representing the uncertainty in the seismic action. However,
2. The methodology in this study, we have focused only on structural modeling uncertainties specic to
existing buildings for a given level of seismic intensity. Therefore, the robust
structural reliability calculated in this work is conditioned on code-based seismic
In the presence of structural modeling uncertainty, instead of a action and limit state capacity assumptions.
unique structural model, a set of plausible structural models can be 3
This paper is based on a probabilistic framework in which the probability is
identied. A robust assessment of structural reliability takes into always conditional on the amount of information available. In this probabilistic
account a whole set of possible structural models that are framework [14], probability represents the degree of belief in a certain outcome based
on the amount of information. We have used the word robust instead of the word
weighted by their corresponding plausibility. A Bayesian updating predictive to describe the updated reliability. That is, the plausibilities of all the
framework can be implemented in order to update both the struc- possible structural models described by M conditional on the amount of information
tural modeling properties and the reliability based on test results available D are taken into account in order to calculate the updated reliability.
222 F. Jalayer et al. / Structural Safety 32 (2010) 220228

2.2. Generating samples according to target PDF phjD; M 3. Structural failure

The MH algorithm can be used to generate samples according to The failure event F can be dened as when structural demand
the target PDF phjD; M. Using Bayes formula one can derive the denoted as Dh exceeds structural capacity Ch : F fh :
PDF as: Dh > Chg. Assuming scalar demand and capacity, the (scalar)
demand to capacity ratio can be dened as Yh Dh=Ch.
pDjh; MphjM Therefore, the failure region F can be dened as F fh : Yh > 1g
phjD; M c1 pDjh; MphjM 4
pDjM and the sequence of embedded intermediate failure regions can
be generated as F i fh : Yh > yi g where 0 < y1 <  < ym = 1.
where phjM is the prior probability distribution for the vector of In this study, the structural capacity is obtained using the push-
parameters h and p(D|h, M) known as the likelihood function is the over analysis as the global displacement at which the rst element
probability distribution for the data specied by parameters h. loses its load bearing capacity (i.e., 3/4th of the of ultimate chord
The MH algorithm can be used to generate samples according to rotation in the member).5 The structural demand is dened as the
the target updated PDF f  phjD; M using the product global displacement corresponding to the intersection of the capac-
p  pDjh; MphjM as the candidate PDF. In order to increase the ity curve for the equivalent SDOF system and the corresponding
acceptance rate for the candidate samples during the Markov Chain code-based seismic response spectra for the seismicity and the soil
simulation process, a sequence of intermediate target PDFs denoted characteristics at the site of the project (a.k.a, capacity spectrum
by fi are introduced which vary gradually between the prior PDF method, [12,9]). It should be mentioned that assumptions involved
phjM and the updated target PDF p(D|h, M). The target fis can be in the choice of the analysis method, the seismic demand and the
modeled as updated PDFs according to Bayes theorem based on structural capacity limit state are considered part of the set of mod-
an increasing amount of data: fi  phjDi ; M were D1  D2 eling assumptions M. Therefore, the estimated probability of failure
     Dn D. That is, at the rst level with a target PDF equal to will be conditional on M.
f1, one could use the prior PDF p1 phjM as the proposal PDF. In
order to approximate f2, the kernel sampling density j1 is con-
4. Modeling of uncertainties
structed as a weighted sum of Gaussian PDFs centered about the
generated samples [4]. The kernel sampling density generated is
As it is mentioned in the previous section, the vector of param-
then used as the proposal density in the second level: p2  j1 .
eters h includes all the uncertain parameters in the problem such
as the uncertainty in the seismic action, the uncertainty in the
