SECED 2015 Conference: Earthquake Risk and Engineering towards a Resilient World

9-10 July 2015, Cambridge UK

USE OF INCREMENTAL DYNAMIC ANALYSES FOR NUCLEAR

FRAGILITY ASSESSMENT
Barbara Polidoro1, Barnali Ghosh2 and Ming Tan3

Abstract: A fragility curve expresses the conditional probability of failure of a structure for a
given input motion parameter, such as the peak ground acceleration or spectral acceleration.
In nuclear engineering practice, fragility curve is evaluated by using margin factors. This
approach represents a very convenient method but it adopts strong simplifying hypotheses.
In the last years, fragility curves have also become very popular for assessing the seismic
vulnerability of civil structures and one of the best current practices is the use of non-linear
dynamic analyses. In this context, incremental dynamic analyses have been used in
conjunction with a pushover analysis to assess a building used as a storage facility for
vehicles and equipment, within a nuclear power plant in UK. The main functional requirement
of the building is to provide a facility to store and maintain back-up and emergency response
equipment that may be required by emergency responders for the nuclear power station.

Introduction
Following the Tohuku earthquake in Japan on 11 March  2011,  a  review  of  the  UK’s nuclear
industry has been carried out by the Office for Nuclear Regulation (ONR) using the lessons
learnt from the Tokyo Electric Power Company (TEPCO) Fukushima-Daiichi station. This
review stated that there were no fundamental   safety   weaknesses   in   the   UK’s   nuclear  
industry, but also concluded that using the lessons learnt from the event the industry can be
made even safer. A document   titled   ‘Japanese   earthquake   and   tsunami:   Implementing   the  
lessons for the UK's nuclear industry’  (ONR, 2012) was produced after the ONR review. This
document identifies the requirement for an emergency equipment store.
The main functional requirement of the building is to provide a facility to store and maintain
back-up and emergency response equipment that may be required by emergency
responders for the nuclear power station. Vehicles and equipment may be required to
respond to Beyond Design Basis (BDB) events, Design Basis events, and to non-nuclear
related events on site. In particular, the facility has to be able to perform its main functions
following a BDB event, and the equipment stored within the facility must be protected so that
it is able to function following the same event. The design life of the building is 70 years.
It is therefore necessary to establish and quantify the applicable BDB events to ensure the
building design demonstrates resilience. The most common method of quantifying a BDB
event is by defining the Design Basis (DB) event and then adding a suitable margin. The
BDB indicates that the building design should demonstrate that there is a High Confidence
(at 95th percentile) of Low Probability (less than 5%) that the building can fulfil its minimum
functions, following a Peak Ground Acceleration (PGA) of 4m/sec2. To reach this purpose a
fragility curve, which expresses the conditional probability of failure of a structure given a
ground motion intensity measure is required.
The most common approach adopted in the nuclear engineering practice to carry out a
fragility curve is   a   simplified   method   called   “response  factor   method”   (Reed et al., 1994). It
represents a very convenient method to calculate the fragility curves, however, it adopts
strong simplifications.
During last years, fragility analyses have become very popular also in the civil engineering
applications to evaluate the vulnerability of structures. Within this context, the growth in

1
Seismic Hazard Specialist, Mott MacDonald, London (UK), barbara.polidoro@mottmac.com
2
Senior Principal Engineer, Mott MacDonald, London (UK), barnali.ghosh@mottmac.com
3
Senior Principal Analyst, Mott MacDonald, Altrincham (UK), ming.tan@mottmac.com
 B. POLIDORO, B. GHOSH and M. TAN

computer processing power has made possible the use of more complex and accurate
analyses. Hence, the state of the art has moved from static to non-linear dynamic analyses.
This paper is structured in a way that, the response factor method, usually adopted to carry
out fragility curves is described first. Then, the procedure to calculate the fragility curve
adopting non-linear dynamic analyses is illustrated and numerical results are provided for the
building in terms of fragility curves at different confidence levels.
Because of the confidentiality of the project, specific information about the location and
geometry of the building will not be provided.

