PDF

Seismic Fragility Applications Guide Update
Seismic Fragility Applications Guide

Update
1019200
Final Report, December 2009
EPRI Project Manager

R. Kassawara
ELECTRIC POWER RESEARCH INSTITUTE

3420 Hillview Avenue, Palo Alto, California 94304-1338 PO Box 10412, Palo Alto, California 94303-0813 USA
800.313.3774 650.855.2121 askepri@epri.com www.epri.com
DISCLAIMER OF WARRANTIES AND LIMITATION OF LIABILITIES
THIS DOCUMENT WAS PREPARED BY THE ORGANIZATION(S) NAMED BELOW AS AN
ACCOUNT OF WORK SPONSORED OR COSPONSORED BY THE ELECTRIC POWER RESEARCH
INSTITUTE, INC. (EPRI). NEITHER EPRI, ANY MEMBER OF EPRI, ANY COSPONSOR, THE
ORGANIZATION(S) BELOW, NOR ANY PERSON ACTING ON BEHALF OF ANY OF THEM:
(A) MAKES ANY WARRANTY OR REPRESENTATION WHATSOEVER, EXPRESS OR IMPLIED, (I)

WITH RESPECT TO THE USE OF ANY INFORMATION, APPARATUS, METHOD, PROCESS, OR
SIMILAR ITEM DISCLOSED IN THIS DOCUMENT, INCLUDING MERCHANTABILITY AND FITNESS
FOR A PARTICULAR PURPOSE, OR (II) THAT SUCH USE DOES NOT INFRINGE ON OR
INTERFERE WITH PRIVATELY OWNED RIGHTS, INCLUDING ANY PARTY'S INTELLECTUAL
PROPERTY, OR (III) THAT THIS DOCUMENT IS SUITABLE TO ANY PARTICULAR USER'S
CIRCUMSTANCE; OR
(B) ASSUMES RESPONSIBILITY FOR ANY DAMAGES OR OTHER LIABILITY WHATSOEVER

(INCLUDING ANY CONSEQUENTIAL DAMAGES, EVEN IF EPRI OR ANY EPRI REPRESENTATIVE
HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES) RESULTING FROM YOUR
SELECTION OR USE OF THIS DOCUMENT OR ANY INFORMATION, APPARATUS, METHOD,
PROCESS, OR SIMILAR ITEM DISCLOSED IN THIS DOCUMENT.
ORGANIZATION(S) THAT PREPARED THIS DOCUMENT
RPK Structural Mechanics Consulting
Simpson Gumpertz & Heger Inc.
NOTE
For further information about EPRI, call the EPRI Customer Assistance Center at 800.313.3774 or
e-mail askepri@epri.com.
Electric Power Research Institute, EPRI, and TOGETHERSHAPING THE FUTURE OF ELECTRICITY
are registered service marks of the Electric Power Research Institute, Inc.
Copyright 2009 Electric Power Research Institute, Inc. All rights reserved.
CITATIONS
This report was prepared by

28625 Mountain Meadow Road
Escondido, CA 92026
Principal Investigator
R. Kennedy
Simpson Gumpertz & Heger Inc.

4000 MacArthur Boulevard, Suite 710
Newport Beach, CA 92660
Principal Investigators
G. Hardy
K. Merz
This report describes research sponsored by the Electric Power Research Institute (EPRI).
This publication is a corporate document that should be cited in the literature in the following
manner:
Seismic Fragility Applications Guide Update. EPRI, Palo Alto, CA: 2009. Product ID Number
1019200.
iii
REPORT SUMMARY
This update of selected seismic fragilities methods provides utilities with in-depth guidance for
performing fragility analysis for a seismic probabilistic risk assessment (SPRA). These cost-
effective and practical procedures for fragility evaluations can be used in performing a SPRA in
support of Risk Informed/Performance Based applications.
Background
EPRI has published two reports in the past which document the seismic fragility methodology:
Methodology for Developing Seismic Fragilities. EPRI TR-103959. June 1994.
Seismic Fragility Application Guide. EPRI Report 1002988. December 2002.
Seismic fragility methodology has advanced since the publication of these two reports and
several of the sections of these reports require updating to reflect changes in the fragility
approach that have occurred and editorial changes that have been implemented.
Objective
Provide utilities with state-of-the-art guidance on seismic fragility analysis methods in support of
regulatory and non-regulatory applications. Specific tasks supporting that global objective
include:
Update existing seismic fragility methodology descriptions to reflect recent advances.
Update existing seismic fragility example problems to reflect recent advances in the seismic
fragility methodology.
Results
This report provides updated implementation guidance for deriving seismic fragilities together
with updates of example fragility calculations. Specifically, the following have been included in
this report:
Updates to the definition of High Confidence of Low Probability of Failure (HCLPF) to
remove the distinction between HCLPF84 and HCLPF50.
Updates to the characterization of peak-to-valley variability treatment in seismic fragility
analyses.
Updates to the use of seismic margin assessment (SMA) screening tables.
Update on seismic fragility methods incorporating the use of seismic experience data.
v
Update of the seismic fragility example of the use of earthquake experience data.
EPRI Perspective
The basic fragility methodology has been documented in selected technical papers and industry
publications (e.g., EPRI TR-103959 and EPRI-TR 1002988). This report updates that fragility
methodology to reflect recent methodological changes in the literature and also updates example
fragility calculations that incorporate these updated methods. This document, along with EPRI
TR-103959 and EPRI TR-1002988, provides methodology, procedures, and an array of example
problems that encompass most situations that typically confront the fragility analyst.
Keywords
Earthquakes
Seismic risk
Fragilities
Individual plant examination for external events
Probabilistic safety assessment
vi
ABSTRACT
Essentially all probabilistic seismic hazard analyses (PSHA) have included the response spectral
peak and valley variability rs as part of the aleatory variability when developing seismic hazard
estimates as a function of the annual frequency of exceedance (AFE). Thus, at any AFE, the
resulting uniform hazard response spectrum (UHRS) already fully includes the effect of rs. This
statement is true irrespective of whether the UHRS is defined at the mean, median, 90%, 84%, or
16% non-exceedance probability (NEP) because the aleatory variability (including rs) similarly
affects the slope of each of these NEP hazard curves. The difference in amplitude of these
various NEP curves is due to the epistemic uncertainty.
The Diablo Canyon PSHA was an exception to this statement since the rs was specifically
removed from the hazard aleatory variability by developing hazard curves for spectral
acceleration averaged over a broad frequency range. However, the above paragraph is applicable
for all PSHA conducted using the Electric Power Research Institute (EPRI) methodology
developed in the 1989 time frame, the EPRI attenuation relationships developed in the 2003 time
frame, or the Lawrence Livermore National Laboratory (LLNL) methodology developed in the
1993 time frame.
Unfortunately, until recently fragility estimates performed in accordance with the standard
Fragility Analysis (FA) Methodology have also included the response spectral peak and valley
variability rs in the fragility estimates. Thus, the effect of rs has been double-counted when the
hazard and fragility curves are convolved together to estimate the seismic risk. This double-
counting of rs has resulted in seismic risk being overstated by as much as a factor of two.
Prior to the Diablo Canyon seismic probabilistic risk assessment (SRPA), this double-counting
of rs was not recognized. As stated above, to remove this double-counting of rs for the Diablo
Canyon SPRA, rs was removed from the aleatory variability of the PSHA. The authors of both
of the previous EPRI seismic fragility reports (EPRI, 1994) and (EPRI, 2002) incorrectly
assumed that rs would be removed in the future from the aleatory variability of the PSHA
similarly to what had been done on the Diablo Canyon SPRA. As a result, both of those
references recommend inclusion of rs in the range of 0.2 to 0.3 in the fragility estimate as had
been the prior common practice. Since it does not appear likely to get the seismic hazard
estimators to remove rs from the aleatory variability included in the PSHA, the current
recommendation is to drop rs from the fragility estimate because it is included in the hazard
estimate. A white paper by Dr. R. P. Kennedy has been included as Appendix A of this report
which addresses the history of the response spectra peak and valley variability treatment within
SPRA studies and the recommendations for how to avoid double counting in future SPRAs.
vii
The key updates provided by this report are associated with the changes to both the (EPRI, 1994)
report and the (EPRI, 2002) report with respect to eliminating the rs from the various associated
criteria presented in each report. In addition, an update to the EPRI 1002988 Seismic Fragility
Applications Guide included revisions to two of the example fragility derivations from that
document:
Estimation of Equipment Capacity Based on Earthquake Experience Data (Appendix C of
EPRI 1002988)
Example Fragility for Instrument Cabinet Derived from Experience Data (Appendix D of
EPRI 1002988)
viii
ACRONYMS
AF Amplification Function
AFE Annual Frequency of Exceedance
AFRS Amplified Floor Response Spectra
CDF Core Damage Frequency
CDFM Conservative Deterministic Failure Margin
COL Collective Operating License
EUS Eastern United States
FA Fragility Analysis
GERS Generic Equipment Ruggedness Spectrum
GIP Generic Implementation Procedure
GMI Ground Motion Incoherence
HFD High Frequency Ductility
IPE Individual Plant Examination
IPEEE Individual Plant Examination of External Events
FOAKE First of a Kind Reactor Engineering
HCLPF High Confidence of Low Probability of Failure
IRS In-Structure Response Spectra
LERF Large Early Release Frequency
NEP Non-Exceedance Probability
NPP Nuclear Power Plant
PGA Peak Ground Acceleration
PRA Probabilistic Risk Assessment
PSHA Probabilistic Seismic Hazard Analysis
RLE Review Level Earthquake
Sa Spectral Acceleration
SSEL Safe Shut-down Equipment List
SPRA Seismic Probabilistic Risk Assessment
SRV Safety Relief Valve
SSCs Structures, Systems and Components
SSI Soil-Structure Interaction
SSMRP Seismic Safety Margin Research Program
SQURTS Seismic Qualification Reporting and Testing Standardization
SSE Safe Shutdown Earthquake
SMA Seismic Margin Assessment
UHRS Uniform Hazard Response Spectrum
UHS Uniform Hazard Spectra
ZPA Zero Period Acceleration
ix
FRAGILITY TERMINOLOGY
A,a Ground Motion Parameter Corresponding to Any Given Frequency of Failure
Am Median Peak Ground Motion Capacity
C Composite Variability = (R2 + U2)0.5
C-C Capacity Factor Variability
C-RS Response Factor Variability
C-S Strength Composite Variability (typical)
R Log Standard Deviation of Randomness
R_S Strength Randomness (typical)
U Log Standard Deviation of Uncertainty (Lack of Knowledge)
U_S Strength Uncertainty (typical)
rs Response Spectra Peak and Valley Variability
R Random Variable with Unit Median and Logarithmic Standard Deviation, R
U Uncertainty Variable with Unit Median and Logarithmic Standard Deviation, U
F Factor of Safety
F Damping Factor of Safety
F Inelastic Energy Absorption Factor of Safety
FC Capacity Factor
FECC Earthquake Component Combination Factor of Safety
FGMI Ground Motion Incoherence Factor of Safety
FM Modeling Factor of Safety
FMC Modal Combination Factor of Safety
FQM Qualification Factor of Safety
xi
FRE Equipment Response Factor of Safety
FRS Structural Response Factor of Safety
FS Strength Factor of Safety
FSA Spectral Shape Factor of Safety
FSSI Soil-Structure Interaction Factor of Safety
FTOTAL Total Factor of Safety
HCLPF84 84% Non-Exceedance Probability High Confidence of Low Probability of Failure
PN Normal Operating Loads
PT Total Load (Sum of Seismic Load and Normal Operating Load)
S Strength of Structural Element
xii
CONTENTS
1 INTRODUCTION .................................................................................................................... 1-1

1.1 Objective of Seismic Fragility Applications Guide Update ........................................... 1-1
1.2 Scope of the Applications Guide Update ..................................................................... 1-1
2 SEISMIC FRAGILITY METHODOLOGY ................................................................................ 2-1

2.1 Screening of High-Capacity Components .................................................................... 2-1
2.2 Seismic Fragility Analysis Methodology ....................................................................... 2-1
2.2.1 Fragility Model ..................................................................................................... 2-2
2.2.2 Failure Modes ..................................................................................................... 2-4
2.2.3 Estimation of Fragility Parameters ...................................................................... 2-5
2.2.3.1 Fragility of Structures .................................................................................... 2-6
2.2.3.2 Fragility of Equipment and Other Components ............................................. 2-8
3 UPDATES TO EPRI TR-103959 ............................................................................................ 3-1

3.1 Basis for Updates......................................................................................................... 3-1
3.2 Updated Text for Section 2 of EPRI TR-103959 .......................................................... 3-2
3.3 Updated Text for Section 3 of EPRI TR-103959 ........................................................ 3-12
3.4 Updated Text for Section 5 of EPRI TR-103959 ........................................................ 3-15
4 UPDATES TO EPRI TR-1002988 .......................................................................................... 4-1

4.1 Basis for Updates......................................................................................................... 4-1
4.2 Updated Text for Appendix C of EPRI TR-1002988 .................................................... 4-1
4.3 Updated Text for Appendix D of EPRI TR-1002988 .................................................... 4-8
5 REFERENCES ....................................................................................................................... 5-1
xiii
LIST OF FIGURES
Figure 2-1 Mean, Median, 5% Non-Exceedance, and 95% Non-Exceedance Fragility

Curves for a Component .................................................................................................... 2-4
xv
1
INTRODUCTION
1.1 Objective of Seismic Fragility Applications Guide Update
Seismic Probabilistic Risk Assessment (SPRA) studies have been conducted for many of the
United States (U.S.) Nuclear Power Plants (NPPs) over the last 20 years. Initially they were
conducted to answer safety concerns in heavily populated areas. The next widespread application
was for satisfaction of the U.S. Nuclear Regulatory Commissions (USNRC) request for
information regarding severe accident vulnerabilities in Generic Letter 88-20, Supplement 4,
(USNRC, 1991a). The USNRC is currently encouraging the use of PRA for making risk-
informed decisions and has developed a Risk-Informed Regulation Implementation Plan
(USNRC, 2000b) and associated regulatory guides. The Licensees in turn are using PRA for
Changes to Licensing Basis, Changes to Technical Specifications, Graded Quality Assurance,
etc. Seismic PRA are also required for each new NPP design before the COL will be granted.
The key elements of a SPRA can be identified as:

Seismic Hazard Analysis: to develop frequencies of occurrence of different levels of ground
motion (e.g., peak ground acceleration (PGA)) at the site.
Seismic Fragility Evaluation: to estimate the conditional probability of failure of important
structures and equipment whose failure may lead to unacceptable damage to the plant (e.g.,
core damage). Plant walkdown is an important activity in conducting this task.
Systems/Accident Sequence Analysis: modeling of the various combinations of structural
and equipment failures that could initiate and propagate a seismic core damage sequence.
Risk Quantification: Assembly of the results of the seismic hazard, seismic fragility, and
systems analyses to estimate the frequencies of core damage and plant damage states.
Assessment of the impact of seismic events on the containment and consequence analyses,
and integration of these results with the core damage analysis to obtain estimates of seismic
risk in terms of effects on public health (e.g., early deaths and latent cancer fatalities).
This report focuses on changes to prior recommendations concerning the generation of seismic
fragilities.
1.2 Scope of the Applications Guide Update
Significant information in the literature exists on how to generate seismic fragilities for use in a
SPRA. The EPRI Methodology for Developing Seismic Fragilities, (EPRI, 1994) and EPRI
Seismic Fragility Application Guide, (EPRI, 2002), contain most of the background and
1-1
Introduction
guidance needed for an analyst to develop seismic fragilities of structures, systems, and
components (SSCs). However, since these reports have been published there have been several
clarifications of the methodology which require that these reports be updated.
Section 2 presents an overall summary of seismic fragility methodology. Past seismic fragility
documents, technical papers and codes/standards have used a variety of seismic fragility
acronyms, and parameter definitions. This use of different parameter definitions is even true for
one of the fundamental references that form the basis for this update document (EPRI 1994). The
background description in Section 2 is intended to provide a consistent set of definitions for the
purposes of this document.
Section 3 discusses the updates that have been taken to the EPRI TR-103959 (EPRI 1994)
document on the methodology for developing seismic fragilities. These updates revolve around
the treatment of spectral peak and valley variability in the development of fragility and HCLPF
capacity estimates. The updates to (EPRI 1994) are documented in italics to designate that these
updated parts of the report can be dropped into that report. Thus, the equation numbers and
reference numbers are formatted to be consistent with the (EPRI 1994) report.
Section 4 documents the updates to the EPRI 1002988 (EPRI 2002) Seismic Fragility
Applications Guide. These updates concentrate on two of the example fragility derivations from
that document:
Estimation of Equipment Capacity Based on Earthquake Experience Data (Appendix C of
EPRI 1002988)
Example Fragility for Instrument Cabinet Derived from Experience Data (Appendix D of
EPRI 1002988)
The updates to (EPRI 2002) are documented in italics to designate that these updated parts of the
report can be dropped into that report. Thus, the equation numbers, figure/table numbers and
reference numbers are formatted to be consistent with the (EPRI 2002) report.
Section 5 contains the references.
1-2
2
SEISMIC FRAGILITY METHODOLOGY
2.1 Screening of High-Capacity Components
The development of fragilities for all components considered in an SPRA is not required. Certain
high-capacity components may be screened-out of the components list based on the review of
seismic qualification criteria and qualification documents and walkdown screening. The decision
to screen components with estimated High Confidence of Low Probability of Failure (HCLPF)
capacities larger than some predetermined value should be based on the seismic hazard and the
associated unconditional failure rate of a component with a fragility corresponding to the
screening level. Based on the results of previous seismic PRAs, the screening level chosen was
often too low and masked the final core damage frequency (CDF) results. Screening for more
seismically active regions (e.g., western US and higher seismic regions in the central and eastern
US) should only be done at a higher HCLPF level. Identification of screening candidates is
primarily done by seismic capability engineers during the plant walkdown phase of the SPRA.
The screening level of the representative surrogate element representing the screened-out
components is determined by fragility analysts using earthquake experience data, plant-specific
qualification data, generic equipment ruggedness spectra, and known fragility test results.
2.2 Seismic Fragility Analysis Methodology
Fragility analysis involves the estimation of the conditional probabilities of structural or

