Probabilistic assessment of the global damage in reinforced concrete

structures
Y.F. Vargas, L.B. Pujades & A.H. Barbat
Universidad Politcnica de Catalua, Barcelona, Spain

J.E. Hurtado
Universidad Nacional de Colombia, Manizales, Colombia

ABSTRACT:
The main objective of this study is to evaluate the expected damage in reinforced concrete structures in a
probabilistic way using Monte Carlo simulation. To do that, the mechanical properties of the materials have been
considered random and a set of real accelerograms are scaled in such a way that the mean of their response
spectra is similar to the elastic response spectra of the Eurocode EC-08. Performing a series of nonlinear
dynamic analysis in the time domain, the probability density function of the maximum displacement at the roof
of the structure is obtained. Afterwards, the mean capacity curve of the structure is calculated in probabilistic
terms. The static and dynamic nonlinear structural analyses are performed with the program RUAUMOKO 2D.
The capacity curve allows defining the states of seismic damage of the structures. According to the
methodology proposed by the ATC 40, the expected seismic damage of a structure can be calculated from its
capacity spectrum and from the horizontal displacement demand at the roof level. In order to calculate the latter,
the equal displacement approximation will be applied and will be compared with the demand of the structure
calculated by means of the before mentioned nonlinear dynamic analysis. The fragility curves of the structures
are then calculated starting from the obtained damage states. The expected structural damage is finally obtained
entering in the fragility curves the spectral displacement obtained by using both, the ATC 40 method and the
nonlinear dynamic method. The results obtained from static and dynamic analyses are finally compared and
discussed from a probabilistic point of view.
Keywords: Reinforced concrete, expected damage, fragility curves, Monte Carlo simulation.

1. INTRODUCTION
In the evaluation of the vulnerability of structures by means of earthquakes, all the involved variables
are random, but they are rarely treated in this way. Due to the current computational capacity, a
massive quantity of calculations can be performed in order to study the behavior of structures from a
probabilistic standpoint. For example, in a reinforced concrete building, the strength of materials, the
loads, and the length of the structural elements among others variables are random. This fact combined
with uncertainty of the seismic hazard, may lead to underestimate or overestimate the actual response
of the structure. Hereby, this study focuses on the analysis of the uncertainties involved in the seismic
response of a reinforced concrete building. To do that, random samples of the mechanical properties
are generated by Monte Carlo method. To obtain the seismic demand of the area in probabilistic terms,
a response spectrum demand is chosen from the EUROCODE 8. Then, a procedure for finding
accelerograms whose response spectrum will be compatible with the selected spectrum is applied. In
this study we carried out two types of analysis, 1) Nonlinear static analysis (NLSA). 2) Nonlinear
dynamic analysis (NLDA). These have been compared in previous studies (see for instance, Mwafy et
al 2001). NLDA does not take into account the mechanical properties as random variables because the
coefficient of variation considered in this study causes that the influence will be negligible. The NLSA
is used to get the capacity curves of the structure and to obtain the expected displacement at the roof of
the building; this methodology is well known (Borzi et al 2007). The displacement obtained with this
procedure will be considered as a random variable and will be compared with the displacement
calculated via NLDA. Finally, the results are discussed from a probabilistic point of view.

2. DESCRIPTION OF THE BUILDING

The building selected to perform the analysis is shown in Figure 1. The location of the building is in
Spain, therefore, some selected accelerogram are taken from the Spanish accelerometric database, and
however, due to the low seismicity in the area, we use additional accelerograms taken from the
European database. The building is regular in plant, this regularity allows using the 2D approach. In
Spain, this type of building is frequently used for family housing. Nevertheless, most of these building
are not framed building; they do not have beams, this elements are replaced by a slab lattice. This kind
of slabs is called waffle slabs. For the purpose of this study we use a framed simplified equivalent
model. Table 1 contains the main features of the building. The Hysteresis rule of structural elements is
Elastoplastic, hardening or softening is not considered. In order to define the yield surfaces of the
columns and beams, is necessary to create a diagrams interaction between moment with axial load,
and moment with the angular deformation respectively. For this, some programs have been developed
in MATLAB in order to calculate the particular points that are necessary when defining the behavior
of structural elements in RUAUMOKO.