The response factor method

A fragility curve expresses the probability of failure (𝑃 ) of a structure as a function of an
intensity measure (e.g., peak ground acceleration (PGA) or Spectral acceleration (S )). The
failure probability conditioned on a ground motion parameter, 𝑎, is given by the cumulative
distribution of the capacity, A, that is considered as a random variable. In fact, a structure
fails if its capacity is equal or less than a given ground motion value.
The functional form usually adopted to describe a fragility curve is the lognormal distribution
(Equation 1) which is defined by two parameters: the median A and the logarithmic
standard deviation β . In the same equation, ϕ(∙) represents the standard Gaussian
cumulative distribution function.
( / )
( / )
𝑃 | (𝑎) = ∫ 𝑒  𝑑𝑥 = 𝜙 (1)
√

Also the response factor method (Reed and Kennedy, 1994) adopts the previous formulation,
but it also distinguishes between epistemic uncertainty (that accounts for the lack of
knowledge about the procedure) and aleatory randomness (e.g., the variability of the 𝑃𝐺𝐴 for
an earthquake of a certain magnitude and at a certain location) by introducing the log-
standard deviations 𝛽 and 𝛽  respectively. In this context, the capacity can be expressed as
in Equation 2, where  𝜀 and 𝜀 are log-normally distributed random variables with median
equal to one and respective log-standard deviations 𝛽    and 𝛽 .
𝐴 = (𝐴 ∙ 𝜀 ) ∙ 𝜀 (2)
Starting from the above hypotheses, it is possible to define a family of fragility curves that
refer to different confidence levels, Q (Equation 3).
( / ) ( )
 𝑃′ | (𝑎) = 𝜙 (3)

Equation 1 (which represents the mean curve) and Equation 3 (which for 𝑄 = 0.5  provides
the median curve) are linked by the following expression (Equation 4).

𝛽≡𝛽 = 𝛽 +𝛽 (4)
In this method, in order to evaluate the fragility parameters, intermediate random variables
(i.e., safety factors) are used. In particular, the capacity, 𝐴,  is expressed as in Equation 5
where adesign is the 𝑃𝐺𝐴 of design and 𝐹 are the random margin factors, lognormally
distributed (with median 𝐹  and log-standard deviation 𝛽 ) that account for the conservatism
and uncertainty in structural response and capacity calculations.
𝐴 = (∏ 𝐹 )𝑎   (5)
Thus, it results that the median capacity can be calculated as  𝐴 = 𝐹 ∙ 𝑎 with  𝐹 =
∏ 𝐹  while the log-standard deviation is given by Equation 6.

𝛽= ∑ 𝛽  𝑤ℎ𝑒𝑟𝑒  𝛽 = 𝛽 + 𝛽   (6)

2
 B. POLIDORO, B. GHOSH and M. TAN

Because of its simplicity, this approach is very very convenient. As this method is based on
strong hypothesis (e.g. lognormally distributed safety factors) and expert judgment it may not
provide robust results.

Fragility curve using nonlinear dynamic analysis

Currently, non-linear dynamic analyses (e.g., Incremental Dynamic Analyses, IDAs) have
become very popular in the civil engineering applications to calculate fragility curves of
structures. These analyses allow evaluation of the seismic response of a structure and in
conjunction with a structural model, allow the calculation of the probability of failure, that is,
the probability of exceedance of the seismic capacity, for different intensity levels. These
probabilities (that already represent an empirical fragility curve) may also be approximated by
a lognormal distribution (Equation 1) whose parameters can be calculated through a
regression process.

