Advanced

Seismic Assessment
Guidelines

Prepared for: Pacific Gas & Electric (PG&E)/PEER Lifelines Program

Task 507

By: Paolo Bazzurro1, C. Allin Cornell1, Charles Menun1,

Nicolas Luco2, Maziar Motahari1

September 30, 2004

1
J. A. Blume Earthquake Engineering Center Department of Civil and Environmental Engineering,
Stanford University, Stanford CA
2
AIR Worldwide Co., 388 Market Street, Suite 610, San Francisco, CA 94111

i
TABLE OF CONTENTS

TABLE OF CONTENTS.................................................................................................................. ii

ACKNOWLEDGEMENTS .............................................................................................................. iii

FORMAT OF THE DOCUMENT ..................................................................................................... 1

INTRODUCTION ............................................................................................................................. 1

PG&E BUILDING INVENTORY ...................................................................................................... 2

PROCEDURE.................................................................................................................................. 4

DISCUSSION OF THE SIX STEPS ............................................................................................... 7

STEP 1: Nonlinear Static Procedure (NSP) curve for the intact building........................... 7

STEP 2: NSP curves for the damaged building ..................................................................... 9

STEP 3: Inferring dynamic response from static response................................................ 12

STEP 4: Occupancy status for damaged building............................................................... 21

STEP 5: Ground motion level associated with a structural limit state .............................. 31

STEP 6: Computation of the fragility curves ........................................................................ 38

APPLICATION OF THE GUIDELINES......................................................................................... 42

Case Study No.1: 3-story Steel-Moment Resisting Frame (SMRF) Building.................... 42

Case Study No.2: Tilt-up Building ........................................................................................ 56

LIST OF REFERENCES ................................................................................................................85

APPENDIX A: Extension to include building-to-building uncertainty within same building

class.............................................................................................................................................. 90

APPENDIX B: Evaluation of epistemic uncertainty for PG&E Buildings .............................. 92

APPENDIX C: Copy of Most Relevant References ...................................................................98

ii
Acknowledgements

The development of these guidelines would not have been possible without the help of
many individuals to whom we are deeply indebted. In particular, Mr. Kent Ferre, Dr.
Norman Abrahamson, and Dr. Lloyd Cluff, of Pacific gas and Electric (PG&E), and Ms.
Maryann Phipps of Estructure have provided the necessary guidance throughout the
project. Mr. William Holmes, Dr. Joseph Maffei of Rutherford & Chekene, Ms. Phipps,
and Dr. Robert Kennedy of SMC have shared their expertise in providing the data needed
for the assessment of the epistemic uncertainty. Dr. Maffei was also instrumental in the
development of Appendix B. A 2002 draft of these Guidelines was used by Dr. Maffei of
Rutherford & Chekene and by the office of Degenkolb Engineers under Mr. James
Malley to conduct trial applications on three PG&E buildings (Tasks 508 and 509,
respectively). Their experience and critiques have been invaluable in modifying and
editing this version. These applications have been reported in detail to PG&E and will be
published as PEER reports.

This project was sponsored by the Pacific Earthquake Engineering Research Center’s
Program of Applied Earthquake Engineering Research of Lifelines Systems supported by
the California Energy Commission, California Department of Transportation, and the
Pacific Gas and Electric Company. This work made use of the Earthquake Engineering
Research Centers Shared Facilities supported by the National Science Foundation, under
award number EEC-9701568 through the Pacific Earthquake Engineering Research
Center (PEER). Any opinions, findings, and conclusions or recommendations expressed
in this material are those of the author(s) and do not necessarily reflect those of the
National Science Foundation.”

iii
Format of the Document

This document comprises text in two different forms:

• Guidelines provide the engineer with the basic information needed to

apply the methodology presented here. Guidelines are written in
normal font.
• Commentary presents relevant notes useful to the understanding of the
guidelines but not strictly necessary to its implementation. The
paragraphs with commentary, which are italicized, indented and
proceeded by a capital C, will appear interspersed with the guidelines
in this draft.

Introduction

The objective of this document is to provide guidelines for the assessment of the seismic
performance of existing structures that are common in the building inventory of major
electric utility such as Pacific Gas and Electric (PG&E), a co-sponsor of the PEER
Lifelines Program. Predicting the post-earthquake functionality of such structures is a
crucial step in evaluating the likelihood that the PG&E power distribution network will
not be able to provide power to customers. As an important by-product, the results of the
procedure outlined in this report also provide a rational support to the engineer inspecting
the facility after an earthquake on whether to permit the occupancy of a building soon
after the occurrence of a damaging earthquake. Currently, the occupancy status of a
damaged building is assessed by an engineer from a firm retained for this purpose by
PG&E based solely on visual inspection.

The final product is cast in terms of the so-called fragility curves for structural limit states
directly related to after-earthquake building occupancy status, namely green, yellow, and
red tags. A limit-state fragility curve provides the conditional probability that the
specified limit state will be reached or exceeded as a function of the severity of the future
ground motion at the site. Besides these three states, this document provides the fragility
curves for two additional structural limit states, the onset-of-damage and the collapse
states, which are not directly related to the building functionality status. The onset-of-
damage state relates to the beginning of detectable structural damage. The five limit
states in increasing order of damage severity are:

• Green tag, G: the building has been inspected and deemed fit for immediate
occupancy.
• Onset of Damage, OD: FEMA 356 and HAZUS manual define the onset of
significant nonlinear behavior (Immediate Occupancy in FEMA 356 and
Slight Damage in HAZUS) for different types of structures, such as those in
the PG&E building stock. No limitations on post-earthquake access are
implied by this limit state.

1
 • Yellow tag, Y: the building has been inspected and deemed fit for restricted
occupancy. The access is limited to specialized personnel only, until detailed
engineering evaluation is completed.
• Red tag, R: the building has been inspected and deemed unsafe. Access is
not allowed until completion of detailed engineering evaluation, retrofit or
rebuilding.
• Collapse, C: the building has collapsed or is on the verge of global instability
or local collapse.

Hence, the Onset-of-Damage limit state lies within the green-tag state boundaries
while the collapse state is, of course, the most severe stage of the red-tag
condition. For post-earthquake structure operability only yellow- and red-tag
states are relevant.

Coupling the fragility curves for PG&E buildings and similar results for other
elements of the PG&E network with probabilistic seismic hazard analysis will
permit assessments of the risk of seismically induced outages in the regional
distribution system. In the aftermath of a severe earthquake the use of limit state
fragility curves along with the observed spectral acceleration values available
from instrumentation networks, such as TriNet, will also provide PG&E with a
rapid damage estimation tool prior to building visual inspection.

The procedure developed in this project is, of course, of more general

applicability. It is also consistent with the broader approach to Performance-
Based Earthquake Engineering put forward by the Pacific Earthquake
Engineering Center (e.g., Moehle and Deierlein, 2004).

PG&E building inventory

The total PG&E building stock comprises more than a thousand buildings, the majority of
which are not deemed critical to the operation of the power distribution network. A
subset of about 200 structures, which includes operation buildings and substations, is
currently insured against earthquake damage. These guidelines are intended to be
applicable to this subset of important buildings, which here will be referred to as the
PG&E building inventory. The inventory comprises many different types of structures
often of older and mixed construction, sometimes built and subsequently expanded and/or
retrofitted in different phases. For the purpose of this document, these buildings can be
categorized into the building classes shown in Table1 (PG&E, 2001c).

These guidelines are structure-independent and, therefore, can be applied, with few
exceptions1, to all buildings in the PG&E inventory. It is important to note, however, that

1
Because this methodology estimates the dynamic response of buildings from static nonlinear pushover
analyses, its application to tall, flexible buildings is not recommended. For this type of structures the
accuracy of nonlinear static pushover analysis in assessing the building seismic response is, in general,
questionable.

2
the guidelines in this present form are not structure-generic, namely they are not readily
applicable to a generic structure belonging to a certain building class. The guidelines are
not structure-generic because, as will become clear in the next section, building-specific
results of nonlinear static procedure (NSP) analysis are necessary for their application.

Description Number
Concrete Tilt-Up Wall Buildings with Wooden Roof Diaphragm 18
Steel Frame Buildings with Concrete Infill Walls (Mill-Type Buildings 35
Prefabricated Metal Buildings 44
Concrete Shear-Wall Buildings 11
Steel Moment-Resisting Frames (SMRFs) 3
Braced Steel Frame Buildings 29
Steel Frames with Infill Unreinforced Masonry Walls 3
Concrete Block Buildings with Wood/Metal Deck Roof 19
Wood-Frame Buildings 34
Concrete Frames with Infill Unreinforced Masonry Walls 1
Other or Unknown 8
205

Table 1: Building types in the PG&E inventory. The number of mill-type buildings is approximate.

The successful application of these guidelines to any specific building depends on the
correct modeling of structural details that appropriately capture the failure modes
observed during damage reconnaissance investigations. Such structural details, in
general, are different for different building types. For demonstration purposes only, the
procedure presented here will be applied (see the example section) to two specific
structures of the PG&E building inventory:
• a steel moment-resisting frame (SMRF) building,
• a tilt-up building.
Additional example applications of these guidelines by practicing engineers can be found
in Degenkolb (2004) and Rutherford & Chekene (2004).

Appendix A includes an overview of the additional steps that may be followed to make
these guidelines applicable to a generic structure of a specific building class. Appendix
A provides the theoretical framework only. No attempt is made to suggest numerical
values for the sources of additional uncertainty defined there. Quantifying such
uncertainties is a major task that is beyond the scope of this document. Appendix B will
provide an overview of the expert elicitation process carried out for acquiring some of the
dispersion parameters necessary for the computation of the fragility curves. Finally,
Appendix C contains a copy of some references that are directly related to these
guidelines. In particular, the reader can find a paper on the methodology devised for these
guidelines (Bazzurro, et al., 2004), a paper on the analyses that supported the calibration
of the static versus dynamic evaluation of residual displacements in damaged buildings
after the mainshock (Luco et al., 2004), and a document on aftershock probabilistic
seismic hazard analysis (APSHA) and its use in the context discussed here (Yeo and
Cornell, 2004).

3
Procedure

In brief, the methodology presented here uses the knowledge of the nonlinear static
behavior of a specific building subject to incremental lateral loads to infer its nonlinear
dynamic response expected for different levels of ground motion severity. Quantitative
measures of the implied degradation in building safety will be used to associate each of
several post-earthquake structural damage states with an appropriate post-earthquake
structural limit state that may imply some degree of occupancy restrictions (e.g., yellow
or red tagging). These restrictions dictate the facility effectiveness in the electrical
distribution network in the immediate aftermath of an earthquake and provide valuable
information to estimate expected downtime. A unique aspect of this procedure is that the
building safety while in a damaged state is measured in terms of its ability to protect
occupants from death or injury due to aftershocks that might occur prior to its repair.
2004 experience in Japan with the Honshu earthquakes shows once again that aftershocks
may cause ground motions as large or larger than the mainshock values. The mainshock
damage may cause either

• decreased global lateral collapse capacity,

• decreased seismic capacity with respect to loss of vertical resistance, locally
or more widely, to gravity and live loads.

Finally, the uncertainty inherent in building response and capacity for different ground
motion levels due to both record-to-record variability and to uncertainty in structural
modeling and evaluation process is used to obtain the desired fragility curves for the five
structural limit states identified in the introduction.

It is not the intent of this document to provide the validation of all of the steps. For the
interested reader, an overview along with a list of references is included in the
commentary.

The procedure has six steps:

• STEP 1: Nonlinear Static Procedure (NSP) for the intact building. Conduct a
conventional nonlinear static procedure (NSP) analysis (also known as static
pushover, SPO) of the selected structure (in each orthogonal direction) in the as-is
condition using a displacement-controlled approach that includes P-delta effects
and strength degradation of structural elements2. Increase the deformation until
either the structure base shear has significantly dropped (e.g., to 20 or 30% or
less) compared to its peak value or until at least one of the structural elements or
connections has reached its ultimate capacity to withstand vertical loads even in
absence of any future ground shaking. The NSP analysis should also be stopped
when the value of global ductility obtained in the analysis becomes unattainable

2
This procedure requires that the applied lateral load be allowed to drop. This can be achieved by including
strength degradation at the element level and by limiting at each step of the analysis the incremental
displacement at one, or more, key node(s) of the structure.

4
 in reality. Identify important damage states along the displacement axis of the
NSP curve(s).

• STEP 2: NSP for the damaged building. For each of the damage states that do
not include loss of local vertical capacity (STEP 1), conduct analogous NSP
analyses assuming the structure is in that damage condition after the mainshock.
These results will be used (STEP 3) to estimate the residual (dynamic) capacity to
resist aftershocks and hence the relative life safety threat of the building when in
one of these potential states. Omit STEP 3 and go directly to STEP 4 for all the
damage states that show at least one element that cannot carry vertical loads.

• STEP 3: Inferring dynamic response from static response. Convert each of the
intact structure’s two NSP curves (STEP 1) into estimates of the peak dynamic
displacement (and, hence, post-earthquake damage state) expected under a range
of potential (mainshock) ground motions, using the SPO2IDA spreadsheet tool
provided (Vamvatsikos and Cornell, 2001b). This step computes the incremental
dynamic analysis (IDA) curve for the intact structure in each direction. The
ordinate of the highest point on the IDA provides the (dynamic) capacity of the
intact building, which is expressed here in terms of (median) ground motion level
necessary to induce either global lateral instability of the structure or local
collapse anywhere in the structure. Then, for the damaged structure in all
damage states (with no local loss of vertical capacity), apply the same tool to each
of the corresponding NSP curves (STEP 2) to estimate the residual capacity in
terms of the (aftershock) ground motion intensity necessary to cause collapse
(either global or local). This residual capacity will depend on the residual
displacement offset expected (or measured) after the mainshock. If necessary,
compute the residual capacity of the damaged structure in both directions. Note
that for the damaged structure the step above provides also the entire IDA, which,
however, will not be used in this procedure.

• STEP 4: Occupancy Status for Damaged Building. Categorize each potential

damage state with no loss of local vertical capacity as one of the following five
structural limit states: green-tag, onset-of-damage, yellow-tag, red-tag, or
collapse. The tagging is based on the corresponding residual capacities identified
in STEP 3 and on the (site-specific) mean annual frequency of exceedance of the
(aftershock3) ground motion values corresponding to such residual capacities.
Damage states, if any, for which vertical capacity is lost at least at one structural
component are associated with the collapse limit state4.

3
Two schemes of tagging are presented below one based explicitly on the aftershock hazard and one only
implicitly using the original, customary mainshock hazard as a proxy for the aftershock hazard, which is
currently less familiar.
4
Local loss of vertical capacity is potentially life threatening and for this reason this condition is associated
here with structural collapse. Note that local loss of vertical capacity does not necessarily implies that the
structure has no global lateral capacity left to resist aftershocks.

5
• STEP 5: Ground motion level associated with a structural limit state. Using the
results of STEPS 3 and 4, identify the four (mainshock) ground motion levels
expected to cause the structure to enter the onset-of-damage state, the yellow-tag
state, the red-tag state, and the collapse state according to a pre-defined criterion
provided in this step. Such ground motion levels are identified by their median
values and related dispersion measure.

• STEP 6: Computation of the fragility curves. Based on the median values and
corresponding dispersion measure of the four ground motion levels number
provided at STEP 5, plot the fragility curves associated with the onset-of-damage,
the yellow-tag, the red-tag, and the collapse states. The fragility curve
corresponding to the green-tag state is equal to unity for all levels of ground
motion.

6
Discussion of the Six Steps

STEP 1: Nonlinear Static Procedure (NSP) curve for the intact building

The knowledge of the post-elastic behavior of the building subject to incremental lateral
seismic loading is the starting point of this procedure. In line with other current
Guidelines (e.g., FEMA 273, 1997; FEMA 356, 2000), it is assumed here that the
engineer is accustomed to performing nonlinear static procedure (or static pushover)
analyses at least for the two main orthogonal directions of the structure. The nonlinear
static procedure normally involves the monotonic application of lateral forces (or
displacements) to a model of the building until a target deformation is reached. The
lateral load is selected to follow a predetermined or adaptive pattern (Kunnath and
Balram, 1999; Balram and Kunnath, 2000) that approximately represents the inertial
forces at the locations of the significant masses. In this application the pushover analysis
should include element strength degradation and should be conducted to failure, namely
until either severe lateral force degradation or local loss of vertical capacity of at least
one structural component (e.g., beam, column, connection, etc.) are reached. The reader
is referred, for example, to Section 3.3 of FEMA-273 for details on how to perform a
NSP analysis.

It is important to emphasize that the seismic evaluations performed within the framework
of these guidelines should use expected strength values rather than nominal or lower-
bound strengths. The intent here is to estimate the realistic response of a structure and
not to apply conservative procedure to achieve a “safe” building design.

The NSP curves expressed in terms of base shear versus roof drift (i.e., roof displacement
divided by the height of the building) for the two main orthogonal directions of the
structure are the main products of the NSP analysis. The engineer identifies on the NSP
curves the major inelastic events that occur in the structure (e.g., failure of a significant
portion of the roof/wall connection of a tilt-up building, fracture of at least 10% of all top
flanges in a steel moment-resisting frame, or local collapse of beam-column connection
due to shear failure) along with associated roof drift levels. These major inelastic events
sometimes cause significant drops in the base shear or changes in the global stiffness of
the building which in turn translate into changes in slope of the NSP curve, as depicted in
Figure 1. In other cases their occurrence may only moderately change the global stiffness
of the structure and, therefore, may not produce sharp kinks in the pushover curve.

7
 BS NSP for E-W direction
DS2
NSP for N-S direction
DS1
DS3 DS4 DS3 DS4=collapse
DS2

DS6=collapse
DS1
DS5

Roof drift
Figure 1: Hypothetical Static Pushover curves for the two orthogonal directions of a building. Legend: BS
stands for base shear.

The engineer needs also to assess whether the structural members that are damaged at any
stage of the NSP curve are still able to withstand gravity and live loads. As mentioned
above, if local loss of vertical capacity at a certain damage state is likely, then there is no
need to continue the NSP analysis is stopped. The decision can be based, for example, on
whether the ultimate deformation of a ductile element (or connection) or the maximum
strength of a brittle member (or connection) have been attained. For example, a steel
beam-column connection that reaches a curvature sufficient not only to fracture the
flanges but also to pull off the shear tab connection may not physically collapse (due, for
example, to membrane effects in the slab) but it cannot be considered suitable for
withstanding vertical loads of any significance. Such ultimate strain and strength values
are sometimes available from laboratory tests published in the literature. If tests are not
available either a detailed finite element analysis of the member (or connection) or
simply engineering judgement need to be used. Although it is outside the scope of these
guidelines to provide such values, the reader can find useful information in FEMA 273
(1997) for several structural components for steel, reinforced concrete, and masonry
constructions; in FEMA 288 (1997), FEMA 289 (1997), and FEMA 355D (2000) for
steel beam-column connections; in Pardoen et al. (1999) for tilt-up wall-roof-
connections; in ATC-11 (1983) and Paulay and Priestley, (1992) for concrete beam-
column connections; and in Duffley et al. (1994) for shear-wall concrete structures.

These inelastic events are to be associated with specific damage states. More formally,
the occurrence of the ith major inelastic event (or a set of events at approximately the
same deformation level) identifies the ith damage state, DSi. DSi is therefore defined by
a) a roof drift value, ∆i, and b) a detailed description of the structural damage associated
with that event, including whether any element has reached ultimate vertical capacity.
The damage description can be valuable to engineers to compare with observable damage
during the inspection of a facility before deciding on the possible building occupancy
restriction after an earthquake. Each one of these damage states will be later associated

8
with one (and only one) structural limit state. The limit states are defined here in terms of
a color tagging condition (Green, Yellow, or Red) or of Onset of Damage or of Collapse
as discussed in STEP 4.

Summary of STEP 1: apply NSP techniques to both directions of the intact building and
identify major damage states DSi that the structure could be in after being hit by an
earthquake mainshock.

Output of STEP 1: two NSP curves of the two main perpendicular directions of the
intact building, a series of structural damage states DSi, a detailed description of the
physical damage for all states DSi, and ∆i values associated with such states.

C: The conventional NSP inherently assumes that the response of the structure
until collapse is dominated by its first mode. A word of caution is given here on
the accuracy of such a hypothesis for some of the building types in the PG&E
inventory. For example, the most frequent failure mode of tilt-up buildings after
the 1994 Northridge earthquake involved the failure of the connections between
the roof diaphragm and the concrete walls. For example, for mill-type buildings,
the high concrete infill walls are expected to be vulnerable when moving in the
out-of-plane direction (e.g., Paulay and Priestley, 1992). This failure mode
cannot be correctly captured using conventional NSP analysis unless particular
care is taken during modeling and load pattern application.

C: If the building is not symmetrical about a plane orthogonal to the applied

lateral loads, in principle the lateral loads should also be applied in both positive
and negative directions for the purpose of estimating the residual capacity.

C: 3D Static Pushover analysis is feasible but not common. For the purpose of
these guidelines the engineer should repeat the analyses for both directions unless
it is evident that one direction is weaker than the other. In such case the analyses
should be performed only on the weaker of the two directions. In the illustrative
example the analyses are carried out for the E-W direction only.

C: If a structure is located close to a fault it may be affected by forward

directivity effects, which cause the ground motion component to be statistically
higher in the direction perpendicular to the fault strike. Hence, for a generic
structure it may not be appropriate to consider only the weaker building direction
because of the statistical difference between the fault parallel and fault normal
ground motions. The engineer should exercise his/her judgment to establish if the
NSP analysis needs to be performed for one of both orthogonal directions.

STEP 2: NSP curves for the damaged building

The NSP analyses performed at STEP 1 identify a suite of potential damage states, DSi,
that the structure may be in after an earthquake. While in that damaged condition the
building is vulnerable and could experience more extensive and severe damage in future

9
events, in particular in aftershocks prior to repair. This life-safety threat drives the
tagging of damaged buildings.

To evaluate the post-earthquake residual capacity of the building to withstand future

lateral loads, this step of the procedure requires obtaining the NSP curves for the building
in every damage state. NSP curves for each damage state are needed for characterizing
(STEP 3) the residual lateral capacity left in the structure after different levels of damage
that might have been inflicted on it by the mainshock. The residual lateral capacity is the
quantity used when assigning a damage state to a structural limit state (STEP 4) and
therefore to a tagging condition. For the purpose of these guidelines, however, the
assessment of the residual lateral capacity of the building in the DSi damage state is of
interest provided that at the local level every damaged structural member is still capable
of carrying vertical loads. If there is loss of local member vertical capacity in the DSi
damage state the building will be assigned a structural limit state that corresponds to
collapse (STEP 4).

Hence, STEP 2 consists of subjecting a model of the building in the ith damage state to the
same NSP analysis described at STEP 1. This exercise is to be repeated for all the
damage states identified at STEP 1 for which the local ability of sustaining gravity and
live loads is not compromised. This process leads to a family of NSP curves, two (one
per orthogonal direction and, possibly, one for positive and negative directions as well)
for each of such damage states.