Figure 1. Frame of the reinforced concrete structure selected for the analysis
Table 1. Main features of the structure
No of levels
High
8
24

Wide
25.65

Period
1.444

3. DATA SIMULATION
3.1 Nonlinear Static Analysis
As mentioned before, the mechanical properties of the materials, specifically the concrete compressive
strength fc and the reinforced yield strength fy will be treated as random variables. The assumed
distribution for these variables is Gaussian; the values that define the distributions of the random
variables and the coefficient of variation are shown in Table 2. Is important to note, that in each
execution of a NLSA, the resistance of structural elements is not constant, because for each sample
new stochastic data is generated.
Table 2. Mean values of the probability distributions
Variable
(Kpa)
(Kpa)
Fc
30000
1000
Fy
411510
22093

(%)
3.33
5.36

1000 executions are performed with the NLSA. When NLSA is performed, the results change
depending on the loading pattern in the height, besides it is very difficult to establish how much to
increase the load and the invariant load distribution cannot consider higher-mode effects (Poursha et

al.2008). To avoid these problems, we use the so calls Adaptative Pushover (PA) method in its version
proposed by Satyarno (Satyarno 1999). The advantage of the PA is the independency of the results of
the loading pattern, because this is calculated as a function of the mass, the equivalent frequency and
displaced shape of the structure, furthermore, the horizontal load limit is controlled by the current
stiffness of the structure. Capacity curves obtained after executing the 1000 PA are shown in Figure 2.
This Figure reflects the uncertainty in the results caused by the random approach.

Figure 2. Capacity curves

3.2. Nonlinear dynamic analysis.

To consider the randomness in the seismic action, the EUROCODE 8, Type 1 for soil type D is taken
as target spectrum. Although we performed several test with the type 2 spectrums, the type 1, soil D
spectrum is used in order to analyze the nonlinear inelastic behavior of the structure. We refer to this
spectrum as code spectrum. A preliminary goal is to select a series of accelerograms with mean elastic
response spectrum compatible with the target spectrum, 20 acceleration records are selected whose
mean 5% damped elastic response spectrum is in the range of +/- 5% of the code spectrum. In Figure
3, the code spectrum and the mean spectrum corresponding to the 20 selected accelerograms are
shown.
Elastic Response Spectrum Type 1 Soil D
Elastic response spectrum
Mean Spectrum selected
1

Sa (Kn)

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

1
1.2
Period (s)

1.4

1.6

1.8

Figure 3. Comparison between Spectrum of EUROCODE 8 and mean

The selected accelerograms are then used for a series of NLDA by scaling them to different levels of
peak ground acceleration (PGA). This procedure is known as incremental dynamic analysis (IDA)
(Vamvatsikos et al 2001). The objective is to obtain the evolution of a variable as a function of
increased PGA. In this case, the PGA is increased to 0.25 g, which is the maximum value for Spain. In
the NLDA we found that the structural response is almost insensitive to the variation of its mechanical
properties shown in Table 2. In the IDA the variable that is matched with the PGA is the expected

spectral displacement (ESD), obviously, as the seismic demand was obtained as a random variable, the
(ESD) will also be, so the values shown are the mean values. Figure 4 shows the variation of (ESD),
when PGA increases, together with +/-1.65 standard deviation intervals (Reliability 95%). Figure 5
shows the evolution of the standard deviation of the ESD. In the Figure 4, there is one major change in
the slope of the curve approximately at 0.16 g, therefore significant changes in the behaviour of the
structure cannot be appreciated in this graph. However, Figure 5 shows three points related to
significant changes in its shape (see Figure 5 and Table 3) the first point is related to the first change
of the slope of the curve. The second one is the first maximum and the third is related to the beginning
of straight line with maximum slope, Later on, we use these points to discuss the behaviour of the
structure.
PGA-ESD

ESD-Standard deviation of ESD

0.35

0.06

0.3
Standard deviation of ESD (m)

0.05

ESD (m)

0.25

0.2

0.15

0.1

0.05

Mean curve
95% Reliability
5% Reliability
0

0.05

0.1

0.15

0.2

0.25

0.04

0.03

0.02

0.01

Figure 4. Relation PGA and mean ESD

0.05

0.1

0.15

0.2

0.25

ESD (m)

PGA (g)

Figure 5. mean ESD and standard deviation

Table 3. Coordinates of the slope change points

Point
X coordinate (m)
Y coordinate (m)
1
0.13
0.0270
2
0.16
0.0294
3
0.23
0.0298