Incremental Dynamic Analysis

An IDA (Vamvatsikos and Cornell, 2002) involves a series of non-linear dynamic analyses of
the structural model under a set of ground motion records, each scaled to several intensity
levels ideally selected to cover the whole range from elastic to non-linear and finally to
collapse of the structure. In particular, the goal of the analysis is to record the damage state
of the structure measured by an engineering demand parameter (EDP, e.g., peak roof drift
ratio or inelastic displacement) for each intensity level which is measured by an intensity
measure (IM, e.g., peak ground acceleration or the 5% damped first-mode spectral
acceleration 𝑆 (𝑇 )). Results can be processed to get the distribution of demand EDP given
the intensity level IM. Hence, the fragility curve can be assessed by calculating for each IM
the probability of exceedance of the seismic capacity.
The most important issue to conduct IDAs is selecting a suitable IM and EDP (Luco et al.,
2007). There are several issues of efficiency and sufficiency associated with the IM selection.
Sufficiency is defined as the independence of the distribution of EDP given the IM from any
other seismological parameters that may characterize the ground motion (e.g., duration,
magnitude or spectral shape). A sufficient IM accounts for all seismological information
needed to determine the effect of a ground motion record on the structure being investigated
and this permits a linear scaling of records to reach the intensity level considered in the
analysis. An efficient IM instead, minimizes the scatter of results, therefore it is required a
reduced number of ground motion records to provide good demand and capacity estimates.
For first-mode-dominated structures, the 5% damped first-mode spectral acceleration,
𝑆 (𝑇 ), is chosen as standard IM (Shome et al 1998, Shome and Cornell 1999).
On the other hand, the EDP has to be selected in order that it can well represent the damage
of the structure. Hence, the peak storey accelerations are usually adopted to describe
contents’   damage,   while   the   maximum   peak   interstorey   drift   ratio   or the inelastic
displacement are used to describe global dynamic instability and several structural
performance limit-states.

Structural model
In order to evaluate the capacity of the structure, a pushover analysis can be adopted (e.g.,
D'Ayala et al. 2014). The pushover analysis of a structure is a static non-linear analysis
under permanent vertical loads and gradually increasing lateral loads (incremental loads) up
to failure. The equivalent static lateral loads approximately represent inertial forces induced
by the earthquake. The output of this analysis is therefore the capacity curve (force-
displacement relationship) of a Multi Degree of Freedom (MDoF) system with the estimated
damage states.
In order to conduct the fragility assessment, this curve has to be converted into a capacity
curve for the equivalent Single Degree of Freedom (SDoF) system. The definition of the
equivalent SDoF from a pushover curve can be obtained by using the approach followed in
the classical N2 method (Fajfar, 2000) and implemented in Eurocode 8 (CEN, 2004).

3
 B. POLIDORO, B. GHOSH and M. TAN

This transformation is made by dividing the base shear and displacement of the MDoF
system with a transformation factor (the first mode participation factor, 𝛤 ) that can be
calculated according to Equation 7, where 𝑚∗   is the mass of the equivalent SDoF (defined as
𝑚∗ = ∑ 𝑚 𝜙 ) while 𝜙 is the fundamental mode shape.
∗
𝛤=∑   (7)

For the calculation of the equivalent mass 𝑚∗   and the factor 𝛤, the assumed displacement
shape 𝜙 is normalized, (i.e., the value at the top is equal to 1).
Then, in order to perform non-linear dynamic analyses, the pushover curve for the equivalent
SDoF has to be idealized. In particular, depending on the shape of the pushover curve, this
can be idealized by a bilinear (an elastic-plastic) or multilinear (elastic-plastic with residual
strength) model. To derive the idealized capacity diagram of the SDoF system, the equal
energy principle can be used, that is, the idealized curve is determined by imposing that the
areas under the actual curve of SDoF and the idealized curve are equal.

Regression process
Once the non-linear dynamic analyses have been performed, the conditional distribution of
the EDP given the intensity level, IM, can be calculated. Hence, the probability of failure, that
is, the probability of exceedance of the seismic capacity for each IM can be assessed.
These probabilities already represent an empirical fragility curve. However, results may also
be fitted by a cumulative lognormal distribution (Equation 1) whose parameters can be
assessed by adopting a regression process (Porter, 2007).
In fact, it is possible to convert Equations 1 to a linear regression problem by taking the
inverse Gaussian cumulative distribution function of each side and fitting a line (Equation 8)
to the data.
𝑦 = 𝑠𝑥 + 𝑐   (8)
In the above equation, s is the slope of the trend line and 𝑐 is the value of 𝑦 where the line
has a x -value of 0 (the intercept). Parameters of fragility curve are related to the fitting line;
in fact, the 𝛽 value correspond to 1/𝑠 , while 𝐴 = −𝑐/𝑠 . Obviously, parameters can be
defined also for different confidence levels.