equipment failure for a given level of seismic ground motion for the screened-in components.
Curves are developed using the fragility model whose parameters are the median acceleration
capacity (Am), and logarithmic standard deviations reflecting randomness in capacity (R) and
uncertainty in the median capacity (U). This task is performed by the seismic fragility analysts.
The objective of a fragility evaluation is to estimate the capacity of a given component relative to
a ground acceleration parameter such as PGA or spectral acceleration. Typically, the seismic
hazard for a plant site is defined by PGA or spectral accelerations (Sa) at different structural
frequencies; hence all fragility estimates are referenced to ground acceleration (peak ground or
spectral acceleration). Although spectral acceleration is the preferred ground motion parameter,
most existing hazard studies focused primarily on PGA, and most SPRAs have been based on
PGA. PGA is used herein as an example indicator only. If the seismic hazard curves are available
in terms of spectral accelerations at different frequencies they could be used as long as
consistency in the hazard and fragility definitions is maintained. In spite of its shortcomings as a
damage measure, PGA is a familiar term for all analysts involved in SPRA (i.e., systems
analysts, hazard analysts, and fragility analysts). Sensitivity studies have indicated that only
minor change results for the CDF calculated using fragilities defined in terms of PGA compared
2-1
Seismic Fragility Methodology
to those defined using average spectral acceleration over a specified frequency range covering
the fundamental frequencies of major structures. The important conclusion is that proper
interface between the analysts (i.e., hazard, fragility, and systems) should take place and it does
not matter so much what parameter the fragility is referenced to as long as the failure mode is
properly defined and the seismic response and capacity values are consistently calculated.
While an analyst can estimate the fragility of a component using a variety of methods, a simple
model for representing a component fragility was developed which has become the preferred
approach. In the following section this fragility model is described.
2.2.1 Fragility Model
The entire fragility family for an element corresponding to a particular failure mode can be
expressed in terms of the best estimate of the median ground acceleration capacity, Am, and two
random variables. Thus, the ground acceleration capacity, A, is given by:
A = Am eR eU, (2-1)
in which eR and eU are random variables with unit medians representing, respectively, the
inherent randomness about the median and the uncertainty in the median value. In this model, we
assume that both eR and eU are lognormally distributed with logarithmic standard deviations, R
and U, respectively. The formulation for fragility given by Eq. 2-1 and the assumption of a
lognormal distribution allow easy development of the family of fragility curves that
appropriately represent fragility uncertainty. For the quantification of fault trees in the plant
system and accident sequence analyses, the uncertainty in fragility needs to be expressed in a
range of conditional failure probabilities for a given ground acceleration. This is achieved as
explained below.
With perfect knowledge of the failure mode and parameters describing the ground acceleration
capacity (i.e., only accounting for the random variability, R), the conditional probability of
failure, fo, for a given PGA level, a, is given by:
a
n( )
Am
fo = (2-2)
R
where
(.) is the standard Gaussian cumulative distribution function. The relationship between fo and a
is the median fragility curve plotted in Figure 2-1 for a component with a median ground
acceleration capacity Am = 0.87g and R = 0.25. For the median conditional probability of failure
range of 5% to 95%, the ground acceleration capacity would range from Am exp (-1.65 R) to Am
exp (1.65 R), i.e., 0.58g to 1.31g.
2-2
When the modeling uncertainty U is included, the fragility becomes a random variable
(uncertain). At each acceleration value, the fragility f can be represented by a subjective
probability density function. The subjective probability, Q (also known as confidence) not
exceeding a fragility f is related to f by:
a
In +U 1(Q)
f = m
A (2-3)
R

where:
Q = P[f < f | a]; i.e., the subjective probability (confidence) that the conditional
probability of failure, f, is less than f for a peak ground acceleration a.
-1(.) = the inverse of the standard Gaussian cumulative distribution function.
For example, the conditional probability of failure f at acceleration 0.6g that has a 95%
nonexceedance subjective probability (confidence) is obtained from Eq. 2-3 as 0.79. The 5% to
95% probability (confidence) interval on the failure at 0.6g is 0 to 0.79.
A mean fragility curve is also plotted in Figure 2-1. This is obtained using Eq. 2-2 but replacing
R with the composite variability C = (R2 + U2)1/2.
2-3
1 Am = 0.87 g
R = 0.25
U = 0.35
0.8
0.79
95% Median Mean
Confidence
0.6
0.4
5%
Confidence
0.20
0.2
0.068
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
HCLPF
0.32g
PEAK GROUND ACCELERATION (g)
Figure 2-1
Mean, Median, 5% Non-Exceedance, and 95% Non-Exceedance Fragility Curves for a
Component
The median ground acceleration capacity Am, and its variability estimates R and U are evaluated
by taking into account the safety margins inherent in capacity predictions, response analysis, and
equipment qualification, as explained below.
2.2.2 Failure Modes
The first step in generating fragility curves such as those in Figure 2-1 is to develop a clear
definition of what constitutes failure for each of the critical elements in the plant. This definition
of failure must be agreeable to both the structural analyst generating the fragility curves and the
systems analyst who must judge the consequences of component failure. Several modes of
failure (each with a different consequence) may have to be considered and fragility curves may
have to be generated for each of these modes. For example, a motor-actuated valve may fail in
any of the following ways:
Failure of power or controls to the valve (typically related to the seismic capacity of such
items as cable trays, control panels, and emergency power). Since these failure modes are not
related to the specific item of equipment (i.e., motor actuated valve) and are common to all
active equipment, such failure modes are most easily handled as failures of separate systems
linked in a series to the equipment.
2-4
Failure of the motor.

Binding of the valve due to distortion and, thus, failure to operate.
Failure of the pressure boundary due to overstress of the flange joint.
It is usually possible to identify the failure mode most likely to be caused by the seismic event by
observations during the walkdown or by reviewing the equipment design and considering only
that mode. Otherwise, fragility curves are developed based on the premise that the component
could fail in any one of all potential failure modes.
Identification of the credible modes of failure is largely based on the analysts experience and
judgment. Review of plant design criteria, calculated stress levels in relation to the allowable
limits, qualification test results, seismic fragility evaluation studies done on other plants, and
reported failures (in past earthquakes, in licensee event reports, and fragility tests) are useful in
this task.
Structures are considered to have failed functionally when they cannot perform their designated
functions. In general, structures are considered to have failed functionally when inelastic
deformations under seismic load are estimated to be sufficient to potentially interfere with the
operability of safety-related equipment attached to the structure, or fractured sufficiently so that
equipment attachments fail. These failure modes represent a conservative lower bound of seismic
capacity since a larger margin of safety against total collapse exists for nuclear structures. Also, a
structural failure is generally assumed to result in a common cause failure of multiple safety
systems, if these safety systems are housed in the same structure. For example, the service water
pumps may be assumed to fail when the enclosure pump house roof collapses. Structures that are
susceptible to sliding are considered to have failed when sufficient sliding deformation has
occurred to fail buried or interconnecting piping or electrical duct banks.
For piping, failure of the support system or fracture or collapse of the pressure boundary, are
credible failure modes. Failure modes of equipment examined may include structural failure
modes (e.g., bending, buckling of supports, anchor bolt pullout, etc.), functional failures (binding
of valve, excessive deflection in rotating equipment), and breaker trip or relay chatter.
Consideration should also be given to the potential for soil failure modes (e.g., liquefaction, toe
bearing pressure failure, base slab uplift, and slope failures). For buried equipment (i.e., piping
and tanks), failure due to lateral soil pressures may be an important mode. Seismically induced
failures of structures or equipment under impact of another structure or equipment may also be a
consideration.
2.2.3 Estimation of Fragility Parameters
In estimating fragility parameters, it is convenient to work in terms of an intermediate random

variable called the factor of safety. The factor of safety, F, on ground acceleration capacity above
a reference level earthquake specified for design; e.g., the safe shutdown earthquake level
specified for design, ASSE, is defined as follows:
2-5
A = FA SSE where A is the actual ground motion acceleration capacity
Actual seismic capacity of element

F=
Actual response due to SSE
Actual capacity Calculated code capacity Design response due to SSE

= X X
Calculated code capacity Design response due to SSE Actual response due to RE
F is further simplified as:
Actual capacity Design response due to SSE

F= X
Design response due to SSE Acctual response due to RE
F =F F (2-4)
C SR
where FC is the capacity factor and FSR is the structural response factor, and RE is the reference
earthquake spectrum derived from the probabilistic hazard study, anchored to the same PGA as
the SSE.
The median factor of safety, Fm, can be directly related to the median ground acceleration
capacity, Am, as:
Am
Fm = (2-5)
A
SSE
The logarithmic standard deviations of F, representing inherent randomness and uncertainty, are
identical to those for the ground acceleration capacity A.
2.2.3.1 Fragility of Structures
For structures, the factor of safety is typically modeled as the product of three random variables:
F = FS FFSR (2-6)
The strength factor, FS , represents the ratio of ultimate strength (or strength at loss-of-function)
to the stress calculated for A SSE . In calculating the value of FS , the non-seismic portion of the
total load acting on the structure is subtracted from the strength as follows:
S - PN
F = (2-7)
S P -P
T N
2-6
where S is the strength of the structural element for the specific failure mode, PN is the normal
operating load (i.e., dead load, operating temperature load, etc.) and PT is the total load on the
structure (i.e., sum of the seismic load for A SSE and the normal operating load). For higher
earthquake levels, other transients (e.g., SRV discharge in BWRs) may have a high-probability
of occurring simultaneously with the earthquake. The definition of PN in such cases should be
extended to include the loads from these transients.
The inelastic energy absorption factor (ductility factor), F, accounts for the fact that an
earthquake represents a limited energy source and many structures or equipment items are
capable of absorbing substantial amounts of energy beyond yield without loss-of-function.
The structure response factor, FSR , is modeled as a product of factors influencing the response
variability:
FSR = FSA FGMI F FM FMC FEC FSSI (2-8)
where:
FSA = spectral shape factor representing the difference and variability in response due to
the difference between the SSE spectrum and the RE spectrum defined by the
hazard analyst.
FGMI = ground motion incoherence factor that accounts for the fact that a traveling seismic
wave does not excite a large foundation uniformly.
F = damping factor representing variability in response due to difference between

actual damping and design damping.
FM = modeling factor accounting for any bias and uncertainty in response due to
modeling assumptions.
FMC = mode combination factor accounting for any bias and variability in response due to
the method used in combining dynamic modes of response.
FEC = earthquake component combination factor accounting for any bias and variability
in response due to the method used in combining earthquake components.
FSSI = factor to account for effect of soil-structure interaction including the reduction of
input motion with depth below the surface.
The median and logarithmic standard deviations of F are expressed as:
Fm = FSm Fm FSAm FGMIm Fm FMm FMCm FECm FSSIm (2-9)
2-7
and
( 2 2 2 2
F = s + + SA + GMI + + SSI )
2 1/ 2
(2-10)
The logarithmic standard deviation F is further divided into random variability, R , and
uncertainty, U . To obtain the median ground acceleration capacity A m the median factor of
safety, Fm , is multiplied by the safe shutdown earthquake peak ground acceleration.
2.2.3.2 Fragility of Equipment and Other Components
For equipment and other components, the factor of safety is composed of a capacity factor, FC ; a
structure response factor, FSR ; and an equipment response (relative to the structure) factor, FRE
Thus,
FE = FCFREFRS (2-11)
The capacity factor FC for the equipment is the ratio of the acceleration level at which the
equipment ceases to perform its intended function to the seismic design level. This acceleration
level could correspond to a breaker tripping in switchgear, excessive deflection of the control rod
drive tubes, or failure of an equipment support. The capacity factor for the equipment may be
calculated as the product of FS and F. The strength factor, FS, is calculated using Eq. (2-7). The
strength, S, of equipment is a function of the failure mode.
Equipment failures can be classified into three categories:

Elastic functional failures.
Brittle failures.
Ductile failures.
Elastic functional failures involve the loss of intended function while the component is stressed
below its yield point. Examples of this type of failure include the following:
Elastic buckling in tank walls and component supports.
Excessive blade deflection in fans.
The load level at which functional failure occurs is considered the strength of the component.
Brittle failure modes are those that have little or no system inelastic energy absorption capability.
Examples include the following:
Anchor bolt failures.
Component support weld failures.
2-8
Shear pin failures.
Each of these failure modes has the ability to absorb some inelastic energy on the component
level, but the plastic zone is very localized and the system ductility for an anchor bolt or a
support weld is very small. The strength of the component failing in a brittle mode is therefore
calculated using the ultimate strength of the material.
Ductile failure modes are those in which the structural system can absorb a significant amount of
energy through inelastic deformation. Examples include the following:
Pressure boundary failure of piping or vessel nozzles.
Structural failure of cable trays and ducting.
Failure of component support members (plastic bending, plastic buckling).
The strength of the component failing in a ductile mode is calculated using the effective yield
strength of the material for tensile loading. For flexural loading, the strength is defined as the
limit load or load to develop a hinge mechanism.
The inelastic energy absorption factor, F, for a piece of equipment is a function of the ductility
ratio, . The median value of F is considered to be 1.0 for brittle and functional failure modes.
The equipment response factor FRE, is the ratio of equipment response calculated in the design to
the realistic equipment response; both responses being calculated for design floor spectra. FRE is
the factor of safety inherent in the computation of equipment response. It depends upon the
response characteristics of the equipment and is influenced by some of the variables listed under
Eq. (2-8). These variables differ according to the seismic qualification procedure. For equipment
qualified by analysis, the important variables that influence response and variability are as
follows:
Qualification method (QM) dynamic analysis vs. static coefficient used, etc.
Spectral shape (SA) including the effects of peak broadening and smoothing, and artificial
time history generation.
Modeling (affects of mode shape and frequency results) (M).
Damping (affects of design damping vs. median damping) ().
Combination of modal responses (for response spectrum method) (MC).
Combination of earthquake components (ECC).
For rigid equipment qualified by static analysis, the variables, except the qualification method,
and combination of earthquake components are not significant. The equipment response factor is
the ratio of the specified static coefficient divided by the zero period acceleration of the floor
level where the equipment is mounted. If the equipment is flexible and was designed via the
static coefficient method, the dynamic characteristics of the equipment must be considered. This
requires estimating the fundamental frequency and damping, if the equipment responds
2-9
predominantly in one mode. The equipment qualification method factor is the ratio of the static
coefficient to the best estimate spectral acceleration at the equipment fundamental frequency.
Where testing is conducted for seismic qualification, the response and capacity may be
determined from specific criteria contained in EPRI (1994).
The overall equipment response factor is the product of these factors of safety corresponding to
each of the variables identified above. The median and logarithmic standard deviations for
randomness and uncertainty are estimated following Eqs. (2-9) and (2-10).
The structural response factor, FSR , is based on the response characteristics of the structure at the
location of the component (equipment) support. The variables pertinent to the structural response
analyses used to generate floor spectra for equipment design are the only variables of interest to
equipment fragility. Time history analyses using the same structural models used to conduct
structural response analysis for structural design are typically used to generate floor spectra. The
applicable variables are as follows:
Spectral shape.
Ground motion incoherence.
Damping.
Modeling.
Mode combination (if mode superposition time history is used).
Soil-structure interaction including reduction with depth of seismic input.
For equipment with a seismic capacity level that has been reached while the structure is still
within the elastic range, the structural response factors should be calculated using damping
values corresponding to less than yield conditions (e.g., about 5% median damping for reinforced
concrete). The combination of earthquake components is not included in the structural response
since the variable is to be addressed for specific equipment orientation in the treatment of
equipment response. Median Fm and variability R and U estimates are made for each of the
parameters affecting capacity and response factors of safety. These median and variability
estimates are then combined using the properties of lognormal distribution in accordance with
Eqs. (2-9) and (2-10) to obtain the overall median factor of safety Fm and variability R and U
estimates required to define the fragility curves for the structure or equipment. For each variable
affecting the factor of safety, the random ( R ) and uncertainty ( U ) variabilities must be
separately estimated. The differentiation is somewhat judgmental, but it can be based on general
guidelines. Essentially, R represents variability due to the randomness of the earthquake
characteristics for the same peak acceleration and to the structural response parameters that relate
to these characteristics. The dispersion represented by U is due to factors such as the following:
Our lack of understanding of structural material properties such as strength, inelastic energy
absorption, and damping.
2-10
Errors in calculated response due to use of approximate modeling of the structure and
inaccuracies in mass and stiffness representations.
Usage of engineering judgment in lieu of complete plant-specific design data for equipment
code capacities and responses.
2-11
3
UPDATES TO EPRI TR-103959
3.1 Basis for Updates
Essentially all PSHAs have included the response spectral peak and valley variability rs as part
of the aleatory variability when developing seismic hazard estimates as a function of the AFE.
Thus, at any AFE, the resulting UHRS already fully includes the effect of rs. Unfortunately,
until recently fragility estimates performed in accordance with the standard Fragility Analysis
Methodology such as detailed in EPRI TR-103959 have also included the response spectral peak
and valley variability rs in the fragility estimates. Thus, the effect of rs has been double-counted
when the hazard and fragility curves are convolved together to estimate the seismic risk. Since
the response spectral peak and valley variability rs is already being included in the aleatory
variability considered in the PSHA, it should not be included in the Fragility Analysis. The rs
recommended on pages 3-4 through 3-7 of EPRI TR-103959 should not be added.
The Conservative Deterministic Failure Margin (CDFM) Methodology defined in EPRI (1991b)
is aimed directly at estimating the 1% failure probability capacity C1% on the composite fragility
curve. Thus:
HCLPFCDFM = C1% = CDFM Capacity (4.1)
However, the CDFM method determines the C1% capacity assuming a smooth target input
response spectrum shape with no peak and valley variability rs about this target response
spectrum shape.
As such, if the target response spectrum shape used in the CDFM evaluation is identical to the
target UHRS obtained from the PSHA that included rs in the aleatory variability, then no
correction should be made to the CDFM computed HCLPF capacity, i.e.:
HCLPFc = HCLPFCDFM (4.2)
No so-called HCLPF50/HCLPF84 correction discussed on pages 2-34 through 2-36, and pages 5-
1 through 5-6 of EPRI TR-103959 should be made to the CDFM computed HCLPF for use with
PSHA hazard curves in which the peak and valley rs has already been included in the hazard
aleatory variability. Making this HCLPF correction double-counts the peak and valley variability
rs.
3-1
Updates to EPRI TR-103959
In addition, results from past PSHA has indicated that the capacity values for surrogate elements
associated with screening HCLPF values have been based on an assumed variability that is too
low. More reasonable generic estimate for variability should be considered.
3.2 Updated Text for Section 2 of EPRI TR-103959
The following represents the changes to the text of Section 2 of EPRI TR-103959 for the
subsection entitled LOGNORMAL DISTRIBUTION FOR FRAGILITY CURVES which
appears from pages 2-24 to 2-36. All tables, figures, equations, and references cited refer to the
original publication. The updates are concerned with notation and the removal of the distinction
between HCLPF50 and HCLPF84:
LOGNORMAL DISTRIBUTION FOR FRAGILITY CURVES
The lognormal distribution has been used in past SPRAs to develop fragility curves for NPP
structures and equipment. Other types of probability distributions (e.g., Weibull or normal) also can
be used; however, the lognormal distribution has particular properties that make it easy to implement
in a fragility analysis. Properties of the lognormal distribution pertinent to fragility curves are discussed
below. More information on the lognormal distribution can be found in Reference 20.
Figure 2-10 shows several example fragility curves (all with the same median capacity) based on the
lognormal distribution which is defined by two parameters:
Figure 2-10
Example Fragility Curves Based on the Lognormal Distribution
the median capacity, am and the logarithmic standard deviation, . The median capacity models
the central tendency while the logarithmic standard deviation represents the dispersion about the
3-2
median. The larger the value of , the more spread out (i.e., flat) a fragility curve becomes. The
curves in Figure 2-10 all have the same median capacity, but with different values. In Figure
2-10 only the randomness aspect of fragility curves and the lognormal model is shown. The
lognormal distribution is related to the normal distribution (also referred to in standard
textbooks as the Gaussian distribution). If the capacity is lognormally distributed then the
logarithms of the capacity values will be normally distributed. Note that all logarithms referred
to in this report are natural logarithms (i.e., to the base e) unless otherwise stated.
By transforming the properties of the lognormal distribution into the normal form, standard
equations and the tables for the normal distribution can be used to solve for the values of the
lognormal distribution. For example, the logarithmic standard deviation, is just the common"
standard deviation of the logarithms of the capacity values. Also, the antilogarithm of the mean
of the logarithms of the capacity values is the median capacity.
Table 2-2 shows a demonstration of these relationships for a data set of 25 capacity values
sampled randomly from a lognormal distribution (they have been sorted in ascending order).
These values were sampled from a lognormal distribution with a median value of 1.0 and a
logarithmic standard deviation of 0.20. As can be seen from this table, the calculated value
(i.e., standard deviation of the logarithms) is 0.20 (i.e., 0.199 rounded), and the median (i.e.,
antilogarithm of 0.000) is 1.0. Note that the median value from the sample capacities (i.e., the
thirteenth value) is 1.0, as expected.
3-3
Table 2-2
Sample Capacity Values from a Lognormal Distribution with a Median of 1.0 and a
Logarithmic Standard Deviation of 0.22
Sample Capacity Logarithms of