2.3. Calculating the failure probability using subset simulation property of the materials and the uncertainty involved in the struc-
tural detailing. The present work focuses on the uncertainty in the
In order to calculate the small failure probabilities encountered parameters of structural modeling of resisting elements, since it is
in the seismic reliability problem, the failure probability can be cal- characterized differently in existing structures and new construc-
culated using a simulation method known as subset simulation [4], tion.6 Therefore, in order to focus on the modeling uncertainties spe-
in which the failure region is modeled as the last element in a se- cic to existing buildings, the seismic assessment is performed
quence of embedded failure regions F = Fm  Fm1    F1. There- conditional on a given level ground motion intensity. The structural
fore, the failure probability can be derived as the following: modeling uncertainty is directly related to the quantity (and the
quality) of information that is available for the structure. In this
Y
m1 study, two different sources of uncertainty are considered: (1) uncer-
PF PF 1 PF i1 jF i 5 tainty in the mechanical properties of materials used in construction
i1 and (2) the reinforcement details that affect the component capacity
in terms of moment-rotation relation; uncertainty in details can be
where F1 is the rst element in the failure sequence (i.e., largest fail- both due to limited information about the design of a specic struc-
ure region) and F = Fm is the target failure region and the last ele- ture and/or local construction practice and also due to low quality
ment in the failure sequence. The rst term in Eq. (5) P(F1) can be control in construction (also known in the engineering jargon as
calculated using standard Monte Carlo simulation, generating sam- structural defects, not uncommon in structures built after the second
ples from the original PDF for the modeling parameters: world war in Italy). As the uncertainties belonging to the second
Z group mentioned above, those related to the percentage of rebar
PF 1 IF 1 hph dh 6 present in the element, rebar diameter (e.g., different from that spec-
ied in the original design notes), the anchorage quality and the cov-
er thickness are considered. The uncertainties in structural detailing
And the intermediate failure probabilities PF i1 jF i is equal to:
are modeled as discrete uncertain variables that can assume a range
Z of possible values with a corresponding plausibility/weight. In the
PF i1 jF i IF i1jF hphjF i dh 7 absence of test results and in situ inspections, the plausibility of each
i
possible value is assigned qualitatively based on engineering consen-
Again the MH algorithm can be used to generate samples as the sus, judgment and experience. It has to be mentioned that once the
states of a Markov Chain with target distribution equal to the con-
ditional PDF phjF i for each intermediate failure region (see [5] for 5
Corresponds to the denition of near-collapse (life-safety) limit state in the Italian
details on how to choose the candidate PDF). The subset simulation code.
6
is shown to be especially efcient for modeling rare failure events It should be noted that the modeling uncertainties in the component capacities
(i.e., when the probability of failure is very small).4 are not taken into account in this study, while this study intentionally focuses on the
effects of uncertainties related to the actual knowledge of structural features of
existing buildings. Moreover, the seismic input representation herein follows a code-
4
It should be mentioned that both the subset simulation algorithm and the based spectral approach not explicitly accounting for variability of the response due
structural model updating algorithm exploit the MH algorithm in order to generate to ground motion. Nevertheless, discussions on how to take into account the
samples as states of a Markov Chain. They differ in that the subset simulation moves modeling uncertainties in the component capacities and how it compares to the
across embedded failure regions whereas the structural model updating evolves uncertainty in the representation of the ground motion, one can refer to, for example,
across increasing levels of information. [15,16].
F. Jalayer et al. / Structural Safety 32 (2010) 220228 223