Results for the case study

Incremental Dynamic Analyses and results of the pushover analyses have been used to
perform the fragility assessment of a storage facility building within a nuclear power plant in
UK. The structure is a regular one floor steel building and has been modeled through the
SAP2000 software. The collapse limit state has been defined according to ASCE/SEI 43-05,
(i.e., it is assumed that the structure reaches the failure when the inelastic energy absorption
factor 𝐹 reaches a value equal to 2).
The elastic modal analysis indicates that the first mode participating mass (𝑚 ) is equal to
82% for the longitudinal direction and equal to 95% for the transversal direction, that is, the
building can be considered as a first-mode-dominated structure. Hence, the 5% damped first-
mode spectral acceleration 𝑆 (𝑇 ) has been selected as intensity measure (sufficient and
efficient given the characteristic of the structure) for conducting the non-linear dynamic
analyses, whereas the inelastic displacement (𝛥 ) has been chosen as EDP.
In order to conduct the fragility assessment, the first mode participation factor (𝛤) has been
evaluated for each direction (according to Equation 7) to transform the pushover curve of the
MDoF system to the capacity curve of the equivalent SDoF (both displacement and base
shear have been divided by 𝛤). According to the shape of the pushover curve, a bilinear
approximation, considering an hardening behaviour after the achievement of the yielding
force, 𝐹 (De Luca et al. 2013) has been used. Figure 1 and Figure 2 show the pushover
curves (idealized) for the equivalent SDOF systems, for the longitudinal and transversal
directions respectively.
4
 B. POLIDORO, B. GHOSH and M. TAN

1800
Longitudinal direction
1600

1400

1200
Force [kN]

1000

800

600

400

200

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Displacement [m]
Figure 1. Linearized pushover curve for the equivalent SDOF (longitudinal direction)

1800
Transversal direction
1600

1400

1200
Force [kN]

1000

800

600

400

200

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Displacement [m]

Figure 2. Linearized pushover curve for the equivalent SDOF (transversal direction)

Table 1 summarizes for each direction, the characteristics of the equivalent SDOF systems,
that is: the first mode participation factor (𝛤), the yielding force (𝐹 ), the capacity force at
collapse (𝐹 ), the yielding and ultimate displacements (𝑑 and 𝑑 respectively), the hardening
value (ℎ), the first mode participating mass (𝑚 ) and finally the stiffness (𝑘 ∗ ), mass (𝑚∗) and
period (𝑇 ∗) of the equivalent SDoF.

5
 B. POLIDORO, B. GHOSH and M. TAN

Table 1. Parameters of the equivalent SDOF

Parameters of the equivalent SDoF Longitudinal direction Transversal direction

𝛤 1.368 1.2
𝐹  [𝑘𝑁] 1465 1518
𝐹  [𝑘𝑁] 1640 1732
𝑑  [𝑚] 0.013 0.012
𝑑  [𝑚] 0.020 0.018
ℎ 0.227 26750
𝑚 [%] 82 95
𝑚∗  [𝑘𝑁 ∙ 𝑠𝑒𝑐 ⁄𝑚] 130.583 100
𝑘 ∗  [𝑘𝑁⁄𝑚] 111832 126500
∗
𝑇  [𝑠𝑒𝑐] 0.215 0.177

The set of ground motions used to perform the non-linear dynamic analyses is based on the
FEMA P695 far-field ground motion set, which includes 22 record pairs, each with two
horizontal components for a total of 44 ground motions. Those ground motions are recorded
at sites located greater than or equal to 10 km from fault rupture; event magnitudes range
from M 6.5 to M 7.6 with an average magnitude of M 7.0.
The scaling factor to be applied at each record for each IM has been calculated through the
evaluation of the exact spectral response at the period of the SDoF by using OpenSees and
Matlab software.
Since the structure can be defined regular, IDAs have been performed independently in the
two directions. Figure 3 and Figure 4 show the IDAs results for the longitudinal and
transversal directions respectively.