Values Capacity Values
1 0.662 -0.412
2 0.732 -0.312
3 0.774 -0.257
4 0.805 -0.216
5 0.833 -0.183
6 0.857 -0.155
7 0.879 -0.129
8 0.900 -0.105
9 0.921 -0.083
10 0.941 -0.061
11 0.960 -0.040
12 0.980 -0.020
13 1.000 0
14 1.020 0.020
15 1.041 0.040
16 1.063 0.061
17 1.086 0.082
18 1.110 0.105
19 1.137 0.128
20 1.167 0.154
21 1.201 0.183
22 1.241 0.216
23 1.292 0.256
24 1.364 0.311
25 1.506 0.410
Mean 1.019 0.000
Standard 0.204 0.199
Deviation
3-4
Figure 2-11 shows a probability density function and a cumulative distribution function for the
capacity of a component expressed in terms of a ground motion variable "a" (e.g., PGA or
average ground spectral acceleration). The total area under the density function is unity, and the shape
of the function represents the relative likelihood of different capacity values. As shown in Figure 2-11 the
cumulative distribution evaluated at acceleration "a" is just the area under the density function between the
acceleration equal to zero and the acceleration equal to "a." The distribution function is the probability
that the capacity is less than or equal to "a." In other words, the distribution function is equal to the
probability of failure at each value of a. Thus, the cumulative distribution function is just the fragility
curve for a component.
Figure 2-11
Example Probability Density Function and Cumulative Distribution Function (i.e., Fragility
Curve)
Notice also in Figure 2-11 that the fragility curve (i.e., the distribution function) is characterized by the
symbol LN(am, ), which is an abbreviated format for saying that the fragility curve is lognormal with
median, am, and logarithmic standard deviation, . This notation will be used throughout the report.
Next, the basic equations for calculating values from a lognormal fragility curve are presented. These
equations are used to obtain the probability of failure, Pf(a), at each value of a, and vice versa. The most
useful equation for calculating Pf(a) is the following:
Pf(a) = (u) (2-3)
where: u = ln(a/am)/
The variable "u" is the transformation of the lognormal parameters to the standardized normal
variable and is necessary in order to be able to use standardized normal distribution tables. The
standardized normal distribution, (u), is tabulated in many handbooks and texts on probability and
statistics and is a common function in many computer programs. Thus, knowing the standardized normal
variable "u" the value for (u) and hence the probability of failure at acceleration, a, can be easily found.
3-5
The inverse operation of knowing (u) and then finding the corresponding value of u, is often represented as
u = (u), where (u) is the inverse standard normal distribution.
-1 -1
The inverse of Eq. 2-3 leads to the capacity value "a" corresponding to a given probability of
failure, Pf (a) as given by the following equation:
a = am [ e u ] (2-4)
where u is the standardized normal variable corresponding to Pf (a). Figure 2-12 shows a
lognormal fragility curve and the associated acceleration values "a" corresponding to specified
probabilities of failure (i.e., 0.05, 0.50 and 0.95). Also, a table shown in Figure 2-12 gives the standardized
normal variable, "u" for common probability-of-failure values. Example calculations are provided
in Figure 2-12 that demonstrates how Eqs. 2-3 and 2-4 can be used.
Figure 2-12
Relationship between Probability of Failure and Capacity Expressed in Terms of
Acceleration
3-6
USE OF THE LOGNORMAL DISTRIBUTION IN SEISMIC FRAGILITY ANALYSIS
The structural analyst works with two types of variability in seismic fragility analysis. The part of
variability that is potentially reducible is defined to be uncertainty. It includes those sources of
variability due to lack of knowledge of structural response and capacity that could be reduced by more
detailed studies. Examples are uncertainties in the dynamic modeling of the structure, lack of
understanding of material capacity, and uncertainties due to use of engineering judgment. An example is the
strength of a shear wall where a model can be tested and thus the uncertainty reduced.
The part of variability that cannot be practically reduced is called randomness. For example, the variability
in structure response for a known PGA is random because of the varying response spectral shape for
different earthquakes with the same PGA. It is unlikely that any amount of testing or analysis will
reduce this randomness, within the confines of the current analytical model.
In seismic fragility analysis the median capacity of a component is considered to be uncertain, which leads
to the need for a double lognormal model. To represent this uncertainty the median capacity is defined to be
lognormally distributed with a "median of medians, (am)m , and a logarithmic standard deviation, u.
Specific median values are selected from the distribution on medians for different confidence values, and
each median value is used to calculate a single fragility curve. The medians are calculated using Eq. 2-4,
where is equal to u. Each fragility curve is then obtained using Eq. 2-3 with the same logarithmic standard
deviation for randomness, r, for each curve. Methods for obtaining estimates for r and u are discussed in
the following sections.
Figure 2-13 shows an example family of fragility curves where the median capacity is uncertain. The density
function shown at the top of Figure 2-13 represents the uncertainty on the median value. In this example
three fragility curves are shown corresponding to the 95%, 50%, and 5% confidence values. It is the
presence of uncertainty that leads to multiple fragility curves as shown at the bottom of Figure 2-13.
Note in Figure 2-13 that the 95% confidence median capacity corresponds to the value where 95% of the
medians are higher (or 5% of the medians are lower). Thus, the analyst says that there is a 95% confidence
that the "true" median capacity is higher than this value.
3-7
Figure 2-13
Family of Fragility Curves at 95%, 50%, and 5% Confidence Levels
For simplicity in notation the distinction between (am)m and am is dropped in the remainder of this
report. The "median of medians" is referred to directly as am, with the understanding that this median is
uncertain and is lognormally distributed with logarithmic standard deviation u.
Useful Properties of the Lognormal Distribution
Although the lognormal distribution is defined with a median and logarithmic standard deviation
these variables can also be expressed in terms of the corresponding mean and "common" standard deviation
values by the following two equations:
3-8
2
aave = am [ e /2 ] (2-5)
a= aave [ (e2/2
1 ] ) (2-6)
where: aave = Mean of the lognormal distribution LN(am, )
a = Standard deviation of the lognormal distribution LN(am, ))
For example, Table 1 can be used to test these relationships where the underlying distribution
from which the 25 values were randomly sampled is LN(1.0, 0.20):
2
aave = 1.00 e ( 0.20 ) /2
= 1.020 (Compare to 1.019)
2
a = 1.020 e ( 0.20 ) = 0.206 (Compare to 0.204)
As discussed in the following sections two approximations that can be used in performing a
fragility analysis are given by the following relationships:
am aave (2-7)
Va (2-8)
Where: Va = Coefficient of variation and is equal to a/.
For the example in Table 1 it can be seen that these approximations are accurate to within a few
percentage points. These approximations are reasonably accurate so long as Va(or ) is less than about
0.4.
The lognormal distribution has unique properties that make it easy to work with when calculating fragility
curves. For a normal distribution the mean of a sum of normally distributed random variables is equal to
the sum of the means. Similarly the median of a product of lognormal distributed random
variables is equal to the product of the medians. This is expressed in equation form as follows:
ym = ( x1 ) m m ( x 2 ) m ( x n ) m (2-9)
Where: y m = median of the lognormal distribution on y which is the product of lognormally

distributed random variables, xi
( x i )m = median of the lognormal random variable xi
3-9
Also, in a manner analogous to the properties of the normal distribution, where the variance of a
sum of normally distributed random variables is equal to the sum of the variances, the lognormal
distribution has a similar property. Note that the variance is just the "common" standard deviation
squared. The square of the logarithmic standard deviation of a product of lognormally distributed random
variables is equal to the sum of the squares of the individual logarithmic standard deviations. This
is expressed in equation form as follows:
2y = 12 + 22 + 2n + (2-10)
Where: y = logarithmic standard deviation of the lognormal distribution on y which

is the product of lognormally distributed random variables, xi
i = logarithmic standard deviation of the lognormal random variable xi
These last two equations are important for many earthquake capacity problems where the effects
of the various underlying variables (i.e., response and capacity) can be put into a multiplicative
format as discussed in the next section.
Figure 2-14 shows an example family of fragility curves. The same three curves from Figure 2-13 are
shown along with a mean fragility curve which is the weighted average of all possible curves. An important
short cut is available for calculating the mean curve without having to average a suite of
individual curves. The mean curve is also lognormal with properties: LN(am, c), where c is given by the
following equation.
c = u2 + 2r (2-11)
As can be seen in Figure 2-14, both the 50% confidence and the mean fragility curves pass through the
median capacity value. However, the mean curve is more spread out compared to the median
curve because the latter has a smaller logarithmic standard deviation (i.e., r compared to c). The
use of the mean fragility curve in SPRA is discussed below in this section. Note that only mean fragility
curves are needed when performing a SPRA for seismic IPE.
Also shown in Figure 2-14 is the HCLPF capacity, which is defined in SPRA to be the 95%
confidence of a 5% probability of exceedance. Using Equation 2-4 the median is selected at the 95%
confidence (i.e., am [ e -1.65 u ]and this median is then multiplied by the factor required to reach a 5%
probability of failure value (i.e., e -1.65 r ). Multiplying together these two terms leads to the following
equation for the HCLPF:
-1.65( r + u )
HCLPF = am [ e ] (2-12)
3-10
Figure 2-14
Example Fragility Curves
Using the definition for c from Eq 2-11 and considering all possible ratios of r to u, the
minimum HCLPF value (i.e., most conservative) is found when r is equal to u. For this case the
HCLPF can be expressed in terms of only c and the median capacity as follows:
-2.3 c
HCLPF = am [ e ] (2-13)
This relationship is only slightly conservative for realistic values of r and p=u found in fragility
analyses.
3-11
subsection entitled BASIC FRAGILITY ANALYSIS VARIABLES FOR STRUCTURES
which appears from pages 3-1 to 3-7. All tables, figures, equations, and references cited refer to
the original publication. The updates are concerned with the removal of the peak and valley
variability in the determination of structure fragility:
BASIC FRAGILITY ANALYSIS VARIABLES FOR STRUCTURES
A list of the basic fragility analysis variables is given in Table 3-1. Each of these variables and
recommended procedures for obtaining their parameter values (i.e., median and logarithmic
standard deviations) are given in the following subsections.
Ground Motion
There are three basic variables which account for the influence of ground motion variability.
Earthquake response spectrum shape
Horizontal direction peak response
Vertical component response
In SPRA a single ground motion parameter is selected for the fragility analysis, which is the same
ground motion parameter used in the hazard analysis for the site. Ultimately, the structure and
equipment fragility values, as well as the plant fragility value, are all expressed in terms of this
parameter. By using the same parameter in both hazard and fragility analyses the integration of the hazard
and fragility curves can be easily performed in a consistent manner.
In a fragility analysis the input consists of three ground response spectra: two horizontal and one vertical. Thus,
the input spectra from the three directions must be correlated to the single ground motion parameter
selected for the SPRA. This leads to variability in the response of structures and equipment since real earthquake
responses will generally be different from the response due to the idealized reference input.
It is assumed in this discussion that the hazard curves for a NPP site are defined in terms of a
single ground motion parameter (e.g., PGA or Sa), which is the average of the corresponding ground
motion parameters from the two horizontal directions. Implied also in this definition is that a UHS is used as
the reference response spectrum shape in the fragility analysis. It should be noted that all modern SPRAs
are now based on seismic hazard studies which include the characterization of the site-specific UHS.
3-12
Table 3-1
Basic Fragility Analyses Variables for Structures
Response
Ground Motion
Earthquake response spectrum shape
Horizontal direction peak response
Damping
Modeling
Frequency
Mode Shape
Torsional coupling
Mode Combination
Time History Simulation
Foundation-Structure Interaction
Ground motion incoherence
Vertical spatial variation of ground motion
SSI analysis
Earthquake Component Combination
Capacity
Strength
Material properties
Strength equation
Inelastic Energy Absorption
It is known by many seismic engineers that PGA is not the best parameter for determining structural
response or damage to NPP structures and equipment. Spectral acceleration averaged over a frequency
range such as 3.0 to 8.5 Hz as was used in the Diablo Canyon SPRA (1) is a better parameter. An average
spectral acceleration between 2.5 and 10 Hz was used in the SPRA for the K-Reactor at the Savannah
River Plant (2). Spectral acceleration-based parameters are believed to provide a better description of
damage, and their use is encouraged in SPRAs.
Each of the three ground motion variables is discussed below, relative to a single ground motion parameter
which is the average of the corresponding parameters from the two horizontal directions. The
reference parameter may be either PGA or a Sa-based parameter.
3-13
Earthquake Response Spectrum Shape. In many early SPRAs, average PGA was used as the ground motion
parameter, and the NUREG/CR-0098 median curve was assumed to be the reference response spectrum
shape for each horizontal direction. The vertical input at the ground level was assumed to be 2/3 of the
horizontal input.
However, it was noted that there was uncertainty in the earthquake signature which resulted in the variability
of the smooth response spectrum shape. In order to account for variability between (1) the smooth reference
response spectrum shape that was assumed in early SPRAs and (2) simulated earthquakes which are
typical of what could potentially occur, a spectral shape uncertainty, u , was specified to be included in
the structure fragility.
In addition, it was noted that, in general, real earthquakes would have spectra different from the
assumed smooth reference spectra used in the early SPRAs. Peaks and valleys in real response spectra
meant that a future earthquake response spectrum, with the same ground motion parameter, would have
spectral ordinates which were either higher or lower than the smooth reference spectrum. This peak and valley
variability, .r , was due to randomness and was also specified to be included in the structure fragility.
In essentially all PSHA conducted to date, this response spectrum peak and valley variability is
included in the development of the seismic hazard estimates. If UHS are used, the uncertainty in
spectral shape is also included in the seismic hazard estimate. Thus, the current recommendation is to
not include peak and valley variability or shape uncertainty in fragility estimates since this would
result in double counting of the variability in a PSHA.
Modern SPRAs utilize UHS as the basis of the reference response spectrum shape. Thus, the typical ranges of
values for peak and valley variability and uncertainty in spectral shape, previously given in Table 3-2,
have been removed. Table 3-2 now only provides recommendations for horizontal direction peak
response variability and for vertical response variability, as discussed in the following subsections.
Table 3-2
Structure Response Basic Variables
Logarithmic Standard Deviation