4. CAPACITY SPECTRUM, DAMAGE STATES AND FRAGILITY CURVES

4.1. Capacity spectrum and damage states
The capacity spectrum is the spectral acceleration-spectral displacement (Sa-Sd) representation of the
capacity curve. Which is often used in its simplified bilinear form, defined by the yielding point
(Dy,Ay) and the ultimate capacity point (Du,Au). Figure 6 shows a capacity spectrum of the RC
building and its bilinear representation In order to analyze expected damage we use simplified
methods allowing to obtain damage states (Ds) thresholds and fragility curves. Four non-null damage
states are considered: (1) slight, (2) moderate, (3) severe and (4) extensive-to-collapse. For a given
damage state, the damage state threshold is defined by the 50% exceedance probability. This damage
state threshold can be defined from the bilinear capacity spectrum, in a simplified way, as follows:
(Lantada et al 2008). In this way, damage states thresholds have been calculated for all the obtained
capacity spectra. Thus, being considered as random variables, Figure 7 shows all the results and the
mean values of each damage state threshold. This figure shows how the dispersion increases with
increasing damage states. This fact may indicate that when the structure enters into nonlinear
behaviour, the damage level has greater uncertainties. The mean and the coefficient of variation of Sa
and Sd of each damage state, are shown in Table 4. It is important noting the agreement with the
values in Table 3 and Table 4. In Table 3, points 1, 2 and 3 correspond to the mean values for the
damage states 2, 3 and 4.

Capacity spectrum and bilinear aproximation

Relation between Spectral acceleration and spectral displacement

0.35

0.35

(Du,Ay)
0.3

0.25

0.25

Spectral acceleration (g)

Spectral acceleration (g)

(Dy,Ay)
0.3

0.2

0.15

0.1

0.05

0.2

0.15

0.1
Ds1
Ds2
Ds3
Ds4

0.05

Capacity spectrum
Bilinear aproximation
0.05

0.1
0.15
0.2
Spectral displacement (m)

0.25

0.3

Figure 6. Capacity spectrum and bilinear curve

0.05

0.1
0.15
0.2
Spectral Displacement (m)

0.25

Figure 7. Damage states as random variables

Ds1 0.7 * Dy. Ds 2 Dy. Ds3 Ds 2 0.25 * ( Ds 4 Ds 2 ). Ds 4 Du.

Table 4. Main features of the structure

0.3

Ds

sd

sd

sa

sa

1
2
3
4

0.0919
0.1314
0.1583
0.2212

0.0129
0.0129
0.0307
0.0668

0.1754
0.2506
0.2531
0.2591

0.0199
0.0199
0.0200
0.0208

(1)

4.2 Fragility curves

For each damage state, the corresponding fragility curve (FC), is defined by the probability of being
exceeded as a function, in our case, the spectral displacement. It is assumed that fragility curves follow
a standard lognormal cumulative distribution function. Each fragility curve is then obtained from the
equation 2, the mean values can be determined from the capacity spectrum and standard deviations can
be estimated by assuming that damage distribution follow a binomial distributions (See Lantada et al.
2008). Figure 8 shows the fragility curves obtained for all the capacity spectra, obviously, according to
Figure 7, as the considered damage state increases the uncertainties involved in the corresponding
fragility curve increases. See fragility curves corresponding to damage sates slight and collapse in
Figure 8. Figure 9 shows the results of a sensitivity test on the influence of the mechanical properties
of materials and the non-damage states thresholds. The stiffness is used as a variable in this test. DS1
and DS2 damage states are practically independent of the stiffness, while for DS4 and DS3 the spectral
displacement decreases with increasing stiffness indicating that the probability of the corresponding
damage states increases with the stiffness. Figure 10 shows the mean fragility curves, and Figure 11
shows the corresponding standard deviations as a function of Sd. This Figure clearly depicts the
dependence of the uncertainties of the damage grades. For instance, the uncertainty of the probability
of exceedance of the DS4 may be greater than 10%, which means that for a reliability of 95% the
increase in the probability of failure will be grater than 16.5%, this increase reaffirms the importance
of working the problem from a probabilistic viewpoint.