30

25

20
Sa(T1) [m/sec ]
2

15

10

0
0 0.02 0.04 0.06 0.08 0.1 0.12
in [m]
Figure 3. Incremental Dynamic Analysis for the longitudinal direction

6
 B. POLIDORO, B. GHOSH and M. TAN

30

25

20
Sa(T1) [m/sec ]
2

15

10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
in [m]
Figure 4. Incremental Dynamic Analysis for the transversal direction

Once the IDAs have been performed, the conditional probability density function (assumed
as lognormal) of the inelastic displacement, Δin, given each intensity level has been
evaluated. Hence, the probability of failure, that is, the probability of exceedance of the
seismic capacity has been assessed.
The empirical fragility curve is then fitted with a lognormal distribution (Equation 1) whose
parameters have been defined by using the regression procedure described in the previous
section.
Curves have been calculated for different confidence levels (i.e., for the 50th 5th and 95th
percentile). Figure 5 and Figure 6 show the family of fragility curves obtained for the
longitudinal and transversal directions respectively.

1
50th Percentile
0.9 5th Percentile
95th Percentile
0.8

0.7

0.6

0.5
Pf

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30 35 40 45 50
Sa(T1) [m/sec2]
Figure 5. Fragility curves for the longitudinal direction

7
 B. POLIDORO, B. GHOSH and M. TAN

1
50th Percentile
0.9 5th Percentile
95th Percentile
0.8

0.7

0.6

0.5
Pf

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30 35 40 45 50
Sa(T1) [m/sec2]
Figure 6. Fragility curves for the transversal direction

In order to perform the fragility assessment of the building, the seismic demand, that is, the
design 𝑆 (𝑇 ) value of the reference code spectrum (with a 𝑃𝐺𝐴 = 4  𝑚/𝑠𝑒𝑐 ) has been
evaluated for each direction. Values are equal to 8.2 m/sec2 for the longitudinal direction and
10.2 m/sec2 for the transversal direction.
For these values, the probabilities of failure have been calculated considering the fragility
curves at 95th percentile. The worst condition (reached for the longitudinal direction) has
provided a probability of failure equal to P = 2%. This result shows that the design criteria of
the building respect the High Confidence of Low Probability (less than 5%).

CONCLUSIONS
A fragility curve expresses the conditional probability of failure of a structure for a given input
motion parameter, such as the peak ground acceleration or spectral acceleration. In nuclear
engineering practice, fragility curve is usually evaluated by using margin factors. This
approach represents a very convenient method but it adopts strong simplifying hypotheses.
In the last years, fragility curves have also become very popular for assessing the seismic
vulnerability of civil structures and one of the best current practices is the use of non-linear
dynamic analyses.
In this context, incremental dynamic analyses (IDAs) have been used to perform the fragility
assessment of a building used as a storage facility for vehicles and equipment within a
nuclear power plant.
As the structure is regular and first mode dominated, IDAs have been performed in the two
directions independently, considering the first mode spectral acceleration value as intensity
measure (IM) while the inelastic displacement has been selected as engineering demand
parameter (EDP). IDAs results have allowed the calculation of the conditional distribution of
the EDP given the IM (assumed as lognormal). Hence, the probability of exceedance of the
seismic capacity has been assessed for each intensity level.
These results, that already provide an empirical fragility curve, have been fitted by a
cumulative lognormal distribution whose parameters have been defined by using a
regression process. Fragility curves have been evaluated for each direction and for three
different confidence levels (50th, 5th and 95th percentile).
Finally, considering the fragility curve at 95th percentile and the design spectral acceleration
value of the reference code spectrum, the probability of failure for each direction has been
calculated. The worst result, reached for the longitudinal direction, has provided a probability
of failure equal to 2%. This has led to the conclusion that the building design, following a

8
 B. POLIDORO, B. GHOSH and M. TAN

Beyond Design Basis event, demonstrates a High Confidence (at 95th percentile) of Low
Probability (less than 5%) of failure.