Figure 2 shows NSP curves for the intact structure and for the damaged structure in
damage states from DS2 to DS5 in the E-W direction (see Figure 1). The NSP curve for
DS6 is missing because it is assumed here that at that damage state some members were
so severely damaged that they could not reliably carry vertical loads. Note that the
pushover curve for a structure at the onset-of-damage state (i.e., DS1) coincides with that
of the intact structure.

10
 BS NSP for intact structure
NSP for structure in DSi

DS3 DS4
DS2
DS2
DS3
DS1 DS5 DS6=collapse
DS4

DS5

Roof drift
Figure 2: Hypothetical NSP curves for structure in the intact condition and at different levels of damage
(i.e., DS2 to DS5).

Summary of STEP 2: apply NSP techniques to the damaged structure in all damage
states DSi for which vertical load carrying capacity is not compromised anywhere in the
structure. Unless one of the two main orthogonal directions of the building is
significantly weaker than the other, perform STEP 2 analyses for both directions.

Output of STEP 2: a family of NSP curves that are associated with different levels of
initial building damage state DSi for which local vertical capacity is preserved
everywhere in the structure. Identify damage states DSj, j>i, that show local loss of
vertical load carrying capacity, if any.

C: In practice the NSP curves for the structure in the damaged state can be
obtained by “re-loading” -- re-imposing an increasing deformation on -- the
model of the structure after it has been quasi-statically unloaded from the ith
damage state5. Obtaining these NSP curves by assuming parallel-to-elastic
unloading/re-loading is a simple alternative that can result in comparable
estimates of residual dynamic capacity (see STEP 3) (e.g., see the case studies).
Another viable alternative for computing reloading stiffness values of damaged
buildings presented in the PEER 508 study (Rutherford & Chekene, 2004).

C: Unloading the structure from a damage state may create an initial offset in the
damaged-structure NSP curve due to the residual permanent displacement in the
5
The use of an unloading procedure that cycles the structure to an equal displacement in the opposite
direction has also been considered. It might be considered to better represent (statically) the behavior and
damage of a structure during an earthquake. As of this date, the simpler unloading procedure has been
found to give results that correlate better with dynamic analyses.

11
 structure. The extent of this permanent displacement is somewhat an artifact of
applying a static procedure to modeling the dynamic response of the structure
subject to ground shaking. The residual displacement obtained from the NSP can
be considered as an upper bound because the structure is not allowed to oscillate
and therefore return to a residual offset closer to its original upright position.
Hence, for the purpose of these guidelines the NSP curves for the structure in the
damaged conditions are assumed to start from the origin of the axes (i.e., no
permanent displacement). The effects of the expected (or measured) dynamic
residual offsets on the residual lateral capacities are accounted for in STEP 3.

C: As illustrated in Figure 2, the horizontal shift of the NSP curves for the
damaged structures back to the origin does not shift the collapse (e.g., DS6)
displacement. In STEP 3, however, the collapse displacement will, in effect, be
reduced by the residual offset.

C: The comment at STEP 1 on unsymmetrical structures applies to a greater

degree to damaged buildings. This may sometimes apply to buildings that are
symmetrical in their intact condition. Note also that certain of these states may
be associated with local collapse and/or with important changes in stiffness or
capacity in the transverse direction. In other words, in some cases (e.g., tilt-ups)
a building characterized by a damage state associated with, say, lateral loads
applied to the N-S direction may need to be “pushed” in the E-W direction as
well.

STEP 3: Inferring dynamic response from static response

The nonlinear dynamic behavior of the structure in both the intact and the damage states
defined at STEP 2 is estimated here using the NSP from STEPS 1 and 2 and the provided
(Microsoft Excel) spreadsheet SPO2IDA (Vamvatsikos and Cornell, 2001b; Vamvatsikos
and Cornell, 2002). The suggested procedure is slightly different for the intact and the
damaged cases and, therefore, they are treated separately below. The dynamic response
of the intact building is evaluated for a full range of potential mainshock ground motion
levels. The result of this operation, which is the dynamic counterpart of the static
pushover curve, is called the Incremental Dynamic Analysis (IDA) curve (Vamvatsikos
and Cornell, 2001a). The dynamic response of the damaged building is of interest here
only at aftershock ground motion levels close to the ultimate capacity of the building and
not for the full spectrum of ground motion intensity.

The use of the SPO2IDA tool requires that the NSP’s used as input be normalized in both
axes (i.e., base shear, BS, vs. roof drift, ∆) by the corresponding quantities at the onset of
damage (or incipient yielding) of the structure. The engineer can use his/her judgement
to identify an appropriate normalization point that can be associated with onset of
damage or incipient yielding in the structure. Selecting an accurate yielding point is
desirable but not crucial. It is, however, important that both the normalizing base shear,
BSy, and roof drift, ∆y, values be consistent with the same selected point on the pushover
curve.

12
However, the IDA’s, which constitute the output of the SPO2IDA tool, have the Y axis
expressed in terms of the ratio of a ground motion parameter rather than of base shear.
The parameter selected here is the spectral acceleration, Sa(T1), at the fundamental
oscillatory period of the intact structure, T1, and the normalizing quantity is Say(T1), the
spectral acceleration at incipient yielding. Of course, the two ratios in terms of base
shear or of spectral accelerations are numerically equivalent. This change of variables on
the Y axis for the IDA case is important in this procedure because the tagging strategy is
based on exceedance of ground motion spectral acceleration (STEP 4) rather than base
shear.

To facilitate the comparison of IDAs corresponding to different damage states, their

ordinates should be “de-normalized” (namely, multiplied by Say(T), where T is equal to T1
for the intact structure IDA and to TDSi for the IDA of structure in damage state DSi).
Under the assumption that the response of PG&E structures is dominated by the first
mode, Sa(T1) for the intact structure (or Sa(TDSi) for the structure in damage state DSi) can
be obtained simply by dividing BS by the building effective modal mass for the
fundamental vibration mode, M 1∗ = ψ 12φ1Τ Mφ1 , where ψ 1 is the first-mode participation
factor, φ1 is the first modal shape, and M is the mass matrix. If the contribution to the
response from higher modes is significant then more modal participation factors come
into play. A detailed procedure on how to transform base shear into spectral acceleration
can be found, for example, in the HAZUS manual (1999).

The procedure above applied to the structure in different states of damage yields,
however, de-normalized IDAs that, in general, are expressed in terms of spectral
accelerations at different oscillatory periods. The commentary provides approximate
methods to convert Sa(TDSi) into Sa(T1) for all damage states DSi. Note, however, that
obtaining IDA’s expressed in terms of spectral acceleration at the same period is
convenient for graphical representation and visual comparison of IDA’s but not strictly
necessary for applying the procedure. More details follow in the commentary.

Numerical examples of base shear to spectral acceleration conversion and axes

normalization are given in the case study section.

Intact structure

An IDA curve is generally considered a tool for predicting the dynamic median
deformation (e.g., inter-story drift or roof displacement) of a structure for a given level of
ground motion severity (here measured by Sa(T1)6) (Vamvatsikos and Cornell, 2002). In
these guidelines, however, the IDA curve is used in the opposite direction, namely to
obtain the mainshock median ground motion level, Sa, that is expected to cause the intact
structure to end up in each of the identified damage states, DSi. Such Sa values can be
read off the IDA curves at the roof drift values ∆i associated with such states. Figure 3
shows a typical output of the SPO2IDA tool where the NSP curve (or, more precisely, its
quadrilinear fit, as discussed in the commentary) is the input and the IDA curve is the
output. Note that in Figure 3 the deformation measure on the X axis is the roof global
6
The period T1 is often dropped hereafter for simplicity.

13
ductility ratio, µ, which is the ratio of the roof drift to the roof drift at the onset of damage
or incipient yielding. Further the IDA curve defines the global collapse capacity
(indicated by a circle in Figure 3) of the intact structure. The basis for IDA curves is
described in the commentary. It is important to emphasize that the IDA curves are to be
considered as median curves. The record-to-record variability will be explicitly dealt
with during the development of the fragility curves in STEPS 5 and 6.

6.00
intact IDA curve
5.00

4.00
R

3.00
DS3 DS4 NSP curve
2.00 DS2
DS5
1.00 DS1 DS6=coll

0.00
0.0 5.0 10.0 15.0
global ductility ratio, µ

Figure 3: Hypothetical NSP and IDA curves for the building in the intact condition. In this case the
abscissa represents the global ductility ratio, µ (namely, the roof drift divided by the roof drift at first
yielding, i.e., at DS1). The ordinate R is equal to BS/BSy for the NSP curve and to Sa(T1)/Say(T1) for the
IDA curve. Note: the dashed line is an example of quadrilinear fit of the pushover curve, as explained in
the commentary to come.

Damaged structure

IDA curves are also part of the procedure applied to the damaged structure. In this case,
however, IDA curves are not used for predicting the mainshock median Sa (given a roof
drift value, ∆i) that will cause the intact structure to end up in damage state DSi, but rather
for estimating the median residual dynamic lateral “capacity”, (Ša,cap)i, of the damaged
structure in each post-mainshock damage state DSi to resist aftershocks. (Ša,cap)i is the
(aftershock) ground motion level that is expected to cause subsequent collapse and,
should the building still be occupied, life loss. The same SPO2IDA spreadsheet provides
an estimate of (Ša,cap)i as one of the points of the IDA for the damaged structure (Figure 4
shows an example for damage state DS3). This capacity is defined either as the condition
where small increases in “load” during aftershocks will cause arbitrarily large increases
in deformation (i.e., where the IDA curve becomes effectively flat; this definition was
also adopted by the SAC/FEMA 350-352 Guidelines (2000)7) or as the aftershock
spectral acceleration that causes local collapse anywhere in the structure.

7
The IDA curve could be also nearly flat close to collapse. SAC used the point at which the local slope is
20% of the initial slope or less.

14
Recall from the commentary of STEP 2 that the NSP curves for the damaged structures
have been shifted to remove the static residual displacement offsets created by the NSP
procedure. The IDA curves derived from those NSP curves, therefore, assume no
residual offset. As one might expect, though, the residual offset after a mainshock
influences the residual capacity to resist aftershocks, (Ša,cap)i (e.g., Luco et al., 2004; see
Appendix C). In these guidelines, an IDA curve for each damaged structure that accounts
for the residual offset is obtained by tracing the IDA curve that assumes no residual offset
up to the point that has a displacement equal to the displacement capacity (defined in the
preceding paragraph) minus the expected (or measured) dynamic residual offset (defined
in the commentary). See the case studies for examples. Equivalently, the collapse
displacement of the NSP curves found in STEP 2 (e.g., the displacement at DS6 in Figure
2) can be reduced by an amount equal to the expected (or measured) dynamic residual
offsets, and these NSP curves that account for the residual offset can be input into the
SPO2IDA spreadsheet. An example of the resulting IDA curve is shown in Figure 4.

Even without considering a residual offset the IDA curves for the damaged structures will
generally provide estimates of the residual capacities, (Ša,cap)i, that are not equal to that of
the intact structure, namely Ša,cap=(Ša,cap)1.8 (the estimates of the median capacity for the
intact structure and for the damage state DS1, which defines the Onset of Damage state,
are the same). In contrast, dynamic analyses of damaged structures (e.g., Luco et al.,
2004) indicate that without a residual offset little or no reduction in the median lateral
capacity relative to intact structures is expected, at least for the structures analyzed here.
To reflect this observation, an adjustment to (Ša,cap)i is made in these guidelines. As
demonstrated in the case studies, the adjustment is achieved simply by multiplying
(Ša,cap)i by the ratio of the capacity of the intact structure to the residual capacity
assuming no residual offset. Hence, if no residual offset is expected (or measured), the
adjusted residual capacity will equal (Ša,cap)1. The NSP curve needs to be obtained for all
the post-earthquake damage states (STEP 2) in which none of the structural elements has
reached its ultimate vertical capacity to withstand gravity and live loads during the
mainshock. All the damage states that do show loss of local vertical capacity will be
automatically associated with collapse (STEP 4).

8
This is a result of differences in the shapes of the NSP curves for the damaged versus intact structures, as
illustrated in Figure 2.

15
 3
2.5 IDA for DS 3
NSP for DS
2 3
(S a,cap ) / S ay
3
1.5
R

1
0.5
0
0 2 4 6 8 10 12 14 16
global ductility ratio,, m

Figure 4: Estimate of the residual median capacity (Ša,cap)3 for damage state DS3 (after accounting for the
expected, or measured, residual offset after the mainshock). As in Figure 3, the ordinate R is equal to
BS/BSy for the NSP curve and to Sa(TDS3)/Say(TDS3) for the IDA curve.

3.00
IDA for intact structure
2.50 DS2
DS3
2.00

1.50 DS4
Sa (g)

1.00
DS5
0.50

0.00
0.0 2.0 4.0 6.0 8.0
Roof drift (%)

Figure 5: IDA curves for the intact structure and for the structure at different levels of damage (after
accounting for the expected, or measured, residual offset after the mainshock and adjusting for the behavior
observed in dynamic analyses of damaged structures). The circles represent the global collapse residual
capacity of each case. All the IDA’s have been de-normalized and scaled to the spectral acceleration at the
same oscillatory period before including them in the same plot (see commentary).

16
 5.0

4.0 residual capacity N-S

3.0
Sa (g)
2.0
residual capacity E-W direction
1.0
Structure in damage state DS i
0.0
0.0 3.0 6.0 9.0
Roof drift (%)

Figure 6: Residual global capacities for the building in damage state DSi estimated from IDA curves
(dotted lines in the figure) for the two main orthogonal directions.

Summary of STEP 3: Using the SPO2IDA spreadsheet provided, estimate the dynamic
response of the structure in its intact and damaged conditions from the corresponding
NSP curves.
Output of STEP 3: the IDA curve for the structure in the intact condition (solid line in
Figure 5) and estimates of the residual capacity (Ša,cap)i for the damaged structure in each
damage state DSi of different severity. The residual capacity is associated with a ground
motion level that causes either global lateral instability (as the circles shown on the dotted
IDA curves in Figure 59,) or local collapse of any element in the structure.

C: It is important to note that the spectral acceleration values corresponding to

the median capacities (Ša,cap)i obtained for damage state DSi (and the entire IDA
curve as well) is in terms of an oscillatory period TDSi which is in general longer
than the initial period of vibration T1 of the intact structure.10 The lengthening of
the fundamental period is due to the damage sustained by the structure. The
value of TDSi can be computed as follows:

K1
TDSi = T1
K DSi

9
The damage state DS6 is not included in the figure because in this illustrative example the local vertical
capacity is assumed to be reached in at least one member at this damage state during the mainshock. Hence,
DS6 is automatically associated with collapse.
10
Obviously this is not the case if, in STEP2, parallel-to-elastic unloading/re-loading is assumed for the
NSP curves, which is another advantage of such an assumption. As demonstrated in the case studies, the
resulting estimates of (Sa,cap)i are comparable to those that make use of NSP curves computed by unloading
and re-loading the structural model.

17
 where K1 and KDsi are the elastic stiffness values of the intact structure and of the
damaged structure in damage state DSi obtained directly from the respective
pushover curves. Numerical examples can be found in the case studies.

C: In order to display all the IDA curves for the intact and the damaged cases in
the same plot (e.g., Figure 5), it is important that they be expressed in terms of the
same reference spectral acceleration, which in this context is chosen to be the
initial elastic fundamental period, T1, of the intact structure. If TDSi is different
than T1, an approximate method to derive Sa(TDSi) from Sa(T1) consists of
multiplying Sa(T1) by an average ratio of Sa(TDSi)/Sa(T1) computed from the
ordinates of a few uniform hazard spectra at appropriate levels of hazard
computed for the building site. Seismic hazard data are easily available from
USGS at the USGS National Seismic Hazard Mapping Project website
(http://geohazards.cr.usgs.gov/eq/index.html) or, at more levels of hazard for
spectral acceleration at additional frequencies, from the CD released by USGS
(2001). An alternative method consists of deriving an average ratio of
Sa(TDSi)/Sa(T1) from an attenuation relationship suitable for the site using a few
pairs of appropriate magnitude and distance values. If the attenuation
relationship does not directly provide values for Sa(TDSi), they can be obtained via
interpolation. Finally, if the periods are in the moderate range, it may be
sufficient to simply assume that the spectral acceleration ratio is proportional to
the inverse of the ratio of the periods, as it would be if the spectrum displayed an
equal spectral velocity in this range.

C: As described above in the guidelines, an adjustment to (Ša,cap)i is made to

reflect the observation that the median residual capacity from dynamic analyses
of damaged structures shows little or no reduction relative to intact structures if
there is no residual offset. This adjustment is achieved by multiplying (Ša,cap)i by
the ratio of (i) the capacity of the intact structure, namely (Ša,cap)1, to (ii) the
residual capacity of the damage structure assuming no residual offset. As will
become clearer in the case studies, this adjustment will, in effect, convert (Ša,cap)i
from a value in terms of TDSi to one in terms of T1. Hence, other than for the
purpose of including the IDA curves for the intact and damaged cases in the same
plot, the conversion described in the preceding comment is not strictly
necessary.11

C: The SPO2IDA spreadsheet requires also that the NSP curve be specified in a
piecewise linear fashion. A quadrilinear fit to the computed NSP curve (see
dotted line in Figure 3) may be necessary before the SPO2IDA tool can be used
(Vamvatsikos and Cornell, 2002). The four linear lines represent the elastic part,
the hardening part, the softening part, and the residual plateau of the pushover

11
In fact, since the effect of the adjustment to (Sa,cap)i is equivalent to finding the residual capacity by
multiplying the intact capacity (Sa,cap)1 by the ratio of the residual capacities with and without a residual
offset (both in terms of TDSi), the ordinates of the IDA curves for the damaged structures do not need to be
de-normalized from R to spectral acceleration.

18
 backbone curve. Some degree of engineering judgment to achieve the best
quadrilinear fit may be needed.

C: The residual capacity of the damaged structure in state DSi may need to be
evaluated for both main orthogonal directions of the building. Figure 6 shows
examples of residual global instability capacities (indicated by circles) estimated
with the aid of IDA’s for both N-S and E-W directions of a building. In this case
the residual capacity for the state DSi will coincide with the residual capacity in
E-W direction, the smaller of the two values. Note that, as explained above, the
IDA curves in Figure 6 are expressed in terms of spectral acceleration at the
same initial period of vibration, T1, of the intact structure in the E-W direction.

C: The dynamic response to ground motion of increasing severity can,

conceptually, be evaluated in a similar manner to that done in the nonlinear static
procedure. Increasingly scaled versions of the same ground motion time history
could be applied to the model of the intact building and the roof drift (or other
appropriate deformation parameter) could be recorded versus Sa(T1)
(Vamvatsikos and Cornell, 2001a). Further repeating the process for a suite of
earthquake records permits the estimation of the median displacement (at each Sa
level) representing the variety of future records of the same intensity. The
resulting curve, namely the Incremental Dynamic Analysis (IDA), is, in effect, the
dynamic counterpart of the NSP curve. An IDA curve provides more accurate
information on how the building will behave dynamically during ground shaking
of different severity levels. Analyses of this kind have been done to validate the
studies in this report. See examples below and the study by Maffei and
Hamburger (2004).

C: Note that IDA and NSP curves can be shown on the same graph because they
share the same quantities12 in both axes. Figure 3 shows one such NSP-IDA pair
for one of the two main orthogonal directions of the building. Of course, a
second different NSP-IDA pair can be constructed for the other orthogonal
direction.

C: The procedure of performing multiple structural response analyses using

scaled ground motion records may, again in principle only, be repeated for the
structure in each of the damage states identified in STEP 1 and analyzed
statically in STEP 2. The damage is typically reflected in decreased
stiffness/capacity of damaged elements in the structural model. Again, if any
damaged element at any damage state is considered to have “failed” (i.e.,
vertical ultimate capacity exceeded) during the mainshock, then the NSP and the
IDA curves need not be computed (as is damage state DS6 in the illustrative
example). The collapse limit state will be assigned to such a damage state
without further analyses. In general, for each structural damage state the result
will be one or more IDA curves (possibly, again, one for each main perpendicular

12
The ordinate R, which is BS/BSy for NSP curves and Sa/Say for IDA curves, is numerically, if not
conceptually, the same.

19
direction of the building, and in principle, one for each of the positive and
negative directions). For example, see Figure 6. For comparison purposes, the
severity of the ground shaking will be expressed in terms of the spectral
acceleration at the fundamental period (in one of the main perpendicular
directions), T1, of the intact building for all damage states.

C: The task described above would require performing many nonlinear dynamic
analyses of the building in both the intact and the several damaged conditions
identified at STEP 2. This has been done for the case study buildings in Phase 2
of this project, and in at least one case in practice (Maffei and Hamburger, 2004),
but it is not expected to be common in even the best engineering practice in the
near future. Instead it is assumed here that the IDA curves of the building (intact
or damaged) can be estimated by those of simple nonlinear oscillators with force-
deformation backbone curves equal to the corresponding NSP curves derived at
STEPS 1 and 2 (coupled with an assumed force-deformation hysteretic behavior).
The spreadsheet provides estimates of the peak dynamic roof drifts using the same
basic assumption of FEMA 273 (1997) and ATC-40 (1996), namely that they can
be adequately predicted by the response of an appropriate nonlinear SDOF
oscillator. The spreadsheet, however, is not limited to bilinear elasto-plastic
oscillators.

C: IDA studies on nonlinear oscillators defined by many different force-

deformation NSP backbone curves and associated hysteretic behavior have been
conducted and synthesized, permitting now the estimation of IDA curves directly
from NSP curves without further need for performing nonlinear dynamic analysis.
The IDA representation can be derived directly from a NSP curve based on
empirical rules tuned to thousands of analyses conducted with different
oscillators and different ground motion. Again, a spreadsheet (Vamvatsikos and
Cornell, 2001b and 2002) is provided along with these guidelines for this
purpose. A version of the SPO2IDA tool for SDOF oscillator of moderate periods
can be found at http://tremble.stanford.edu/cgi-bin/spo2ida-mt.pl.

C: The SPO2IDA spreadsheet does not consider oscillators that have an initial
displacement offset. Hence, the manner in which SPO2IDA is used in these
guidelines to estimate the residual collapse capacity of a damaged structure with
a residual offset is an approximation that may not agree well with the results of
nonlinear dynamic analyses. To some extent, the adjustment to the estimate of
(Sa,cap)i produced by SPO2IDA that is made in these guidelines corrects for this
discrepancy.

C: The "expected" residual displacement offset referred to in these guidelines can

be based on residual-given-peak displacement data from nonlinear dynamic
analyses (of intact structures). Some such data is presented in the case studies,
but a more extensive source of information on the subject can be found in Ruiz-
Garcia, 2004). Alternatively (or as an update), the residual offset after a
mainshock can be measured directly in the field. In this latter case, portions of

20
 STEPS 3-6 will need to be completed after a mainshock has occurred, or else they
can be completed a priori for a range of potential residual offset values.
Fortunately these steps are not very time-consuming, particularly in comparison
to the steps that can be completed before knowing the value of the residual offset
(i.e., STEPS 1-2 and most of STEP 3).