Basic Variable r u
Horizontal direction peak response 0.12 to 0.14 0

Ground vertical equals 2/3 ground horizontal 0.22 to 0.28 0.20 to 0.26
Site-specific analysis 0.22 to 0.28 less than
generic values
3-14
subsections entitled USE OF SCREENING TABLES IN SPRA, OBTAINING FRAGILITY
PARAMETER VALUES FROM SMA HCLPF CALCULATIONS, and ESTIMATING
MEAN CORE DAMAGE FROM RESULTS OF SMA which appear within pages 5-1 to 5-10.
All tables, figures, equations, and references cited refer to the original publication. The updates
are concerned with the removal of the peak and valley variability in the determination of HCLPF
screening values. Also, excess conservatism is removed from the capacity estimates made based
on HCLPF screening values:
USE OF SCREENING TABLES IN SPRA
In SMA structures and equipment are screened-out by verifying during the plant walkdown that
these rugged elements comply with the caveats in Tables 2-3 and 2-4 of the EPRI Report NP-6041 (1).
This process minimizes the number of components for which HCLPF calculations must be performed.
It is important to note that the screening step implies that the plant HCLPF capacity cannot
exceed the capacity level on which the screening guidance is based. For example, the success path
approach treats the safe shut-down equipment list (SSEL) as a series of components, like the links in a chain.
If any single link fails then the chain fails. Thus, the strength of the chain is equal to the strength of the
weakest link. In an analogous manner the SSEL is only as strong as the weakest component. When a
component is screened-out it is based on the assumption that the strength is greater than the
screening level, but it is not known how much stronger. Even if all the calculated HCLPFs are
greater than the screening level the plant HCLPF capacity for an SMA can only be stated as exceeding the
screening level.
In SPRA the screening tables can also be used to screen relatively strong elements out of the
analysis. To account for the capacity level assumed in the screening tables there are two approaches:1) an
additional element may be added to the SPRA to represent all the screened-out components, or 2) an
element added for each individual screened component . Otherwise, if all components were to be
screened-out, and no additional elements added to the SPRA, it might be concluded that the frequency of
failure is zero.
The purpose of the single additional or surrogate element is to represent all the structure and equipment
failure modes that are screened-out. It is judged by engineers who have experience in fragility analysis
that the median capacity is at least a factor of two greater than the HCLPF (e.g., see Reference 2). If the
ratio of median to HCLPF is conservatively assumed to be equal to 2, the corresponding
logarithmic standard deviation using the procedure in Section 4 is just:
1
c = ln(2) = 0.3
2.33
However, for adding several screened elements to a SPRA, the use of c = 0.3 is too low. For
individual SSCs, a more reasonable range for c is 0.4 to 0.5. Using a generic estimate for c of
0.45 for individual SSCs the ratio of median to HCLPF becomes:
3-15
Median/HCLPF = e 2.325( 0.45) = 2.85

The SMA screening tables are purposely conservative since they are applied to many situations. In
order to make the screening tables practical it was necessary to cover all potential situations without
making the caveats and restrictions difficult to use. In most cases, the components that are screened-out
have HCLPF capacities that are significantly higher than the screening table capacity levels.
Consistent with the assumption that SPRA is to be realistic, and not conservative nor optimistic, it would be
incorrect to replace each screened-out component by a conservative surrogate element. This is also consistent
with the results of past SPRAs that have shown that only a few components are significant contributors to
the frequency of core damage.
What this means in a practical sense in a SPRA is that each fault tree whose top event terminates at a
branch on an event tree has one surrogate element. This implies that in any series of elements in a fault
tree there is generally only one surrogate element. However, sometimes two fault tree branches are
connected in parallel and possibly could be composed of distinctly different types of elements (i.e., not
redundant trains). For this case it would be appropriate to place a surrogate element in each of
the two parallel branches since the surrogate elements would be independent of each other as
discussed in Section 2.
Placing a surrogate element in each fault tree is appropriate when evaluating a single fault tree,
but is not generally appropriate when evaluating the entire event tree. For this case the conservative
surrogate element should be placed as a single top event. However, if the results are evaluated for each
sequence in the event tree individually or for a set of sequences in a plant damage state, the conservative
surrogate element should only occur once for each case. The largest contribution that the surrogate element
should make to any sequence, plant damage state or core damage mean frequency is equal to the value
obtained by convoluting the mean fragility curve for the surrogate element with the mean site hazard
curve. This result tests the adequacy of using the surrogate element and the screening process in the
SPRA. If the mean frequency from the surrogate element is too large, then the site hazard is too high to
permit screening.
In modern SPRAs, the inclusion of individual screened out components in the analysis does not entail a large
level of effort. Thus, rather than consider the appropriateness of surrogate placement in various event trees,
the analysis may be conducted with all components represented individually.
For plants that perform screening using the rules in the GIP (3) the peak spectral acceleration of
the bounding spectrum is 0.8g, which is the same as the screening level for the first column in the EPRI
report NP-6041 Tables 2-3 and 2-4. The caveats and inclusion rules for the GIP are more conservative
than for NP-6041. On this basis the median spectral acceleration capacity of individual structures
and equipment screened-out are at least 2.85x0.8g = 2.3g, for equipment screened-out using the GIP.
This is on the conservative side.
If the second column in Tables 2-3 and 2-4 in EPRI NP-6041 are used for screening at higher
seismic sites then a higher median capacity can be used for the individual element. The HCLPF
capacity level for the second column in these two tables is 1.2g spectral acceleration at the ground. Using
the second column gives a higher median capacity (2.85x1.2g = 3.4g) for the individual element, but the
caveats required to satisfy the screening level are more stringent.
3-16
The screening tables of EPRI NP-6041 are defined in terms of peak 5% damped spectral acceleration of the
ground. For components mounted at elevation in structures it would be preferable for these screening tables to
be defined in terms of peak 5% damped spectral acceleration input to the component. Such an approach will be
considered in a follow-on document.
In situations where the mean hazard curve is expressed in terms of PGA, the median capacity of
the surrogate or individual element, which is based on the peak spectral acceleration value, can be
easily converted. The response spectrum shape assumed in the fragility analysis is used to make this
conversion. For example, if the second column in Tables 2-3 and 2-4 of EPRI NP-6041 is used for screening,
then the median PGA capacity is multiplied by PGA / peak Sa obtained from the 5% damped response
spectrum curve. Alternately, if the mean hazard curve is expressed in terms of average spectral acceleration a
similar conversion can be made.
The same screening procedures and benefits found in SMA can be transferred to SPRA. This
screening step provides an efficient means to minimize the level of analysis effort for determination
of HCLPF values. Anchorage must always be considered in addition to the element per se, but hopefully
through use of plant-specific calculations or conservative bounding calculations many anchorage details can
be shown to have relatively high capacities.
OBTAINING FRAGILITY PARAMETER VALUES FROM SMA HCLPF CALCULATIONS
Because the SPRA is usually controlled by a few elements the fragility analysis can be first
performed by scaling SMA HCLPF values to obtain median capacities. Typical logarithmic standard
deviation values from past SPRAs can be assigned to each component. These initial fragility values then would
be input into the SPRA systems model, and the dominate contributors identified. For those elements that
control the SPRA a more rigorous fragility analysis then would be performed for those few components.
This strategy enables a utility to utilize their in-house staff to do the screening (see discussion
above) and perform SMA HCLPF calculations for elements not screened-out. The CDFM method for
calculating HCLPFs is a deterministic approach that follows the same procedures that are
normally used in design, but with rules that have been liberalized to reflect the SMA philosophy. These
are procedures that many utility engineers are now familiar with.
The following are the steps utilizing this strategy that would be performed for a seismic IPE review
for a plant located at a 0.3g PGA site:
Set the screening limit to 1.2g spectral acceleration and use the second column in Tables 2-3
and 2-4 in EPRI NP-6041 to screen.
Walkdown the plant and verify that the rugged structures and equipment that are screened-
out satisfy the caveats.
Add the surrogate element to the appropriate places in the systems model as discussed above
or incorporate all screened elements in the system model. As suggested above individual
elements should have a median Sa capacity of 3.4g and a c value of 0.45 while surrogate
elements should have a median Sa capacity of 2.4g and a c value of 0.30.
Compute CDFM HCLPF capacities for all structures and equipment not screened-out.
Obtain approximate fragility curves for those elements not screened-out from:
3-17
Median = 2.5 CDFM HCLPF
c = 0.4
Use the approximate fragility curves in the systems analysis to identify the dominant
contributors to the SPRA.
For the few components that dominate the seismic risk, obtain more accurate fragility
parameter values and perform a new systems analysis to obtain a more accurate mean core
damage frequency and to confirm that the dominant contributors have not changed.
In the philosophy for the SMA methodology the HCLPF is defined to be at about the 1%
exceedance probability when there is no separation of variability into randomness and uncertainty parts. This
is equivalent to the HCLPF being defined at approximately the 95% confidence of about a 5% probability of
not being exceeded when the variability is separated. In past SPRAs it has been found that the
composite logarithmic standard deviation ranges from about 0.3 to about 0.5. Considering these two
results gives the following lower and upper bound median/HCLPF factors:
Median
lower bound: = e 2.33( 0.3) = 2.0
HCLPF
Median
upper bound: = e 2.33( 0.5) = 3.2
HCLPF
There have been a few cases where c has been larger (i.e., 0.6); however using a range of 0.3 to
0.6 would not change the conclusions in this section.
Note that the lower bound ratio of 2.0 is consistent with the statement used in the previous section
that the median is at least a factor of 2 greater than the HCLPF. If a c value is taken equal to 0.40, then the
median/HCLPF factor is 2.5.
ESTIMATING MEAN CORE DAMAGE FREQUENCY FROM RESULTS OF SMA
A simple approach to obtain a generally conservative estimate of the mean CDF is to first perform a
SMA and determine the plant HCLPF. Note that in this analysis the systems engineers develop
an SSEL that has a high-reliability. They have taken into consideration the reliability of the equipment
to operate on demand and the possible human errors that might occur when responding to a plant
accident. Although these probabilities are not formally included in the SMA when the EPRI success
path approach is used, they still guide the systems analyst as to whether a particular component should
be placed in the SSEL. The goal in SMA is to produce an overall operating reliability on demand of 99%
for the plant, which considers the two success paths that are required.
In this approach for obtaining an approximate mean CDF, the CDF is set equal to the value
corresponding to the HCLPF read from the mean hazard curve divided by 2. This is the essence of this
simplified approach.
For example, assume that the HCLPF capacity for the plant is determined to be 0.39 g. The annual
frequency of exceedance read from the western U.S. hazard curve shown in Figure 5-1 at 0.39 g
-4 -4
is 4.6x10 . Following the above recommendation, the approximate CDF is equal to 2.3x10 per year.
3-18
In general, this approach will be conservative because the SMA may not have taken credit for the
strongest success path. Also, since most hazard curves are concaved downward the recommended
factor of two is on the conservative side. For the above example, if the fragility curve were integrated directly
with the hazard curve then the rigorous mean CDF is found to be 1.6x10-4.
Figure 5-1
Typical Probabilistic Seismic Hazard Curves (4)
The basis for the recommendation for obtaining an approximate value for the mean CDF can be
seen from the rigorous solution. If the mean seismic hazard curve can be expressed in the form of a
power equation, i.e.,
kH
a
H (a ) = H (a m ) (5-1)
am
where: H(a) = annual frequency of exceeding ground acceleration, a
H(am) = annual frequency of exceeding median capacity, am
kH = hazard slope parameter
and the fragility curve is represented by the lognormal equation LN(am,c), the mean probability of
failure is given exactly by the following equation (4).
_ 2
pf = H (a m )e1 / 2 ( k Hc ) (5-2)
Combining these two equations the mean probability of failure can be written as follows where a* is a
specific acceleration value at which the hazard curve is to be evaluated:
3-19
_ a k H 1 / 2 ( k ) 2
p f = H (a*) e H c
(5-3)
a *
Now if a* is defined to be equal to a factor f times the HCLPF, i.e.,
a* = f {HCLPF} (5-4)
and remembering that the approximate, but reasonably accurate, relationship between the HCLPF
and the median is:
HCLPF = a me-2.3c (5-5)
then the sub-expression in Equation 5-3 becomes:

kH
a e 2.3k Hc
=
a * f kH
Finally, Equation 5-3 can be written as:
e 2.3k Hc +1 / 2 ( k Hc )
2
_
p f = H (f {HCLPF}) (5-6)
f k H
It was found in investigating various capacity levels (i.e., f {HCLPF}) at which the hazard is read
that the capacity corresponding to a 10% probability of failure gave the most reasonable and consistent
results. Since the HCLPF is defined at about the 1% probability of failure the factor f is given by the
following equation which is used with c values of 0.3, 0.4 and 0.5 in Table 5-1.
f = e c ( 2.31.282 ) (5-7)
From Eq. 5-7 the following values of factor f for the three c values are used in Table 5-1.
c
_0.3_ _0.4_ _0.5_
Factor, f 1.35 1.50 1.65
_
Table 5-1 evaluates the ratio equal to p f / H (f {HCLPF}) using Eq. 5-6 for the three cases that
cover the range of c and for various hazard curve slopes (see Section 2 for typical values of AR and kH).
Note from Table 5-1 that a factor of 0.5 [(i.e., divide H(f {HCLPF}) by 2] applied to the hazard frequency to
obtain an approximate mean CDF works very well for all cases, except for very large KH and c
values. The results for a slight additional simplification are shown in Table 5-2 where in all three c cases a
single factor, f, equal to 1.5 is used. Here also the factor of 0.5 on the hazard frequency used to obtain an
approximate mean CDF also works very well.
This simple procedure that uses the results of the SMA analysis provides a quick means for
estimating the frequency of core damage. This approach also ties together the SMA and SPRA analysis
procedures.
3-20
Table 5-1
Ratio of Probability of Failure to Hazard Probability Evaluated at fx (HCLPF)
Hazard Shape Factor Pf/H(fxHCLPF)
f = 1.35 f = 1.50 f = 1.65
AR KH c = 0.3 c = 0.4 c = 0.5

3.75 1.74 0.581 0.52 0.472
3.25 1.95 0.555 0.497 0.454

2.75 2.28 0.519 0.469 0.436
2.25 2.84 0.475 0.442 0.434
2.05 3.21 0.455 0.437 0.451

1.85 3.74 0.437 0.447 0.507
1.65 4.60 0.431 0.51 0.711
1.50 5.68 0.466 0.711 1.412
Table 5-2
Ratio of Probability of Failure to Hazard Probability Evaluated at 1.5x(HCLPF)
Hazard Shape Factor Pf/H(1.5xHCLPF)
AR KA c = 0.3 c = 0.4 c = 0.5

3.75 1.74 0.698 0.52 0.4
3.25 1.95 0.681 0.497 0.377
2.75 2.28 0.66 0.469 0.351
2.25 2.84 0.641 0.442 0.331

2.05 3.21 0.638 0.437 0.332
1.85 3.74 0.647 0.447 0.355
1.65 4.60 0.7 0.51 0.458

1.50 5.68 0.848 0.711 0.822
3-21
4
UPDATES TO EPRI TR-1002988
4.1 Basis for Updates
Appendix C of EPRI TR-1002988 was developed to demonstrate how capacity estimates for
components could be inferred using survival statistics of a number of components that had been
undamaged at HCLPF input motion levels. The prior Appendix C was formulated with a
conservative bias concerning assumed variability and a ground mounting location for all
components. For improved fragility estimates it is desirable to remove this conservative bias.
Appendix C of EPRI TR-1002988 has been revised to provide component capacity estimates for
ground-mounted components that are at least 2.85 times HCLPF levels. The additional
consideration of components mounted within structures allows the estimation of component
capacities that can be compared to clipped in-structure spectra values.
Appendix D of EPRI TR-1002988 was developed to provide an example of how the Appendix C
capacity estimates could be used to develop component fragilities. The revised Appendix D
considers the development of a fragility for the same example component using the new capacity
estimates considered in the revised Appendix C.
4.2 Updated Text for Appendix C of EPRI TR-1002988
The following represents the changes to the text of Appendix C of EPRI TR-1002988
ESTIMATION OF EQUIPMENT CAPACITY BASED ON EARTHQUAKE EXPERIENCE
DATA. All tables, figures, equations, and references cited refer to the Appendix of the original
publication. The updates are concerned with revisions to the capacity estimation procedures for
both ground mounted and structure mounted components:
C
ESTIMATION OF EQUIPMENT CAPACITY BASED ON EARTHQUAKE
EXPERIENCE DATA
Earthquake experience data was used to develop the SQUG Reference Spectrum (Reference C1)
which is implied to be a conservative capacity spectrum for verification of seismic adequacy.
This appendix utilizes survival analysis to determine the distribution on capacity for use in
development of seismic fragilities.
Let the average spectral capacity of a given equipment class, defined as a 5% damped spectral
acceleration value averaged over the 3-8 Hz frequency range, be represented by the random
variable C. The distribution of C is taken as log-normal with a known (assumed) log-normal
4-1
standard deviation,c , but an unknown log-normal mean, ln(Cm), where Cm represents the
median capacity.
Let the average spectral demand that the equipment class has been subjected to, defined as a 5%
damped free-field spectral acceleration value averaged over the 2.5-8 Hz frequency range, be
represented by the random variable D. The distribution of D is taken as log-normal with a
known (assumed) log-normal standard deviation,D, but an estimated log-normal mean, ln(Dm),
where Dm represents the median demand.
Next we consider n independent equipment items from the equipment class, with known free-field
spectral demand {D1 , --, Di , --, Dn} resulting in an average Reference Spectrum value,
Dave = RS. Each of the n items has survived the respective input motion represented by Di without
damage. Here, caveats (installation specifications) are used to define the equipment class which
excludes items with damage due to non-engineered attributes such as lack of anchorage or
inadequate restraint. Since we are interested in obtaining the highest estimate of capacity, we
also assume that D is representative of the strong ground motion which occurs within the
epicentral region of a major earthquake.
EVALUATION USING FREE-FIELD SPECTRAL DEMAND
To start this evaluation, we assume that the equipment has been directly subjected to a level of
mounting point motion equivalent to the free-field ground motion without building amplification
(or deamplification) effects. Much of the earthquake experience data equipment was actually
housed within structures that had mounting point motions that were amplified beyond the free-
field motion. Thus, the equipment survived motions greater than the free-field and the
assumption that the equipment has survived each of the known free-field spectral demand is a
conservative evaluation premise.
If now we consider the ratio of capacity to demand for each of the n items, Ci/Di, we conclude
that all n ratios are greater than unity or,
Ci / Di > 1,
since no damage has been observed in any of the n equipment items belonging to the equipment
class. We also note that the ratio of spectral capacity to spectral demand, X = C/D, is a log-
normal variable with mean, ln( X m ) = ln(C m / D m ) , and log-normal standard deviation,
{
X = ( D ) + (C )
2
}
2 1/ 2
. The probability of failure for an item of equipment is given by
PF = P(X < 1) = F(X = 1) ,
where F is the Cumulative Distribution Function (CDF) of X. If a reduced variate is defined as
u = ln( X ) / X , u 0 = ln( X m ) / X ,
4-2
then, given z = u u0 , we may write F(X) = (z) , where is the standard normal CDF. Thus
PF = F(X = 1) = P(u < 0) = P(z < u0 ) = (u0 )
The probability of survival for an equipment item is then,
Ps = 1 PF
Now, given n pairs of independent Di , Ci with known Di and average RS but unknown Ci , we
apply the constraint, Xi = Ci / Di > 1, since no failure has been observed in the n equipment
items. If the Xi are ordered such that, X1 < Xi < Xn , then the minimum probability of survival is
given by
P(Xi > 1) = i {1 F(Xi )}xi =1 = (1 PF )