1
Sd

PDs / Sd
Ln

Sd
Ds
,

Ds

(2)

Fragility Curves

Sensitivity analysis for the Ds

1
0.3

0.9
0.8

0.25

0.7
0.2
Sd (m)

P[Ds/Sd]

0.6
0.5

0.15

0.4
0.1

0.3
0.2

FC1
FC2
FC3
FC4

0.1
0

0.1

0.2
0.3
0.4
Spectral displacement (m)

0.5

Ds1
Ds2
Ds3
Ds4

0.05

0
7800

0.6

Figure 8. Fragility curves like random variables

7900

8000

8100
8200
Stiffness (Kn/m)

Mean fragility Curves

Standard deviation of P[Ds/Sd]

0.12

0.9
0.1

0.8
0.7

P[Ds1/Sd]
P[Ds2/Sd]
P[Ds3/Sd]
P[Ds4/Sd]

0.08

P[Ds/Sd]

0.6
P[Ds/Sd]

8400

Figure 9. Sensitivity analysis for the Ds

0.5
0.4

0.06

0.04

0.3
0.2

FC1
FC2
FC3
FC4

0.1
0

8300

0.1

0.2
0.3
0.4
Spectral displacement (m)

Figure 10. Mean fragility curves

0.5

0.6

0.02

0.1

0.2
0.3
0.4
Spectral displacement (m)

0.5

0.6

Figure 11. Standard deviation of the fragility curves

5. EXPECTED SPECTRAL DISPLACEMENT AND DAMAGE INDEX

Expected spectral displacement (ESD), is the maximum spectral displacement expected in a building
due to the seismicity of the area, in this study, this value was obtained in section 4 through NLDA, the
results were presented in Figure 4 and Figure 5. Different studies have searched for simplified
procedures to estimate the ESD, (Kim et al 2008), a much more simplified procedure is the so called
equal displacement approximation, (EDA) which is well described in ATC-40. (See also Mahaney
1993). The EDA is performed by using the spectra corresponding to the selected accelerograms in
order to a better comparison with the results coming from the NLDA. In order to obtain the ESD as a
function of PGA the spectra are scaled to 0.25 g obtaining the mean and standard deviation (SEDR) of
ESD. These results are shown in Figure 12 and Figure 13 respectively. These figures also show the
NLDA results. The main conclusion of this analysis is that the EDA methodology is an adequate
approximation for the ESD in the building, because this not undervalue the expected displacement,
moreover from a probabilistic viewpoint this method is also conservative because, in the nonlinear
range, the standard deviation with EDA is higher than with NLDA. In the following, we analyze the
damage index (DI), which is defined by the equation 3.

DI

1 4
iP(DS i )
4 i 0

(3)

PGA-ESD

PGA-Standard deviation of ESD

0.35

0.06

0.3
Standard deviation of ESD (m)

0.05

0.25

ESD (m)

0.2

0.15

0.1

0.03

0.02

0.01

0.05

0.04

NLDA
EDA
0

0.05

0.1

0.15

0.2

0.25

NLDA
EDA
0

0.05

0.1

PGA (g)

0.15

0.2

0.25

PGA (g)

Figure 12. Comparison between function PGA-ESD

Figure 13. Comparison between function PGA-SEDR

P(DSi) is the probability of the damage state i and can be easily calculated from fragility curves.
Figure 14 shows this procedure for calculating these probabilities. DI can be understood as the
normalized mean damage grade and it is a measure of overall damage in the structure. DI can be also
plotted as a function of ESD. Thus, the DI can be calculated for any spectral displacement but the
comparison of DI obtained with EDA and NLDA, requires computing the PGA corresponding to each
ESD. Figure 15 shows the obtained results: Namely, mean values and 95% reliability curves. Again,
our results confirm that the EDA is conservative respect to NLDA, even when considering a reliability
of 95% for random variables, but, if the variables were no treated from a probabilistic approach, this
would result in an underestimation of the actual damage that may occur in the building. In the case of
the building analyzed in this work, the damage index estimated by using a deterministic approach
could be 0.25 lesser than the one computed from a probabilistic viewpoint