(a) (b)
(1) (2) (3) (4)

2.20 m
(IV)

3.90 m
(III)

3.90 m
(II)

3.50 m
(I)

Y Z X
4.20 m 4.20 m 4.20 m 3.50 m

Fig. 1. (a) The tri-dimensional view of the scholastic building. (b) The central frame of the case-study building.

test results are available on the quantity in question, they can be reinforced concrete, assuming that the concrete is not conned,7
used applying the Bayesian methodology described in the previous and that the reinforcing steel behavior is elasticperfectly-plastic.
sections to update the plausibility of each possible value for the cor- The behavior of the plastic hinge is characterized by four phases,
responding discrete uncertain variable. As it regards the correlation namely: rigid, cracked, post-yielding, and post-peak. In addition to
between different uncertain parameters, a simplied model of corre- exural deformation, the yielding rotation takes into account also
lation is constructed by classifying different sets of correlated uncer- the shear deformation and the deformation related to bar-slip based
tain parameters within groups that are not cross-correlated [5]. on the code recommendations [21]. Moreover, the shear span used in
the calculation of the plastic rotation is based on the code formulas.
As it regards the post-peak behavior, it is assumed that the section
5. Case-study
resistance drops to zero, resulting in a tri-linear curve which is
sketched in Fig. 2.8
As the case-study, an existing school structure located in Avel-
lino (Italy) is considered herein. The structure is situated in seismic
Zone II according to the Italian seismic classication of the OPCM
3519 [22]. The structure consists of three stories and a semi- 5.1. In situ test methods to quantify material properties
embedded story and its foundation lies on soil type B according
to Eurocode 8 [8]. For the structure in question, the original design For the case-study building described above, results of both
notes and graphics have been gathered. The building is constructed destructive and non-destructive test results are available. As
in the 1960s and it is designed for gravity loads only, as it is fre- destructive test methods, compression core tests are performed
quently encountered in the post second world war construction. in order to determine the in-place compressive strength of con-
In Fig. 1a, the tri-dimensional view of the structure is illustrated; crete and the tensile test is performed in order to determine the
it can be observed that the building is irregular both in plane and in situ tensile yield and ultimate strength of reinforcing steel.
elevation. The main central frame in the structure is extracted The drilled core test consists of removal of standard cores from pri-
and used as the structural model (Fig. 1b). The columns have rect- mary concrete components followed by the laboratory testing of
angular section with the following dimensions: rst storey: these samples in order to obtain the core-strength. The core-
40  55 cm2, second storey: 40  45 cm2, third storey: strength should be converted to the in situ strength following an
40  40 cm2, and forth storey: 30  40 cm2. The beams, also with approved procedure. In this work, the procedure suggested by
rectangular section, have the following dimensions: 40  70 cm2 the British Standard 1881 [3] is used. In the case of destructive ten-
at rst and second oors, and 30  50 cm2 for the ultimate two sile tests, reinforcing steel bar samples are removed from the struc-
oors. The reinforcement ratios, as dened in the original notes, ture and tested in laboratory in order to determine both the tensile
range from about 0.2% to 0.7% in the columns and 0.7% to 1% in yield and ultimate strength [11]. As non-destructive test methods,
the beams. It can be inferred from the original design notes that the ultrasonic test results are performed in order to determine the
the steel rebar is of the type Aq42 (nominal minimum yield resis- concrete compressive strength [26]. Ultrasonic testing uses high
tance fy = 2700 kg/cm2 or 265 MPa) and the concrete has a mini- frequency sound energy to make measurements of the velocity of
mum resistance equal to 180 kg/cm2 (17.7 MPa, [23]). The nite the sound wave through the reinforced concrete medium. The re-
element model of the frame is constructed assuming that the sults of ultrasonic testing are often calibrated against drilled core
non-linear behavior in the structure is concentrated in plastic test results performed at the exact same location in order to con-
hinges located at the element ends. Each beam or column element vert the sound velocity into concrete in situ strength.
is modeled by coupling in series of an elastic element and two ri-
7
gidplastic elements (hinges). The stiffness of the rigidplastic ele- For a building that is not designed according to seismic provisions, it is unlikely
the number of stirrups in and outside the section are sufcient to assure that the
ment is dened by its moment-rotation relation which is derived
concrete is conned everywhere inside the stirrups, see also Fardis [10].
by analyzing the reinforced concrete section at the hinge location. 8
It should be mentioned that, in this case-study example, the section resistance is
In this study, the section analysis is based on the widely adopted in evaluated in terms of the exural response only and the shear resisting capacity has
current practice ManderPriestly [18] constitutive relation for not been considered.
224 F. Jalayer et al. / Structural Safety 32 (2010) 220228

Ultimate moment

Yielding moment

Cracking moment

Yield rotation Ultimate rotation

Fig. 3. Schematic diagram of the rotation capacity for limit state of severe damage.
Fig. 2. Schematic diagram of the typical tri-linear behavior characterizing the rigid
plastic hinge.
5.3. Case 1: structural detailing consistent with design specications