n
Since Cm is unknown, it can only be specified by the assignment of a confidence coefficient. The
lower confidence limit on PF is found by considering the probability of an assumed failure for an
n+1 item of equipment. This probability of failure is taken as the confidence-level, , such that
the observed result of n cases of no failure is the best that could have occurred. Thus,
= 1 (1 PF )
n+1
is the probability of failure for at least one item given the survival of n items. Now we can
estimate the population mean, ln(Xm), which assures that, for a given level of confidence, , the
lowest capacity demand ratio of n equipment items will be greater than unity by requiring
PF = 1 (1 ) = ( u0 ) , or
1/(n+1)
u0 = 1 {1 (1 )1 /( n +1) }
where 1 is the standard normal inverse CDF.
Since u0 = ln( X m ) / X , we may write
X m = C m / D m = e uox
If the median demand, Dm, is estimated as D m Dave = RS , then the capacity associated with
95% confidence is given by
4-3
C 95 = RS e uox
where u0 is evaluated at = 0.95.
The High Confidence Low Probability of Failure (HCLPF) value, or 95% confidence of less than
a 5% failure, is given by the 5% capacity level on the 95% confidence CDF, or
C HCLPF = RS euox 1.645c = RS Fk
where we identify the factor, FK = euox 1.645c , as the reduction or knockdown factor applied to
the Reference Spectrum to achieve a HCLPF capacity value.
To complete the evaluation, the variability of the demand, D, and the variability of the capacity,
C, must be estimated. It is unlikely that equipment was subjected to ground motion more than
twice the Reference Spectrum level or less than one-half of the Reference Spectrum level. If the
log-normal standard deviation associated with the demand variable is set as D = 0.3, then less
than 2% of the equipment would have been subjected to motion outside the range 0.5 x RS to 2 x
RS. The selection of the log-normal standard deviation associated with the capacity variable is
more judgmental. Based on past studies, a reasonable range for C is 0.4 to 0.5, which yields a
generic mid-range estimate of C = 0.45.
Given D = 0.3 and C = 0.45 as representative log-normal standard deviations for spectral
demand and capacity, then X = 0.54 , and we obtain the following tabulation of
capacity/demand ratio for a confidence coefficient, = 0.95, or a 95% confidence-level, for
equipment survival for class group sizes ranging from 75 to 15.
n PF (-uo) Xm = Cm/Dm FK
75 0.038651 -1.76656 2.600 1.240
60 0.047924 -1.66533 2.461 1.174
55 0.052090 -1.62492 2.408 1.149
50 0.057048 -1.58005 2.350 1.121
45 0.063049 -1.52967 2.287 1.091
40 0.070461 -1.47237 2.217 1.058
35 0.079847 -1.40610 2.139 1.020
30 0.092114 -1.32785 2.051 0.978
25 0.108830 -1.23277 1.948 0.929
20 0.132946 -1.11257 1.825 0.871
15 0.170750 -0.95121 1.673 0.798
From this table, we note that a class group size of 30 is the minimum number of items necessary
to demonstrate that the Reference Spectrum level represents a HCLPF capacity level for an
approximate unity or greater knockdown factor.
The development outlined above provides an estimate of the population mean, ln(Cm), which, for
high levels of confidence, will be conservative (i.e., low) compared to the true population mean.
4-4
Our problem, as a set of n observations of no damage for the demand level recorded or
estimated for each observation, may be interpreted as a sample taken from a large population of
equipment meeting the attribute limits or caveats of the equipment class. We wish to infer an
estimate of the sample mean capacity, or ln(Cm), for which the conservatism is removed. This
would then provide an estimate of the true median capacity of the equipment to be used in risk
informed seismic evaluations of equipment.
One method of achieving this capacity estimate is to consider the HCLPF values computed
above, RS FK , as one-sided lower tolerance limits based on the sample size and sample mean
value. This may be represented by
ln (C np ) = ln(C m ) k np C
where Cnp is the lower tolerance limit , such that the probability is p that at least a proportion
lies below Cnp (or a proportion 1 lies above Cnp ) and where knp is the tolerance factor
based on p, , and sample size, n. In general, for the case of a known (or assumed) standard
deviation (Reference C2)
k np = 1 ( ) + 1 ( p ) /(n )1 / 2
If p = 0.95 and = 0.05, and we identify, Cnp = CHCLPF = RS FK , then
(Cm / RS)tol = FK e( knp ) c
and the following tabulation is obtained using the prior results for FK :
(knp)c
n FK e (Cm/RS)tol
75 1.240 2.283 2.832
60 1.174 2.307 2.708
55 1.149 2.316 2.661
50 1.121 2.328 2.610
45 1.091 2.341 2.554
40 1.058 2.357 2.493
35 1.020 2.376 2.424
30 0.978 2.400 2.347
25 0.929 2.431 2.259
20 0.871 2.474 2.154
15 0.798 2.538 2.025
These estimates of sample median capacity are viewed as low bound estimates of the true median
spectral capacity.
Another estimate of the population median spectral capacity may be achieved by noting that the
HCLPF capacity may be approximated by the product of the 1% value (-1(0.01) = -2.326) of
capacity (Reference C3) and a slight bias factor:
4-5
-2.326C
CHCLPF Cm (e ) 1.02
Using the sample HCLPF values as estimates of the population HCLPF, we may infer the
population median capacity. Again we let CHCLPF = RS FK , then
2.326C
{Cm/RS}1% = FK e /1.02
resulting in the alternate tabulation:
e
2.326 C
n FK {Cm/RS}1%
75 1.240 2.849 3.464
60 1.174 2.849 3.279
55 1.149 2.849 3.208
50 1.121 2.849 3.131
45 1.091 2.849 3.047
40 1.058 2.849 2.954
35 1.020 2.849 2.850
30 0.978 2.849 2.732
25 0.929 2.849 2.595
20 0.871 2.849 2.432
15 0.798 2.849 2.228
These estimates of the population median capacity are viewed as upper bound estimates of the
true median spectral capacity.
If these two median capacity estimates are taken as upper, U = {Cm/RS}1% , and lower,
L = {Cm/RS}tol ,bounds then we may estimate the median capacity by the geometric average of the
two bounds:
1/2
n L={Cm/RS}tol U={Cm/RS}1% (UL)
75 2.832 3.464 3.132
60 2.708 3.279 2.980
55 2.661 3.208 2.922
50 2.610 3.131 2.859
45 2.554 3.047 2.790
40 2.493 2.954 2.714
35 2.424 2.850 2.629
30 2.347 2.732 2.532
25 2.259 2.595 2.421
20 2.154 2.432 2.289
15 2.025 2.228 2.124
The FK , (C m / RS) tol , and (C m / RS)1% values obtained are based on an assumed C value of
0.45. Changes in the assumed C affect all of the estimated values. An increase in the assumed
C , lowers the HCLPF and increases the median capacity. Conversely, a decrease in assumed
C increases the HCLPF and decreases the median value. Thus, within a reasonable range of
C , we would not expect the unconditional failure rate of a derived fragility to vary by a
significant amount.
4-6
The sensitivity of C on the results is checked for n=50 which is a reasonable estimate of
equipment items within each equipment class that was subjected to the reference ground motion:
D C X
0.3 0.500 0.58
0.3 0.450 0.54
0.3 0.400 0.50
C
1/2
n FK {Cm/RS)tol {Cm/RS)1% (UL)
50 0.500 1.104 2.823 3.463 3.127
50 0.450 1.121 2.610 3.131 2.859
50 0.400 1.141 2.418 2.837 2.619
It is concluded that for the base case n=50, D = 0.3, and C = 0.45, the best estimate of median
spectral capacity for ground mounted equipment is Cm= 2.86 RS 2.85 RS.
EVALUATION USING IN-STRUCTURE SPECTRAL DEMAND
In the above development, it was conservatively assumed that all equipment was directly
subjected to the free-field demand motion without consideration of supporting structure
amplification. As noted previously, much of the earthquake experience data equipment was
housed within structures that that have amplified mounting point input motion. This more
realistic demand may be represented by D =D (AF), where (AF) is an amplification function.
The new ratio of capacity to demand would be C/D = C/[(AF)D] and n would now be
associated with the equipment items with structure borne mounting point demand. If we identify
C = C/(AF), then the prior development for estimation of the mean capacity still applies with C
= C (AF).
It is to be noted that the median free-field spectral demand, Dm= RS, is a wide-band spectral
acceleration value averaged over the 2.5 - 8 Hz frequency range. Amplified in-structure spectra,
however, have narrow-band response characteristics. As noted in Reference C4, the comparison
of dynamically determined capacity levels should be conducted on an effective wide-band basis.
Narrow-band earthquake in-structure spectra require clipping prior to comparison with
estimates of median spectral capacity. Thus, the amplification function, (AF), represents an
average value of the clipped ratio of in-structure spectra to free-field spectra over the frequency
range of the Reference Spectrum. A reasonable range for the amplification function, (AF) =
D/D, for the set of structures housing equipment at the reference data sites, is 1.0 to 1.67. If
(AF) is taken as a log-normal variable for which (AF) = 1.0 corresponds to a minus two-sigma
value and (AF) =1.67 corresponds to a plus one-sigma value, then the log-normal standard
deviation associated with the amplification variable may be estimated as:
AF = {ln[1.67/1.0]}/3 = 0.171
The median value of (AF) may then be estimated as:
(AF)m = e = 1.407
2 AF
The mean capacity of in-structure mounted equipment is then given by:

Cm= (AF)m Cm= 1.407(2.86 RS) = 4.02 RS 4.0 RS
4-7
For the development of the mean spectral capacity referenced to the free-field, both free-field
and in-structure reference data were represented, thus the estimated variability included the AF
as part of the variability C. To account for this inclusion, the variability associated with the
mean spectral in-structure capacity should be reduced to yield C = {(C)2 - (AF)2}1/2 = {(0.45)2 -
2 1/2
(0.17) } = 0.416. Then, using the 1% capacity value to estimate the HCLPF capacity we obtain
CHCLPF Cm (e-2.326c) = 1.53 RS 1.5 RS
SUMMARY
For ground-mounted items, the HCLPF capacity for equipment may be taken as CHCLPF = 1.1 RS
and the median capacity for equipment may be taken as Cm= 2.85 RS. The SQUG Reference
Spectrum has a value RS =1.2 g which makes the HCLPF capacity CHCLPF = 1.3 g and the median
capacity Cm = 3.4 g, for equipment screening and fragility estimates, respectively. These values
may be conservatively compared to either free-field demand levels or clipped in-structure
demand levels.
For structure-mounted items, the HCLPF capacity for equipment may be taken as CHCLPF = 1.5
RS and the median capacity for equipment may be taken as Cm = 4.0 RS. Using the SQUG
Reference Spectrum value RS =1.2 g, the resulting HCLPF capacity is CHCLPF = 1.8 g and the
median capacity is Cm = 4.8 g, to be used for equipment screening and fragility estimates,
respectively. These values are to be compared to clipped in-structure demand levels.
REFERENCES
C1. Generic Implementation Procedures (GIP) for Seismic Verification of Nuclear Plant
Equipment, Revision 2, Seismic Qualification Utility Group (SQUG), June, 1991.
C2. Hald, A., Statistical Theory with Engineering Applications, John Wiley & Sons, 1952.
C3. Kennedy, R. P., Overview of Methods for Seismic PRA and Margins Analysis Including
Recent Innovation, Proceedings of OECD/NEA Workshop on Seismic Risk, Aug. 10-12,
1999, Tokyo, Japan.
C4. Methodology for Developing Seismic Fragilities, EPRI Report TR-103959, Electric
Power Research Institute, Palo Alto, California, June 1994.
4.3 Updated Text for Appendix D of EPRI TR-1002988
The following represents the changes to the text of Appendix D of EPRI TR-1002988
EXAMPLE FRAGILITY FOR INSTRUMENT CABINET DERIVED FROM EXPERIENCE
DATA. All tables, figures, equations, and references cited refer to the Appendix of the original
publication. The updates are concerned with the application of the in-structure capacity estimates
developed in the revised version of Appendix C. EPRI TR-1002988 Appendix B, however,
includes a description of the development of in-structure response spectra for seismic margin or
seismic PRA evaluation by scaling. The data included within Appendix B as a scaling of
response spectra example is also used as the basis for the starting point for the Appendix D
4-8
example on the development of example fragility for an instrument cabinet derived from
experience data. Rather than repeat all of the data within Appendix B, references will be made to
the specific data and information used for the Appendix D example.
D
EXAMPLE FRAGILITY FOR INSTRUMENT CABINET DERIVED FROM
EXPERIENCE DATA.
In this example a fragility will be developed for an instrument cabinet located in Building E as
described in Appendix B. The instrument cabinet is welded to embeds and the anchorage and
base stiffness are deemed to be rugged relative to the demand. GERS are not available for
instrument cabinets, therefore seismic experience is selected for derivation of a fragility. The
instrument cabinet has no electro-mechanical relays, computers, programmable recorders or
strip chart recorders and meets the caveats of the GIP, Reference D1.
D.1 Demand
For this example we will use the in-structure spectrum for node 162610 shown in Figure B-11 in
Appendix B. This in-structure response spectrum reflects three reductions in the original
Uniform Hazard ground motion spectrum. The original ground motion UHS for the area was
anchored to 0.12 g and was used to develop amplified in-structure response spectra. Subsequent
site-specific studies determined that 85% of the area spectrum was more appropriate as a
median ground motion input. The resulting PGA was 0.102g as shown in the UHS spectrum,
Figure B-3 of Appendix B.
The ground motion spectrum peaks at 20 Hz and further reduction can be taken for Ground
Motion Incoherence (GMI) and for High Frequency Ductility (HFD) effects. Since the
instrument cabinet does not contain any electro-mechanical relays, the high frequency ductility
reduction is appropriate to reflect the effective reduction in damage at high frequency
acceleration due to small amounts of ductility that may be experienced in the cabinet base. The
resulting spectra shown in Figure B-11 of of Appendix B are the result of the three reductions
and are considered to be median centered response spectra. Since the site was a low seismic
hazard site and the structures and equipment were expected to have HCLPF values above the
1E-5 median UHS PGA, the 1E-5 UHS spectrum was considered the best estimate spectral shape
to define the input motion.
The UHS were only defined out to1E-5/year frequency of occurrence. Peak ground acceleration
was defined beyond 1E-5/year so the fragilities are anchored to PGA in lieu of spectral
acceleration.
In accordance with the procedure in Reference D4, the narrow band in-structure demand
spectrum can be clipped to represent a more realistic demand on the instrument cabinet for
comparison to the broad banded seismic experience spectrum. Referring to Figure B-11 of
Appendix B, and the clipping procedure in Reference D4, the following parameters and clipping
factor are derived.
4-9
The parameter f 80 is the frequency band width for the spectrum taken at 80% of the peak and is
about 2 Hz. The factor fc is the 14.5 Hz central frequency of the demand spectrum. The
bandwidth factor, B, is computed as
B = f0.8 / fc = 0.138
For B less than 0.2, the clipping factor, CC is:
CC = 0.3 + 0.86 B = 0.42
The peak spectral acceleration of the demand spectrum at 14.5 Hz is 1.5g. The clipped peak is
then:
SaC = 0.42(1.5) = 0.63g .
The uncertainty in the clipping factor, UCc is defined in Reference D4 as:
UCc = 0.37 0.5 B = 0.30

D.2 Capacity
The instrument cabinet has not been tested nor is there any available data on similar models.
Therefore the capacity is determined from earthquake experience. Reference D1 provides a
Reference Spectrum that was determined to be applicable to generic classes of equipment and
subsystems. In support of using seismic experience for verification of new and replacement
equipment in newer non A-46 plants, the SEQUAL Owners Group has done an extensive
reevaluation of the seismic experience data and has presented the results in a topical report,
Reference D2. A weighting of database site spectra was conducted, accounting for the number of
representative examples at each database site. The resulting average of weighted database
spectra very closely matches the GIP Reference Spectrum in the 2.5 to 7.5 Hz range. The
comparison of the SEQUAL experience spectrum to the SQUG Reference Spectrum is shown in
Figure D-1 for instrument cabinets and panels. It closely matches the SQUG database spectrum.
There are 46 documented independent examples of instrument cabinets and panels in the
SEQUAL database. Using the survival analysis methodology described in Appendix C, the
median and HCLPF capacity spectra can be derived. For 46 examples, the multiplier, FK, is
interpolated from Appendix C to be 1.1. Thus, the HCLPF spectrum for ground mounted
components is 1.1 times the SQUG spectrum. The median capacity multiplier for ground
mounted components, (Cm /RS) , is 2.8. It is logical to assume that the reference set of SEQUAL
seismic experience cabinets were housed within structures, thus the structure mounted capacity
factors apply. Using the median amplification function, (AF)m =1.4, the HCLPF spectrum for
structure mounted components is 1.54 times the SQUG spectrum and median capacity multiplier
for structure mounted components, (Cm /RS) , is 3.9.
4-10
The actual fundamental frequency of the instrument cabinet is not known. Based on the guidance
in Reference D3, the side to side and front to back fundamental modes would be greater than 8
Hz for a cabinet with welded anchorage and stiff base. Internal panels could also be expected to
be above 8 Hz. Since the demand spectrum in Figure B-11 of Appendix B peaks at 14.5 Hz, the
direct application of the SQUG Reference Spectrum (or SEQUAL spectrum) indicates a
reduction in apparent demand compared to the plateau region of the SQUG Reference Spectrum.
Reference D7 demonstrated that the characteristic shape of the Reference Spectrum beyond 7.5
Hz was due to the energy content of the records used to construct the Reference Spectrum and
that there was considerable evidence that the 1.2g spectral level, as a HCLPF measure, could be
conservatively extended into the high-frequency range beyond 7.5 Hz as a demonstrated
functional level for anchored components. The extended Reference Spectrum is indicated in
Figure D-1. The capacity factors developed in Appendix C are valid for defining equipment
capacity in the high frequency range (>8 Hz).
Figure D-1
Equipment Class 20 Control and Instrumentation Panels and Cabinets
The median capacity is then:
C m = (C m /RS)( RS) = 3.9(1.2 g ) = 4.68 g
The uncertainty in the capacity is as the minimum used in Appendix C
U C = 0.42
The uncertainty in the reference demand measure used in Appendix C is taken as randomness of
capacity
4-11
RC == 0.30
D.3 Capacity Factor