C: It is worth emphasizing again that for the purpose of these guidelines, it is only
the residual collapse capacities (either global or local) of the damaged structures
that are of major interest. These are defined, as for the intact structure, as the
ground motion that causes either global lateral instability or local collapse
anywhere in the structure. Figure 5 shows the former case where global
instability is reached. The residual capacity values measured in terms of spectral
acceleration will be used in STEP 4 to evaluate the post-earthquake tagging
condition designation. The spreadsheet provided, however, estimates the entire
IDA and not only the residual capacity.

STEP 4: Occupancy Status for Damaged Building

The criteria that follow are used to recommend the tagging condition (i.e., the limit state)
to the structure in any specific damage state, DSi, for which no loss of local or global
vertical capacity is observed during the mainshock. If loss of vertical capacity is
observed, then that damage state is associated with collapse13.

The proposed criteria for tagging damaged buildings, in whichever damage states they
may be, are expressed in terms of:

• P0, the building-site-specific mean annual frequency (MAF) of exceedance of the

ground motion corresponding to the median capacity, (Ša,cap)1, of the building in
its intact conditions. P0, which refers to the pre-earthquake conditions, can be
obtained using conventional Probabilistic Seismic Hazard Analysis (PSHA) codes
or directly from the USGS website (http://eqhazmaps.usgs.gov/) for a selected set
of oscillator periods and firm-soil to soft-rock conditions.
• P, is the building-site-specific MAF14 of the aftershock ground motion
corresponding to the median capacity, (Ša,cap)i, of the building in the damage state
DSi. P is, strictly speaking, a time-varying quantity that decreases with time
elapsed from the mainshock and, therefore, it is better computed using an
aftershock PSHA (APSHA) approach, such as that presented in Wiemer (2000)
13
Local collapse potentially represents a life-threatening condition. A damaged building that was made
unstable by the mainshock may in fact become progressively unsafe or may (partially or globally) collapse
under its own weight and live loads without the occurrence of any aftershock (e.g., ATC TechBrief2,
1999).
14
The primary criteria proposed first in fact implicitly assume that one can use the pre-mainshock hazard or
MAF, which is more familiar and more readily available, as a proxy for the post-mainshock or “aftershock
hazard”. The alternative criteria discussed later in the commentary are based more explicitly on the
aftershock hazard as is more consistent with the intent of the procedure. In that case a new concept,
equivalent constant rate, ECR, analogous to MAF is introduced.

21
 and Yeo, 2004. (For an application of APSHA, see also Yeo and Cornell, 2004,
which is attached for convenience in Appendix C.) However, although it requires
only minor modifications to conventional PSHA codes, software for performing
aftershock hazard is not yet widely used. For this reason, the primary criteria
proposed here implicitly assume that one can use the pre-mainshock hazard or
MAF, which is more familiar and more readily available, as a proxy for
evaluating the post-mainshock or “aftershock hazard”. The alternative criteria,
discussed later in the commentary, are based more explicitly on the aftershock
hazard as is more consistent with the intent of the procedure. The alternative
tagging criteria account for the increased probability of collapse from pre- to post-
earthquake conditions due both to decreased capacity and to increased
(aftershock) seismicity
• Estimates of Ša,cap for the intact building and of (Ša,cap)i for all the damage states
DSi’s.

The proposed primary tagging criteria are displayed in graphical form in Figure 7. The
figure has two scales for the ordinates, the percentage of loss in Ša,cap and the ratio of
P/P0 that measures the increase in frequency of exceeding the median residual capacity of
the building damaged by the mainshock15.
10
8
50
Capacity
RED
Loss of Sa cap (% )
5
Loss (%) P/P0
40
2 1.1
P/P0

4
5 1.2
3 10 1.4
YELLOW 5%/50yrs 20 2.0
30 2.9
2 20
40 4.6
2%/50yrs 50 8.0
GREEN 10
5
2
1 0
1.E-04 5.E-04 1.E-03

P0

(a) (b)
Figure 7: (a) Graphical representation of the recommended tagging criteria. (b) Average relationship for
loss of ground motion capacity and rate of increase in mean annual frequency of exceedance of ground
motion for coastal California sites for which the absolute value of the (log-log) slope of an average ground
motion hazard curve in the surroundings of 10-3 annual frequency of exceedance is about three16.

15
The values of the quantities in Figure 7 and in the text are proposed by the authors and should be re-
evaluated and customized by PG&E to fit their specific needs. Note that these values can also be tailored
differently for structures of different importance.
16
The hazard curve slope, which depends on the magnitude distribution and on the rate of decay of seismic
waves with distance, is lower in Eastern United States, for example. Hence, in principle caution should be
exercised when using this same chart in other areas of the world. The abscissa refers to the pre-earthquake

22
The criteria can be summarized as follows. Any building is identified by a particular
value of P0 that can be computed during “peace” time before any earthquake has
occurred. A larger value of P0 implies that the building is either relatively “weak”, or
that it is located in an area of higher seismic hazard compared to that considered in the
original design, or a combination of both. The opposite is true for lower values of P0.
How the color of the tag changes with capacity loss can be found by searching on a
vertical line at that specific value of P0. Therefore a building whose P0 is equal, for
example, to 3 x 10-4 needs to be damaged severely enough to loose about 5% of its initial
capacity before it is tagged Y and about 30% before it is tagged R. If the intact building
had been much weaker or in a harsher seismic environment such that its value of P0 were
equal, for example, to 1 x 10-3, then a nominal loss of lateral capacity of only 2% or
larger would cause the building to be red-tagged. No yellow tag could be assigned in this
case, the tag would be either green or red. Conversely a much stronger building with P0
smaller than 2 x 10-4 would need to lose 20% of its original capacity before being tagged
Yellow and 40% to be tagged Red. An explanation of the genesis of such values is
provided in the commentary.

The illustrative example shown in Figure 8 refers to a structure located at a site for which
P0 ≈ 2.2x10-4. For this pre-mainshock hazard level the structure is yellow-tagged if it
looses about 15% of the original capacity and it is red tagged if it looses about 35% of the
original capacity. In this case the damage state DS2, which shows a residual capacity,
(Sa,cap)2, greater than 85% of the capacity of the intact structure, (Ša,cap)1, is green-tagged.
Similarly, the damage states DS3 and DS4 are tagged Yellow and Red, respectively,
because 0.65x(Ša,cap)1<(Ša,cap)3<0.85x(Ša,cap)1 and (Ša,cap)4<0.65x(Ša,cap)1. The damage
state DS1 can be associated with the onset of damage (OD) limit state, which is within the
green tag state boundaries regardless of the value of P0. Similarly, the damage state DS6,
which corresponds to incipient (local and global) collapse limit state is always red-
tagged.

conditions only; it is simply the long-term MAF of exceedance of the intact building ground motion
capacity
17
The values of the quantities in Figure 7 and in the text are proposed by the authors and should be re-
evaluated and customized by PG&E to fit their specific needs. Note that these values can also be tailored
differently for structures of different importance.
18
For California sites, it is assumed here that the absolute value of the (log-log) slope of an average ground
motion hazard curve in the neighborhood of 10-3 annual probability of exceedance is equal to 3.

23
 3.00
100% IDA for intact structure
2.50 85% GREEN DS2
DS3
2.00 65% YELLOW

1.50 DS4
S a (g)
RED
1.00
DS5
0.50

0.00
0.0 2.0 4.0 6.0 8.0
Roof drift (%)

Figure 8: Tagging of different damage states within the illustrative example. The yellow tagging given to
DS3 assumes that the P0 of exceeding the spectral capacity (Ša,cap)1 at the building site is about ≈ 2.2x10-4.

Summary of STEP 4: apply criteria based on global residual capacity reduction and site-
specific aftershock19 ground motion hazard for mapping damage states into structural
limit states.

Output of STEP 4: a one-to-one map of each damage state DSi into one of the following
four limit states: Onset of Damage (OD), Yellow tag (Y), Red tag (R), and Collapse (C).

C: The objective of this step is to determine whether a building that has sustained
damage as defined by one of the damage states DSi identified in the previous
subsections can be safely occupied or whether access should be limited or
forbidden. More formally, this step assigns structural limit states to the damage
states identified in STEP 2.

C: The post-earthquake occupancy status of the building is historically defined in

terms of color tags posted to the building (e.g., ATC-20, 1989; ATC TechBrief 2,
1999; FEMA 352, 2000). In these guidelines the tagging definitions customized
to PG&E buildings are as follows (PG&E, 2001b):

• Green tag (“inspected and safe”) implies that the building may
require some repairs but is safe for immediate occupancy by
PG&E personnel, pending completion of detailed evaluations.
• Yellow tag (“restricted occupancy”) implies that access to
PG&E personnel in any hazardous area of the building should
be restricted until repairs or stabilization can be implemented
The building is open only for structural repairs, emergency
operations, and retrieval of equipment. For buildings of
critical importance, these operations may require emergency
19
See Footnote 3.

24
 personnel access inside the buildings (or parts thereof) for
essentially the duration of the emergency (i.e., 72 hours as per
PG&E, 2001b).
• Red tag (“unsafe”) implies that the building is unsafe and it
should not be occupied for at least the entire emergency period
of 72 hours. After 72 hours, entry would be permitted for
emergency stabilization of structural elements (after a
structural review). In extreme cases, stabilization of a red-
tagged building may not be possible or economical and the
building may need to be demolished.

C: If needed, these designations could be further subdivided into more specific

categories based on whether the damage is non-structural only or structural as
well and on whether the building can pose a threat to occupants. FEMA 352,
Section 3.3, suggests the following color tag definitions: Green 1 if the damage is
only non-structural and the structure is safe Green 2 if the structure has suffered
non-structural and structural damage and the structure is safe, Yellow 1 if the
structure has sustained non-structural damage to a level that could be unsafe for
occupancy, Yellow 2 if the structure has suffered non-structural as well as
structural damage that pose a limited safety hazard, Red 1 if the structure is
severely damaged but repairable, and, finally, Red 2 if the structure has sustained
a great deal of damage to be a potential collapse hazard and should be
demolished. This refinement, however, is not currently included in these
guidelines.

C: Historically, trained engineers have assigned the tagging categories only on

the basis of visual inspection of the building. A number of different analytical
approaches have been proposed, for use when time permits, to supplement the
tagging procedure based on building inspection. Whereas the implied time-
consuming engineering analyses are not feasible in the immediate post-
earthquake period in which PG&E facility decisions are urgently needed, these
previous proposals do give guidance as to how to make tagging decisions in the
new context defined by the present guidelines. In these guidelines such analyses
(STEPS 1 through 3) can be made prior to the earthquake during the assessment
of the power system reliability. These same assessments can be made available to
the inspectors of PG&E buildings in order for them to be able to make more
informed decisions immediately after the event. For example, the FEMA 352
Guidelines for post-earthquake evaluation of steel moment-resisting frames
suggest three different analysis-based procedures:

• Determine the capacity of the damaged building relative to

current code requirements
• Determine the capacity of the damaged building relative to
pre-earthquake conditions
• Determine the probability of earthquake-induced collapse of
the damaged building in a specified exposure period

25
C: The first method has the clear drawback that some of the old buildings may not
be fulfilling the requirements of the modern codes even in their intact state. This
procedure may in principle lead to yellow- and red-tagging buildings even before
the occurrence of damage caused by any earthquake. This option does have the
benefit, however, of being associated with an absolute life safety criterion. This
first option is considered here only in the alternative criteria below.

C: The second method, despite some obvious caveats (e.g., the same capacity
reduction from an originally “strong” and an originally “weak” building should
not necessarily lead to the same tagging color), is appealing for its simplicity.

C: A combination of the second and third method has been adopted as the
“primary” criteria of these guidelines. The likelihood of building collapse is
considered here to prevent the possibility that the second method alone awards a
green tag to relatively lightly or moderately damaged, older, originally relatively
weak buildings despite unsafe occupancy conditions. A tagging strategy based
simply on reduction of lateral capacity with respect to that of the building in its
pristine condition (i.e., the second method alone) may also fail to differentiate
between initially “strong” and initially “weak” structures and, therefore, only
partially promote seismic retrofit of buildings.

C: As stated above, the primary tagging criteria explicitly consider the likelihood,
P, that the damaged-building capacity is exceeded but the computation of P is
performed for simplicity using pre-mainshock PSHA. For these guidelines for the
boundaries between green and yellow tags and between yellow and red tags,
respectively, we selected P values to be equal to 2% in 50 years (mean return
period, MRP, of 2,475 years) and 5% in 50 years (MRP of 975 years). These
values may appear too restrictive when compared to building code requirements
for new buildings that prescribe life safety as performance objective for a 10% in
50 years (i.e., MRP of 475 years) ground motion level. We selected low values to
implicitly and partially account for the increased aftershock hazard that the
damaged building is subject to when the inspection may take place perhaps one
or two days after the earthquake. These values, which represent quantitative
measures of acceptable risk, should be modified according to the building
importance and severity of failure consequences.

C: If the proposed tagging criteria were simply based on maximum acceptable

collapse risk of the partially damaged building, however, the green, yellow, and
red tag areas in Figure 7 would be oblique bands delimited by straight lines of
constant P values. The diagonal bands in Figure 7, however, are delimited by
horizontal lines drawn somewhat arbitrarily at constant values of capacity loss of
2%, 20%, and 40%. “Weak”, under-designed buildings that are potentially
unsafe (i.e., larger P0) even in pre-earthquake condition would not be tagged Y or
R and possibly, at a later stage, retrofitted unless some identifiable physical
damage occurred in the building. Tagging Y or R an undamaged building would
be difficult to accept by owner and occupants. A hardly detectable capacity loss of

26
only 2% encourages Y and R tagging and it is meant here to simply serve as a
trigger for action for such buildings. The lines at 20% and 40%, that increase the
Y and R tagging areas of “strong” (smaller P0) buildings, have been dictated by a
different concept. Results from dynamic analyses (see Luco et al., 2004 in
Appendix C) have shown, in general, that a rather widespread damage in the
building is needed before the capacity drops by these amounts. In cases of
widespread damage the assessment of the true building capacity is more
uncertain and it is conservative to force some restriction of occupancy until
further more detailed analyses are performed.

C: Note that a collection of figures such as Figure 7 could be produced to

account for the time-varying nature of aftershock hazard. Such tagging criteria
could be applicable after the earthquake at different snapshots in time when more
detailed inspections and/or improved capacity analyses may take place. To mimic
the decreasing aftershock hazard with time one could devise criteria that are less
stringent as time goes by. This would be reflected by a change in the constant-P
lines that separate green, yellow, and red tagging areas. Criteria applicable, for
example, one month after the mainshock may have constant P lines demarcating
the G, Y, and R areas at higher values than those displayed in Figure 7, which is
meant to be applicable immediately after the mainshock. The increased
knowledge about building capacity deriving from more detailed inspection and
further engineering analyses may also call for the removal of the conservative
lines at 20% and 40% capacity drops in the criteria applicable at a later stage.

C: A final remark is in order. It is conceptually preferable to develop tagging

criteria in terms of MAF of building collapse rather than MAF of median ground
motion capacity. This entails considering the ground motion capacity as a
random variable as opposed to a constant. Under certain tenable assumptions the
former MAF can be computed by multiplying the latter by a “correction factor”,
CF, larger than one that accounts for the aleatory and epistemic uncertainties in
the ground motion capacity (called β in the section to come) and the slope of the
hazard curve in the neighborhood of that ground motion level. Figure 7 could
still be used as is with this alternative interpretation. As a first approximation, CF
can be considered to be the same for the capacities in both the intact and the
damaged cases and therefore it does not have any impact on the values of the
oblique lines of constant P/P0 that divide green, yellow, and red tag regions. For
a particular application, the only change would be in the value, P0’, of the
abscissa to be used to enter the graph in Figure 7, which will be larger than the
MAF of exceeding the median ground motion capacity currently used. This new
interpretation is briefly discussed again in the alternate criteria below.

C: As anticipated in the text, a second set of alternate criteria is based on

aftershock PSHA. Although these alternative criteria are more in tune with the
spirit of these guidelines, at the time of this writing the use of aftershock PSHA is
still not common. Hence, they have been proposed here only as a viable

27
alternative although we recommend that in the future its use is considered by
PG&E.

C: STEP 4 Alternate: Occupancy Status for Damaged Building

C: In this alternate approach to determining the tagging state of the structure the
criteria are based on an assessment of the aftershock ground motion hazard at the
site, HA(Sa), which equals an “equivalent” MAF. Analogous to the discussion
above we denote the MAF associated with the median aftershock capacity,
(Ša,cap)i, of the building in damage state DSi as PA when it is based on this
aftershock hazard assessment. In this case the tagging criteria are stated as:

• If PA is less than α1P’0 then the tagging condition for DSi is green.
• If PA is more than α1P’0 but less than α2P’0 then the tagging condition is
yellow.
• If PA is more than α2P’0 the tagging condition is red.

C: The MAF level P’0 is the presumed new building design ground motion MAF.
Subject to further debate, it is suggested that this value be taken as 2% in 50
years or 0.0004 per annum even in near-fault zones where other, lower definitions
of the design ground motion may apply. The proposed values for post-
earthquake-recovery-critical PG&E facilities are:

• α1 = 3
• α2 = 10

Such values are selected arbitrarily and can be used as default values before
PG&E defines its own ones. These criteria plot as oblique bands in Figure 7 at
levels about a factor of three higher than those shown there. The additional
marginal adjustments might be made as well. These criteria do not, however,
depend on the P0 of the existing intact structure, which may be substantially
higher or lower than P’0 for new buildings. See the comments below for further
discussion.

C: The aftershock PA is not, however, a simple constant. The rate of aftershocks

depends on the mainshock magnitude and the rate decays as a function of the time
lag, τ, after the mainshock. To a lesser degree it also depends on the duration of
the interval in the future during which the building will remain in the post-
mainshock damaged state.

C: For current purposes, before explicit aftershock ground motion hazard results
are available, we suggest the following approximations. Consider a Base Case,
in which the mainshock moment magnitude is 7.0, the tagging decision is made
within one day of the mainshock, and the duration of interest is in the order of a

28
 one year. In that base case the effective20 PA is approximately21 equal to the pre-
mainshock hazard (for any Sa capacity level). So, for example, if the pre-
mainshock MAF of exceeding a particular Sa capacity value is 0.001 per annum,
then so is the effective PA. By the criteria above a building with this Sa capacity
would by green tagged. The pre-mainshock value may be obtained from an
existing site-specific study or from the previously cited USGS web site. However,
the PA is approximately 3 times higher for a magnitude 7.5 event and about 6
times higher for an 8. For these mainshock values the same Sa capacity would
have effective aftershock hazard, PA, of 0.003 and 0.006 respectively. By the
criteria above this building would be yellow tagged in the first case and red
tagged in the second. (For a magnitude 6 PA is 1 to 2 orders of magnitude lower
than for a 7 and may be safely ignored.)

C: If a tagging decision can be delayed to a later date after the mainshock or if a

decision is to be made a second time at a later date, the value of PA will go down
by a factor of about 0.75 at 7 days, a factor of 0.6 at 30 days, and 0.3 at one year
after the mainshock. So for example the building with the Sa capacity with a
0.001 pre-mainshock hazard would find its effective aftershock hazard reduced
from 0.006 to 0.0036 some 7 days after the mainshock, and, according to the
criteria above, its red tag changed to yellow. For the most common early
decisions the PA is quite insensitive to the duration of interest provided it is about
6 months or longer.22

C: The analysis here is designed to echo the primary criterion. So, for example, it
too makes the tagging criteria dependent on simply PA, the MAF associated with
exceedance of the median Sa capacity, and not on the MAF of collapse. The
paper underlying the recommendations here, Yeo and Cornell (2004) is, however,
based on the MAF of collapse, comparing this via α1 and α2 to the pre-mainshock
collapse MAF. As discussed there the MAF of collapse exceeds the MAF
associated with exceedance of the median capacity by a factor exp[k2β2/2]. This
factor depends on the slope of the hazard curve, k, and the β (uncertainty
measure) of the capacity. The k values of the pre-mainshock and post-mainshock
(aftershock) hazard curves are very nearly equal. If the β’s for the intact and
damage-state capacities were approximately the same then the MAF’s for
collapse would have the same ratio as the MAF’s associated with median

20
The analysis here follows Yeo and Cornell (2004), whose article is included in Appendix C. By
“effective” we mean here what that reference calls the “Equivalent Constant Rate” associated with a social
discount factor of about 3.5%.
21
The simplified procedure here is based on numerical experience with aftershock PSHA (Yeo, 2004). It is
valid for the typical San Francisco Bay Area case where a single nearby fault (segment) is the dominant
source of a damaging mainshock ground motion. In other areas, such as the Los Angeles region where
several faults may contribute to the pre-mainshock hazard, the aftershock hazard will be a smaller portion
of the pre-mainshock hazard. In time it is anticipated that aftershock hazard maps will be provided by a
USGS web site in real time after an event to facilitate tagging on an event and site-specific basis. A test site
has been operating for some time. Prior to an event the aftershock PSHA can also be done site-specifically
for anticipated scenario mainshocks by simple modifications to existing PSHA codes.
22
Values for other magnitudes, time lags and durations can be found in Yeo (2004).

29
 capacity exceedance, and the α1 and α2 values should be the same on either basis.
While the aftershock capacity uncertainty β may well be larger than that for the
intact structure, especially when the building has not yet been carefully inspected
and/or analyzed, the approximation may be close enough for current purposes. If
so it implies that decisions about the α values can be thought of in relative MAF of
collapse terms, e.g., how much larger an MAF of collapse (and potential
fatalities) can be tolerated in post-quake circumstances.

C: The cited paper discusses circumstances and conditions that should be

considered for setting the α levels, such as the importance of the structure to post-
quake recovery. The values above are the highest recommended there, but the
subject deserves broader professional discussion and consensus. The paper also
discusses ways in which tagging restrictions might be eased for volunteer workers
whose exposure is limited.

C: To get a sense of the implications of the proposed α and P’o levels consider the
following examples. The criteria imply that if a critical PG&E building designed
to just meet current IBC standards has a MAF of exceeding the median intact23 Sa
capacity of 2% in 50 years (0.0004 per annum), then it may can be green tagged
immediately after a magnitude 7.5 or less if it is undamaged, and yellow tagged
for such an event even if it has lost some significant fraction of its capacity. On
the other hand, if it were designed for the “deterministic” near-fault MCE with an
implicit higher MAF of, say, 6% in 50 years (0.0012 per annum), then if
undamaged it would be green tagged after a magnitude 7 but yellow tagged after
a 7.5. Damage to the structure would reduce the Sa capacity and increase PA for
all these cases, potentially changing the tagging to a more restrictive level. Older
buildings with lesser Sa capacities before and/or after the main event are also
more at risk. But an undamaged building with a pre-mainshock “MAF capacity”
or a damaged building with a post-mainshock “MAF capacity” of as high as
0.004 (20% in 50 years) would still be accessible on a restricted basis after a
magnitude 7. (These numbers are appropriate, recall, for cases when the site
mainshock hazard is dominated by a single nearby fault.)