5.2. Evaluation of structural demand and capacity As the rst case, the reliability of the structure is calculated
based on the state of knowledge about the building before in situ
In accordance with the Eurocode 8 and Italian seismic code [20] inspections and tests are conducted. Moreover, in order to evaluate
the severe damage limit state is reached at the instance when the the signicance of considering the uncertainty in structural detail-
rst element experiences 3/4th of the of its ultimate chord rotation ing parameters in Case 2 described below, the uncertainty is as-
in the member, as shown in Fig. 3. sumed to be only in material mechanical properties. The prior
The structural lateral load resistance curve (a.k.a., capacity probability distributions for concrete and steel are constructed
curve) is represented herein in terms of the base-shear and roof- using the results of a statistical study done on the characteristics
displacement. The non-linear static analyses are performed using of the steel and concrete used in the constructions of the time
the SAP2000 (Ver. 10) structural analysis commercial software. [24,25]. As it regards the correlation among material mechanical
The information regarding the gravity loading is extracted from properties, it is assumed that the material properties for steel
the original design notes. The horizontal forces are calculated and concrete are uncorrelated. Moreover, the concrete resistance
based on the code procedure using the rst fundamental mode for the RC elements on each oor is considered to be uncorrelated
shape as the lateral displacement prole along structural height. with those of the other oors.
The same displacement prole is also used in order to calculate Since the probability of failure in this case is predicted to be
the rst-mode participation factor for the structure. The capacity very small, the subset simulation procedure described in the meth-
curve is transformed into that of an equivalent SDOF system utiliz- odology is used in three levels in order to calculate the probability
ing the modal participation factor and is successively bi-linearized of failure. In the rst level 500 and 200 analyses in the second and
as it is prescribed in capacity spectrum method. The displacement- third level are performed with intermediate failure probabilities
based structural capacity Ch corresponds to the point on the equal to 0.10, 0.20 and 0.005 respectively. The probability of failure
capacity curve in which the rst element reaches the threshold is calculated to be equal to 0.0003 with a coefcient of variation
of severe damage as described above. The ultimate deformation equal to 1.57. The coefcient of variation is calculated based on a
capacity point is followed by a sharp drop in the structural resis- procedure described in [5] where the failure probability is equal
tance represented by the base-shear. to the expected value for the average of a set of correlated
The seismic action is modeled based on the code-specied elas- Bernoulli variables generated as members of a Markov Chain
tic spectrum modied to account for the soil condition at the site of sequence.
the structure (type B). This spectrum is anchored about the peak
ground acceleration on rock, ag, corresponding to the seismic zone
in which the site is classied, with a return period equal to 5.4. Case 2: no test results but considering the uncertainty in the
475 years (i.e., 10% probability of exceedance in 50 years). For the structural detailing
site of the case-study structure, this results in ag equal to 0.25 g.
The displacement-based structural demand Dh is then calculated As the second case, the reliability of the case-study frame is cal-
as the displacement at which the capacity curve for the equivalent culated based on the state of knowledge about the building before
SDOF system intersects the spectrum after it is modied based on a in situ inspections and tests are conducted. The sources of uncer-
factor reecting the inelastic properties of the structure in tainty are assumed to be the material mechanical properties and
question. the structural detailing. A list of the possible sources of uncertainty
As it is mentioned in the section on the methodology applied, in the reinforced concrete section detailing has been constructed
the simulation-based methods (e.g., Monte Carlo, Monte Carlo by identifying the various possibilities, their relative plausibility,
Markov Chain) are used in order to calculate the structural reliabil- and their correlation with other sources.
ity. For each realization within the simulation process, rst, the Table 1 demonstrates a list of possible sources for structural
parameters dening the plastic hinge moment-rotation diagram modeling uncertainty represented by discrete probability mass
are calculated and implemented in the structural nite element functions and the corresponding correlation structure. The vari-
model. In the next step a non-linear static analysis is performed ables representing uncertainty in the quality of anchorage (in this
on the (simulated) structural model followed by the application case 180 hooks) are modeled by considering two possibilities,
of the capacity spectrum method in order to determine the de- either the hooks are done according to the specications or not
mand/capacity ratio corresponding to this simulation realization. done at all. In the case where they are done well, it is assumed that
F. Jalayer et al. / Structural Safety 32 (2010) 220228 225

the full area of the present rebar at the ends is effective; otherwise, Table 2
in the absence of the hooks, only half of the rebar area is consid- Probabilistic characterization of the mechanical property of RC.