The median value of the capacity factor FC , is the ratio of the median capacity to the median
clipped demand.
FC = 4.68/0.63 = 7.43
The uncertainty and randomness in the capacity factor is derived from the uncertainty and
randomness in the capacity and the uncertainty in the clipped demand.
UFC = (0.30 2 + 0.42 2 )1 / 2 = 0.52
RFC = 0.30
D.4 Structural Response Factor

There are several variables that contribute to randomness and uncertainty in development of the
in-structure response spectra that define the demand. The original development of in-structure
response spectra was conducted using a state of the art three dimensional model of the
reinforced concrete structure. Since the model was three dimensional, torsional coupling effects
are automatically included.
The original in-structure spectra were developed by mode superposition time history analysis
using three simultaneous independent time histories whos response spectra closely matched the
ground motion UHS. At the 14.5 Hz fundamental frequency that results in the narrow banded
peak of the in-structure spectrum, the time history spectrum was about 10% higher than the
target UHS spectrum.
The structure is robust relative to the demand and is stressed to less than half of the yield
capacity, therefore, a best estimate of structural damping of 4 % was used in the analysis.
The resulting in-structure spectra were then scaled to account for ground motion incoherence,
(GMI) and for high frequency ductility (HFD) effects using the random vibration methodology
described in Appendix B.
In most cases, the parameters used are considered to be median centered thus the response
factors are unity unless otherwise noted. The variables to be addressed in deriving the structural
response factor and it variability are:
Spectral Shape (SS)
Damping (D)
Modeling (M)
Frequency
Mode Shape
4-12
Mode Combination (MC)
In addition, there is some uncertainty associated with the scaling of in-structure spectra to
account for:
Ground Motion Incoherence (GMI)
High Frequency Ductility Effects (HFD)
Scaling using random vibration theory (RV)
D.4.1 Spectral Shape (SS)
At the frequency of interest, the response spectrum resulting from the time history input motion
was about 10% greater than the target UHS. In addition, probabilistic response invariably
results in a lower peak response than a so called median centered deterministic response. In
Reference D4, a demand reduction factor of 0.92 is suggested to account for this. The FOAKE
study (Reference D5) showed a demand reduction of at least 10% for the three models used on
rock sites. For soil sites the demand reduction was much larger. For the rock site in this
problem, the demand reduction is considered to be 0.9, resulting in a demand reduction factor
of:
FDR = 1/0.9 = 1.11
If no reduction is considered to be about a 1% probability case (2.33 R case), then
1
RDR = ln 1.11 = 0.04
2.33
In addition, the input motion is defined at the average of the two orthogonal components of
earthquake. One direction could have a peak greater than the other. The random variability, R ,
in the ratio of horizontal components is about 0.13. The resulting spectral shape factor and
variability are:
FSS = 1.1(1.11) = 1.22
RSS = (0.04 + 0.13 ) = 0.14

2 2 1/2
D.4.2 Damping (D)

Structural damping used in deriving the in-structure spectra was 4 %. For structures at less
than yield, Regulatory Guide 1.60 would recommend 4% for reinforced concrete. Reference
D4 recommends 5% as a median value for reinforced concrete with considerable cracking.
Considerable cracking at less than yield was not expected for the building E structure and a
compromise median damping value of 4 % was used as a best estimate.
FD = 1.0
4-13
Reference D4 suggests that 3% is a 1U value for damping. In the amplified portion of the
UHS, we can estimate the uncertainty is spectral acceleration due the uncertainty in damping as
the logarithm of the square root of the damping ratio.
UD = ln (4.5/3)1/2 = 0.2
D.4.3 Modeling (M)

There are two variables to consider under modeling, structural frequency and mode shape. The
model is considered to be a best estimate of the structural load path from roof to foundation. Per
Reference D4, when code properties are used for concrete, the calculated and actual stiffness
are similar, thus we consider the calculated fundamental frequency to be median centered. For a
detailed model such as being considered, Reference D4 suggests that the uncertainty, f , on
frequency is about 0.15. The -1 frequency is then:
f1 = e0.15 (14.5) = 12.48 Hz
The UHS ground motion response spectrum shown in Figure B-3 is fairly flat in this region and
the change in spectral acceleration is only about 3%.
Uf = ln (1.03) = 0.03
For complex structures, Reference D4 recommends that the uncertainty on mode shape,
UMS = 0.15.
The resulting uncertainty in the modeling is then:
UM = (0.03 2 + 0.15 2 )1 / 2 = 0.15
D.4.4 Mode Combination (MC)

Mode superposition time history analysis was conducted. The combination of modal responses
by SRSS was considered to be median centered. From the response spectra in Figure B-11, the
dominant direction X spectrum appears to be dominated by a single mode. Some contribution
from higher modes would be present but not very influential for this case. The mode combination
randomness is small and is estimated to be:
RMC = 0.05
D.4.5 Ground Motion Incoherence (GMI)

From Appendix B, the reduction factor for ground motion spectral acceleration at 14.5 Hz is
about 0.87. If we consider the no reduction case to be an upper limit on input motion of about
3, the uncertainty for GMI is:
UGMI = (1/3) ln (1/0.87) = 0.05
4-14
D.4.6 High-Frequency Ductility Reduction (HFD)

From Appendix B, the high-frequency reduction factor for the ground motion spectrum at 14.5
Hz is about 0.7. With no reduction being an extreme 3 limit,
UHFD = (1/3) ln (1/0.7) = 0.12
D.4.7 Scaling Using Random Vibration Theory (RV)

The scale factors developed using random vibration theory are considered to be quite accurate
since the scale factors were developed from a ratio of response between the original and reduced
ground motion spectra rather than comparing a single response analysis for the reduced ground
motion spectrum to the original response analysis. Some uncertainty is of course present in any
analysis and this uncertainty in the method of scaling is estimated to be:
URV = 0.10
D.4.8 Structural Response Factor (FRS)

All variables for structural response were unity except for spectral shape ( FSS = 1.22).
FRS = 1.22
= SRSS of individual s
= SRSS (SS , D , M, MC , GMI, HFD , RV )
RRS = (.142 + 0 + 0 + 0.052 + 0 + 0 + 0)1 / 2 = 0.15
U RS = (0 + 0.202 + 0.152 + 0 + 0.052 + 0.12 2 + 0.102 )1 / 2 = 0.30
D.5 Fragility
The median capacity is the product of the capacity factor, structural response factor and the
PGA of the 1E-5 UHS spectrum used for the determination of structural response level.
A m = FC FRS ( PGA )
A m = 7.43(1.22)(0.102 g ) = 0.92 g
R = (0.302 + 0.152 )1 / 2 = 0.34
U = (0.52 2 + 0.30 2 )1 / 2 = 0.60
A HCLPF = 0.92 g (e) 1.65( 0.34+0.60 ) = 0.195g
4-15
Note in this case that the HCLPF capacity derived from seismic experience data, in terms of
PGA, is a factor of 1.9 greater than the 1E-5 UHS evaluation spectrum. In this particular case,
the seismic hazard is very low and if convolved with the fragility, the resulting unconditional
failure rate would be reasonably low.
D.6 References
D1. SQUG, Generic Implementation Procedure (GIP) For Seismic Verification of Nuclear
Power Plant Equipment, Revision 2, Corrected, Seismic Qualification Utility Group, June
28, 1991.
D2. SEQUAL, Topical Report, Basis for Adoption of the Experience-Based Seismic Equipment
Qualification (EBSEQ) Methodology by Non-A46 Nuclear Power Plants, SEQUAL
Owners Group, April 2001.
D3. EPRI TR-102180, Guidelines for Estimation or Verification of Equipment Natural

Frequency, Electric Power Research Institute (EPRI), Palo Alto, California, March
1993.
D4. EPRI TR-103959, Methodology for Developing Seismic Fragilities, Electric Power
Research Institute (EPRI), Palo Alto, California, June 1994.
D5. FOAKE (1993), Technical Core Group Report on FOAKE Task E-1, ASME Piping,
Appendix M, Quantification of the Margin of Conservatism in Seismic Design In-Structure
Response Spectra, January, 1993.
D6. EPRI TR-1002988, Seismic Fragility Application Guide, Electric Power Research
Institute (EPRI), Palo Alto, California, December 2002.
D7. EPRI NP-7498, Recommended Procedures to Address High-Frequency Ground Motions

in Seismic Margin Assessment for Severe Accident Policy Resolution, Appendix B of
Industry Approach to Seismic Severe Accident Policy Implementation, Electric Power
Research Institute (EPRI), Palo Alto, California, November 1991.
4-16
5
REFERENCES
1. EPRI (1991b) A Methodology for Assessment of Nuclear Power Plant Seismic Margin,
EPRI NP-6041SL, Revision 1, Electric Power Research Institute (EPRI), Palo Alto,
California, June 1991.
2. EPRI (1994) Methodology for Developing Seismic Fragilities, EPRI TR 103959,

Electric Power Research Institute (EPRI), Palo Alto, California, June 1994.
3. EPRI (2002) Seismic Fragility Application Guide, EPRI 1002988, Electric Power
Research Institute (EPRI), Palo Alto, California, December 2002.
4. USNRC (1991a) Individual Plant Examination of External Events (IPEEE) for Severe
Accident Vulnerabilities 10 CFR 50.54 (f) (Generic Letter No . 88-20, Supplement 4)
June, 1991.
5. USNRC (2000b), Risk-Informed Regulation Implementation Plan, October 2000.
5-1
A
R. P. KENNEDY WHITE PAPER ON PEAK AND VALLEY
VARIABILITY TREATMENT
A-1
RPK-90108
Issues Concerning the Inclusion of Response

Spectral Peak and Valley Variability rs in
Development of Fragility and HCLPF Capacity Estimates
Robert P. Kennedy
January 16, 2009
1. Introduction
Essentially all probabilistic seismic hazard assessments (PSHA) have included the response
spectral peak and valley variability rs as part of the aleatory variability when developing seismic hazard
estimates as a function of the annual frequency of exceedance (AFE). Thus, at any AFE, the resulting
uniform hazard response spectrum (UHRS) already fully includes the effect of rs. This statement is true
irrespective of whether the UHRS is defined at the mean, median, 90%, 84%, or 16% non-exceedance
probability (NEP) because the aleatory variability (including rs) similarly affects the slope of each of these
NEP hazard curves. The difference in amplitude of these various NEP curves is due to the epistemic
uncertainty.
I know of only one exception to the above paragraph. The exception is the Diablo Canyon PSHA
which removed rs from the hazard aleatory variability by developing hazard curves for spectral
acceleration averaged over a broad frequency range. However, the above paragraph is applicable for all
PSHA conducted using the EPRI 1989 Methodology, the EPRI 2003 attenuation relationships, or the LLNL
1993 Methodology.
Unfortunately, until recently, fragility estimates performed in accordance with the standard
Fragility Analysis (FA) Methodology have also included the response spectral peak and valley variability
rs in the fragility estimates. Thus, the effect of rs has been double counted when the hazard and fragility
curves are convolved together to estimate the seismic risk. This double counting of rs has resulted in
seismic risk being overstated by as much as a factor of two.
Prior to the Diablo Canyon seismic probabilistic risk assessment (SRPA), this double counting of
rs was not recognized. As stated above, to remove this double counting of rs for the Diablo Canyon SPRA,
rs was removed from the aleatory variability of the PSHA. The authors of EPRI TR-103959 Methodology
for Developing Seismic Fragilities (Ref. 1) incorrectly assumed that rs would be removed in the future
from the aleatory variability of the PSHA similarly to what had been done on the Diablo Canyon SPRA. As
a result, pages 3-4 through 3-7 of Ref. 1 recommend inclusion of rs in the range of 0.2 to 0.3 in the fragility
estimate as had been the prior common practice. Although Ref. 1 recommended defining the seismic hazard
over a broad frequency band similar to the Diablo Canyon SPRA, no caution was provided against double
counting of rs if this was not done. Unfortunately, it has not been done in other PSHA.
Since Ref. 1 was written, PSHA have continued to include the response spectral peak and valley
variability rs in the aleatory variability included in the hazard estimate. Thus, inclusion of rs in the
fragility estimate as recommended in Ref. 1 has continued to result in double counting rs. Furthermore, in a
number of more recent fragility estimates, rs has been increased to as much as 0.4 which compounds the
double counting even more.
Since it does not appear likely to get the seismic hazard estimators to remove rs from the aleatory
variability included in the PSHA, the current recommendation is to drop rs from the fragility estimate
because it is included in the hazard estimate.
2. Recommendation for Modifying Existing Fragility Analyses
In the Fragility analysis (FA) Methodology, the Demand Analysis is either performed by multiple
time-history analyses to estimate both the median and variability of the Demand, or by response spectra

28625 Mountain Meadow Road, Escondido, CA 92026
(760)751-3510 (760) 751-3537 (Fax)
email: bob@rpkstruct.com
analysis to estimate the median Demand and then estimate the overall Demand variability by an SRSS
combination of parameter variabilities. In the response spectra method, the response spectral peak and
valley variability rs is specifically defined as one of these parameter variabilities. Therefore, it is easy to
determine the rs value used in the fragility evaluation.
When the multiple time-history analysis method is used to determine the Demand median and
variability, it is more difficult to separate out the rs included in the analysis. The multiple time-histories
used as input are selected and possibly conditioned so as to produce median spectral accelerations at all
natural frequencies in the frequency range of interest (generally 2 to 20 Hz) that closely match the target
Uniform Hazard Response Spectrum (UHRS). However, unless the individual records are strongly
conditioned to closely match the target UHRS so that variability from the UHRS shape is small, the
spectral accelerations at each frequency do not match the target UHRS for each individual record. For the
suite of individual records, it is then necessary to compute the rs at each natural frequency and average
these rs values over the frequency range of interest. Within my experience of reviewing the input to these
multiple time-history analyses, the rs values have typically lay in the 0.2 to 0.4 range and have often been
close to the upper end of this range unless care was taken in selecting and conditioning the individual time-
history records.
Once the rs used in the fragility analysis has been determined, the composite (mean) variability c
that doesnt include rs can be computed from:
[2
c = crs 2rs ]
0.5
(2.1)
where crs is the reported composite variability which included rs . Then the corrected mean fragility curve
to be convolved with a mean hazard curve that includes rs in the aleatory variability is:
C p = C 50% exp x p c ( ) (2.2)
where Cp is the capacity at the p failure probability, C50% is the median capacity and xp is the standard
normal variant corresponding to the p non-exceedance probability (NEP).
Since the HCLPF capacity is defined by the 1% failure probability capacity on the composite
fragility curve, the corrected HCLPF capacity is given by:
HCLPFc = C1% = C 50% exp( 2.326 c ) (2.3)
The HCLPFrs reported in a fragility analysis that included rs in the fragility evaluation needs to be
increased by a factor Fc to avoid double counting of rs, where:
HCLPFc = Fc HCLPFrs (2.4)
Fc = exp[2.326( crs c )] (2.5)
Table 1 shows the required correction factor Fc for the typical range of c from 0.3 to 0.6 and rs from 0.2 to
0.4.
For rs=0.2, the increase factor Fc for a HCLPF capacity that doesnt double count rs already
included in the aleatory variability considered for the hazard curve ranges from 1.08 to 1.15, i.e., the
increase factor is relatively moderate. However, for rs=0.4, this increase factor ranges from 1.33 to 1.59.
For simplicity, the discussions in this and subsequent sections are written in terms of the
composite variability c. However, the same approaches apply when variability is subdivided into random
variability R and uncertainty U. See Appendix Section B.1 for discussion of the correction to R and U.
2
3. Recommendation for New Fragility Analyses
Since the response spectral peak and valley variability rs is already being included in the aleatory
variability considered in the PSHA, it should not be included in the Fragility Analysis. The rs
recommended on pages 3-4 through 3-7 of Ref. 1 should not be added.
When the multiple time-history analysis method is used to determine the Demand median and
variability either each of the time-histories used should be strongly conditioned so that the response spectra
for each time-history closely match the target UHRS at all frequencies, or the rs should be computed as
discussed in the previous section, and the composite (mean) variability c should be corrected by the use of
Eqn. 2.1. So long as rs is less than 0.2, the correction is small and can be ignored with only a small
conservative bias being introduced to the computed HCLPF capacity and the computed seismic risk.
However, if rs exceeds 0.2, this correction should be made in order to avoid excess conservatism.
4. Recommendation for Use of HCLPF Capacities Computed by the Conservative Deterministic Failure
Margin Methodology
The Conservative Deterministic Failure Margin (CDFM) Methodology defined in Refs. 2 through
4 is aimed directly at estimating the 1% failure probability capacity C1% on the composite fragility curve.
Thus:
HCLPFCDFM = C1% = CDFM Capacity (4.1)
However, the CDFM method determines the C1% capacity assuming a smooth target input response
spectrum shape with no peak and valley variability rs about this target response spectrum shape. An
updated description of the CDFM Method is presented in Appendix A.
As such, if the target response spectrum shape used in the CDFM evaluation is identical to the
target UHRS obtained from the PSHA that included rs in the aleatory variability, then no correction should
be made to the CDFM computed HCLPF capacity, i.e.:
HCLPFc = HCLPFCDFM (4.2)
No so-called HCLPF50/HCLPF84 correction discussed on pages 2-34 through 2-36 and pages 5-1 through
5-6 of Ref. 1 should be made to the CDFM computed HCLPF for use with PSHA hazard curves in which
the peak and valley rs has already been included in the hazard aleatory variability. Making this correction
double counts the peak and valley variability rs.
However, in many cases, the CDFM method computed HCLPF capacity was computed using a
target response spectrum shape that is substantially different from the target* UHRS shape defined by the
PSHA. When these two response spectra shapes differ significantly, the HCLPF capacity should be
modified to be appropriate for the target UHRS shape. This modification is especially important when the
HCLPF capacity is defined in terms of a Peak Ground Acceleration (PGA) capacity since structure and
component failures are primarily induced by spectral accelerations (SA) in the 2.5 to 10 Hz range as briefly
discussed on page 3-3 of Ref. 1.
*
The term target UHRS shape is used herein to avoid distinguishing between the mean, median, or 84%
NEP UHRS shapes. For estimating mean risk, it would be most rigorous to use the mean UHRS shape
corresponding to an annual frequency of exceedance (AFE) less than ten times the computed mean risk.
However, when anchored to a common spectral acceleration in the 5 Hz to 10 Hz frequency range, the
differences between the mean, median, and 84% NEP UHRS shapes are negligible over the 2 Hz to 20 Hz
frequency range of greatest interest. Thus, from a practical standpoint it is immaterial as to whether the
mean, median, or 84% NEP UHRS shape is used. The differences are larger in the 2 Hz to 20 Hz range
when the mean, median, or 84% NEP UHRS shapes are anchored to a common PGA.
-3-
A reasonable approximate correction is to report the HCLPF capacity in terms of the 5 Hz spectral
acceleration SA5 using the spectral shape used in the CDFM based HCLPF calculations. This SA5 HCLPM
capacity can then be converted to a fragility and subsequently convolved with the 5 Hz hazard curve to
obtain an estimate of the mean seismic risk.
For plant damage state risk a more rigorous approach is to report the HCLPF capacity in terms of
the 1, 2.5, 5, 10, and 25 Hz spectral acceleration capacities as well as the PGA capacity based on the
spectral shape used in the HCLPF calculations. Next, each of these HCLPF capacities are convolved with
the corresponding hazard curve to obtain failure probabilities P1, P2.5, P5, P10, P25, and Ppga. Then the best-
estimate of the mean seismic risk P is given by:
f1 P1 + f 2.5 P2.5 + f 5 P5 + f10 P10 + f 25 P25 + f pga Ppga