Probability of each damage state

1

Damage Index obtained with EDA and NLDA

1

P Null

0.9

0.9
P Slight

0.8

0.7

0.7

0.6

0.6

0.5

DI

PDs/Sd

0.8

P Moderate

0.5

0.4

0.4

0.3

0.3

0.2

0.2

P Extensive

0.1
0

NLDA
EDA
NLDA 95% Reliability
EDA 95% Reliability

0.1
P Complete
0

0.05

0.1

0.15

0.2

0.25

ESD

Figure 14. Probability of Ds depending ESD

0.05

0.1

0.15

0.2
PGA

0.25

0.3

0.35

0.4

Figure 15. DI obtained with NLDA, EDA

6. DISCUSSION AND CONCLUSIONS

This paper contains the assessment of the vulnerability fragility and expected damage in a
reinforced concrete building. However, the results here obtained go further and they concern
to the comparison of nonlinear static and dynamic analysis procedures. Here, we use both
types of analysis for the calculations. Furthermore, we face the problem from a probabilistic
point of view, since the parameters related to the mechanical properties of the materials and
the seismic demands are considered as random variables. A first hint is that notwithstanding
that NLDA is a powerful tool in the assessment of the structural behaviour of buildings when
submitted to seismic actions, this procedure has little or no sense if the seismic demand is not

carefully and properly selected. We put special care in the selection of the accelerograms used
in this study. We have selected accelerograms corresponding to seismic events from the
Spanish and European strong motion records databases. In order to observe a wide range of
spectral displacements the Eurocode type 1 design spectrum for soils type D has been taken as
target demand spectrum. The accelerograms have been selected according to this criterion and
have been scales to cover PGA values until 0.25 g. We use standard pushover analysis to
obtain probabilistic capacity curves. A modified adaptive technique has been used to define
the horizontal incremental load limit to automatically stop the pushover analysis in a great
number of massive runs. Starting from capacity spectra, simplified methods allow obtaining
damage states thresholds and probabilistic fragility curves. An interesting conclusion of this
exercise is that uncertainties increase in the nonlinear range. For the collapse damage state,
the uncertainties in the fragility curves may be greater than 10%. EDA and NLDA are used to
obtain the expected spectral displacement ESD and their standard deviations as a function of
the PGA. Again, uncertainties increase with increasing PGA. This fact can be attributed to the
increase of the inelastic behaviour of the building. EDA is a successful approach because does
not underestimate the actual displacement, but in structures with higher ductility can be much
conservative. Furthermore, the fact that both ESD and SEDR are greater in EDA than in
NLDA confirms that EDA is conservative. In the NLDA, the seismic action is the main
responsible for uncertainties in the spectral response, being less significant the influence of
the uncertainties in the mechanical properties of the building. However, as the damage state
increases a sensitivity test shows a correlation between stiffness and spectral displacement.
For damage states DS3 and DS4 the spectral displacement decreases with increasing stiffness
indicating that the probability of the corresponding damage state increases with the stiffness.
This result is important since the damage states 3 and 4 have a high influence in the
calculation of the damage index. Finally, the comparison of the damage index as a function of
PGA and the corresponding uncertainties shows that, for damage states from severe to
collapse and for a reliability of 95%, the uncertainties in the damage index may be higher than
0.25 units or 42% of the damage index. Thus, perhaps the most important conclusion is that,
both static and dynamic structural analyses should be faced by using probabilistic approaches.
REFERENCES
ATC-40. 1996. Seismic evaluation and retrofit of concrete buildings, Applied Technology Council, Redwood
City, California
Borzi B, Phino R, H. Crowley 2007. Simplified Pushover analysis for large-scale assessment of RC buildings.
Engineering Structures 30 (2008) 804-820.
Eurocode 8. 2004. Design of structures for earthquake resistance. Part 1: General rules, seismic actions and rules
for building
Kim S.P & Y.C. Kuruma 2008. An alternative pushover analysis procedure to estimate seismic displacement
demands. Engineering structures 30 (2008) 3793-3807.
Lantada N, Pujades LG & AH. Barbat. Vulnerability index and capacity spectrum based methods for urban
seismic risk evaluation. A comparison. Natural Hazards 51:501-524
MATLAB. The mathworks inc.
Mahaney J.A. Paret T.F. Kehoe B.E. & S.A. Freeman, 1993 The capacity spectrum method for evaluating
structural response during the Loma Prieta earthquake. National earthquakes conference, Memphis.
Mwafy A.M & A.S Elnashai. 2001. Static pushover versus dynamic collapse analysis of RC buildings.
Engineering Structures 23 (2001) 407424.
Poursha M, Khoshnoudian F & AS. Moghadam. 2009. A consecutive modal pushover procedure for estimating
the seismic demands of tall buildings. Engineering Structures 31 (2009) 591-599
Vamvatsikos D and CA. Cornell 2002. The Incremental Dynamic Analysis and its application to PerformanceBased Earthquake Engineering. Proceeding of the 12th European Conference on Earthquake Engineering,
London September 2002