C: It should be clearly understood that these aftershock hazard rate are what is
referred to as ECR’s or “effective constant rates” (Yeo and Cornell, 2004). They
are much smaller than the immediate post-mainshock aftershock hazard rates.
For example, as discussed above after a magnitude 7 the “effective MAF” is
about equal to the pre-mainshock MAF for any Sa level and any site. However, in
the first year after the mainshock the actual total probability of an aftershock
exceeding that Sa level is about 30 times the annual probability that a mainshock
will do the same. In the following year the aftershock probability will have

23
The median capacity may be higher than this; the SAC/FEMA study of steel moment-resisting frame
buildings (FEMA 350) found that IBC implied high confidence failure probabilities of 2% in 50 years or
less, implying that the median capacity has an MAF of considerably less than 2% in 50 years.

30
 decayed significantly while the mainshock probability will be unchanged24. The
process of translating this transient aftershock rate into an equivalent constant
rate (for the purposes of comparing the ECR with conventional tolerable constant
incident rates) is approximately equal to simply multiplying the total expected
number of aftershock events (e.g., collapses or exccedances of the median Sa
capacity) by a “social discount factor” which may have a value of about 3.5%.
Hence the factor 30 (which equals about 1/0.035). The user is cautioned that this
ECR approach is not a practice that has been peer-reviewed, much less widely
used. There is in fact little previous decision theoretical literature or practical
guidance for dealing with such transient hazards in any public safety policy
arena.

STEP 5: Ground motion level associated with a structural limit state

The proposed tagging approach (STEP 4) coupled with the IDA curve for the intact
structure (STEP 3) leads to the identification of the median spectral capacity values
associated with the onset of post-earthquake tagging status, denoted ŠaLS for limit state LS
(e.g., LS equal to OD, onset of damage; Y, yellow; R, red; or Coll, collapse). Identifying
the median spectral capacity value for the green tag state is not necessary for reasons that
will become clear in STEP 6. The process is shown in Figure 9.

From Figure 8 it follows that for this illustrative example the damage state DS1
correspond to incipient OD, DS2 is in Green, DS3 is in Yellow, DS4 and DS5 are in Red,
and DS6 refers to incipient Collapse. Figure 9 shows how this information can be used to
determine the median spectral capacity value, ŠaLS, for any limit state, LS. The value of
ŠaLS can be found immediately for OD and C whose onset correspond to one of the
damage states (here, DS1 and DS6, respectively) by reading them off the IDA for the
intact structure in Figure 9 at the drift value of ∆OD and ∆C, respectively. The values of
ŠaY and ŠaR, however, are not readily available because the damage states DSi (and,
therefore, ∆i) from the NSP curve in general do not correspond to the inception of the Y
and R limit states. The onset of R, for example, occurs for a damage state in between DS3
and DS4 (Figure 8). In this case either the procedure is repeated for one or more
intermediate DS’s until the computed spectral acceleration capacity is, for all practical
purposes, reasonably close to the target threshold, or, alternatively but less accurately, an
interpolation scheme is used instead25. Finally, the value of ŠaG is zero; any level of
ground motion will generate a G tag or worse.

As mentioned earlier, this IDA-based procedure has identified the median spectral
acceleration value corresponding to the onset of a given structural limit state. The ground
motion intensity (i.e., spectral acceleration) at which the limit state (and associated

24
These discussions assume that the mainshocks are Poissonian, i.e., “memoryless” and hence future
mainshock event probabilities are not affected by the occurrence of a mainshock.
25
For any given P0 the loss in capacity that corresponds to the target boundary between two limit states is
known from Figure 7. Therefore, the interpolation procedure will provide the target drift that corresponds
to the desired drop in capacity. This drift is that associated with the inception of the limit state that was
sought.

31
tagging) will occur cannot, in fact, be predicted perfectly. The value just identified is a
“best guess”, and it is assumed therefore that there is a 50-50 chance that the
limit/tagging state (or worse) will be observed if this ground motion occurs at the site.
There is a smaller chance at lower ground motion levels and a larger chance at higher
levels.
3.00
Intact IDA curve
^SaC
2.50
^R
Sa
Sa(T1) (g)

2.00
^Y
Sa
1.50
DS3 DS4
1.00 DS2

DS5
^ OD 0.50 DS1 DS6=C
Sa SPO curve

0.00
Onset of Damage

Onset of Yellow

Onset of Red

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0

Collapse
Roof drift (%)

Figure 9: Median spectral acceleration capacity associated with all the structural limit states except the
green tag state.

These chances are first quantified by estimating the dispersion, β, which is a combined
measure of two basic kinds of uncertainty: aleatory uncertainty (or randomness) and
epistemic uncertainty. The former kind manifests itself, for example, in the variability in
the dynamic displacements produced by different ground motion records (even though
they may have the same intensity value). Aleatory uncertainty is intrinsic in the random,
unpredictable nature of earthquakes and cannot be reduced. The epistemic kind of
uncertainty stems, for example, from both the limited accuracy of the selected response
analysis approach and the imperfect knowledge of parameter values of the adopted
mathematical model of the structure. Within the limits associated with current scientific
knowledge, this second type of uncertainty can be reduced, at some expense, with more
detailed investigation of the structure, more refined models, more testing of in-place
material properties, etc.

Aleatory uncertainty (βR)

Based on the results of many nonlinear dynamic analyses, it is known (e.g., Miranda;
2000) that the aleatory portion of the dispersion, βR, depends on the initial period of

32
vibration of the structure. On average, the values of βR follow the trend shown in Figure
10, namely short-period (high-frequency) structures show more record-to-record
variability in their displacements (or correspondingly, in the spectral acceleration at
which a given displacement or damage state is first observed). Further the response
dispersion is larger for larger degrees of nonlinearity, especially when the ground motion
is near the collapse capacity. This difference is captured in the three curves presented in
Figure 10, one corresponding to collapse, one to the Red- and Yellow-tag conditions, and
one to the less severe Onset-of-Damage limit state.

The curves in Figure 10 are based on average results of nonlinear dynamic analyses of
many nonlinear oscillators with different NSP backbone curves and to somewhat
arbitrary definitions of limit states based on average structural deformation levels.
Hence, they are suitable as βR default values. When applying the procedure in these
guidelines, however, the provided SPO2IDA spreadsheet will provide structure-specific
βR, values for each limit state that should be used in place of those shown in Figure 10.

βR

Period, T (sec)
Figure 10: Recommended (default) values for βR.

Epistemic uncertainty (βU)

The epistemic part of the uncertainty, βU, reflects the professional confidence that the
selected model and the analysis procedure will predict accurate results. For example, the
values for βU will be larger (i.e., one will have lessor confidence) for complex, older
structures (e.g., mill-type buildings) modeled using a simplified model and untested
material properties than for a clean, modern steel moment-resisting frame whose

33
properties have been well determined and whose model has been developed with extreme
care to details. Similarly the confidence in the results increases with the level of effort in
the structural modeling (e.g., in a SMRF a center-line model yields less reliable results
than a model with panel zones).

A review of the PG&E building stock (Table 1) justifies dividing the inventory into four
categories of buildings for the purpose of estimating βU values:

1. Tilt-up or concrete block buildings (retrofitted and un-retrofitted);

2. Mill-type buildings;
3. Prefabricated metal buildings
4. All other buildings

A detailed description of the type of buildings included in each category is provided in

Appendix B.

Tables 2a through 2e provide two different sets of βU values26 to be applied to all

buildings in the four categories. Each set is referred to here as either Baseline or
Improved. The values for the Baseline case should always be used unless:

• the building is relatively simple (e.g., no structural irregularities), or

• the development of NSP curves accounts for specific characteristics of the
building, such as structural irregularities or the effects of elements not typically
considered part of the seismic-force-resisting system.

Appendix B gives conditions and requirements to change the evaluation uncertainty from
Baseline to Improved for each building category. The conditions and requirements are
intended to identify the characteristics that most significantly affect the uncertainty in a
seismic evaluation for a given building type. The terms Baseline and Improved reflect a
relative measure of uncertainty specific to the building category. They are not intended
to have an identical meaning when applied to different building categories. The terms do
not indicate an absolute measure of epistemic uncertainty.

Note that Table 2e is directly applicable to steel moment-resisting frames (SMRFs),

which are included in the “all other buildings” category. For all the other types of
buildings included in this category, the epistemic uncertainty is presumably higher. On a
case-by-case basis, the engineer should exercise judgement to apply βU values that are
appropriate to the case under consideration. The values provided for other buildings in
Tables 2a to 2d can be used as guidance, with βU values in Table 2c (mill-type buildings)
as an upper bound.

26
The values of βU were obtained by interviewing expert practicing engineers. The experts elicited in this
study were Mr. William Holmes and Dr. Joe Maffei of Rutherford & Chekene, Oakland, CA, and Dr.
Maryann Phipps of Estructure, El Cerrito, CA. These values were also compared with those provided by
several Probabilistic Risk Assessment studies and by Dr. Robert Kennedy. A brief description of the βU
elicitation process can be found in Appendix B. Some of the values are quite large compared to others in
the literature and may need revision with time.

34
 Limit State Evaluation Uncertainty
Baseline Improved
Onset of Damage 0.6 0.35
Yellow Tag 0.6 0.4
Red Tag 0.6 0.4
Collapse 0.5 0.4

Table 2a: Recommended (default) βU values for tilt-up or concrete-block buildings (retrofitted).

Limit State Evaluation Uncertainty

Baseline Improved
Onset of Damage 0.7 0.4
Yellow Tag 0.85 0.65
Red Tag 0.85 0.65
Collapse 0.8 0.5

Table 2b: Recommended (default) βU values for tilt-up or concrete-block buildings (unretrofitted).

Limit State Evaluation Uncertainty

Baseline Improved
Onset of Damage 0.7 0.4
Yellow Tag 0.8 0.6
Red Tag 0.9 0.6
Collapse 1.0 0.6

Table 2c: Recommended (default) βU values for mill-type buildings.

Limit State Evaluation Uncertainty

Baseline Improved
Onset of Damage 0.7 0.35
Yellow Tag 0.8 0.5
Red Tag 0.8 0.5
Collapse 0.9 0.4

Table 2d: Recommended (default) βU values for prefabricated metal buildings.

Limit State Evaluation Uncertainty

Baseline Improved
Onset of Damage 0.3 0.25
Yellow Tag 0.6 0.5
Red Tag 0.6 0.5
Collapse 0.5 0.4

Table 2e: Recommended (default) βU values for SMRF buildings within the “all other buildings”
category.

35
Total uncertainty (β)

Finally the value of net dispersion, β, to be used in the determination of the fragility
curve (STEP 6) is the SRSS value:

β = β R2 + β U2
Given the values of βR and βU provided, the total dispersion ranges from
0.2 2 + 0.252 = 0.32 when assessing the onset of damage of very simple, moderate-
period SMRF structures that are carefully modeled and analyzed, to 1.352 + 1.0 2 = 1.68
when estimating the collapse of older, fairly stiff complex mill-type buildings modeled
and analyzed with limited effort.

Summary of STEP 5: Find the median and dispersion values for the ground motion
levels corresponding to the onset of the OD, Y, R, and Coll limit states.

Output of STEP 5: One pair of values, ŠaLS (median) and β (dispersion), for each of the
four limit states corresponding to the onset of OD, Y, R, and Coll limit states.

C: In the literature several studies have dealt with the aleatory uncertainty in the median
(nonlinear) response given a level of ground motion (or, alternatively, in the median
ground motion necessary to induce a specified structural deformation level). Most of
these studies considered either SDOF nonlinear oscillators or MDOF structures such as
SMRFs or steel jacket-type offshore platforms. To our knowledge, no studies have
investigated this issue for less “clean” structures, such as tilt-ups or steel frame
buildings with concrete infill walls. The SAC project is one good example (e.g., FEMA
355F) where “hard” numbers of βR based on multiple MDOF nonlinear dynamic
analyses are available for several low-rise, mid-rise, and high-rise SMRF buildings.
More recently uncertainty values have been obtained for a wide range of structural frame
models by researchers such as Prof. H. Krawinkler and students (e.g., Medina, 2002) and
Ibarra, 2003). Regarding βR values of nonlinear SDOF oscillators, the most exhaustive
effort in terms of breadth of structural periods and of hysteretic backbone and cyclic
behavior analyzed is the work that forms the basis of the SPO2IDA spreadsheet
(Vamvatsikos and Cornell, 2001b) provided here. Such values are, again, consistent with
the assumption adopted both here and by FEMA 273 and ATC-40 that the nonlinear
MDOF response of fairly stiff structures can be can be adequately predicted by the
response of an appropriate nonlinear SDOF oscillator.

C: Very limited research has been devoted to date to a systematic estimation of the
epistemic uncertainty in the structural nonlinear deformation caused by a given ground
motion or, dually, in the ground motion necessary to induce a given structural
deformation. Although limited to SMRF structures, the SAC project (e.g., FEMA 351 and
FEMA 355F) is the most widely available reference for gathering information on βU
values. In the SAC reference βU is intended to represent mainly the epistemic uncertainty
in the estimation of demands and capacities. Although equations are provided to
compute βU values (e.g., see Appendix A in FEMA 351, Yun et al., 2000) from the

36
uncertainty in each separate component, the values of βU for SMRF structures in Table
A-1 of FEMA 351 range from 0.15 to 0.35 for the Immediate Occupancy Performance
Level and from 0.25 to 0.60 for the Collapse Prevention Performance Level.

C: Except for the SAC project, the only available studies dealing with quantification of
uncertainty can be found in the so-called “gray” literature pertinent to seismic fragility
of structures common either in nuclear power plant Probabilistic Risk Assessment (PRA)
and Seismic Margin Analysis (SMA) practice (e.g., containment structures and reinforced
concrete shear-wall buildings), or in Department of Energy (DOE) hazardous facilities
(e.g., waste storage tanks). Some examples are Kennedy et al., 1980; Kennedy and
Ravindra, 1984; Kennedy et al. (1989); Bandyopadhyay et al. (1993); Reed and Kennedy
(1994); Duffley et al. (1994); Klamerus and Cherry (2001); Kennedy (2001). We also
had access to a PG&E document on the seismic evaluation of the Diablo Canyon Power
Plant (PG&E, 2001a).

C: Regarding such nuclear PRA/SMA and DOE studies some comments are in order:

• The buildings of interest to the nuclear industry and to DOE may not
necessarily be very similar to those in the PG&E building stock. Hence, the
βU values provided in those studies are to be considered as illustrative but not
necessarily strictly applicable to the PG&E buildings.
• Given the high-level consequences in case of structural failures of those
facilities (e.g., release of radionuclides), often in such studies the limit state of
interest is more related to crack-tightness in the concrete than to structural
collapse. For example, the limit state considered for the Diablo Canyon
turbine building (PG&E, 2001a) was defined as “onset of severe structural
distress and significant structural degradation”. Consequently, the βU values
provided there refer to levels of structural deformation that this current
document would define as onset of damage (or somewhat beyond that). No
information is available for more severe deformation levels associated here
with Red-tag or Collapse limit states.
• In those studies the researchers use results from engineering analyses (e.g.,
Klamerus and Cherry, 2001) for obtaining βU values but engineering
judgement often plays an important role in the definition of those numbers.
• The values of βU in different studies were not consistently intended to cover
the same sources of epistemic uncertainty or were not found using the same
techniques. Examples of sources of epistemic uncertainty considered there
are “uncertainty in the dynamic modeling of the structure, lack of
understanding of material capacity, and uncertainty due to the use of
engineering judgment to supplement an inadequate amount of hard statistical
data” (Kennedy, 2001).
C: Despite these shortcomings, the βU values provided by the studies mentioned above
consistently range between 0.3 and 0.4. Such values are somewhat consistent with the βU
values for Improved evaluation provided in Tables 2a to 2e.

37
C: Given the importance of the structure under consideration, in the PRA/SMA and DOE
studies the amount of information about material strength and ductility capacities is often
far superior to that for other civil structures. This factor has to be kept into
consideration when setting the βU values to be used in these guidelines.

C: In reality an additional source of uncertainty due to the “mis-tagging” of damaged

buildings by engineers after inspection may have a potentially large impact on this
procedure. The likelihood of “mis-tagging” a damaged building is expected to be lower
if the engineer can have access to the results of studies based on these guidelines (which
in turn are based on residual capacity associated with damaged states that can be
observed during the inspection), rather than if he/she uses only engineering judgement
based on experience. Mis-tagging in the conservative direction may be an engineer’s
tendency without the structure-specific guidance that analyses such as those proposed
here provide. Mis-tagging in the unconservative direction may be more likely in
buildings such as industrial steel frames or tilt-ups where the structural members are
exposed to fast, easy inspection rather than in concrete frames where more intrusive and
costly actions needed to expose members and joints are seldom taken. As per the
advisory committee decision, the uncertainty due to mis-tagging is not accounted for in
these guidelines.

STEP 6: Computation of the fragility curves

The fragility curve (Figure 11 to come) for a given structural limit state LS (LS equal to
onset of damage, green, yellow, red-, or collapse state) provides the annual probability
that the intact building will end up in the specified limit state (or worse) given the
occurrence at the site of an earthquake ground motion of intensity Sa. The fragility curve,
for the yellow-tag state, for example, is denoted as FY(Sa).
Based on the common lognormal assumption, the curve’s estimation for the generic
structural limit state LS requires two parameter values, a median ŠaLS value and a measure
of dispersion, β. The former is the central value of the curve that correspond to an
exceedance probability of 50%, the latter controls its slope (the larger the β value, the
flatter the curve). Values of ŠaLS and β are provided in STEP 5. The former parameter is
referred to as the median spectral capacity value of that limit state.
The fragility curve, FLS(Sa), for the generic structural limit state LS is determined by
plotting the values of probability p = {0.05, 0.25, 0.5, 0.75 and 0.95} versus the
corresponding values, Sa:
(
S a = S aLS e xβ

for the values of x equal to {–1.65, -0.67, 0.0, 0.67 and 1.65}, respectively.27

27
Additional values of p and x can be found in any table of the Gaussian distribution function (e.g.,
Benjamin and Cornell, 1970).

38
 1

0.75
FLS(Sa)
0.5

0.25

0
0 ^OD 1
Sa ^Y 2
Sa Sa ^
^ SaR C3 4 5 6

Sa(T1) (g)
Figure 11: Fragility curves for onset of damage, green, yellow, and red tags, and for collapse of the
building.

We assume that this illustrative example consists of a prefabricated metal building with
fundamental period of one second analyzed according to the specifications of the
Baseline evaluation (Appendix B). We used the ŠaLS from Figure 9, the βR values from
Figure 10 for a structure of 1 sec fundamental period (i.e., 0.23 for OD, 0.25 for Y, 0.28
for R, and 0.40 for Coll), and the βU values from Table 2d, Baseline case (i.e., 0.7 for
OD, 0.8 for Y and R, and 0.9 for Coll). According to the SRSS operation defined above,
the resulting β values are, therefore, 0.74 for OD, 0.84 for Y, 0.85 for R, and 0.98 for
Coll. Figure 11 shows the resulting fragility curves for all the limit states.

Note that in Figure 11 the fragility curve for the green-tag state is equal to unity for all
values of ground motions. This is because, by definition, the fragility curve provides the
likelihood that the building will be in the given limit state or worse if a ground motion
with specified Sa were to occur at the site. Because there are no structural limit states less
severe than green, it follows that the building will have a green tag or worse with
certainty for any level of ground motion. Note that fragility curve for onset of damage
(OD) is the steepest of the four because the value of β is the smallest. The opposite is
true for the fragility curve corresponding to the collapse state.

Summary of STEP 6: Compute and plot the fragility curves for the four limit states OD,
Y, R, and Coll.

Output of STEP 6: fragility curves for the four limit states OD, Y, R, and Coll.

39
C: In some applications it may be convenient to keep the epistemic uncertainty,
βU, separated from the aleatory uncertainty, βR when computing fragility curves.
If this is done, a family of fragility curves rather than one (mean) fragility curve is
associated with any structural limit state. The central point of each fragility
(
curve in this family can be computed by applying again S aS, y = S aLS e yβU . For
example, the central value, S aS, y , of the median (50th), the16th, and the 84th
percentile fragility curves can be found by replacing y with the values of 0, -1,
and +1 in the equation above. Each fragility curve in the family can be computed
using S a = S aS, y e yβU where now, unlike before, β is replaced by βU, and the slope is
simply given by βR. Figure 12 shows the mean fragility curve (βU and βR
combined) along with the corresponding 16th, 50th, and 84th percentile fragility
curves (βU and βR separated) obtained for the yellow tag condition in the
illustrative example. The mean fragility curve in Figure 12 coincides with the
fragility curve for the yellow-tag condition shown in Figure 11.

C: The interpretation of different fragility curves for the same limit state is
straightforward. For example, the engineer is 84% confident that the true
fragility curve for the yellow tag limit state does not exceed the 84th-percentile
fragility curve shown in the figure. More precisely, this confidence applies at any
one Sa value, and then the locus of such points is the 84th-percentile curve. To be
above at one point does not necessarily insure that the entire curve is above.
Figure 12 also shows the fragility curve one obtains from combining the two β’s;
this produces a mean estimate of FLS(Sa). This latter curve coincides with the
fragility curve for the yellow-tag state in Figure 11.

C: The IDA for the intact structure is used in this procedure to assess the median
ground motion, ŠaLS, at which a certain tagging state LS (or worse) will be
observed. As observed previously in the text, this assignment implies that the
engineer should avoid using either a pessimistic – conservative – or optimistic
assessments and judgments in STEPS 1 through 5.

C: The fragility curve, FLS(Sa), (or the family of fragility curves obtained if βR and
βU are dealt with separately), in subsequent system reliability studies, may be
multiplied by the likelihood of occurrence of ground motion level Sa(T1) at the
fundamental period of vibration of the intact structure and then integrated over
all values of Sa(T1). This operation can be described in discrete form, for
example, for the Yellow limit state by the following equation:

N N
P[ LS ≥ Y ] ≈ ∑ P[ LS ≥ Y | S a (T1 ) = ai ]P[(T1 ) = ai ] = ∑ FY (ai )P[ S a (T1 ) = ai ]
i =1 i =1

The first term in the right hand side can be read off the fragility curve for the
Yellow tag state for Sa(T1)=ai. The term P[ S a (T1 ) = ai ] , which should be
interpreted as the likelihood that Sa(T1) is in the neighborhood of ai, can be found

40
 by numerically differentiating the hazard curve for all the meaningful values of ai.
The result will be the annual probability of experiencing this tagging state or
worse in the structure. Note that if the epistemic uncertainty is explicitly
accounted for in the probabilistic seismic hazard analysis of the site, multiple
hazard curves for Sa(T1) will be available besides the mean hazard curve. In this
case the integration over all values of Sa(T1) will be repeated as many times as the
number of hazard curves and the results of each integration will be weighted by
the weight assigned to each hazard curve.

16th median (50th)

1

0.75
FLS(Sa)

0.5 mean
(combined)
0.25
84th
0
0 1 2 3 4 5 6
Sa(T1) (g)
Figure 12: Fragility curves obtained both by keeping βR and βU separated (median (50th), 16th, and 84th) and
by combining them (mean or combined estimate).