FURTHER THOUGHTS
Resilience of critical facilities has always been one of the major challenges for Facility
Owners of structures exposed to extreme events, as the sole reliance on National Codes is
insufficient. Eurocode, itself, recommends the use of a risk analysis approach together with
hazard identification for assessment of extreme events. The most common method of
qualifying the resilience of critical facilities in extreme events is by using simplified
approaches that adopts safety factors to define parameters of the fragility curves. However,
recent research has highlighted the potential use of non-linear dynamic analyses for the
following reasons:

 Extra rigour and thought is required to highlight failure modes and weaknesses in the
structure.
 It allows a comparison of different design solutions which leads to a more transparent
decision making and robust design.
 The  approach  permits  the  main  uncertainties  (and  “weak  links”)  to  be  identified  for  the  
specific structure, which can be dealt with directly by practical risk reduction
measures.

One of the recommended probabilistic approaches to assess a structure is the calculation of

fragility curves by using non-linear dynamic analysis. This method is a robust and elegant
procedure compared to the current practice as it accounts for the uncertainty related to the
seismic ground motion, and eventually, epistemic uncertainties related to the geometry or the
characteristics of the materials of the specific structure. This approach has been extensively
used in the insurance sector for financial modelling (especially for civil structures) and can be
used for other high risk industry.

REFERENCES

ASCE/SEI 43-05 (2007) Seismic Design Criteria for Structures, Systems, and Components in Nuclear
Facilities, American Society of Civil Engineers, 1801 Alexander Bell Drive, Reston, VA 20191 USA
CEN 2004 European Standard EN1998-1:2004 (2004) Eurocode 8: Design of structures for
earthquake resistance. Part 1: general rules, seismic actions and rules for buildings, Comité Européen
de Normalisation, Brussels.
D’Ayala  D,  Meslem  A,  Vamvatsikos  D,  Porter  K, Rossetto T, Crowley H, Silva V (2014) Guidelines for
Analytical Vulnerability Assessment, Vulnerability Global Component project [Available at:
http://www.nexus.globalquakemodel.org/gem-vulnerability/posts/guidelines-for-analytical-vulnerability-
assessment].
De Luca F, Vamvatsikos D, Iervolino I (2013) Near-Optimal piecewise linear fits of static pushover
capacity curves for equivalent SDOF analysis, Earthquake Engineering and Structural Dynamics,
42(4): 523-543.
Fajfar P (2000) A non linear Analysis Method for Performance based Seismic Design, Earthq. Spectra
16(3): 573-592
Federal Emergency Management Agency (2009) Recommended Methodology for Quantification of
Building System Performance and Response Parameters, Report No. FEMA P695, Prepared by
Applied Technology Council, Prepared for Federal Emergency Management Agency, Washington,
DC.
Luco N and Cornell CA (2007) Structure-specific scalar intensity measures for near-source and
ordinary earthquake ground motions, Earthquake Spectra 23(2): 357-392.

9
 B. POLIDORO, B. GHOSH and M. TAN

ONR Report ONR-FRREP- 12-001 Revision 0 (2012) Japanese earthquake and tsunami: Implementing
the lessons for the UK's nuclear industry
Porter K, Kennedy R, Bachman R (2007) Creating fragility functions for performance-based
earthquake engineering, Earthquake Spectra 23(2):471–489
Reed J W and Kennedy R P (1994) Methodology for Developing Seismic Fragilities, TR-103959,
Electric Power Research Institute, Palo Alto, California, USA.
Shome N and Cornell C A (1999) Probabilistic seismic demand analysis of nonlinear structures, RMS
Program, Report No. RMS35 (Ph.D. Thesis), Stanford University, CA.
Shome N, Cornell C A, Bazzurro P, Carballo J E (1998) Earthquakes, records and nonlinear
responses, Earthquake Spectra, 14(3), 469–500.
Vamvatsikos D, Cornell C, (2002) Incremental Dynamic Analysis, Earthquake Engineering and
Structural Dynamics, 31(3): 491-514.

10