ered to be effective at member ends. It is further assumed that in Var. Dist. Mean (kg/cm2) COV
90% of the cases the hook is done according to the specications. fc LN 165 (16 Mpa) 0.15
Moreover, anchorage quality is supposed to be uniform throughout fy LN 3200 (314 Mpa) 0.08
each oor; that is the corresponding variables are fully correlated.
In the next category, the error in diameter is considered. This type
of error usually occurs when rebar diameters are close-enough to Monte Carlo Simulation with 200 simulations (the pushover curves
be mistaken visually. This has been modeled only for columns of corresponding to a few simulation realizations are shown in Fig. 4).
the ground oor where both diameter 14 and 16 rebar were spec- The probability of failure is calculated to be equal to 0.005 with a
ied in the design notes. It is supposed that in 95% of the cases the coefcient of variation equal to 1.0. The coefcient of variation is
correct diameter is being placed. The next category addresses the calculated based on a procedure described in [5] where the failure
possible errors in superposition of columns across oors. It is as- probability is equal to the expected value for the average of a set of
sumed that, if the overlapping splice length is sufcient, the whole independent Bernoulli variables (binary outcome 1 with probabil-
area of column rebar extending from the story beneath is effective. ity equal to Pf and 0 with probability 1  Pf).
Otherwise, as a possible variation, a case is considered in which
75% of the rebar area is effective. This type of error is supposed 5.5. Case 3: using test results to update predictions
to occur uniformly across each oor. Another type of error in
detailing can take place when the rebar in column section is posi- Table 3 demonstrates the actual test results available for the
tioned without particular attention to whether it is oriented to- case-study structure which consist of (non-destructive) ultrasonic
wards the weak direction or otherwise. In such case, two results (6 data per oor), (destructive) drilled core tests (2 data per
possibilities can be considered, one (the more plausible one) in oor) for determining the concrete resistance and the tension test
which the rebar is positioned so that the section has more exural for reinforcing steel (1 data). It should be noted that the standard
resistance in the weak direction and its opposite case (less plausi- error assigned to the ultrasonic tests is larger than that assigned
ble case) where the section has more exural resistance in the to the drilled core tests to take into account the fact that the ultra-
strong direction. This kind of error is supposed to be uniform sonic results are calibrated (using regression analysis) with respect
across the same type of section throughout the same oor. The to the drilled core tests.
kind of human error related to missing rebar is taken into account The results of the tests are used in two levels in order to update
only in beams where a large quantity of short-length diagonal re- the probability distribution for concrete and reinforcing steel resis-
bar are specied. This could cause accidental loss of a rebar. It is as- tance at different oors in the structure and to calculate the robust
sumed that any specied rebar is present as specied with 90% reliability. In the rst level the destructive test results are imple-
probability. This kind of error is also modeled as uniform across mented and in the second level the non-destructive test results
section type and the oor to which it belongs. The last category ad- are used. The test results are implemented using the MH algorithm
dresses the errors in concrete cover which is discretized by consid- with 200 simulations at each level in order to update the probabil-
ering only three possible values. The most likely value is taken to ity distribution for concrete and reinforcing steel resistance and
be 3 cm with a lower limit of 2 cm and higher limit of 4 cm. This also to update the structural reliability. However, in the rst level
kind of error is assumed to be systematic across each oor. before the data are employed, the same 200 samples generated
It should however be emphasized that probabilities/weights as- employing standard Monte Carlo simulation in Case 2 are used.
signed to each possibility are herein determined qualitatively and
based on judgment and experience. Table 2 demonstrates the 6. The results
parameters for constructing a prior probability distribution for
the steel yielding strength and concrete strength in compression Fig. 5a and b demonstrates the histograms for structural perfor-
as material properties. The prior probability distributions for con- mance variable (the demand to capacity ratio) corresponding to
crete and steel are modeled the same as in Case 1. the Case 1 (uncertainty in only material properties) and Case 2
Since in this case, there are additional sources of uncertainty (uncertainty in both material properties and defects) described
compared to in Case 1, the failure probability is predicted to be above, respectively. The lognormal PDFs t to the two histograms
large enough to be estimated using standard Monte Carlo simula- are illustrated in Fig. 5c. It can be observed that taking into account
tion. Hence, the reliability of the frame is calculated using the the uncertainty in structural defects leads to a signicant increase

Table 1
Probabilistic characterization of the structural detailing parameters.

Defects Possibilities Prob. Type


Insufcient anchorage (beams) Sufcient (100% effective) 0.900 Systematic over oor
Absent (50% effective) 0.100
Error in diameter (columns) /16 0.950 Systematic over oor and section type
/14 0.050
Superposition (columns) 100% of the area effective 0.950 Systematic over oor
75% of the area effective 0.050
Errors in conguration (columns) More plausible conguration 0.950 Systematic over oor and section type
Less plausible conguration 0.050
Absence of a bar (beams) Absence of a bar 0.100 Systematic over oor and section type
Presence of a bar 0.900
Conceret cover 2 cm 0.125 Systematic over oor
3 cm 0.750
4 cm 0.125
226 F. Jalayer et al. / Structural Safety 32 (2010) 220228

Prior Distribution
4
x 10 60 0.02
5 exp(ln f )=162
c
40 ln fc = 0.144
4.5 0.01
20
4
0 0
3.5 100 150 200 250 50 100 150 200 250
base shear [kg]