C: Note that the fragility curve of a particular state is, strictly, the probability that
that state or worse, will be the outcome of experiencing a specified level of
ground motion, as characterized by spectral acceleration, Sa. As implied by some
of the terminology used above the fragility curve can also be interpreted as the
(cumulative) probability distribution function, FLS(Sa), of a limit state “capacity”
measured in ground motion intensity terms. In this interpretation the LS state will
be observed if the ground motion input level (demand) exceeds this (random)
capacity (measured in ground motion terms). To estimate the probability
distribution of this random variable one can, in principle, run for a suite of
ground motion records a series of incremental dynamic analyses (IDA’s),
observing for each record the first Sa level at which the limit state is observed.
These represent observations of this “capacity”. One then can refer to the
median, ŠaLS, and dispersion28, β, of this presumably lognormally distributed
random variable.

28
The dispersion is approximately equal to the coefficient of variation, but strictly it is defined here as the
standard deviation of the natural log.

41
Application of the guidelines

Case Study No. 1: 3-story Steel Moment-Resisting Frame (SMRF) Building

The SMRF building analyzed here is the San Francisco Service Center Operations
Building located at 2180 Harrison Street, San Francisco. The building is a 3-story
structure built approximately in 1989 with a floor area of 62,600 square feet and outside
dimensions of 98 feet by 217 feet.

The building is symmetric and has two sets of two EW frames and two sets of two NS
frames. The application of these guidelines and the subsequent validation via nonlinear
dynamic analyses of the IDA curve computed by the SPO2IDA spreadsheet was
performed on only one of the E-W frames (Figure 13).

W 24x162 Bo x 18x18x 3/ 4
W 24x76

14’
W33x118

14’
W 33x118

15’6”

32’6’’ 28’ 32’6’’

Figure 13: Schematic model of the frame

Structure Model:

The E-W moment-resisting frame of the structure is modeled with RAM-Perform V1.04,
a commercially available structural analysis program. Beam-column connections are
modeled as fracturing connections (described in next section), and the loading used is
summarized in Table 3 below.

1st Floor 2nd Floor Roof

Dead Load (psf) 82 82 48 (53 for mass)
Perimeter Walls (lb/ft) 106 105 175
Live Load (psf) 50-150 50-150 40

Table 3: Values of dead loads and live loads used in the analysis

42
Connection Model:

Beam-column connections are modeled as point-hinge fracturing connections.

Connection (flange) fracture is assumed to occur at 0.01 plastic hinge rotation, when the
moment drops to 30% of its initial value. The connections are assumed to completely fail
when the shear tab ruptures. This is expected to occur at 0.07 plastic hinge rotation
resulting in the resisting moment and the vertical shear capacity dropping to effectively
zero, as indicated by the dashed vertical line. These latter features, however, were not
explicitly included in the model; due to software limitations the model continues
horizontally indefinitely. This does not affect the conclusions, however, because, as
discussed below, the occurrence of such an event will be considered as “collapse”. A
diagram of the connection moment-curvature curve is provided in Figure 14. Note: panel
zone effects are not considered, and the yield strengths for the beams and columns are
36ksi and 50ksi, respectively.

Typical Beam-Column Connection

25000
Moment ( Kips-in )

20000

15000

10000

5000

0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10

Plastic Hinge Rotation

Figure 14: Beam-Column moment-rotation relationship.

Application of the guidelines:

STEP 1: NSP of the intact structure and identification of damage states

The bold line in Figure 15 shows the results of the NSP performed on the intact structure.
The following damage states are identified and specified on this curve:

1. DS1, or Onset of Damage. This state is defined where nonlinear behavior starts
on the curve. The corresponding roof drift is ∆1=0.9%

2. DS2 is defined where the first considerable drop is noticed on the curve, related to
the fracture of the exterior beam-column connections of the first floor. The
corresponding roof drift is ∆2=1.65%

43
 Intact SPO and Damage States
1000

900

800

DS1 DS2

Base Shear ( Kips )

700

600
DS4
500 DS3

400

300

200 Intact SPO

Quadilinear approximation
100

0
0.00 0.01 0.02 0.03 0.04 0.05

Roof Drift

Figure 15: NSP curves for the intact structure. The NSP curve for DS1 is identical to that for the intact
structure. [See Figure 1]

3. DS3 represents the second considerable drop on the curve, and is related to the
fracture of interior connections. This is observed at a roof drift of ∆3=2.4%

4. DS4 is the drift where the first shear-tab failure is observed within the structure,
i.e., a local plastic rotation of 0.07. The roof drift causing this damage state is
∆4=4.8%. For reasons discussed above the NSP results are not valid beyond this
point; we indicate this by the dashed vertical line at 4.8%.

Notice that these choices of damage states are not unique. Any other point relating to a
change in any structure properties can be chosen as a damage state. Later, these damage
states will be associated to different structural limit states.

Some of the values obtained from this Step 1 analysis are the initial stiffness of the intact
structure and modal properties. The initial stiffness (K1) is 93,300 Kips/rad, and the
period of the intact structure is T1=0.73 seconds.

STEP 2: NSP curves for the damaged structure

To obtain the NSP curves for each damage state DSi, the intact structure is first "pushed"
to that damage state (i.e., until the roof drifts specified in Step 1 are reached), then it is
unloaded to zero base shear. After unloading, the structure, which in general experiences
a certain level of static offset drift, is loaded again until failure, to be defined as the first
local shear tab failure. The NSP curve obtained is the NSP curve for the damage state
DSi. The NSP curves for the damage states DS2 and DS3 are shown in Figures 16 and 17.
Note that the model within the software used for the NSP analyses is such that the static
pushover curves for DS2 and DS3 return to, and follow the original NSP curve for the

44
intact structure. (This, however, may not always be the case.) For this reason the first
shear tab failure occurs again at a roof drift of 4.8%, beyond which the model is not valid,
as implied by the dashed vertical lines.

From these curves the elastic stiffness of the damaged structure are estimated as KDS2=
68500 Kips/rad, and KDS3= 54500 Kips/rad, for damage states DS2 and DS3, respectively.
These stiffness values and the fundamental period of the intact structure are used to find
the fundamental period of vibration of the structure in these two damage states. Such
periods of vibration will be used in Step 3.29
NSP Curves
1000
900
Base Shear ( Kips )

800
700
600
500
400
300 Intact SPO
200 DS2 SPO
100 Quadrilinear approx.
0
0.00 0.01 0.02 0.03 0.04 0.05
Roof Drift

Figure 16: NSP curve for DS2 [See Figure 2]

NSP Curves
1000
900
Base Shear ( Kips )

800
700
600
500
400
300
Intact
200 DS3 SPO
100 Quadrilinear Approx.
0
0.00 0.01 0.02 0.03 0.04 0.05
Roof Drift

Figure 17: NSP curve for DS3 [See Figure 2]

29
Note that the period of vibration of the structure in the two damage states is equal to that of the intact
structure if, instead, parallel-to-elastic unloading/re-loading is assumed for the NSP curves of the damaged
structure. In this case study, these "assumed" NSP curves are similar to those shown in Figures 16 and 17,
and in STEP 3 we will demonstrate that the resulting estimates of the residual lateral capacity are almost
identical.

45
STEP 3: From SPO to IDA

In this step we obtain the dynamic IDA estimates from the static NSP curves using the
SPO2IDA software. To match the input requirements for this software we must conduct
several simple steps of straight-line approximation of the NSP’s and normalization of the
axes. The normalized IDA that is output must then be “de-normalized” for proper
interpretation. We go through these steps in detail in what follows.

In order to be used in the SPO2IDA spreadsheet, the NSP curves need be approximated
by no more than four straight lines. The piecewise linear fit of the NSP curves is also
shown in Figures 15 to 17 and in Table 4.

The damaged structure NSP’s display a large initial residual static offset that is an
artificial product of the static way in which the structure has been loaded and unloaded to
reflect the mainshock response. For the same maximum displacement, on average a
comparatively much smaller residual offset would be expected in a proper dynamic
analysis. We have chosen therefore in these guidelines to remove that offset by shifting
the damaged NSP’s back to zero offset, as shown in Figure 18. Note that this shift does
not reduce the roof drift at which collapse is predicted. To accomplish this shift, the
value of the offset (given by the drift for point 1 in Table 4) is subtracted from all drifts in
the other columns with the exception of the collapse point, as shown in Table 5. Later, in
Step 3, the expected (or measured) dynamic residual offset is taken into account.

NSP Curves
900

800

700
Base Shear ( Kips )

600

500

400

300

200
DS2 NSP
100 Without offset

4.8
0
0.00 0.57 1.00 2.00 3.00 4.00 5.00
Roof Drift

Figure 18: DS2 NSP with and without the offset. Note that local collapse state

46
 Point 1 Point 2 Point 3 Point 4 Point 5
Drift BS Drift BS Drift BS Drift BS Drift BS
(%) (K) (%) (K) (%) (K) (%) (K) (%) (K)
Intact 0 0 0.9 840 1.65 890 2.8 630 4.8 630
DS2 0.57 0 1.65 740 2.10 800 2.8 630 4.8 630
DS3 1.30 0 2.45 630 4.8 630

Table 4: Quadrilinear approximation of the un-normalized NSP curves with static residual offsets (Figures
15-17).

Point 1 Point 2 Point 3 Point 4 Point 5

Drift BS Drift BS Drift BS Drift BS Drift BS

(%) (K) (%) (K) (%) (K) (%) (K) (%) (K)

Intact 0 0 0.9 840 1.65 890 2.8 630 4.8 630

DS2 0 0 1.08 740 1.53 800 2.23 630 4.8 630
DS3 0 0 1.15 630 4.8 630

Table 5: Quadrilinear approximation of the un-normalized NSP curves with the static residual offset
removed. This offset is removed from the NSP curves of the damage states DS2 and DS3 by subtracting
from the drift values in Table 4, excluding the collapse points (i.e., last value in each row), the values in the
first column of each pertinent row.

Point 1 Point 2 Point 3 Point 4 Point 5

µ R µ R µ R µ R µ R

Intact 0 0 1 1 1.83 1.06 3.11 .75 5.33 .75

DS2 0 0 1 1 1.42 1.08 2.07 .85 4.44 .85
DS3 0 0 1 1 4.17 1

Table 6: Quadrilinear approximation of the normalized NSP curves (without static residual offset). The
normalization is obtained by dividing all the numbers under Points 3-5 in Table 5 by the corresponding
numbers under Point 2 (i.e., the yielding point).

Furthermore, the NSP curves in Figures 15 to 18 are in terms of base shear and roof drift.
As explained in the text, the SPO2IDA spreadsheet requires normalized NSP curves
expressed in terms of global ductility ratio, µ, and the ratio, R, of the base shear, BS, to
the base shear, BSyi, at incipient yielding. Hence, the following normalization step is
performed on NSP results from Step 2. Each NSP curve is considered separately, and the
point corresponding to first yielding is selected on each of them (DS1, DS2, and DS3
marks in Figure 15). The ordinates of these points are called BSyi and ∆yi. These points
are associated with column 2 in Table 5. Using Ri=BSi/BSyi, and µi = ∆i/∆yi, the input for
SPO2IDA is as shown in Table 6.

47
Using the NSP curves in Table 6 as an input to the SPO2IDA spreadsheet produces the
dynamic µ vs. R relationships (i.e., the IDA curves) for the intact structure and the
damaged state structures (Table 7). The normalized NSP and IDA curve for the intact
structure are shown in Figure 19. The SPO2IDA output is in terms of normalized drift (µ
= ∆/∆y) and normalized spectral acceleration, R = Sa/Say, where Say, is the yield
acceleration to be discussed below.

Point 1 Point 2 Point 3 Point 4 Point 5

µ R µ R µ R µ R µ R
Intact 0 0 1 1 2 2.1 4 3.3 5.33 3.9
DS2 0 0 1 1 2 2.05 4 3.2 4.44 3.6
DS3 0 0 1 1 2 2.15 4 4.0 4.17 4.2

Table7: Normalized (median) IDA curves obtained via the SPO2IDA spreadsheet.

As stated, the IDA curves in Table 7 are in terms of µ and R, where R is now to be
interpreted as Sa/Say. A “de-normalization” process should be performed to produce
more meaningful IDA curves directly in terms of roof drift vs. Sa. Tables 8 and Tables 9a
and 9b demonstrate this procedure. Table 8 shows the results of de-normalizing the
abscissa or displacement axis, by multiplying the µ’s by the yield drifts from Point 2 of
Table 5. Tables 9a and 9b show the ordinate transformation. For the intact structure, R is
multiplied by Say, which is estimated by BSy (840 K) divided by the product of the mass
of the structure (1380 K) and the first-mode participation factor (0.88). The result is
Table 9a.
For the damaged structures, the R columns in Table 8 are also multiplied by their Say
values, i.e., by BSy levels (740 and 630 for DS2 and DS3, respectively) divided by the
same product of the mass and participation factor. The results are shown in Table 9b.
(Note that these Say values have the same relative values as the BSy values used to
normalize the NSP abscissa.) The resulting spectral accelerations are associated with the
periods of the damaged structures (TDS2 and TDS3), which are 0.84 and 0.93 secs
respectively. These were obtained by multiplying the intact structure’s period by the
square root of the inverse ratio of the initial stiffnesses discussed in Step 2.

To compare and plot all the IDA curves on the same figure it is necessary to use a
common spectral acceleration, i.e., one associated with the same period for all cases,
intact and damaged. We chose to transform the spectral acceleration at fundamental
period of vibration, TDSi, of the damaged structure in damage state DSi into Sa(T1) where
T1 is the fundamental period of vibration of the intact structure. Hence, the ordinates
(spectral accelerations) of the IDA curve associated with Table 9b for damage states DS2
and DS3 should be multiplied by Sa(T1)/Sa(TDSi). This value can be estimated from
uniform hazard spectra at an appropriate level of hazard for the building site. In this
example, we used the uniform hazard spectrum (UHS) with 10% probability of
exceedance (PE) in 50 years for San Francisco provided by USGS. For these periods the

48
spectral acceleration ratios might also be estimated to close enough approximation by
assuming the spectrum of future earthquakes will have the same spectral velocity in the
period range of interest here, i.e., simply by the ratio of the periods. From the selected
UHS, Sa(0.73)/Sa(0.84)=1.12, and Sa(0.73)/Sa(0.93)=1.26. Hence, the spectral
accelerations in Table 9b are multiplied by 1.12 for DS2, and by 1.26 for DS3 (Table
10)30.

Point 1 Point 2 Point 3 Point 4 Point 5

Drift R Drift R Drift R Drift R Drift R
Intact 0 0 0.9 1 1.8 2.1 3.6 3.3 4.8 3.9
DS2 0 0 1.08 1 2.2 2.05 4.3 3.2 4.8 3.6
DS3 0 0 1.15 1 2.3 2.15 4.6 4.0 4.8 4.2

Table 8: IDA curves in roof drift (in percent) versus R terms. The µ coordinate is transformed into drift by
multiplying every µ cell in Table 7 by the yield drift corresponding to that state, i.e., the drifts in Table 5
for Point 2.

Point 1 Point 2 Point 3 Point 4 Point 5

Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g)
Intact 0 0 0.9 0.69 1.8 1.45 3.6 2.28 4.8 2.7

Table 9a: Un-normalized IDA curve for the intact structure in terms of roof drift and Sa at the fundamental
period of vibration of the intact structure.

Point 1 Point 2 Point 3 Point 4 Point 5

Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g)

DS2 0 0 1.08 0.61 2.2 1.25 4.3 1.95 4.8 2.20

DS3 0 0 1.15 0.52 2.3 1.12 4.6 2.08 4.8 2.18

Table 9b: Un-normalized IDA curves for the structure in damage states DS2 and DS3. In this table Sa is the
spectral acceleration at the fundamental periods of vibration of the structure in damage states DS2 or DS3.

Point 1 Point 2 Point 3 Point 4 Point 5

Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g) Drift Sa(g)
DS2 0 0 1.08 0.68 2.2 1.40 4.3 2.18 4.8 2.46
DS3 0 0 1.15 0.66 2.3 1.41 4.6 2.65 4.8 2.75

Table 10: Un-normalized IDA curves for structure in damage states DS2 and DS3. Sa is the spectral
acceleration at the period of the intact structure for all cases.

30
Note: the ratio of periods is 1.15 and 1.27 implying the equal spectral velocity approximation would have
been adequate in this case.

49
 Normalized Intact NSP and IDA
4.5
4
Normalized NSP
3.5
Normalized IDA
3
2.5
R

2
1.5
1
0.5
0
0.00 1.00 2.00 3.00 4.00 5.00 6.00
µ

Figure 19: Normalized NSP and IDA for the intact structure. Here R is BS/Bsy for the NSP, and Sa/Say
for the IDA [See Figure 3]

The IDA curves for the intact structure and for structure in different damage states
expressed all in terms of the spectral acceleration at the fundamental period of the intact
structure, are shown in Figure 20(a). Note that despite the somewhat more severe
damage in DS3, the IDA of DS3 crosses the IDA curves for both the intact structure and
DS2. Like the NSP curves (e.g., Figure 18) they are derived from, however, these IDA
curves assume no residual offset after the mainshock. Based on dynamic analyses of a
similar structure carried out during Phase II of this project, the median residual offset (in
terms of roof drift) for the structure in damage state DS2 and DS3 is 0.07% and 0.30%,
respectively. (Note that these dynamic residual roof drifts are much smaller than those
implied by the NSP curves.) As explained in the guidelines, IDA curves that account for
this expected dynamic residual offset are obtained by tracing the IDA curves in Figure
20(a) to the point that has a roof drift equal to the roof drift capacity minus the residual
offset – namely, 4.8%-0.07%=4.73% and 4.8%-0.30%=4.50% for DS2 and DS3,
respectively. The resulting IDA curves are shown in Figure 20(b).

As seen in Figure 20(a), even without a residual offset the IDA curves for the structure in
damage state DS2 and DS3 yield estimates of the residual capacities (2.46g and 2.75g,
respectively) that are different than that of the intact structure (2.7g). In contrast,
dynamic analyses of the structure in damage states DS2 and DS3 carried out during Phase
2 of this project indicate that without a residual offset little or no change in the median
lateral capacity relative to the intact structure is expected. Hence, as explained in the
guidelines, the residual capacity estimates from Figure 20(b) (i.e., after accounting for the
residual offset), namely 2.45g for DS2 and 2.58g for DS3, are adjusted to reflect this
observation. This is accomplished by multiplying these estimates by the ratio of the
intact capacity to the residual capacity assuming no residual offset, namely 2.7g /2.46g
and 2.7g /2.75g for DS2 and DS3, respectively. The final estimates of the residual
capacities that will be used in STEP4 are, therefore, (Sa,cap)2=2.45g*2.7g /2.46g=2.69g
and (Sa,cap)3=2.58g*2.7g /2.75g=2.53g. As noted in the commentary of the guidelines,
these estimates are equivalent to multiplying the capacity of the intact structure (2.7g) by
the ratio of the residual capacities with and without a residual offset (2.45g/2.46g=0.996

50
for DS2 and 2.58g/2.75g=0.94 for DS3). As alluded to in Step 2, the residual capacities
estimated under the assumption of parallel-to-elastic unloading/re-loading for the
damaged NSP curves are nearly identical to those estimated here (i.e., 2.68g instead of
2.69g for DS2 and 2.56g instead of 2.53g for DS3).

Intact and DS IDA's

Intact IDA
4.0
DS2 IDA
DS3 IDA

3.0
Sa ( g )

2.0

1.0

0.0
4.8
0 1 2 3 4 5 6
Roof Drift ( % )

Figure 20(a): IDA curves for the intact structure and for the damaged structure in damage states DS2 and
DS3 before accounting for the expected dynamic residual offset. Again, the IDA curve for DS1 is identical
to that for the intact structure. The Sa's are all at the period of the intact structure.

Intact and DS IDA's

Intact IDA
4.0
DS2 IDA
DS3 IDA

3.0
Sa ( g )

2.0

1.0

0.0
4.8
0 1 2 3 4 5 6
Roof Drift ( % )

Figure 20(b): IDA curves for the intact structure and for the damaged structure in damage states DS2 and
DS3 after accounting for the expected dynamic residual offset, but before adjusting for the behavior
observed in dynamic analyses of damaged structures. Again, the IDA curve for DS1 is identical to that for
the intact structure. The Sa's are all at the period of the intact structure.

51
Step 4: Occupancy Status for Damaged Building

In this step we determine what limit or tagging states to associate with each of the
damage states. The simplest case, the onset of damage limit state, is predicted to occur at
a drift of 0.9%.

Based on Figure 20a the intact structure will reach the 4.8% drift that implies collapse at
a Sa value of 2.7g. From the results of STEP 3, the (median) Sa capacity estimates of the
damaged structures are 2.69g and 2.53g. These are the aftershock ground motion
intensities (as measured in terms of 0.73 sec period Sa) that are expected to cause (local)
collapse of the damaged structures.

To determine the tagging states associated with DS2 and DS3 we must consider the loss in
capacity and the probability (MAF) of an aftershock causing a collapse of the damaged
structure. The spectral acceleration capacity reductions for DS2 and DS3- are 0.4% and
6.3% respectively. If the MAF of exceeding the median capacity of the intact building at
the building location, P0, were, for example, equal to 5 x 10-4, then the damage state DS2
is green-tagged whereas DS3 is yellow-tagged. The damage state DS4, which is
associated with local collapse, is by default tagged red and it is used here as describing
the incipient collapse for the purpose of developing fragility curves.

Step 5: Ground motion level at incipient structural limit state

In this step we identify the mainshock ground motion intensity that is expected to bring
the intact structure to the verge of each relevant structural limit state. For the P0 equal to
5 x 10-4 assumed above, the loss in capacity computed for damage sates DS2 and DS3
(0.4% and 6.3%, respectively) locates them well within the boundaries of Green and
Yellow tags. Additional damage states should be considered in between DS2 and DS3 and
in between DS3 and DS4 until the computed capacity of the building in these additional
damage states has dropped by the quantity suggested by the tagging criteria (i.e., 2% and
20%) for the value of P0. Luco et al. (2004) (see Appendix C) have considered a damage
state in between DS3 and DS4 for this very same structure. Here, however, we estimated
the values of the median roof drift, and from them, the values of the median spectral
acceleration, ŠaY and ŠaR, corresponding to the incipient Yellow and Red tag states via
interpolation. The resulting main shock ground motions causing the onset of all limit
states, including Y and R whose values were estimated via interpolation, are shown in
Table 11. These correspond to the median values of the limit state fragility curves.

Structural Limit State: OD Y R C

Median Roof drift: 0.9% 2.09% 3.4% 4.8%
Median ground motion:
0.65g 1.6g 2.25g 2.7g

OD
Table 11: Median roof drifts and median Sa(T1) corresponding to incipient structural limit states (i.e, Ša ,
ŠaY, ŠaR, and ŠaC).

52
The values of βR and βU are taken from Figure 10 (for T=0.73 sec) and from Table 2e, for
SMRFs, assuming the baseline analysis. The total dispersion, β, is the square root of
sum of the squares of βR and βU. Values of βR, βU, and β for this example are shown in
Table 12.