P= (4.3)
f
where the various f factors are judgement based weighting factors based on the importance of a particular
natural frequency on the estimate of damage. In this approach, in my judgement the sum of the f2.5, f5, and
f10 weighting factors should add to about 65% to 70% of the total sum of f factors in order to account for the
importance of the 2.5 to 10 Hz frequency range on damage.
For individual structures, systems, and components (SSCs), the approach described in the prior
paragraph is impractical to apply to every SSC. For individual SSCs, if the UHRS shape is significantly
different than the reference spectral shape used to develop HCLPFs or fragilities, a new structural response
analysis can be performed to define the demand on SSCs in the structures. The new demand is then
compared to the demand used to define the HCLPF or fragility in order to scale the existing
HCLPF/fragility.
This same issue of differences in the target response spectrum shape used in the fragility analysis
from the target UHRS shape defined by the PSHA can also exist for fragilities computed by the Fragility
Analysis (FA) Methodology described in Ref. 1. If these two spectral shapes differ significantly, the above
described corrections can also be applied in the FA Methodology.
5. Discussion of the Terms HCLPF50 Versus HCLPF84
The terms HCLPF50 and HCLPF84 used on pages 2-34 through 2-36 and pages 5-1 through 5-6 of
Ref. 1 are based on an outdated concept and should be abandoned. The term HCLPF84 referred to a HCLPF
value computed without consideration of the response spectral peak and valley variability rs and most
commonly referred to a HCLPF value computed by the CDFM Method because this method does not
include any consideration of rs. This HCLPF84 value was initially intended for use in comparison with a
deterministic design or evaluation response spectrum defined at the 84% NEP amplification since this
response spectrum included consideration of the peak and valley variability rs.
Conversely, the term HCLPF50 referred to a HCLPF value computed with inclusion of the peak
and valley variability rs in the fragility and HCLPF computation. This HCLPF50 term was reserved for
HCLPF values computed by the standard Fragility Analysis (FA) Methodology described in Ref. 1 which
included consideration of rs. This HCLPF50 value was initially intended for use in comparison with a
deterministic design or evaluation response spectrum defined at the median (50%) amplification which did
not include consideration of rs.
The concept of HCLPF84 and HCLPF50 was reasonably valid so long as one was dealing with
deterministic design or evaluation response spectra defined by either 84% or 50% amplification factors.
However, this concept is not relevant when dealing with UHRS determined from a PSHA evaluation. The
mean, median 16%, 84%, and 90% UHRS all include the same aleatory variability. Therefore, the peak and
valley variability rs is equally considered whether one is dealing with a mean, median, 16%, 84%, or 90%
UHRS. The difference in these UHRS is due to the epistemic uncertainty being considered.
4
The key issue now is whether the response spectral peak and valley variability rs has been
included in the aleatory variability considered in the PSHA. Since rs has been included in essentially all
PSHA evaluations, this rs should not be included in the fragility evaluation on the HCLPF capacity
evaluation.
So long as rs has been included in the aleatory variability considered in the PSHA, the correct c
to be used in the fragility analysis should not include rs so as to avoid double counting the effect of rs. If
rs has been included in the fragility analysis, then the corrected c should be computed from Eqn. (2.1).
Similarly, the corrected HCLPFc should not include rs and should be obtained from Eqn. (2.4) and (2.5).
The appropriate HCLPF capacity to be used with PSHA hazard curves that are based on rs being
included in the aleatory variability (i.e., all currently available hazard curves except the Diablo Canyon
hazard curves) is defined herein as HCLPFc. This HCLPFc is equal to the following older HCLPF terms
used in the literature:
HCLPFc = HCLPFCDFM = HCLPF84 (5.1)
If one has a HCLPF value that includes rs in its computation (called HCLPFrs herein), this
HCLPFrs should be converted to HCLPFc using Eqns. (2.4) and (2.5). Even though HCLPFrs is equal to the
outdated term HCLPF50, it is not correct to convert HCLPF50 to HCLPF84 using Eqn. (2-14) of Ref. 1 and
then define HCLPFc by HCLPF84. The Eqn. (2-14) of Ref. 1, i.e.:
HCLPF84 = HCLPF50 exp (rs)
provides an incorrect conversion. Table 2 shows the ratio (HCLPF84/HCLPF50) computed by the incorrect
Eqn. (2-14) of Ref. 1. This ratio does not match very well the correct correction factor Fc shown in Table 1.
The ratio (HCLPF84/HCLPF50) is too high when rs is less than about 0.3.
6. Use of Screening Tables in SPRA
Pages 5-1 through 5-4 of Ref. 1 describe the use of screening tables from Ref. 2 in a SPRA
evaluation. This section recommends that all screened out components be replaced by a surrogate element
in the SPRA with the surrogate element having the following fragility parameters:
SA50 = 2SASL exp(-rs) (6.1)
c = 0.3
where SA50 is the median peak spectral acceleration capacity of the surrogate element, SASL is the screening
level peak 5% damped spectral acceleration of the ground motion, rs is the response spectral peak and
valley variability included in the fragility evaluation, and c is the composite (mean) variability assigned to
the surrogate element. The low c of 0.3 is recommended in Ref. 1 for the surrogate element because the
single surrogate element is used to replace a number of components whose failures are combined by OR
statements in the system model. The combined HCLPF capacity is not significantly reduced by this OR
combination of individual component fragilities. However, the combined median capacity is significantly
reduced. The resulting median/HCLPF factor of 2 in Eqn. 6.1 is obtained from:
exp[2.326(0.3)] = 2.0
Equation (6.1) remains valid. However, when the peak and valley variability is already included in
the aleatory variability used in the PSHA, the peak and valley variability rs to be included in the fragility
analysis should be set to zero in order to avoid double counting this source of variability. Therefore Eqn.
(6.1) becomes:
SA50 = 2SASL (6.2)
-5-
c = 0.3
HCLPF = SASL
Although Eqn. (6.1) is appropriate, the recommendation on Page 5-2 of Ref. 1 to use rs in the
range of 0.2 to 0.3 is not appropriate so long as the peak and valley variability is included in the aleatory
variability used in the PSHA. This recommendation results in double counting of rs in the SPRA.
The screening tables of Ref. 2 can also be used to estimate the HCLPF capacity of individual
SSCs in lieu of screening these SSCs out of the system model and replacing them by a surrogate element.
However, for this application, the use of c=0.3 is too low. For individual SSCs, a more reasonable range
for c is 0.4 to 0.5. Using a more reasonable generic estimate for c of 0.45 for individual SSCs:
Median/HCLPF = exp[2.326(0.45)] = 2.85
Thus, for individual SSCs:
SA50 = 2.85 SASL (6.3)
c = 0.45
HCLPF = SASL
is more appropriate.
A note of caution on the use of the Ref. 2 screening tables to estimate HCLPF and fragility
capacities is that these screening tables do not directly address anchorage capacities in many cases.
Anchorage HCLPF capacities need to be individually assessed, and often control.
Another note is that these screening tables are defined in terms of peak 5% damped spectral
acceleration of the ground. For components mounted at elevation in structures it would be preferable for
these screening tables to be defined in terms of peak 5% damped spectral acceleration input to the
component. This issue is addressed in Appendix Section B.2.
7. Summary of Recommendations
The response spectral peak and valley variability rs is being included as part of the aleatory
variability considered in the PSHA. Therefore all resulting hazard curves and all resulting UHRS (mean,
median, 84% NEP, etc.) already account for the peak and valley variability rs. As a result, this peak and
valley variability rs should be removed from estimates of the fragility variability c and the HCLPFc
capacity so as to avoid the double counting of rs.
When an existing fragility analysis has included rs, the corrected c can be obtained from Eqn.
(2.1). Similarly when the existing HCLPFrs capacity has included the effect of rs, the corrected HCLPFc
capacity can be obtained from Eqns. (2.4) and (2.5). Representative examples of the resulting increase
factor Fc are shown in Table 1. This topic is further discussed in Section 2.
New fragility analyses should not include the peak and valley variability rs since it is already
included in the aleatory variability used in the PSHA. Recommendations for avoiding inclusion of rs when
multiple time-history analyses are performed are presented in Section 3.
The Conservative Deterministic Failure Margin (CDFM) Methodology defined in Refs. 2 through
4 estimates the 1% failure probability capacity C1% on the composite fragility curve without inclusion of the
peak and valley variability rs. Therefore HCLPFCDFM can be used directly to define C1%. It should not be
modified by any correction factor associated with rs. This topic is further discussed in Section 4.
When the target response spectrum shape used in either the CDFM analysis or Fragility Analysis
(FA) differs significantly from the target UHRS shape develop from the PSHA evaluation, a correction in
6
the HCLPF capacity and fragility estimate needs to be made before convolving fragility and hazard
estimates to obtain the estimate of seismic risk. Recommendations for correcting the problem are also given
in Section 4. These recommendations are equally applicable for both CDFM and FA evaluations. However,
in general, this problem is likely to be more significant in existing CDFM evaluations than it is in existing
FA evaluations. Even so, the issue should be considered for both types of existing evaluations.
The terms HCLPF50 and HCLPF84 are outdated terms applicable when the HCLPF capacity was
being compared to deterministic design or evaluation response spectra defined with either 50% or 84%
amplification, respectively. These terms have no relevance when the HCLPF capacity is being compared to
a PSHA determined UHRS irrespective of whether this UHRS is a mean, median, or 84% NEP UHRS. All
of these UHRS have included the peak and valley variability rs as part of the aleatory variability
considered in the UHRS. This issue is amplified upon in Section 5.
The relevant issue is whether the peak and valley variability rs has been included in the FA
evaluation. If it has been included, the HCLPFrs capacity needs to be increased by the correction factor Fc as
discussed in Section 2. The HCLPF84/HCLPF50 ratio defined by Eqn. (2-14) of Ref. 1 does not provide an
appropriate correction to account for the effect of rs on the HCLPF capacity as can be seen by comparing
the ratio HCLPF84/HCLPF50 in Table 2 with the correct Fc factor in Table 1 for various rs and c values.
Section 6 revises the discussion given on pages 5-1 through 5-4 of Ref. 1 on the appropriate
median value to use for the surrogate element to account for screened out components. Since rs has already
been included in PSHA hazard curves, it should not also be included in developing the median capacity of
the surrogate element. Thus, the median value of the surrogate element should be set at twice the screening
level.
The overall summary is that for the FA method, every place that Ref. 1 discusses the peak and
valley variability rs, the appropriate value to use is rs=0 since this variability has already been included in
the aleatory variability of the PSHA. Including rs in the fragility analysis conservatively double counts this
source of variability.
-7-
8. References
1. Methodology for Developing Seismic Fragilities, EPRI TR-103959, Electric Power Research
Institute, June 1994
2. A Methodology for Assessment of Nuclear Power Plant Seismic Margin, EPRI NP-6041-SL,
Revision 1, Electric Power Research Institute, August 1991
3. Kennedy, R.P., Overview of Methods for Seismic PRA and Margin Assessments Including Recent
Innovations, Proceedings of the OECD-NEA Workshop on Seismic Risk, Tokyo, Japan, August
1999
4. Seismic Design Criteria for Structures, Systems and Components in Nuclear Facilities, American
Society of Civil Engineers, ASCE/SEI Standard 43-05, 2005
5. Kennedy, R.P., Various Types of Reported Seismic Margins and Their Use, Proceedings:
EPRI/NRC Workshop on Nuclear Power Plant Reevaluation to Quantify Seismic Margins, EPRI
NP-4101-SR, Electric Power Research Institute, August 1985
6. Prassinos, P.G., et al, Recommendations to the Nuclear Regulatory Commission on Trial

Guidelines for Seismic Margin Reviews of Nuclear Power Plants, NUREG/CR-4482, U.S. Nuclear
Regulatory Commission, March 1986
7. Kennedy, R.P., A Seismic Margin Assessment Procedure, Proceedings-Symposium on Current

Issues Related to Nuclear Power Plant Structures, December 1986
8. Seismic Analysis of Safety Related Nuclear Structures and Commentary, American Society of
Civil Engineers, ASCE Standard 4-98, 1998
8
Table 1: HCLPF Correction Factor Fc
Fc = (HCLPFc/HCLPFrs)
c
rs = 0.2 rs = 0.3 rs = 0.4
0.3 1.15 1.34 1.59
0.4 1.12 1.26 1.47
0.5 1.09 1.21 1.39
0.6 1.08 1.18 1.33
Table 2: HCLPF84/HCLPF50 Ratio
rs = 0.2 rs = 0.3 rs = 0.4

HCLPF84/HCLPF50 1.22 1.35 1.49
-9-
A.1
Appendix A
Updated Description of the Conservative Deterministic

Failure Margin (CDFM) method for Computing
Seismic Capacity
A.1 Background
The Conservative Deterministic Failure Margin (CDFM) Method was first proposed in Refs. 5
through 7 as a deterministic method for estimating seismic capacity and was aimed at achieving a seismic
capacity corresponding to about the 1% non-exceedance probability (NEP) for a specified target response
spectrum. The method does not consider any response spectral peak and valley variability about the target
response spectrum shape. Thus, the 1% NEP seismic capacity statement is conditional on the input motion
matching the target response spectrum shape anchored to the reported CDFM capacity.
Many example applications of the CDFM Method are presented in Ref. 2 which therefore
represents the most extensive tutorial on the application of the method. A theoretical derivation that
demonstrates that the CDFM Method reasonably achieves its stated goal of estimating the 1% NEP seismic
capacity is presented in Ref. 3. Recently, ASCE/SEI 43-05 Seismic Design Criteria for Structures, Systems,
and Components in Nuclear Facilities (Ref. 4) has adopted the CDFM Method for seismic design of
nuclear facilities. Therefore, Ref. 4 represents the most recent criteria for the CDFM Method.
Initially the CDFM Method was proposed to compute a 1% NEP seismic capacity for comparison
with a deterministic 84% amplified design or evaluation response spectrum. For this reason, no
consideration of the response spectral peak and valley variability rs was included in the CDFM Method.
The effect of this variability has already been included in the deterministic 84% amplified design or
evaluation response spectrum.
In more recent years, the design or evaluation response spectrum has come to be defined in terms
of a Uniform Hazard Response Spectrum (UHRS) produced by a Probabilistic Seismic Hazard Assessment
(PSHA) and the CDFM capacity is now being compared against an appropriately defined UHRS, such as
for example the mean 1x10-4 annual frequency of exceedance UHRS. Furthermore, the CDFM computed
1% NEP seismic capacity C1% is being used to define fragility curves to be convolved with mean PHSA
derived hazard curves to estimate the mean seismic risk.
The concept of deterministic 84% amplified design or evaluation response spectra is no longer in
vogue. Having the CDFM Method described in terms of a deterministic 84% amplified response spectrum
as is done in Refs. 2 and 3 has introduced considerable confusion about how to use the CDFM defined
capacity C1% for comparison with a UHRS or for development of a fragility curve. The description of the
CDFM Method needs to be updated so as to be consistent with its current usage.
The key issue is that the CDFM defined capacity C1% is conditional on a specified response
spectrum and no peak and valley variability rs about this shape is included in the defined C1% capacity. It is
irrelevant as to whether this shape is a deterministic 84% amplified response spectrum shape, or is a mean,
median, or 84% non-exceedance probability (NEP) UHRS shape.
Since the CDFM Method does not consider the peak and valley variability rs of the response
spectrum, the C1% capacity computed by the CDFM Method will differ from the C1% capacity computed by
the Fragility Analysis (FA) Method described in Ref. 1 when rs is included in the FA Method. If the FA
Method has included a rs=0.2, the CDFM computed C1% should be a factor of 1.07 to 1.15 higher than the
C1% computed by the FA Method. If the FA method has included a rs larger than 0.2, the difference
between the CDFM computed C1% and the FA computed C1% will be larger. The key difference is in the
treatment of rs.
This appendix will demonstrate that the CDFM Method reasonably estimates C1% for any specified
response spectrum shape conditional on no peak and valley variability rs about this shape being
considered.
A.2 General Criteria of CDFM Method
In order to achieve a 1% NEP probability capacity C1%, the CDFM Method attempts to
approximate the following levels of conservatism by the specified deterministic criteria:
1. For the specified seismic margin earthquake ground motion level SME, the elastic computed
response (SME Demand) of structures and components mounted thereon should be defined at
about the 84% nonexceedance probability (NEP).
2. Strengths for most components should be defined at about the 98% exceedance probability so
that even if the SME demand slightly exceeds this CDFM strength by more than a permissible
conservatively specified inelastic energy absorption capability, there will still result a very
low probability of failure. However, for the CDFM strength of very brittle failure modes
(weld failure, relay chatter, etc.) which have no inelastic energy absorption capability, the
conservatism at which the strength is defined should be increased to about the 99%
exceedance probability.
3. Inelastic distortion associated with a Demand/Strength ratio greater than unity is permissible.
The permissible level of inelastic distortion should be specified at about the 5% failure
probability level. The inelastic energy absorption capability FN should be slightly
conservatively estimated at about the 84% NEP for this permissible level of inelastic
distortion.
4. Finally,
Seismic Demand/Strength FN (A.1)
where FN is the inelastic energy absorption factor.