60 0.03
3 exp(ln f )=168

First Level
c
40 0.02 =0.1038
2.5 ln fc

20 0.01
2
0 0
1.5 100 150 200 250 50 100 150 200 250

100 0.04

Second Level
1 exp(ln f )=173
c
0.5 ln fc =0.0757
50 0.02
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
0 0
roof displacement [cm] 50 100 150 200 250
100 150 200 250
Fig. 4. Pushover curves generated for different realizations of the simulation Basement fc
procedure (Case 1).
Fig. 6. The distribution of the concrete strength in the basement: (a) the rst level
corresponds to the prior distribution for the concrete strength before test results are
Table 3 taken into account (uncertainty in both material properties and structural defects,
Test results available for the structure. Case 1). The histogram is based on 200 samples and the rightmost diagram
demonstrates the lognormal PDF t to the data, (b) the second level corresponds to
Test # data Type Standard error
the updated distribution of the basement concrete strength after the destructive
Drilled core test Basement 2 Destructive 0.15 test results for both concrete and steel are implemented, and (c) the third level
Drilled core test Ground oor 2 Destructive 0.15 corresponds to the updated distribution for basement concrete strength after the
Drilled core Fisrt oor 2 Destructive 0.15 non-destructive test results are also considered.
Ultrasonic test Basement 6 Non-destructive 0.335
Ultrasonic test Ground oor 6 Non-destructive 0.335
Ultrasonic test First oor 6 Non-destructive 0.335
Tension test reinforcing steel 1 Destructive 0.08 material properties and structural defects are taken into account.
The rst row of the gure represents the prior distribution for
the concrete strength before the test and inspection results are
150 Case 2 known or are taken into consideration. The rightmost column in
D/C=0.75
the gure represents the lognormal curves t to the histograms.
100 D/C=0.05
It can be observed that Bayesian updating of the concrete strength
50 based on the destructive test results (most accurate) leads to an in-
crease in the mean value for concrete strength and to a decrease in
0 its standard deviation. The addition of the non-destructive tests in
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1
the third level leads to further increase in the mean value and
60 Case 1 reduction the standard deviation.
D/C=0.82 Fig. 7 demonstrates the data for the steel yielding strength for
40 D/C=0.066
the case where the uncertainty in both material properties and
20 structural defects are taken into account. The second row repre-
sents the updated data for steel yield strength after the destructive
0 test results are implemented in the Bayesian framework. It can be
0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1
observed that the addition of test results (in this case only one data
10 point for steel strength) leads to an increase in the mean value and
case 2
case 1
to a decrease in the dispersion.
5 Fig. 8 demonstrates the histograms and the lognormal curves
tted to the demand to capacity ratio for the three increasing levels
0 of data. The rst level corresponds to the prior lognormal probabil-
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 ity distribution for the demand to capacity ratio before taking into
D/C Ratio consideration the test results. The second level corresponds to the
updated distribution after considering the drilled core test results
Fig. 5. (a) The histogram for the demand to capacity ratio when only the for concrete and the tension test result for reinforcing steel. The
uncertainty in the material properties are considered (Case 2, based on 500
last level illustrates the updated distribution for structural perfor-
samples). (b) The histogram for the demand to capacity ratio when both the
uncertainty in material properties and structural defects are considered (Case 1, mance variable after considering also the ultrasonic test results for
based on 200 samples). (c) The lognormal probability distribution functions t to concrete.
the demand to capacity ratio data presented in Case 2 (a) and Case 1 (b). For all three values of condence level suggested by the code
(i.e., FC = 1, 1.2, 1.35) the corresponding demand to capacity ratios
both in the mean value of the demand to capacity ratio and also in for the structure is calculated. The resulting three values for de-
the standard deviation for the demand to capacity ratio. mand to capacity ratio are marked on the curves illustrated in
Fig. 6 demonstrates the data for concrete strength at the base- Fig. 8. Note that the failure threshold is also marked at the value
ment of the structure for the case where the uncertainty in both ln Y = ln 1.0 = 0.0. The condence factors can be estimated, for
F. Jalayer et al. / Structural Safety 32 (2010) 220228 227

x 10
-3 exceedance corresponds to a condence factor between FC = 1.0
60 2 and FC = 1.20. In Fig. 8c, after the non-destructive test results are
f = 3200
Prior Distribution

y
1.5 f =240
also considered, the demand to capacity ratio with 5% probability
40 y of exceedance corresponds to a condence factor slightly greater
1 than 1.0 which corresponds to the code-recommended value for
20 the most complete level of knowledge.
0.5