Limit State
OD Y R Collapse
βU 0.3 0.6 0.6 0.5
βR 0.25 0.28 0.32 0.45
β 0.39 0.66 0.68 0.67

Table 12: Values of βU , βR , and β used to obtain the fragility curves shown in Figure 22.
3.5

3.0 2.7 Collapse

C

2.5 2.25 R Red Tag

Sa ( g )

2.0
1.6 Y Yellow Tag
1.5

1.0
0.65 OD Intact SPO
0.5

0.0
0.0 1.0 2.0 3.0 4.0 5.0

Roof Drift ( % )
Figure 21: Median value of Sa causing the structure to enter or exceed Onset of Damage, Yellow Tag, or
Collapse state. [See Figure 9]

0.75
FLS(Sa)

0.5

0.25

0
0 1 2 3 4 5 6

Sa (g)
ŠaOD =0.65 ŠaY =1.6 ŠaR =2.25 ŠaC =2.7

Figure 22: Fragility curves for the Onset of Damage state, the Yellow-tag state, and the Collapse state for
this SMRF structure. [See Figure 11]

53
Step 6: Computation of the Fragility Curves

The fragility curves for each structural limit state (Figure 22) are found using the median
spectral accelerations, ŠaLS, and the β values from STEP 5. Recall that the fragility
(
curves are found using the equation S a = S aLS e xβ , where the x values are reported in any
table of the Gaussian distribution function.

Validation:

To confirm the accuracy of the SPO2IDA tool used in these guidelines, a suite of 30
ground motions (magnitude ranging from 6.5 to 6.9) was used to perform nonlinear
dynamic analyses on the intact structure. The records are first scaled to have the same
spectral acceleration, Sa(T1), at the first-mode period of the structure (T1=0.73 sec) and
then run through the structural model. This procedures was used to obtain so-called
Sa(T1) versus maximum roof drift "stripes" at four different levels of Sa(T1). The stripes,
together with the median IDA (solid black line) estimated from the nonlinear dynamic
analyses, are drawn in Figure 23. The dashed line in the figure is the IDA produced from
the NSP curves by the SPO2IDA spreadsheet. The agreement between the two IDA
estimates is excellent. In particular, note that at a roof drift of 4.8% the median Sa(T1)
from the nonlinear dynamic analyses is nearly equal to the (local) collapse capacity of the
intact structure estimated (in STEP 3) via the nonlinear static procedure put forth in these
guidelines.

To verify the validity of estimating the dynamic spectral acceleration capacities of the
intact and damaged structures by the Guideline’s nonlinear static procedure the following
was done. Dynamic analyses such as those described above were conducted on the intact
and on the damaged structures at spectral acceleration levels near the anticipated capacity
values. The percentage of the 30 records in which local collapse (shear tab failure) is
reached is plotted versus spectral acceleration level in Figure 24. The median (local)
collapse capacity is obtained by interpolation from this curve. The dynamic capacities
are 2.75, 2.4 and 2.05g for intact, DS2, and DS3 respectively. The agreement between
these values and the Guideline’s NSP procedure’s estimates of capacities (2.7g for the
intact structure, and 2.69g for both DS2 and 2.53g for DS3) is good for the intact
structure but not satisfactory for the damage states DS2, and DS3. There is reason to
believe that the damaged state dynamic capacities have been somewhat underestimated,
because the model adopted to estimate them was based on the loaded and unloaded static
pushover (NSP) which left a very asymmetrical fracture/damage state in the structure
used for the dynamic analyses. This asymmetry has the tendency to cause larger
nonlinear dynamic drifts (than would a more realistic model of the post-mainshock
damaged structure); these larger drifts imply smaller Sa capacities.

In Phase 2 of this project (refer to Luco et al., 2004 in Appendix C for more details), the
residual capacities of a similar structure in damage states DS2 and DS3 were computed
using back-to-back (mainshock-aftershock) dynamic analyses. First the intact structure
was analyzed dynamically under the 30 earthquake records (considered as mainshocks)
scaled to produce the roof drift associated with each damage state (i.e., 1.65% for DS2

54
and 2.4% for DS3). Then these 30 structures in each damage state were each analyzed
dynamically under the same 30 earthquake records (now considered as aftershocks)
scaled to produce the roof drift associated with collapse (i.e., 4.8%). The spectral
acceleration of each scaled aftershock, which is the residual capacity of the structure in
each damage state, was normalized by the spectral acceleration capacity of the intact
structure under the same earthquake record. The median of this ratio across all 30
aftershocks by 30 damaged structures was found to be 0.98 and 0.92 for damage states
DS2 and DS3, respectively. These ratios of dynamic capacities are within a few percent
of those estimated using the nonlinear static procedure put forth in these guidelines.

IDA validation
3.5

3.0

2.5
Sa ( g )

2.0

1.5
Median IDA
1.0
Estimated IDA
0.5

0.0
0 4 8 12 16 20 24
Roof Drift ( % )

Figure 23: Validation of IDA for intact stucture. Comparison of the proposed NSP-based estimation
procedure with the median of multiple nonlinear dynamic runs.

Local Collapse
3.1

2.9
2.75
2.7
Sa ( g )

2.5
2.4

2.3
Intact
DS2
2.1 DS3
2.05
1.9
30 40 50 60 70 80
Percentage of records causing failure

Figure 24: Validation of median spectral acceleration capacities based on multiple nonlinear dynamic.

55
Case Study No. 2: Tilt-up Building
To develop and validate the proposed procedure for use with tilt-up buildings, the
response of a warehouse in Hollister CA, which has previously been studied by other
researchers (Hamburger et al., 1996; Wallace et al. 1999), is considered. The warehouse
is 300 feet by 98.5 feet in plan and 30 feet in height. The building was instrumented by
the California Strong Motion Instrumentation Program (CSMIP) and has been subjected
to ground shaking during the Morgan Hill (1984), Alum Rock (1986) and Loma Prieta
(1989) earthquakes.
Based on the damage to tilt-up structures documented by the EERI reconnaissance team
(1995) following the 1994 Northridge earthquake, our study focuses on predicting the
ground motion intensity at which the roof diaphragm separates from the tilt-up panels on
the perimeter of the building due to the failure of the roof-wall connections. Application
of the proposed procedure for this purpose is described below followed by a validation of
the results by means of a comprehensive series of IDAs performed with an ensemble of
ten ground motions.

Structure Model
A plan view of the structure model used in the analyses described herein is shown in
Figure 25. Currently, only ground motions acting parallel to the x-axis are considered.
The model, which is in most respects similar to that used by Hamburger et al. (1996),
consists of three key components: (1) the roof diaphragm, (2) the walls perpendicular to
the ground motion (i.e., parallel to the y-axis in Figure 25) and (3) the connections
between the roof diaphragm and these walls. The walls along gridlines 1 and 2 in Figure
25 are modeled as rigid supports, reflecting the fact that these elements are so stiff that
they do not amplify the input ground motions at their bases. We assume that the shear
capacity of the roof-wall connections along gridlines 1 and 3 is sufficient to transfer the
shear forces from the roof diaphragm to the walls along these gridlines.

Roof diaphragm
The horizontal truss shown in Figure 25 is used to model the roof diaphragm. This truss
model consists of linear-elastic and nonlinear bar elements. The chord elements along
gridlines B and C and the web elements spaced at 18' 9" on center parallel to the x-axis
are modeled as stiff linear-elastic elements so that all of the diaphragm deformation
occurs in the nonlinear diagonal elements shown in Figure 25. With appropriate values
assigned to the material properties of these nonlinear elements (i.e., those related to the
size and spacing of the nails and the thickness of the plywood), the experimental results
reported by Pardoen et al. (1999) can be reproduced, as shown in Figure 26. However,
for the analyses reported herein, the material properties of these diagonal elements are
selected to better represent the construction of the warehouse roof, including the presence
of roofing material.

Walls perpendicular to ground motion

The walls perpendicular to the ground motion (along gridlines A and D) are modeled
with standard linear-elastic beam elements. These elements are intended to represent the

56
30 ft high × 5.5 in thick reinforced concrete tilt-up panels at these locations. Hamburger
et al. (1996) adopted a similar model. A truer representation of these walls would account
for the fact that they are pinned at the base. However, preliminary tests with structure
models that include a better representation of the wall indicate that the simplified wall
model shown in Figure 25 is sufficiently accurate. As expected, modeling the walls with
pinned bases stiffens the structure model slightly. This stiffening effect is accounted for
in our simplified wall model by factoring the stiffness of the beam elements used to
represent the wall.

Roof-wall connections
The roof-wall connections have been modeled in several ways in past studies. Hall (1999)
and Wallace et al. (1999) represent the roof-wall connections with pairs of springs that
produce a ductile response in tension and a nearly rigid response in compression.
Hamburger et al. (1996) model the roof-wall connections with linear spring elements. A
shortcoming of the connection models used in these previous studies is their inability to
represent the brittle behavior of the roof-wall connections observed in past earthquakes.
In particular, there are many documented instances from Northridge (1994) where the
roof diaphragm separated from the wall, causing a portion of the roof to collapse that
these models cannot properly represent.
In spite of recent experiments conducted to characterize the properties of the roof-wall
connections used in tilt-up buildings (e.g., Pardoen et al., 1999), there is still little data to
support one model over another. Consequently, in this study, the roof-wall connections
are modeled with nonlinear bar elements that are capable of reproducing two types of
behavior, as shown in Figure 27. Note that in all cases the connection fractures when
elongated beyond a prescribed value. The models shown in Figure 27 are intended to
represent the behavior of roof-wall connections commonly used in practice, such as those
shown in Figure 28.

Masses
The mass that drives the seismic response of the tilt-up building under consideration is
associated with the roof diaphragm and the walls along gridlines A and D. The roof mass
is based on a uniformly distributed roof load of 12 psf and is evenly distributed to the
nodes on either side of the diaphragm along gridlines B and C (see Figure 25). The
resulting seismic mass at each of these nodes is 0.029 k·s2/in. The seismic mass
associated with each wall is evenly distributed to the nodes along gridlines A and D in
Figure 25 and is based on normal weight concrete (145 pcf) and a tributary wall height of
20 ft (two-thirds the total height of the wall). This tributary wall height is chosen to
approximate the reaction at the top of the wall due to a triangular distribution of inertial
forces over its height when it pivots about its base. The resulting seismic mass at each
node along gridlines A and D due to the walls is 0.064 k·s2/in. Note that the walls account
for approximately 70% of the total seismic mass.

Modal properties
A modal analysis of the structure model described above yields the modal properties
summarized in Table 13. The fundamental period of the model is 0.57s, which agrees

57
with the value computed by Hamburger et al. (1996) and measured by CSMIP
accelerometers located at the center of the diaphragm. Note that only the “odd” modes of
vibration have non-zero participation factors. This is because the even modes are anti-
symmetric shapes that cannot be excited by a ground motion that acts through the center
of rigidity of the model. Also note that the first four symmetric modes (1,3,5,7) account
for 98.1% of the total effective seismic mass.

Application of the Proposed Procedure

Application of the proposed procedure is demonstrated here in detail for the Hollister tilt-
up model shown in Figure 25 with ductile roof-wall connections shown in Figure 27a. It
is assumed that the building has been retrofitted and that some of the conditions listed in
Appendix B for the evaluation of tilt-up buildings have not been considered (in particular,
the evaluation of foundation overturning and the evaluation of wall in-plane behavior).
For brevity, the procedure is only presented for ground motions acting parallel to the x-
axis shown in Figure 25; however, in practice, the analysis should also be conducted for
ground motions acting parallel to the y-axis. Application of the procedure to a building
with brittle connections is similar to that described below. Consequently, the details of
the analysis of the brittle structure are not described; however, the resulting fragility
curves for this case are presented in Step 6 in order to understand the how the assumed
ductility of the roof-wall connections influence the fragility curves.

Step 1: Nonlinear static procedure (NSP) curve for the intact building
The NSP curve for the intact structure, which relates the base shear, BS, to a control
displacement, ∆C , is obtained by determining the required intensity of a prescribed load
pattern, F , for a series of monotonically increasing values of ∆C . In this example, ∆C is
the displacement of the top of the wall at the intersection of gridlines 2 and D in Figure
25 and the load pattern is that associated with the first mode shape of the intact structure,
i.e., F = λMφ1 , where M is the mass matrix, φ1 is the mode shape associated with the
first mode of vibration and λ is an intensity factor. The resulting NSP curve is plotted in
Figure 29.
Referring to Figure 29, three inelastic events associated with the following damage states
are identified.
1. Onset of damage. The onset of damage in the structure is identified in Figure 29 as
DS1, which corresponds to the wall displacement ∆C = 0.9" at which the NSP curve
first begins to depart from linearity. At wall displacements greater than ∆C = 0.9", the
roof diaphragm sustains damage. Observable structural damage associated with
damage state DS1 includes slight permanent deformation of some nails that fasten the
plywood diaphragm to the framing members and possibly some minor damage to
strain-sensitive roofing materials. All structural members are still able to support
gravity and live loads.
2. Significant roof diaphragm damage. At damage state DS2 in Figure 29, which
manifests itself as a slight kink in the NSP curve at ∆C = 7.5", the roof diaphragm

58
 shows signs of permanent deformation including yielding and/or pull-out of nails, the
enlargement of nail holes in the plywood diaphragm, the formation of gaps between
adjacent sheets of plywood and noticeable damage to the roofing material. In
addition, the roof-wall connections may show some signs of distress including
enlargement of the nail holes in the framing members and/or metal connections, and
yielding and/or pull-out of the nails. The building may also be leaning slightly.
However, in this damage state, the roof-wall connections have not fractured and all
structural members are still able to support gravity and live loads.
3. Fracture of the roof-wall connections. Damage state DS3 in Figure 29, which
coincides with the sharp and significant drop in the base shear at ∆C = 11.9", is
associated with the fracturing of a large number of roof-wall connections along one or
both walls. For the structure considered in this example, where the strength of the
roof-wall connections was considered deterministic, dynamic analyses at finely
incremented S a levels have shown that the initial failure of a small number of
connections quickly leads to a total loss of lateral and gravity load carrying capacity.
This is because the load carried by the fractured connections is transferred to adjacent
connections (with same strength) that become overloaded as a result and also fail.
Consequently, in this damage state, the structure has at least partially collapsed and
many structural members are no longer capable of resisting gravity or lateral loads. At
the end of this application section, a few results are included on the effects of
explicitly considering the variability in the strength of roof-wall connections.

Step 2: NSP curves for the damaged building

To evaluate the post-earthquake residual capacity of the structure to withstand future
seismic loads, a NSP curve is developed for each damage state identified in Step 1 that is
capable of supporting gravity loads, i.e., damage states DS1 and DS2 in this case. DS3 is in
fact associated with incipient collapse. For damage state DS1 (onset of damage), this NSP
curve is identical to that of the intact structure; thus, only damage state DS2 must be
considered in this example.
Referring to Figure 30, the NSP curve for damage state DS2 is obtained by (1) displacing
the structure model to ∆C = 7.5" (point 1), (2) quasi-statically unloading it to zero base
shear (point 2) and (3) reloading it to failure (point 3). The sequence of points 2-3-4-5
defines the shape of the NSP curve for damage state DS2. As discussed in the guidelines,
the residual displacement associated with this pushover curve ( ∆C = 2.8") is ignored in
this step and we horizontally shift the NSP curve for damage state DS2 until its origin
(point 2 in Figure 30) coincides with (0,0). As illustrated in Figure 31, we do not shift
the failure displacement ( ∆C = 11.9"); the resulting disconnect is bridged by simply
extending the last segment of the pushover curve.31 The expected (or measured) residual
displacement will be accounted for in Step 3.

31
The resulting disconnect is bridged by extending the last segment of the pushover rather than assuming
perfectly-plastic behavior across the gap because the latter cannot easily be input into SPO2IDA. The
effect on the final estimate of residual capacity is not expected to be significant.

59
The NSP curves for the intact structure and damage state DS2 are plotted together in
Figure 31. Note that the NSP curves are estimated to have the same initial slope; hence,
f DS 2 = f 1 , where f1 and f DS 2 are the initial first mode frequencies of the intact and
damaged structures, respectively32. Also note that the linear-elastic region of the NSP
curve for damage state DS2 is larger that that of the intact structure due to the strain
hardening present in the roof diaphragm model (see Figure 26). The peak base shear is
also larger for damage state DS2 than it is for the intact structure, only because the last
segment of the damaged pushover curve is extended in shifting the NSP back to zero
residual displacement.

Step 3: Inferring dynamic response from static response

In order to use the SPO2IDA spreadsheet tool to predict the peak dynamic response of
the intact and damaged structure, the NSP curves obtained in Steps 1 and 2 must be
modified as follows.
1. Piecewise linear idealization of the NSP curves. The SPO2IDA spreadsheet requires
that the NSP curves be defined in a piecewise linear fashion. Figure 32 shows
plausible idealizations for the NSP curves obtained for the intact and damaged
structures in steps 1 and 2 (assuming no residual displacement offset).
2. Normalization of the base shear and displacement axes. The SPO2IDA spreadsheet
assumes that the base shear and displacement axes of the NSP curves are normalized
by the corresponding quantities at the onset of damage. Thus, for the intact structure
the displacement ordinates are expressed in terms of the global ductility ratio,
µ = ∆C ∆Cy , where ∆Cy = 1.4″ is the yield displacement and the base shear ordinates
are expressed as the ratio R = BS BS y , where BS y = 315 kips is the base shear that
corresponds to ∆Cy in Figure 32a. Similar calculations are made for the damaged
structure using the base shear and displacement values shown in Figure 32b. Plots of
the normalized NSP curves for use with the SPO2IDA spreadsheet are shown in
Figure 33.
Using the normalized NSP curves plotted in Figure 33 as input, the SPO2IDA
spreadsheet predicts the dynamic response of the structure in its intact and damaged
conditions. The resulting normalized median IDA curves associated with the intact and
damaged structures (but assuming no residual displacement offset) are also plotted in
Figure 33. Note that the IDA results are expressed in terms of R = S a ( f1 ) / S ay ( f1 ) for
the intact structure and R = S a ( f DS 2 ) / S ay , DS 2 ( f DS2 ) for the damaged structure, where
S ay ( f 1 ) and S ay , DS 2 ( f DS 2 ) are the first-mode pseudo-acceleration values that correspond
to ∆Cy in Figure 32. Based on the results summarized in Table 13, which suggest that the
response of the tilt-up is dominated by its first mode, we assume S ay ( f1 ) = BS y / M 1∗ for
the intact structure, where M 1∗ = ψ 12φ1Τ Mφ1 is the effective modal mass of the first mode

32
The same is true if parallel-to-elastic unloading/re-loading is assumed in order to approximate the NSP
curve for damage state DS2. The effect of this approximation on the NSP curve, and more importantly on
the residual capacity estimated in Step 3, are small.

60
of vibration and ψ 1 is the first mode participation factor. M 1∗ and ψ 1 are listed in Table
13. Similarly, S ay , DS 2 ( f DS 2 ) = BS y , DS2 / M 1∗, DS 2 for the damaged structure. Figure 31
suggests that the initial elastic response of the damaged structure is identical to that of the
intact structure; therefore, M 1∗, DS 2 = M 1∗ for this building.

For the purposes of developing the fragility curves, it is beneficial to plot the IDA curves
in terms of ∆C and S a ( f1 ) for the intact structure or S a ( f DS 2 ) for the damaged structure.
These plots are shown in Figure 34. The displacement values in Figure 34 are obtained
by multiplying the global ductility values shown in Figure 33 by ∆Cy = 1.4′′ for the intact
structure and by ∆Cy = 2.0′′ for the damaged structure. The S a ( f1 ) ordinates shown in
Figure 34a for the intact structure are obtained by multiplying the R = S a ( f1 ) / S ay ( f1 )
ordinates in Figure 33a by S ay ( f 1 ) . Similarly, the S a ( f DS 2 ) ordinates plotted in Figure
34b for the damaged structure are obtained by multiplying the
R = S a ( f DS 2 ) / S ay ,DS 2 ( f DS2 ) ordinates in Figure 33b by S ay ,DS 2 ( f DS 2 ) .

Note that the spectral acceleration values shown in Figures 34a and 34b are those
associated with the fundamental frequencies f1 and f DS 2 of the intact and damaged
structures, respectively. In general, f DS 2 ≠ f1 .33 However, for the purposes of
determining the tagging condition in Step 4, the capacities for the intact and damaged
structures must be expressed in terms of the same reference spectral acceleration, which
is chosen to be that associated with the first mode of the intact structure, S a ( f1 ) . Hence,
the S a axis for the damaged structure in Figure 34b is scaled by S a ( f1 ) / S a ( f DS 2 ) .34
However, because the first-mode frequencies of the intact and damaged buildings are
identical in this example, S a ( f1 ) / S a ( f DS 2 ) = 1.0 . We reiterate that this will not be the
case in general. The resulting IDA curves for the intact and damaged structures are
plotted in Figure 35. Also indicated in Figure 35 are the capacity of the intact structure
and the residual capacity of the damaged structure (assuming no residual displacement
offset).
An IDA curve for the damaged structure that accounts for the expected (or measured)
residual displacement offset is also shown in Figure 35. In this example, the expected
residual displacement is assumed to be equal to the static offset ( ∆C = 2.8"). As
explained in the guidelines and illustrated in Figure 35, the IDA curve that accounts for
this residual displacement is obtained by tracing the IDA curve that assumes no residual
offset up to the displacement equal to the collapse capacity minus the expected residual
displacement, or 11.9"-2.8"=9.1". Accounting for the residual displacement offset
reduces the residual capacity of the damaged structure, as one might intuitively expect.

33
Unless, of course, parallel-to-elastic unloading/re-loading is assumed to approximate the NSP curve for
the damaged structure.
34
As explained in the commentary to the guidelines, this conversion is not strictly necessary because it is
captured by the final adjustment that is made to the estimates of the residual capacity for the damaged
structures.

61
Finally, the estimate of the residual capacity that accounts for the expected (or measured)
residual displacement is adjusted to reflect the observation that, without a residual offset,
little or no reduction in the residual capacity relative to that of the intact structure is
expected. As explained in the guidelines, this is accomplished by multiplying the
residual capacity estimate from Figure 35 (accounting for the expected residual
displacement) by the ratio of the capacity of the intact structure to the residual capacity of
the damaged structure assuming no residual displacement. The final estimate of the
(
residual capacity for the structure in damage state DS2, therefore, is ( S a ,cap ) 2 = 2.12g *
2.63/2.36 = 2.36g.35 Note that it is only a coincidence that this final estimate of (Ša,cap)2
happens to equal the residual capacity assuming no residual displacement offset.

Step 4: Occupancy status for the damaged building

As outlined in the text of the guidelines, the tag assigned to a structure in damage state
DS2 is based on
a) the reduction in the residual median lateral capacity relative to that of the intact
structure, and
b) the MAF of exceeding the spectral capacity (Ša,cap)2, which we denote P, and the
MAF of exceeding the ground motion corresponding to the median capacity,
(Ša,cap)1, of the building in its intact conditions, which we call P0.

Considering the capacities identified in Figure 35, we see that the reduction in the median
capacity is approximately 10.2% and (Ša,cap)2=2.36g. Referring to Figure 7, this
reduction in capacity results in a green tag if P0< 2.9x10-4, a yellow tag if 2.9x10-4≤P0<
7.3x10-4, and a red tag if P0≥7.3x10-4. For the purposes of demonstrating the procedure,
we assume that the MAF of exceeding (Ša,cap)2 is about 2.9x10-4. At this level of P0 a
drop of capacity of about 10% means onset of yellow tag36. In other words, if the tilt-up
is deemed to be in damage state DS2 following an earthquake, it should be yellow-tagged.
Note that because damage state DS3 is associated with at least a partial collapse of the
structure, it is automatically red-tagged.