Because of the conservatism introduced at the various steps, the result is a high-confidence-of-a-
low-probability-of-failure (HCLPF ) when Eqn. (A.1) is satisfied. Any seismic evaluation which
introduces approximately the level of conservatism as defined in the above four steps meets the intent of
the CDFM approach and would be expected to achieve a 1% NEP capacity C1% (commonly called the
HCLPF capacity).
In this approach, most of the conservatism is introduced in the capacity evaluation and in the
degree of inelastic distortion that is permitted. Only sufficient conservatism is introduced in the computed
demand so as to achieve computed demands at about the 84% NEP level conditional on design or
evaluation response spectrum with no peak and valley variability rs included.
Table 2-5 of Ref. 2 provides a detailed summary of the original CDFM criteria. This table has
been updated herein as Table A-1 to make it consistent with current application of the CDFM method. The
revisions have been italicized so as to clearly indicate the changes.
A.3 Estimation of the Conservatism Introduced by the CDFM Method Generally Applied
A.3.1 Basic Approach
This section is a revision of Section A.2 of Ref. 4 made to revise and clarify the median demand
conservatism ratio RD.
2
A.3
The median seismic capacity C50% can be estimated from:
S50%
C 50% = FN SME (A.2)
D 50% 50%
where S50%, D50%, FN50% are median estimates of the component seismic strength, seismic demand for a
specified SME input, and inelastic energy absorption (nonlinear) factor, respectively. In turn, the CDFM
seismic capacity CCDFM is given by:
S CDFM
C CDFM = FN SME (A.3)
D CDFM CDFM
where SCDFM DCDFM, and FN CDFM are the deterministic strength, demand, and nonlinear factors defined in
accordance with the CDFM method. Defining RS, RD, and RN as the median conservatism ratios associated
with the CDFM method, then:
S50% = RS SCDFM
D50% = DCDFM/RD (A.4)
FN50% = RN FN CDFM
and
C50% = RC CCDFM
(A.5)
RC = RS RD RN
where RC is the overall median conservatism ratio associated with the CDFM method. The ratios RS, RD,
and RN will be estimated in the following three subsections.
A.3.2 Median Strength Conservatism Ratio
The CDFM strength is normally computed using code specified allowable ultimate (maximum)
strengths.
Based upon a review of median capacities from past seismic probabilistic risk assessment studies
versus US code specified ultimate strengths for a number of failure modes, it is judged that for ductile
failure modes when the conservatism of material strengths, code strength equations, and seismic strain-rate
effects are considered, the code ultimate strengths have at least a 98% probability of exceedance. For a
low ductility failure modes, an additional factor of conservatism of about 1.33 is typically introduced.
Thus:
(Ductile) R S = e 2.054S
(A.6)
2.054S
(Low Ductility) R S = 1.33e
where S is the strength logarithmic standard deviation (typically in the 0.2 to 0.4 range), and 2.054 is the
standardized normal variable for 2% NEP.
A.3.3 Median Demand Conservatism Ratio
Seismic demands for CDFM evaluations are typically computed in accordance with the
requirements of the current version of ASCE 4 which is ASCE 4-98 (Ref. 8). In the Preface, ASCE 4-98
states that the standard specifies:
methods for analysis with essentially no conservative bias except for small levels of
conservatism added only to account for modeling uncertainties such as selection of
material properties, mass, geometry, and damping. For example, use of this standard will
produce seismic responses that have about a 90% chance of not being exceeded for an
input response spectrum specified at the 84th percentile non-exceedance level.
This statement is out of date because input response spectra are no longer being defined at the 84% NEP
level. The current draft version of ASCE 4-09 states in the Forward:
The intent of this standard is to provide criteria for seismic analysis such that the
resulting seismic response of SSCs is at about the 80% non-exceedance level for a given
seismic input motion. That is, there is less than a 20% probability that the responses
computed in accordance with this standard will be exceeded for the given input motion.
This statement from the draft of ASCE 4-09 reflects the current expectation about the degree of
conservatism introduced in demand analyses performed in accordance with ASCE 4. Thus, the median
demand ratio RD can be estimated from:
0.842
RD = e (A.7)
where D is the seismic demand logarithmic standard deviation for a specified seismic input (typically in
the 0.2 to 0.4 range).
A.3.4 Median Nonlinear Conservatism Ratio
In the CDFM method, the nonlinear factor is expected to be specified at about the 5% NEP level.
Thus for ductile failure modes, the median nonlinear factor ratio RN should be:
1.645 N
Ductile R N = e (A.8a)
where N is the logarithmic standard deviation for the nonlinear factor (typically in the 0.2 to 0.4 range for
ductile failure modes) and 1.645 is the standardized normal variable for 5% NEP.
However, for low ductility (brittle) failure modes, no credit is taken for a nonlinear factor, i.e.:
FN50% FNCDFM 1.0

(A.8b)
R N 1.0
A.3.5 Resulting CDFM Capacity Conservatism
Combining Eqns. (A.5) through (A.8), the median CDFM capacity ratio RC is estimated to be:
(Ductile Failures) R C = e 2.054S + 0.842D +1.645 N

(A.9)
2.054S + 0.842D
(Low Ductility) R C = 1.33e
and from Eqn. (A.5):
4
A.5
C1% = RC CCDFM e-2.326 (A.10)
[
= S2 + 2D + 2N ]
1
2
(A.11)
Table A.2 presents the ratio of (C1%/CCDFM) for typical values of S, D, and N as computed from
Eqns. (A.9) through (A.11). It can be seen that over this entire range of values:
C1% CCDFM (A.12)
with the ratio (C1%/CCDFM) ranging from 0.94 to 1.29 with a median value of 1.14.
The CDFM capacity can also be used to estimate the 10% probability of unacceptable
performance capacity C10%. From Eqn. (A.5):
C10% = RC CCDFMe-1.282 (A.13)
Table A.3 presents the ratios of (C10%/CCDFM) for typical values of S, D, and N. It can be seen that:
C10% 1.5 CCDFM (A.14)
However, the approximation of Eqn. (A.14) is not as good as that of Eqn. (A.12).
A.4 Conclusion
The CDFM Method can provide a realistic estimate of the 1% NEP seismic capacity C1% for a
specified response spectrum shape without inclusion of the peak and valley variability. Furthermore, the
10% NEP seismic capacity is at least 1.5 times the CDFM capacity. This C10% capacity controls the fragility
curve for composite c values less than 0.39. The resulting fragility can be defined by the following seismic
margin factors F1%, F5%, F10%, and F50% times the CDFM capacity:
F1% F5% F10% F50%

0.30 1.10 1.35 1.50 2.20
0.40 1.00 1.31 1.52 2.54
0.50 1.00 1.41 1.69 3.20
0.60 1.00 1.50 1.87 4.04
Table A.1
SUMMARY OF CONSERVATIVE DETERMINISTIC FAILURE MARGIN APPROACH
Load Combination: Normal + SME
Ground Response Spectrum: Anchor CDFM Capacity to defined response spectrum shape without
consideration of spectral shape variability
Seismic Demand Perform seismic demand analysis in accordance with latest version of ASCE
4 (Ref. 8)
Damping Conservative estimate of median damping
Structural Model: Best Estimate (Median) + Uncertainty Variation in Frequency
Soil-Structure-Interaction: Best Estimate (Median) + Parameter Variation
Material Strength: Code specified minimum strength or 95% exceedance actual strength if test
data are available.
Static Strength Equations: Code ultimate strength (ACI), maximum strength (AISC), Service Level D
(ASME), or functional limits. If test data are available to demonstrate
excessive conservatism of code equation then use 84% exceedance of test
data for strength equation.
Inelastic Energy Absorption: For non-brittle failure modes and linear analysis, use appropriate inelastic
energy absorption factor from ASCE 43-05 (Ref. 4) to account for ductility
benefits, or perform nonlinear analysis and go to 95% exceedance ductility
levels.
In-Structure (Floor) Spectra Generation: Use frequency shifting rather than peak broadening to account for uncertainty
plus use conservative estimate of median damping.
6
A.7
Table A.2 Ratio of (C1%/CCDFM)
Strength Demand Low Ductility Ductile Failure Modes

Variability Variability Failure Modes
S D N=0.2 N=0.4
0.2 0.2 1.23 1.11 1.10
0.3 1.12 1.03 1.07
0.4 1.00 0.94 1.01
0.3 0.2 1.26 1.17 1.21
0.3 1.18 1.11 1.19
0.4 1.08 1.03 1.13
0.4 0.2 1.27 1.20 1.29
0.3 1.22 1.16 1.27
0.4 1.14 1.10 1.23
Table A.3 Ratio of (C10%/CCDFM)
Strength Demand Low Ductility Ductile Failure Modes

Variability Variability Failure Modes
S D N=0.2 N=0.4
0.2 0.2 1.66 1.59 1.84
0.3 1.63 1.59 1.88
0.4 1.59 1.57 1.89
0.3 0.2 1.84 1.80 2.12
0.3 1.85 1.82 2.18
0.4 1.82 1.81 2.20
0.4 0.2 2.02 2.00 2.41
0.3 2.06 2.04 2.49
0.4 2.06 2.05 2.53
B.1
Appendix B
Other Related HCLPF and Fragility Issues
B.1 Correction of Randomness R and Uncertainty U
When the total randomness variability Rrs includes the random spectral peak and valley variability
rrs, the corrected total random variability R is given by:
[
R = 2Rrs 2rrs ]
0.5
(B.1.1)
similarly to Eqn. (2.1) for c. The corresponding correction for the uncertainty U is given by:
[
U = 2Urs 2urs ] 0.5
(B.1.2)
where urs is the spectral shape uncertainty included in the total computed uncertainty Urs.
Typically, in most fragility evaluations all of the spectral shape variability rs is attributed to
randomness rrs. Thus typically:
rrs = rs (B.1.3)
urs = 0
so that all of the correction can be made to Rrs in accordance with Eqn. (B.1.1) and U does not need to be
corrected. However, Eqns. (B.1.1) and (B.1.2) cover the full range of possibilities.
B.2 Conversion of Ref. 2 Screening Tables for Components Defined in Terms of Peak 5% Damped Peak
Spectral Acceleration at Base of Component
The component screening levels presented in Table 2-4 of Ref. 2 were developed by Robert
Campbell, John Reed, and myself based on the consideration of a diverse body of information summarized
in Appendix A of Ref. 2. These component screening levels were defined in terms of the peak 5% damped
spectral acceleration (SAG) of the ground instead of the peak 5% damped spectral acceleration (SAS) at the
base of the component. Most of the data base summarized in Appendix A of Ref. 2 was reported in terms of
ground motion instead of in-structure response spectra (ISRS). Furthermore, it was judged that realistic
median ISRS might not exist in many cases where these component screening levels might be used.
However, footnote (y) in Table 2-4 cautions against the use of the SAG screening level for situations where
the SAS level from realistic ISRS exceeds 1.67SAG (i.e. SAS exceeds 2.0g for SAG of 1.2g).
In my judgement, the following represents a reasonable approach for estimating HCLPF and
generic fragility levels for individual components in terms of SAS at the base of the component from the
screening level SASL given in Table 2-4 of Ref. 2. For individual SSCs, the generic fragility estimate in
terms of SAG given by Eqn. (6.3) herein in Section 6 is:
SAG50 = 2.85 SASL (B.2.1)

SAG = 0.45
HCLPFSAG = SASL
A reasonable range for the amplification factor AF=(SAS/SAG) for the data from which SASC was
developed is 1.0 to 1.67. Reasonably estimating that AF=1.0 corresponds to minus 2, and AF=1.67
corresponds to plus 1, then:
AF = [n (1.67 )]/ 3 = 0.17 (B.2.2)
AF50 = exp(2AF) = 1.4 (B.2.3)
The median SAS50, variability SAS, and HCLPFSAS in terms of the ISRS SAS for an individual
component can be estimated by:
SA50 = AF50 SAG50 = 4.0 SASL (B.2.4)
2
[
SAS = SAG 2AF ]
0.5
= 0.42
HCLPFSAS = SAS50 exp(-2.326SAS) = 1.5 SASL
since AF is included as part of the SAG estimate. Thus, for SASC=1.2g, the corresponding HCLPFSAS is
reasonably estimated to be 1.8g.
2
Export Control Restrictions The Electric Power Research Institute Inc., (EPRI, www.epri.com)
Access to and use of EPRI Intellectual Property is granted with the spe- conducts research and development relating to the generation, delivery
cific understanding and requirement that responsibility for ensuring full and use of electricity for the benefit of the public. An independent,
compliance with all applicable U.S. and foreign export laws and regu- nonprofit organization, EPRI brings together its scientists and engineers
lations is being undertaken by you and your company. This includes as well as experts from academia and industry to help address challenges
an obligation to ensure that any individual receiving access hereunder in electricity, including reliability, efficiency, health, safety and the
who is not a U.S. citizen or permanent U.S. resident is permitted access environment. EPRI also provides technology, policy and economic
under applicable U.S. and foreign export laws and regulations. In the analyses to drive long-range research and development planning, and
event you are uncertain whether you or your company may lawfully supports research in emerging technologies. EPRIs members represent
obtain access to this EPRI Intellectual Property, you acknowledge that it more than 90 percent of the electricity generated and delivered in the
is your obligation to consult with your companys legal counsel to deter- United States, and international participation extends to 40 countries.
mine whether this access is lawful. Although EPRI may make available EPRIs principal offices and laboratories are located in Palo Alto, Calif.;
on a case-by-case basis an informal assessment of the applicable U.S. Charlotte, N.C.; Knoxville, Tenn.; and Lenox, Mass.
export classification for specific EPRI Intellectual Property, you and your
Together...Shaping the Future of Electricity
company acknowledge that this assessment is solely for informational
purposes and not for reliance purposes. You and your company ac-
knowledge that it is still the obligation of you and your company to make
your own assessment of the applicable U.S. export classification and
ensure compliance accordingly. You and your company understand and
acknowledge your obligations to make a prompt report to EPRI and the
appropriate authorities regarding any access to or use of EPRI Intellec-
tual Property hereunder that may be in violation of applicable U.S. or
foreign export laws or regulations.
Programs:
Nuclear Power
Safety Risk Technology
2009 Electric Power Research Institute (EPRI), Inc. All rights reserved. Electric Power
Research Institute, EPRI, and TOGETHER...SHAPING THE FUTURE OF ELECTRICITY are
registered service marks of the Electric Power Research Institute, Inc.
1019200
Electric Power Research Institute

3420 Hillview Avenue, Palo Alto, California 94304-1338 PO Box 10412, Palo Alto, California 94303-0813 USA
800.313.3774 650.855.2121 askepri@epri.com www.epri.com

PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PDF

Uploaded by

Copyright:

Available Formats

Seismic Fragility Applications Guide Update

Seismic Fragility Applications Guide

Final Report, December 2009

EPRI Project Manager

ELECTRIC POWER RESEARCH INSTITUTE

(A) MAKES ANY WARRANTY OR REPRESENTATION WHATSOEVER, EXPRESS OR IMPLIED, (I)

(B) ASSUMES RESPONSIBILITY FOR ANY DAMAGES OR OTHER LIABILITY WHATSOEVER

ORGANIZATION(S) THAT PREPARED THIS DOCUMENT

RPK Structural Mechanics Consulting

Simpson Gumpertz & Heger Inc.

This report was prepared by

RPK Structural Mechanics Consulting

Simpson Gumpertz & Heger Inc.

A,a Ground Motion Parameter Corresponding to Any Given Frequency of Failure

Am Median Peak Ground Motion Capacity

C Composite Variability = (R2 + U2)0.5

C-C Capacity Factor Variability

C-RS Response Factor Variability

C-S Strength Composite Variability (typical)

R Log Standard Deviation of Randomness

R_S Strength Randomness (typical)

U Log Standard Deviation of Uncertainty (Lack of Knowledge)

U_S Strength Uncertainty (typical)

rs Response Spectra Peak and Valley Variability

R Random Variable with Unit Median and Logarithmic Standard Deviation, R

U Uncertainty Variable with Unit Median and Logarithmic Standard Deviation, U

F Damping Factor of Safety

F Inelastic Energy Absorption Factor of Safety

FECC Earthquake Component Combination Factor of Safety

FGMI Ground Motion Incoherence Factor of Safety

FM Modeling Factor of Safety

FMC Modal Combination Factor of Safety

FQM Qualification Factor of Safety

FRS Structural Response Factor of Safety

FS Strength Factor of Safety

FSA Spectral Shape Factor of Safety

FSSI Soil-Structure Interaction Factor of Safety

FTOTAL Total Factor of Safety

HCLPF84 84% Non-Exceedance Probability High Confidence of Low Probability of Failure

PN Normal Operating Loads

PT Total Load (Sum of Seismic Load and Normal Operating Load)

S Strength of Structural Element

1 INTRODUCTION .................................................................................................................... 1-1

2 SEISMIC FRAGILITY METHODOLOGY ................................................................................ 2-1

3 UPDATES TO EPRI TR-103959 ............................................................................................ 3-1

4 UPDATES TO EPRI TR-1002988 .......................................................................................... 4-1

5 REFERENCES ....................................................................................................................... 5-1

Figure 2-1 Mean, Median, 5% Non-Exceedance, and 95% Non-Exceedance Fragility

1.1 Objective of Seismic Fragility Applications Guide Update

The key elements of a SPRA can be identified as:

1.2 Scope of the Applications Guide Update

Section 5 contains the references.

2.1 Screening of High-Capacity Components

2.2 Seismic Fragility Analysis Methodology

Fragility analysis involves the estimation of the conditional probabilities of structural or

2.2.1 Fragility Model

-1(.) = the inverse of the standard Gaussian cumulative distribution function.

2.2.2 Failure Modes

Failure of the motor.

2.2.3 Estimation of Fragility Parameters

In estimating fragility parameters, it is convenient to work in terms of an intermediate random

A = FA SSE where A is the actual ground motion acceleration capacity

Actual seismic capacity of element