0 0 7. Conclusions
2500 3000 3500 4000 2000 2500 3000 3500 4000
-3
x 10 This study aims to characterize, to quantify and to update the
80 3
First Level Updating

f = 3380 uncertain modeling parameters, namely, mechanical properties of


y
60 f =184 materials and the structural detailing specic to existing RC build-
2 y
ings, as a function of the amount of information available. The
40
motivation behind this research effort is to create a benchmark
1
20 against which the condence factors (FCs) recommended by inter-
national codes for seismic assessment of existing buildings can be
0 0 evaluated. A Monte Carlo Markov Chain (MCMC) algorithm is em-
2500 3000 3500 4000 2000 2500 3000 3500 4000
fy fy ployed inside the Bayesian updating framework in order to both
update the structural modeling parameters and reliability as a re-
Fig. 7. The distribution of the steel yielding strength: (a) the rst level corresponds sult of in situ tests and inspections. The Bayesian updating algo-
to the prior distribution for the steel yield strength before test results are taken into rithm implemented is general with respect to the type of in situ
account (uncertainty in both material properties and structural defects, Case 1). The test performed and the standard error associated with it. The pro-
histogram is based on 200 samples and the rightmost diagram demonstrates the
cedure is applied to the seismic assessment of an existing school
normal PDF t to the data and (b) the second level corresponds to the updated
distribution of the steel yield strength after the destructive test results for both structure in Italy which serves as case-study, given the code-based
concrete and steel are implemented. seismic action and capacity models.

Considering the uncertainty in structural detailing parameters


FC
C= 1 FC
C= 1.2 FC
C= 1.35 FA
AILURE
E increases both the mean and standard deviation of the demand
6
to capacity ratio for the structure.
prior

4 area=5% The quality and quantity of the acquired in situ information


2 affects signicantly the structural reliability assessment of the
existing building.
0
-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 With regard to benchmarking of the European Code FCs, an opti-
mal FC is dened as the value which leads to a demand to capac-
posterior 2nd level posterior 1st level

10 ity ratio with a specied probability of being exceeded (e.g., 5%).


For the case-study analyzed, a certain degree of consistency was
5 area=5%
= found between these optimal FCs, with respect to the code-
specied FC values dened as a function of increasing knowl-
0 edge levels.
-0.6
- -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2

15
Acknowledgments
10
area=5%
5 This work was supported in part by the second working group
within the project ReLUIS 2005/2008 Dipartimento della Protezi-
0
-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 one Civile. This support is gratefully acknowledged. Any opinions,
ln D/C ndings and conclusions or recommendations expressed in this
material are those of the authors and do not necessarily reect
Fig. 8. Distribution of the demand to capacity ratio when both the uncertainty in
those of the ReLUIS.
the material property and in the structural defects are taken into account: (a) the
prior lognormal PDF t to the demand to capacity ratio before test results are being
considered, (b) the updated lognormal PDF t to the demand to capacity ratio after References
u implementing the destructive test results, (c) the updated lognormal PDF t to the
data after the non-destructive test results are also implemented. The probability of
[1] Beck JL, Katagiotis LS. Updating models and their uncertainties. I: Bayesian
failure is indicated by the area under the curve for demand to capacity ratios greater statistical framework. J Eng Mech 1998;124(4).
than unity (0 on the logarithmic scale). The demand to capacity ratios correspond- [2] Beck JL, Au SK. Bayesian updating of structural models and reliability using
ing to the code-specied values for the condence factor are also marked on the Markov chain Monte Carlo simulation. J Eng Mech 2002;128(4).
gure. [3] BS 1881-124. Testing concrete. Methods for analysis of hardened concrete. The
British Standards; November 1988.
[4] Au SK, Beck JL. First excursion probabilities for linear systems by very efcient
importance sampling. Probab Eng Mech 2001;16:193207.
example, as the value of FC that leads to a demand to capacity ratio [5] Au SK, Beck JL. Subset simulation and its application to probabilistic seismic
performance assessment. J Eng Mech 2003;16(4).
with say 5% probability of exceedance. In the prior stage (Fig. 8a),
[8] CEN, European Committee for Standardisation TC250/SC8/. Eurocode 8: design
the condence factor corresponding to a value of demand to capac- provisions for earthquake resistance of structures, part 1.1: general rules,
ity ration with 5% probability of exceedance is larger than (but seismic actions and rules for buildings. PrEN1998-1; 2003.
close to) FC = 1.35. In Fig. 8b, where the distribution for demand [9] Fajfar P. Capacity spectrum method based on inelastic demand spectra.
Earthquake Eng Struct Dynam 1999;28:97993.
to capacity ratio is updated after the destructive test results are [10] Fardis MN. Seismic assessment and retrotting of existing buildings according
considered, the demand to capacity ratio with 5% probability of to Eurocode 8. In: Fifth national conference on earthquake engineering; 2003.
228 F. Jalayer et al. / Structural Safety 32 (2010) 220228