Step 5: Ground motion level associated with a structural limit state

The median mainshock ground motion intensities associated with the damage states
identified in Step 1 are determined from the IDA curve associated with the intact
structure plotted in Figure 35 (or Figure 34a). Referring to this figure, we see that ŠaOD=
0.22g for the onset of damage limit state, ŠaY=1.85g for damage state DS2, which was
assigned a yellow tag in Step 4, and ŠaC=2.63g for the collapse limit state DS3.
The dispersion of these capacities, β, is composed of an aleatory component, βR, which is
estimated from Figure 10, and an epistemic component, βU, which is based on the

35
Note that if parallel-to-elastic unloading/re-loading is assumed to approximate the NSP curve for the
damaged structure, the residual capacity for damage state DS2 is 2.31g, which is within about 2% of that
computed in this example.
36
If 2.9x10-4≤P0< 7.3x10-4, then DS2 should still be yellow-tagged but an interpolation scheme should be
used to find ŠaY in STEP 5.

62
baseline values listed Table 2a. The net dispersion used in Step 6 is the square-root-sum-
of-squares of these two components, i.e., β = β R2 + βU2 . The β values for the three
damage states are summarized in Table 14.

Step 6: Computation of the fragility curves

Using the median mainshock ground motion intensities and dispersions obtained in Step
5 for the onset-of-damage, yellow-tag and collapse limit states, the fragility curves for
these damage states are computed as described in the guidelines. The resulting fragility
curves are plotted in Figure 36. Also shown in Figure 36 are the fragility curves
computed for the tilt-up building assuming brittle roof-wall connections (the details of the
calculations leading to the brittle-connection fragility curves have not been presented
since they are similar in nature to those used to generate the fragility curves for the
ductile connection case). Note that the onset-of-damage and yellow-tag fragility curves
are identical for the two types of connections16, but the fragility curves for the collapse
limit state differ, reflecting the improved performance of the building when the roof-wall
connections are ductile. As can be seen in Figure 36, the median collapse capacity of the
building with ductile connections is approximately 10% greater than that of the building
with brittle connections.

16
At damage state DS2, the roof-wall connections are in their linear-elastic range; hence, the behavior of
the building is indistinguishable for the two connection types and the median mainshock ground motion
intensities associated with DS2, which depend only on the IDA results for the intact structure up to the
displacement that defines this damage state, are identical. For the building with brittle connections, the
reduction in the residual capacity for DS2 (23%) is a little greater than that obtained for the building with
ductile connections (19%); however, this difference is not enough to change the yellow-tag designation or
the β-values for this damage state.

63
Validation of the Results

The median IDA curves predicted by SPO2IDA for the intact tilt-up building in the above
example are validated by a comprehensive series of dynamic time-history analyses
conducted with the ensemble of ground motions listed in Table 15. Each ground motion
is scaled by the factor indicated in Table 15 so that the pseudo-acceleration at the
fundamental period of the building ( T1 = 0.57 s) is S a (T1 ) = 2.0 g .
For each of the ground motions listed in Table 15, nonlinear dynamic analyses of the tilt-
up model were conducted for 0.4 g ≤ S a (T1 ) ≤ 4.8 g by scaling the record appropriately in
accordance with the standard procedure for generating IDA curves. For each time-
history analysis, the peak wall displacement at the intersection of gridlines 2 and D in
Figure 25 (i.e., the control displacement, ∆C , in the above example) was recorded. The
peak wall displacements obtained for each accelerogram at the levels of ground motion
intensity considered are plotted in Figure 37 for a tilt-up building with ductile roof-wall
connections and a building with brittle roof-wall connections. Also plotted in Figure 37
is the median IDA curve based on the results of the time-history analyses and the 16th ,
84th and 50th percentile (median) IDA curves predicted by SPO2IDA. Note that the axes
in Figure 37 have been normalized as described earlier for use with SPO2IDA.
Figure 37 shows that for the building with ductile roof-wall connections, the median IDA
curve predicted by SPO2IDA closely matches that based on the time-history results for
all values of µ . The median collapse capacity predicted by SPO2IDA for this case
(ŠaC/Say =7.8) is approximately 8% less than that suggested by the median curve based on
the time-history results (ŠaC/Say =8.5). Thus, SPO2IDA appears to be an appropriate tool
for predicting even the severe dynamic response of this building.
For the building with brittle connections, the agreement between the SPO2IDA results
and the time-history results is excellent for µ < 5 , but weakens for larger ductility values
with SPO2IDA consistently over-predicting the dynamic response of the system. One
must recognize however that the time-history results are based on a relatively small
sample size (ten recorded ground motions), so discrepancies should be expected in this
region where the drift dispersion is large. Also, note that the median curve based on the
time-history results falls well within the band defined by the 16th and 84th percentile
curves predicted by SPO2IDA. Consequently, we conclude that for the building with
brittle connections, the SPO2IDA results are reasonable and acceptable for the purposes
of use with these guidelines.

64
Wall:
• 5½" thick 0" 98' 6" 0"
• 30' 0" high Ground motion in
x-direction.
• Modeled with
horizontal linear-
1
elastic beam
elements.
• 2/3 of wall mass Glulam beams are
(20') evenly modeled with linear-
distributed to roof elastic bar elements.
nodes shown.

16 @ 18' 9"
= 300' 0"
x
2

End walls are

Roof chord modeled
modeled as rigid
with stiff linear-elastic
elements.
bar elements to
account for the in-
plane rigidity of the
concrete walls.
3
Half of roof mass
evenly distributed to
nodes along each chord. A B C D

y Roof-wall connections modeled

with zero-length fracturing
Roof diaphragm modeled with elements. See Figure 27.
nonlinear elements capable of
reproducing the pinched
behavior observed by Pardoen

Figure 25. Plan view of Hollister tilt-up.

65
 30

20

10
Lateral load (kips)

-10

-20

-30
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Lateral displacement (in)

Figure 26. Load-displacement response of diaphragm element. Element calibrated to reproduce

UC Irvine test results (NC Control – 16’ x 20’ specimen).

Fu = 1667 k/ft Fu = 1667 k/ft

δy = 0.25" δfrac = 0.75"

linear-elastic in linear-elastic in δfrac = 0.25"
compression compression

(a) Ductile (b) Brittle

Figure 27. Roof-wall connection behaviors considered.

66
 shot pins or screws
light gage steel strip
twisted metal
strap plywood

joist hanger steel angle

ledger purlin or joist hanger
wall panel subpurlin
wall panel

Figure 28. Roof-wall connections commonly used in current practice.

BS (kips)
1000

DS2 DS3
800

600

DS1
400

200

0
0 2 4 6 8 10 12 14 16 ∆C (inches)

Figure 29. NSP curve for intact structure.

67
BS DS 2 (kips)
1000
3

800
1

600

400

200

2 4 5
0
0 2 4 6 8 10 12 14 16 ∆C (inches)

residual ∆ = 2.8"
Figure 30. Development of the NSP curve for damage state DS2.

BS or BS DS 2 (kips)
1000

Damaged structure
800

600

400
Intact structure

200

0
0 2 4 6 8 10 12 14 16 ∆C (inches)

Figure 31. NSP curves for intact structure and damage state DS2.

68
 (a) Intact structure
BS frac = 935 kips

Piecewise linear 0.16k 0

idealization

0.32 k 0
BS y = 315 kips

Original NSP curve

k0

∆C y =1.4" ∆C ,hard = 8.1" ∆C , frac = 11.9" ∆C (inches)

(b) Damaged structure

BS frac , DS 2 = 935 kips Original NSP curve

Piecewise linear
idealization 0.18k0

BS y , DS 2 = 450 kips 0.50 k0

k0

∆C y =2.0" ∆C ,hard = 4.8" ∆C , frac = 9.1" ∆C (inches)

Figure 32. Idealized piece-wise linear NSP curves for intact and damaged structures.

69
 R (a) Intact structure
10

9
S a , cap ( f1 ) S ay ( f1 ) = 7.8
8
Median IDA curve
7
predicted by SPO2IDA

5
Piecewise linear
4 NSP curve

0
0 2 4 6 8 10 12 µ = ∆C / ∆Cy

R (b) Damaged structure

10

7 Median IDA curve

predicted by SPO2IDA
6

5 S a , cap ( f DS 2 ) S ay , DS 2 ( f DS 2 ) = 4.4
4

3
Piecewise linear
2 NSP curve

0
0 2 4 6 8 10 12 µ = ∆C / ∆Cy

Figure 33. Normalized NSP curves and median IDA curves predicted by SPO2IDA for the intact and
damaged structures. For the intact structure, R = S a ( f1 ) S ay ( f1 ) for the IDA curve and
R = BS / BS y for the NSP curve. For the damaged structure, R = Sa ( f DS 2 ) S ay , DS 2 ( f DS 2 )
for the IDA curve and R = BS DS 2 BS y , DS 2 for the NSP curve.

70
S a ( f 1 ) (g) (a) Intact structure
3
S a ,cap = 2.63 g

0
0 2 4 6 8 10 12 14 16 ∆C (inches)

S a ( f DS 2 ) (g) (b) Damaged structure

3

( S a ,cap ) 2 = 2.12 g
2

0
0 2 4 6 8 10 12 14 16 ∆C (inches)

Figure 34. Median IDA curves for the intact and damaged structures.

71
 S a ( f 1 ) (g)
3

SˆaColl
, cap = 2.63 g
S a ,cap = 2.63 g

IDA for intact ( S a ,cap ) 2 = 2.12 g

structure
2

Sˆ aY,cap = 1.85 g

IDA for damaged

structure
1

Sˆ aOD
,cap = 0.22 g
0
0 2 4 6 8 10 12 14 16 ∆C (inches)
DS1 DS2 DS3
Figure 35. Median IDA curves for the intact and damaged structures in terms of S a ( f1 ) . Note that
for this example, f DS 2 = f 1 so S a ( f DS 2 ) = S a ( f1 ) .

FS ( S a )
1

Onset of damage

Yellow tag
Collapse
0.5
(brittle connections)

Collapse
(ductile connections)
Median collapse capacities:

SˆaColl
, cap = 2.40 g (brittle)

SˆaColl
, cap = 2.63 g (ductile)

0
0 1 2 3 4 5 S a ( f1 ) (g)
Figure 36. Fragility curves for the onset-of-damage, yellow-tag and collapse limit states. Note that for
the onset-of-damage and yellow-tag limit states, the fragility curves for ductile and brittle
connections are identical.
72
S a ( f1 ) S ay ( f1 ) Ductile roof-wall connections
15
Data points from dynamic
14
analyses 16th percentile IDA curve
13 predicted by SPO2IDA
12
11
10
9
Median IDA curve from
dynamic analyses
8
7 Median IDA curve
predicted by SPO2IDA
6
5
4
84th percentile curve
3 predicted by SPO2IDA
2
1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 µ = ∆C / ∆Cy

S a ( f1 ) S ay ( f1 ) Brittle roof-wall connections

15
14 Data points from dynamic
analyses
13
16th percentile IDA curve
12 predicted by SPO2IDA
11
10
9
8
Median IDA curve
predicted by SPO2IDA
7
6
Median IDA curve from
5 dynamic analyses
4
84th percentile curve
3
predicted by SPO2IDA
2
1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 µ = ∆C / ∆Cy

Figure 37. Comparison of median IDA curves obtained from dynamic analyses to that predicted
by SPO2IDA. 73
Table 13. Modal properties of the intact tilt-up building model.

Effective modal mass

Mode Period, Frequency, Damping Participation
n Tn (s) f n (rad/s) ratio, ς n factor, ψ n
M n∗ M n∗
∑n M n∗
1 0.573 11.0 0.046 1.56 2.418 0.858
2 0.307 20.5 0.037 0.00 0.000 0.000
3 0.227 27.7 0.039 0.49 0.240 0.087
4 0.193 32.6 0.041 0.00 0.000 0.000
5 0.175 35.9 0.043 0.27 0.073 0.026
6 0.163 38.5 0.044 0.00 0.000 0.000
7 0.155 40.5 0.045 0.17 0.029 0.010
Total 0.981

Table 14. Aleatory ( β R ) and epistemic ( β U ) uncertainty associated with limit states.

Limit state βR βU β = β R2 + βU2

Onset of damage 0.27 0.6 0.66
Yellow tag 0.30 0.6 0.67
Collapse 0.46 0.5 0.68

74
Table 15. Recorded ground motions used for nonlinear dynamic analyses of the tilt-up building.

Distance S a (T1 ) Scale factor

Ground motion MW (km) (2.0/ S a (T1 ) )
(g)

1. Loma Prieta, Halls Valley (1989) 6.9 31.6 0.19 10.5

2. Landers, Palm Springs Airport (1992) 7.3 37.5 0.25 8.0
3. Northridge, Camarillo (1994) 6.7 36.5 0.44 4.5
4. Kobe, Tadoka (1995) 6.9 30.5 1.07 1.9
5. Kern Country, Taft Lincoln School (1952) 7.4 41.0 0.40 5.0
6. Coalinga, Gold Hill 2W (1983) 6.4 36.6 0.19 10.5
7. Imperial Valley, Niland F.S. (1979) 6.5 35.9 0.17 11.8
8. Morgan Hill, Hollister City Hall (1984) 6.2 32.5 0.14 14.3
9. Whittier Narrows, Saticoy St. (1987) 6.0 39.8 0.17 11.8
10. North Palm Springs, Indio (1986) 6.0 39.6 0.49 4.1

75
Effects of Random Connection Strengths
As mentioned in the description of damage state DS3, when the roof-wall connections
have the same strength and ductility, as assumed in the above analyses, the initial failure
of a small number of connections usually leads to the total loss of lateral and gravity load
carrying capacity as the load carried by the fractured connections is transferred to
adjacent connections that become overloaded as a result and also fail. Consequently, the
time-history analyses used to plot Figure 37 rarely fractured a fraction of the roof-wall
connections. In particular, for those time-history analyses that used ground motions with
intensities that were within ±30% of the intensity that caused all the roof-wall
connections to fail, approximately 10% of the analyses performed on the structure
modeled with ductile connections and 9% of the analyses performed on the structure
modeled with brittle connections resulted in a damaged structure in which the percentage
of fractured connections was less than 67% (but more than 0%). Furthermore, the
intensity of those ground motions that produced a damaged structure in which a fraction
of the roof-wall connections were fractured was usually close to the intensity required to
cause all the connections to fail; i.e., for each ground motion there is only a small range
of intensities in which a fraction of the roof-wall connections fail. Similar conclusions
can be drawn from the nonlinear pushover analysis plotted in Figure 29, in which a sharp
and significant drop in the base shear is observed at the roof displacement ∆ C = 11.9"
due to all the roof-wall connections fracturing.
Unfortunately, the above observations do not agree well with what was observed
following the 1994 Northridge earthquake, during which the roofs of tilt-up buildings
partially collapsed as a result of localized roof-wall connection failures. It is speculated
that the discrepancy between the results of the time-history analyses used to produce
Figure 37 and the failures reported after Northridge may be due to the variability of the
roof-wall connection strengths in the buildings that failed during the Northridge
earthquake. For example, a connection that has a higher capacity than its neighbors may
be able to arrest the progressive failure of all the roof-wall connections after one
connection fails. In this section, we explore the plausibility of this explanation for the
localized roof-wall connection failures observed in Northridge with a second, but no less
important, objective of providing insight into how the analysis procedure for tilt-up
buildings may be modified to include damage states characterized by the fraction of the
roof-wall connections that failed during the mainshock.

Structure Model and Random Connection Strengths

In the following analyses, we consider the tilt-up building shown in Figure 25 with brittle
roof-wall connections, whose behavior under load is as shown in Figure 27b. We assume
that the tensile connection strength indicated in Figure 27b (Fu = 1667 k/ft) is the mean
value of a Weibull random variable with a coefficient of variation (c.o.v.) of δ = 20%.
Furthermore, we assume that the strength of each connection is independent of all other
connections in the building.

76
Ground Motions

We are interested in the percentage of fractured roof-wall connections caused by an

earthquake when the 30 roof-wall connections shown in Figure 25 are randomly assigned
strengths drawn from the Weibull distribution described above. Due to the random
assignment of the connection strengths, the percentage of fractured connections caused
by an earthquake is also random. To quantify the variability in the percentage of fractured
connections, 20 time-history analyses (each with a different realization of connection
strengths) are performed for each ground motion listed in Table 15. Naturally, the
intensity of the ground motions to use for the dynamic analyses is an important
consideration. For example, if the applied ground motion is very weak, then none of the
roof-wall connections will fail, while if the ground motion is very strong, then all the
connections are likely to fail, regardless of the assumed distribution of their tensile
strengths. This is because the common ground motion introduces a correlation between
the collapses of different connections. Consequently, rather than scaling each ground
motion to have the same Sa(T1) value, each ground motion is scaled relative to the
spectral acceleration (denoted S a∗ (T1 ) ) at which all the connections fracture when their
strengths are set to their mean value (Fu = 1667 k/ft). Table 16 lists S a∗ (T1 ) for each
ground motion. In the following analyses, we let the ground motion intensity scale factor
γ = 1 coincide with S a∗ (T1 ) . Note that with this definition of γ, when the strengths of all
the connections are assigned their mean values, they will all fail when γ ≥ 1.

Results of the Time-History Analyses

The results of the time-history analyses described above are summarized for each ground
motion in Figures 38a – 38j. Plotted in these figures for γ = 0.7, 0.9, 1.0 and 1.1 are (1)
the empirical cumulative distribution functions (ECDFs) of the percentage of fractured
connections and (2) the distribution of the damage along the length of the wall.

Empirical cumulative distribution functions

The ECDFs plotted on the left-hand side of Figures 38a – 38j quantify the probability that
the percentage of fractured connections is less than or equal to some prescribed fraction
0 ≤ α ≤ 1 for each ground motion when it is scaled by γ = 0.7, 0.9, 1.0 and 1.1. The data
points plotted in these figures correspond to the percentage of fractured connections
recorded for each of the 20 realizations of connection strengths at the four levels of
ground motion intensity considered. The lines are beta CDFs fitted using the method of
moments to the 20 data points collected for each value of γ.

It is evident from the ECDF plots that as the intensity factor γ is decreased, a partial
failure of the roof-wall connections is more likely to occur. In particular, we note that for
γ = 0.7, the median of α lies between 15% and 90% for all ground motions considered;
compare this result to those obtained from the time-history analyses performed with the
roof-wall connection strengths set equal to their mean values, in which none of the
connections fail when γ = 0.7, except for one ground motion (#4). Consequently, when

77
the roof-wall connection strengths are variable, the range of ground motion intensities
that cause partial roof-wall connection failures is wider than that observed for the case in
which all the connection strengths are set equal to their mean values. Furthermore, it is
evident from the ECDFs plotted in Figures 38a – 38j that when γ ≥ 1, which will cause all
the connections to fail when their strengths are set equal to their mean values, there are
many cases in which only a partial failure occurs. Therefore, we conclude from these
plots that the variability of the roof-wall connection strengths is a plausible explanation
for the partial failures observed after the Northridge earthquake.

Damage distribution plots

To gain additional insight into the performance of the tilt-up building shown in Figure 25
when the roof-wall connection strengths are variable, consider the damage distribution
plots shown on the right-hand side of Figures 38a – 38j. The horizontal axis of these plots
indicates the location along the length of the wall and the vertical axis measures the
fraction of the simulations in which the connection at a particular location fails.

The damage distribution plots suggest that, in general, the connections near the middle of
the wall are more susceptible to fracturing. This observation is reasonable when one
assumes that the building responds in its fundamental mode of vibration, which is plotted
in Figure 39 along with the third, fifth and seventh modes37. However, there are
exceptions to this observation, notably ground motions 4 and 10 (Figures 38d and 38j,
respectively), which appear to cause the connections near the ends of the wall to fail more
often, and ground motion 9 (Figure 38i), which appears to create complex patterns of
damage along the length of the wall. We remark that for each ground motion, the
connection strengths are randomly selected for each simulation at each intensity level.
Therefore, it is unlikely that the damage patterns observed for ground motions 4, 9 and 10
are due to “odd” combinations of connection strengths since the patterns can be seen at
all four values of γ considered; instead, the unusual damage patterns seen in Figures 38d,
38i and 38j are likely due to unusual characteristics of these ground motions.

To verify this assertion consider Table 17, which summarizes the intensity of the
response spectra at the fundamental mode of vibration, S a (T1 , ζ 1 ) , to that of the third,
fifth and seventh modes ( S a (Tn , ζ n ) , n = 3,5,7 ) for each ground motion plotted in Figure
38. Studying this table, we see that the three ground motions that exhibit the unusual
damage patterns in Figure 38 (ground motions 4, 9 and 10) are also the “weakest” at the
fundamental period of vibration of the structure relative to the higher modes of vibration.
As indicated in Figure 39, unlike the case of the fundamental mode of vibration, the
connection elongations near the ends of the wall for the higher modes of vibration are
comparable to those near the midpoint of the wall; consequently, it is reasonable to
expect damage patterns that are not concentrated near the midpoint of the wall when the
ground motion has significant frequency content near the higher modes of vibration of
the structure relative to the fundamental mode, as is the case for ground motions 4, 9 and
37
Recall that only the “odd” modes of vibration have non-zero participation factors because the “even”
modes are anti-symmetric shapes that cannot be excited by a ground motion that acts through the center of
rigidity of the model.

78
10. We conclude therefore that the unusual damage patterns caused by ground motions 4,
9 and 10 are the result of the frequency content of these ground motions relative to the
natural modes of vibration of the structure.

79
 Percentage of fractured connections Distribution of damage
1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8

Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

γ = 0.9
(a) 0.5 0.5
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(b) 0.5 0.5
γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0
Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(c) 0.5 0.5
γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Figure 38. Empirical CDFs of the percentage of fractured connections and the distribution of the
damage along the length of the wall for ground motions 1, 2 and 3.

80
 Percentage of fractured connections Distribution of damage
1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )
0.7 0.7

0.6 0.6 γ = 0.7

γ = 0.9
(d) 0.5 0.5
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(e) 0.5 0.5

γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0
Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(f) 0.5 0.5
γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Figure 38 (con’t). Empirical CDFs of the percentage of fractured connections and the
distribution of the damage along the length of the wall for ground motions 4, 5 and 6.

81
 Percentage of fractured connections Distribution of damage
1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

Cumulative probability, F(α ) 0.8 0.8

0.7 0.7

0.6 0.6 γ = 0.7

γ = 0.9
(g) 0.5 0.5
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(h) 0.5 0.5

γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Percentage of fractured connections Distribution of damage

1.0 1.0
Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

0.8 0.8
Cumulative probability, F(α )

0.7 0.7

0.6 0.6 γ = 0.7

(i) 0.5 0.5

γ = 0.9
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Figure 38 (con’t). Empirical CDFs of the percentage of fractured connections and the
distribution of the damage along the length of the wall for ground motions 7, 8 and 9.