[11] Federal Emergency Management Agency (FEMA) 356. Pre-standard and Gazzetta Ufciale della Repubblica Italiana no. 105 del 8-5-2003. Suppl.
commentary for seismic rehabilitation of buildings. FEMA-356; November 2000. Ordinario no. 72; 2003 [in Italian].
[12] Freeman SA, Nicoletti JP, Tyrell JV. Evaluations of existing buildings for seismic [21] Ordinanza del Presidente del Consiglio dei Ministri (OPCM) no. 3431,
risk a case study of Puget sound naval shipyard, Bremerton, Washington. In: Ulteriori modiche ed integrazioni allordinanza del Presidente del
Proceedings of US national conference on earthquake engineering, Berkeley, Consiglio dei Ministri no. 3274 del 20 marzo 2003. Gazzetta Ufciale della
USA; 1975. p. 11322. Repubblica Italiana no. 107 del 10-5-2005. Suppl. Ordinario no. 85; 2005 [in
[13] Hastings WK. Monte-Carlo sampling methods using Markov chains and their Italian].
applications. Biometrika 1970;57(1):97109. [22] Ordinanza del Presidente del Consiglio dei Ministri (OPCM) no. 3519. Criteri
[14] Jaynes ET. Probability theory: the logic of science. Cambridge University Press; per lindividuazione delle zone sismiche e la formazione e laggiornamento
2003. degli elenchi delle medesime zone. Gazzetta Ufciale della Repubblica Italiana
[15] Jalayer F, Beck JL. Effect of the alternative representations of ground motion no. 108; 2006 [in Italian].
uncertainty on seismic risk assessment of structures. Earthquake Eng Struct [23] Regio Decreto Legge (R.D.L.) 2229. Norme per lesecuzione delle opere in
Dynam 2008;37(1):6179. conglomerato cementizio semplice o armato; 1939 [in Italian].
[16] Jalayer F, Franchin P, Pinto PE. A scalar decision variable for seismic reliability [24] Verderame GM, Manfredi G, Frunzio G. Le propriet meccaniche dei
analysis of RC frames. Earthquake Eng Struct Dynam 2007;36(13):205079 calcestruzzi impiegati nelle strutture in cemento armato realizzate negli
[Special issue on structural reliability]. anni 60. In: X Congresso Nazionale Lingegneria Sismica in Italia, Potenza
[17] Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E. Equations of Matera, 913 Settembre; 2001 [in Italian].
state calculations by fast computing machines. J Chem Phys 1953;21:108792. [25] Verderame GM, Stella A, Cosenza E. Le propriet meccaniche degli acciai
[18] Mander JB, Priestley MJN, e Park R. Theoretical stressstrain model for impiegati nelle strutture in cemento armato realizzate negli anni 60. In: X
conned concrete. J Struct Eng ASCE 1988;114(8):180425. Convegno Nazionale LIngegneria Sismica in Italia, Potenza e Matera, 913
[19] Min.LL.PP DM. Norme Tecniche per le Costruzioni (NTC). Gazzetta Ufciale Settembre; 2001 [in Italian].
della Repubblica Italiana, 29, 14 gennaio; 2008 [in Italian]. [26] UNI 9524:1989 + A1. Calcestruzzo Indurito. Rilievi Microsismici Mediante
[20] Ordinanza del Presidente del Consiglio dei Ministri (OPCM) no. 3274, Norme Impulsi d Onde Vibrazionali ad Alta Frequenza, in Campioni o Strutture di
tecniche per il progetto,la valutazione e ladeguamento sismico degli edici. Calcestruzzo Semplice, Armato o Precompresso; 1992 [in Italian].

You might also like