82
 Percentage of fractured connections Distribution of damage
1.0 1.0

Avg. damage (0 = no damage; 1 = fractured)

0.9 0.9

Cumulative probability, F(α ) 0.8 0.8

0.7 0.7

0.6 0.6 γ = 0.7

γ = 0.9
(j) 0.5 0.5
γ = 1.0
0.4 0.4 γ = 1.1
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Percent fractured (α ) Connection number

Figure 38 (con’t). Empirical CDFs of the percentage of fractured connections and the
distribution of the damage along the length of the wall for ground motion 10.

Mode 1 T1 = 0.573 s
ζ1 = 0.046
ψ1 = 1.56 4 8 12
connection number

Mode 3 T3 = 0.227 s
ζ3 = 0.039
ψ3 = 0.49 4 8 12
connection number

Mode 5 T5 = 0.175 s
ζ5 = 0.043
ψ5 = 0.27 4 8 12
connection number

Mode 7 T7 = 0.155 s
ζ7 = 0.045
ψ7 = 0.17 4 8 12
connection number

Figure 39. Modal connection elongations.

83
Table 16. Spectral acceleration, S a∗ (T1 ) , at which all the connections fracture when their
strengths are set to their mean value (Fu = 1667 k/ft).

Ground Motion S a∗ (T1 ) (g)

(see Table 15)
1 1.9
2 2.4
3 3.1
4 4.0
5 2.3
6 3.0
7 1.8
8 1.6
9 1.2
10 4.4

Table 17. Intensity of the response spectrum at the fundamental mode of vibration of the
example tilt-up building relative to that corresponding to the third, fifth and seventh modes.

Ground Motion S a (T1 , ζ 1 ) S a (T1 , ζ 1 ) S a (T1 , ζ 1 )

S a (T1 , ζ 1 ) (g)
(see Table 15) S a (T3 , ζ 3 ) S a (T5 , ζ 5 ) S a (T7 , ζ 7 )

1 0.19 1.75 1.49 1.69

2 0.25 0.91 0.94 1.14
3 0.44 1.18 1.84 2.02
4 1.07 0.50 0.54 0.92
5 0.40 0.60 0.95 0.95
6 0.19 1.46 1.74 2.14
7 0.17 0.68 0.47 0.61
8 0.14 1.10 0.89 1.15
9 0.17 0.32 0.49 0.67
10 0.49 0.39 0.40 0.26

84
References

ATC-11, 1983. “Seismic Resistance of Reinforced Concrete Shear Walls and Frame Joints:
Implications of Recent Research for Design Engineers”, Applied Technology Council, 555 Twin
Dolphin Drive, Suite 550, Redwood City, California 94065

ATC-20, 1989. “Procedures for Postearthquake Safety Evaluation of Buildings”, Applied

Technology Council, 555 Twin Dolphin Drive, Suite 550, Redwood City, California 94065

ATC-40, 1996. “Seismic Evaluation and Retrofit of Concrete Buildings”, Vol. I-II, Applied
Technology Council, 555 Twin Dolphin Drive, Suite 550, Redwood City, California 94065

ATC, 1999. Earthquake Aftershocks - Entering Damaged Buildings, TechBrief 2, Applied

Technology Council, 555 Twin Dolphin Drive, Suite 550, Redwood City, California 94065

Bandyopadhyay, K., Cornell, C.A., Costantino, C., Kennedy, R.P., Miller, C., and A. Veletsos,
1993. “Seismic Design and Evaluation Guidelines for the Department of Energy High-Level
Waste Storage Tanks and Appurtenances”, Report BNL 52361 UC-406 UC-510, Department of
Nuclear Energy Brookhaven National Laboratory, Upton, NY 11973

Bazzurro, P., Cornell, C.A., Menun, C., and M. Motahari, 2004. “Guidelines for Seismic
Assessment of Damaged Buildings", Proceedings of 13th WCEE, Paper 1708, Vancouver, B.C.,
Canada, August 1-6.

Benjamin, J.R., and C.A. Cornell, 1970. “Probability, Statistics, and Decision for Civil
Engineers”, Mc-Graw Hill, Inc., New York.

Cornell, C.A., Vamvatsikos, D., Jalayer, F., and N. Luco, 2000. “Seismic Reliability of Steel
Frames”, Proc. IFIP WG 7.5, Working Conference on Reliability and Optimization of Structural
Systems, Ann Arbor, MI, September.

Cornell, C.A., Jalayer, F, Hamburger, R.O. and Foutch, D.A., “The Probabilistic Basis for the
2000 SAC/FEMA Steel Moment Frame Guidelines,” Journal of Structural Engineering, Vol.
128, No. 4, pp. 526-533, April 2002.

Degenkolb Engineers, 2004, “Application and Evaluation of Advanced Seismic Assessment

Guidelines for Concrete Tilt-up Buildings”, Report Prepared for PG&E/PEER Lifelines
Program, Task 509, San Francisco, CA, December.

Duffley, T.A., Goldman, A., and C.R. Farrar, 1994. “Shear Wall Ultimate Drift Limits”, Report
NUREG/CR-6104 LA-12649-MS, Los Alamos National Laboratory, Los Alamos, NM 87545

Earthquake Engineering Research Institute (EERI), 1996, “Tilt-up-wall buildings”, Earthquake

Spectra, Supplement C to Vol. 11, pp 99-123.

HAZUS, 1999. “Natural Hazard Estimation Methodology”, Federal Emergency Management

Agency, FEMA (http://www.fema.gov/hazus/hazus6.htm)

85
FEMA 178, 1992. “NEHRP Handbook for the Seismic Evaluation of Existing Buildings”,
Building Seismic Safety Council, Washington, D.C., June.

FEMA 273, 1997. “NEHRP Guidelines for the Seismic Rehabilitation of Buildings”, Building
Seismic Safety Council, Washington, D.C., October.

FEMA 288, 1997. “Background Reports: Metallurgy, Fracture Mechanics, Welding, Moment
Connections and Frame System Behavior”, Report No. SAC-95-09, SAC Joint Venture,
Sacramento, California, March.

FEMA 289, 1997. “Connection Test Summaries”, Report No. SAC-96-02, SAC Joint Venture,
Sacramento, California, June.

FEMA 306, 1999. “Evaluation of Earthquake Damaged Concrete and Masonry Wall Buildings’,
Basic Procedures Manual, prepared by ATC (ATC-43 Project), May.

FEMA 307, 1999. “Evaluation of Earthquake Damaged Concrete and Masonry Wall Buildings’,
Technical Resources, prepared by ATC (ATC-43 Project), May.

FEMA 350-352, 2000. “Recommended Seismic Design Criteria for New Steel Moment-Frame
Buildings (350); Recommended Seismic Evaluation and Upgrade Criteria for Existing Welded
Steel Moment-Frame Buildings (351); Recommended Post-earthquake Evaluation and Repair
Criteria for Welded Steel Moment-Frame Buildings (352)”, SAC Joint Venture, Sacramento,
California, July.

FEMA 355D, 2000. “State of the Art Report on Connection Performance”, SAC Joint Venture,
Sacramento, California, September.

FEMA 355F, 2000. “State of the Art Report on Performance Prediction and Evaluation of Steel
Moment-Frame Buildings”, SAC Joint Venture, Sacramento, California, September.

FEMA 356, 2000. “Prestandard and Commentary for the seismic rehabilitation of buildings”,
Building Seismic Safety Council, Washington, D.C., November.

Gupta, B., and S. Kunnath, 2000, “Adaptive Spectra-based Pushover Procedure for Seismic
Evaluation of Structures”, Earthquake Spectra, Vol. 16, No. 2, pp 376-391

Hall, J.F., 1999. "Seismic Response of Tilt-up Buildings" Draft report to PG&E.

Hamburger, R.O., Harris, S.K., Martin, S.C., McCormick, D.L., and P.G. Somerville, 1996.
"Response of Tilt-up Buildings to Seismic Demands: Observations and Case Studies from the
1994 Northridge Earthquake", EQE International, Inc., San Francisco, CA.

Ibarra, L., 2003. “Global Collapse of Frame Structures Under Seismic Excitations”, Ph.D.
Dissertation, Dept. of Civil and Environmental Engineering, Stanford University, Stanford,
California, September.

ICBO, 2000. “2000 International Building Code (IBC)”, International Conference of Building
Officials Publisher, Whittier, CA, 756 pp.

86
Kennedy, R. P., Cornell, C.A., Campbell, R.D., Kaplan, S. and H.F. Perla, 1980. “A Probabilistic
Seismic Safety Study of an Existing Nuclear Power Plant”, Nuclear Engineering and Design,
Vol. 59, No. 2, pp. 315-338, , North Holland, Amsterdam, August.
Kennedy, R. P. and M.K. Ravindra, 1984. "Seismic Fragilities for Nuclear Power Plant Risk
Studies", Nuclear Engineering and Design, V. 79, No. 1, pp. 47-68, North Holland, Amsterdam,
May.

Kennedy, R.P., 2001. “Seismic PRA Fragility Course”, Yucca Mountain Nuclear Waste
Repository Project, Department of Energy, Course held in Las Vegas for AIC/BECHTEL
engineers.

Kennedy, R.P., Short, S.A., McDonald, J.R., McCann, M.W., and R.C. Murray, 1989, “Design
and Evaluation Guidelines for Department of Energy Facilities Subjected to Natural Phenomena
Hazards”, Interim Report DOE UCRL-15910, prepared for The Office of the Assistant Secretary
for Environment, Safety & Health Office of Safety Appraisals United States Department of
Energy, Lawrence Livermore National Laboratory, Livermore, CA 94550.

Klamerus, E.W., and J.L. Cherry, 2001. “Structural Seismic Fragility Analysis of the Zion
Containment”, Report NUREG/CR-6740 SABD2001-2504P, Sandia National Laboratories,
Albuquerque, NM 87185.

Kunnath, S., and B. Gupta, 1999. “Spectra-compatible Pushover Analysis of Structures”, U.S.-
Japan Workshop on Performance-Based Earthquake Engineering Methodology for Reinforced
Concrete Building Structures, Maui, Hawaii, 13 September.

Luco, N., Bazzurro, P., and C.A. Cornell, 2004. “Dynamic Versus Static Computation of the
Residual Capacity of a Mainshock-Damaged Building to Withstand an Aftershock”, Proceedings
of 13th WCEE, Paper 2405, Vancouver, B.C., Canada, August 1-6.

Maffei, J., and R. Hamburger, 2004. “Seismic Assessment of the Building Located at 111
Almaden Street, San Jose, California”, by Rutherford & Chekene and Simpson Gumpertz &
Heger, Prepared for PG&E, August 6, 2004.

Medina, R. A., 2002. “Seismic demands for Non-Deteriorating Frame Structures and Their
Dependence on Ground Motions”, Ph.D. Dissertation, Dept. of Civil and Environmental
Engineering, Stanford University, Stanford, California, November.

Miranda, E., 2000. “Inelastic displacement ratios for structures on firm sites”, ASCE, Journal of
Structural Engineering, Vol. 126, No. 10, pp. 1150-1159.

Moehle, J., and G.G. Deierlein, 2004. “A Framework Methodology for Performance-Based
Earthquake Engineering”, Proceedings of 13th WCEE, Paper 679, Vancouver, B.C., Canada,
August 1-6.

Pardoen, G.C., Del Carlo, D. and R.P. Kazanjy, 1999. "Stiffness of Timber Diaphragms and
Strength of Timber Connections" Report No. PEER-99/04, Pacific Earthquake Engineering
Research Center, University of California, Berkeley.

Paulay, T., and M.J.N. Priestley, 1992. “Seismic Design of Reinforced Concrete and Masonry
Buildings”, John Wiley & Sons, Ltd., New York.

87
Paulay, T., 2000. “A Simple Displacement Compatibility-Based Seismic Design Strategy for
Reinforced Concrete Buildings”, Proceedings of the 12th World Conference on Earthquake
Engineering, Paper No. 0062, Auckland, New Zealand, January 30-February 4.

Pacific Gas & Electric (PG&E), 2001a. “Diablo Canyon Power Plant Long Term Seismic
Program”, San Francisco, CA 94105.

Pacific Gas & Electric (PG&E), 2001b. Personal Communication, Electronic mail message from
Kent Ferre dated December 4, 2001.

Pacific Gas & Electric (PG&E), 2001c. Personal Communication, Electronic mail message from
Kent Ferre dated December 11, 2001

Reed, J.W., and R.P. Kennedy, 1994. “Methodology for Developing Seismic Fragilities”, TR-
103959 Research Project, prepared for Electric Power Research Institute (EPRI), 3412 Hillview
Avenue, Palo Alto, CA 94304.

Ruiz-Garcia, J. (2004). “Performance-based assessment of existing structures accounting for

residual displacements,” Ph.D. Dissertation (in preparation), Dept. of Civil and Environmental
Engineering, Stanford University, Stanford, California.

Rutherford & Chekene Consulting Engineers, 2004. “Test Application of Advanced Seismic
Assessment Guidelines 3-Story Steel Moment-Resisting Frame Building”, Report Prepared for
PG&E/PEER Lifelines Program, Task 508, Oakland, CA, October.

Structural Engineers Association of Northern California (SEAONC), 2001. “Guidelines for

Seismic Evaluation and Rehabilitation of Tilt-up Buildings and Other Rigid Wall/Flexible
Diaphragm Structures”, SEAONC, San Francisco, CA.

Uniform Building Code (UBC), 1997. “Structural Engineering Design and Provisions”, Vol. 2,
International Conference of Building Officials (ICBO), Whittier, CA.

United States Geological Survey (USGS), 2001. “Seismic Hazard Curves and Uniform Hazard
Response Spectra for the United States", Compact Disc, Version 3.10, by A.D. Frankel and E.V.
Leyendecker, March.

Vamvatsikos, D. and C. A. Cornell, 2001a. “Incremental Dynamic Analysis”, Earthquake

Engineering and Structural Dynamics, Vol. 31, No. 3, John Wiley & Sons, Ltd., New York,
April, URL: http://nausika.stanford.edu/docs/IDA-paper.pdf

Vamvatsikos, D. and C. A. Cornell, 2001b. “SPO2IDA: Excel spreadsheet”,Web

implementation, URL: http://tremble.stanford.edu/cgi-bin/spo2ida-mt.pl

Vamvatsikos, D. and C. A. Cornell, 2002. “Practical Estimation of the Seismic Demand and
Capacity of Oscillators with Multi-Linear Static Pushovers through Incremental Dynamic
Analysis”, Proceedings of 7th U.S. National Conference on Earthquake Engineering, Boston, July
21-25.

88
Wallace, J.W., Stewart, J.P. and A.S. Whittaker, (eds) 1999. “Building Vulnerability Studies:
Modeling and Evaluation of Tilt-up and Steel Reinforced Concrete Buildings.” PEER-PG&E
Cooperative Research Program, University of California, Los Angeles, CA.

Wiemer, S., 2000. “Introducing probabilistic aftershock hazard mapping”, Geophysical Research
Letters, Vol. 27, pp. 3405-3408.
URL: http://www.statsresearch.co.nz/statsei2/Abstracts/Wiemer/Wiemer.html

Yeo, G.-L., 2004. Stochastic Characterization and Decision Bases under Time-Dependent
Aftershock Risk in Performance-Based Earthquake Engineering”, Ph.D. Dissertation, Dept. of
Civil and Environmental Engineering, Stanford University, Stanford, California, December.

Yeo, G.-L., and C.A. Cornell, 2004. “Building Tagging Criteria Based on Aftershock PSHA",
Proceedings of 13th WCEE, Paper 3283, Vancouver, B.C., Canada, August 1-6.

Yun, S.-Y., Hamburger, R.O., Cornell, C.A. and D. Foutch (2002). “Seismic Performance
Evaluation for Steel Moment Frames”, ASCE Journal of Structural Engineering, Vol. 128, No. 4,
pp. 534-545

89
 Appendix A

Extension to include building-to-building uncertainty within same building class

These guidelines are directly applicable to specific buildings for which a detailed NSP
curves are obtained according to the specifications provided here. The PG&E building
inventory considered here consists of about 200 structures belonging to several buildings
types (Table 1). It is likely that the cost of the engineering analyses necessary for the
application of these guidelines may be so high that these detailed studies might be
performed only for subset of such buildings. In this case, the procedure presented in this
document needs to be expanded to incorporate an extra source of uncertainty due to the
variability in response between buildings belonging to the same class (e.g., tilt-ups or mill
types). This appendix will briefly outline the issues involved, should this source of
uncertainty be included. No attempt, however, will be made to quantifying such
uncertainty. PG&E may want to conduct a separate study with this specific scope.

Any building within a building class is characterized by a “true” (but unknown) fragility
curve for each limit state, LS. This curve is fully characterized by a “central” (or median)
point, ŠaLS, and by the slope of the curve, which is controlled by the aleatory uncertainty
term, βR. The central point of this curve, ŠaLS, however, is itself uncertain in an epistemic
sense. The customary assumption is that ŠaLS follows a lognormal distribution with a
median value equal to (ŠaLS)med and a dispersion measure given by the epistemic
uncertainty term, βU. As briefly described in the commentary of Step 6, this provides a
first-order treatment of the uncertainty in the fragility curve and gives rise to a “family”
of fragility curves (Figure 12). The central point of each fragility curve in this family can
(
be computed by applying the equation S aS, y = ( S aLS ) med e yβU , where values of y are
tabulated in any Gaussian distribution table. For example, the median (50th), the16th-,
and the 84th-percentile estimates of this central value, S aS, y , can be found by replacing y
with the values of 0, -1, and +1, respectively, in the equation above. The slope of each
fragility curve in this building-specific family is assumed to be equal to βR. Therefore,
one could define a “median fragility curve” for a limit state LS as being the median
estimates of this family of fragility curves. The median fragility curve is characterized by
the median estimate of the central (or median) point, (ŠaLS)med, and by the slope which is
controlled by βR.

All the buildings within a building class can be considered as a statistical population.
Each building as a member of this population is characterized, as described above, by a
“true” but unknown fragility curve for the same generic limit state, LS. As explained
above, having done the proposed guidelines analysis, our available information about
each building can be described in terms of a median estimate of the fragility curve with
parameters, (ŠaLS)med and βR, and by an uncertainty measure of its central point given by
βU. In general the values of these three parameters vary from building to building within
the same class. In particular, here we call βpop the building-to-building variation of

90
 (
(ŠaLS)med across the buildings of the same class, and we call ( S aLS ) class
med the median value of

(Ša )med across the class of buildings. We also assume that we could compute (βR)avg and
LS

(βU)avg, the average value of the βR.and βU values for all the buildings in the class.
(
med , (βR)avg,, (βU)avg, and βpop, then we could represent the mean fragility
If we knew ( S aLS ) class
(
curve for the entire class as the curve whose central (median) point is given by ( S aLS ) class
med

and whose slope, βclass, is obtained via an SRSS operation of three β’s, namely
β class = ( β R ) 2avg + ( β U ) avg
2
+ β pop
2
. This curve would be similar to one of those
represented in Figure 11 but flatter to reflect the increased variability. If instead we
prefer to keep aleatory uncertainty separated from epistemic uncertainty, then the generic
building within a building class can be represented by a family of fragility curves whose
(
median fragility curve, as defined above, whose central point is ( S aLS ) class
med and its slope is
( LS class
controlled by (βR)avg. The building-to-building dispersion around ( S a ) med of the median
point of the fragility curves in this building-class-specific family is now
β Utot = ( β U ) avg
2
+ β pop
2
.

If the guidelines were to be applied to a specific building for which no engineering

analyses are performed, then the mean fragility curve derived for its building class could
(
be computed as explained above. However, we will not know the values of ( S aLS ) class med ,

(βR)avg,, (βU)avg, and βpop unless we sample the population (i.e., consider different
buildings in the class), compute the fragility curves (using these guidelines) of the sample
structures, and then use the results to estimate the four parameters.

In practice because each class consists of many buildings, a brute-force random sample of
the population may not be practical. Hence, it may be advisable to break each building
class in sub-populations by identifying parameters (e.g., age, size, irregularities, etc.) that
might affect the fragility curves of buildings within each sub-population, and then use
more advance sampling techniques, such as stratified or latin-hypercube sampling. Such
techniques may diminish the number of samples within a building class to a manageable
number to be feasible in practice. A preliminary study should be conducted to identify
the most effective parameters to be used to identify sub-population of buildings within
each building class.

91
 Appendix B

Evaluation of Epistemic Uncertainty for PG&E Buildings

Building categories

Category Criteria for Inclusion Approximate Number

Tilt-up or block • Identified as Tilt-Up or Concrete Block 18 tilt-up plus 15 block
with Wood/Metal Deck Roof in the buildings in the
PG&E inventory, and prioritization inventory
• 1 story or 1 story plus mezzanine, and included in Table1.
• ≥3000 square feet in area, and
• Structure does not have a diaphragm
discontinuity or non-parallel systems
(Table 16-M in UBC (1997))
Mill-type Identified as Mill-Type in the PG&E 35 older urban
inventory, consisting of older steel framing substations.
with concrete walls.
Prefabricated Identified as Prefabricated Metal Building 44 in the prioritization
metal in the PG&E inventory. inventory of Table 1.
All other Not meeting criteria for the above three 99 in the prioritization
categories. inventory of Table 1.

Evaluation uncertainty

As described in the guidelines, the minimum level of seismic evaluation for all buildings
includes a nonlinear static procedure analysis carried out according to the procedure
described in the main body of this document. The seismic evaluation includes a
performing nonlinear static procedure on the intact structure and additional nonlinear
static procedures on the structure assuming it has been subjected to different identifiable
damage states. Again, all seismic evaluations according to these guidelines should use
expected strength values rather than nominal or lower-bound strengths.

The levels of epistemic uncertainty evaluation are defined as either Baseline or Improved.
The Baseline case is expected to be used more often than the Improved case. Tables for
each building category give conditions and required seismic evaluations to change the
evaluation uncertainty level from Baseline to Improved. The conditions and required
evaluations are intended to identify the characteristics that most significantly affect the
uncertainty in a seismic evaluation for a given building type.

92
For all building categories, however, an Improved uncertainty evaluation always requires
that:

• The structural drawings be available.

• Building inspection be carried out and consistency with drawings be checked
and, in some cases, enforced38.
• Testing of archaic materials be performed in cases where the material
properties affect the seismic response.

If these requirements are not met, the epistemic uncertainty evaluation is Baseline.

Evaluation of uncertainty and βU values

Tilt-up or block buildings

The evaluation uncertainty is Improved if all the required evaluations that the engineers
deems important for estimating the structural response are carried out when the
conditions identified below are present. Otherwise the evaluation uncertainty is Baseline.
Note that some of the evaluations listed in the table may have to be considered in a
Baseline evaluation as well to ensure reasonably accurate response estimates

The βU values for this building category are reported in Table 2a for (recently) retrofitted
buildings and in Table 2b for un-retrofitted buildings. . The βU values given for Baseline
assume that some but not all of the evaluations have been carried out. . In intermediate
cases the engineer may use his/her judgment whether interpolation between Baseline and
Improved βU values are justified for the case considered.

Evaluations for TILT-UP OR BLOCK BUILDINGS

Condition Required evaluation
All buildings in Evaluate the expected strength of the roof diaphragm in relation
category to the expected strength of the wall-to-roof tension connection.

All buildings in Evaluate the potential for foundation overturning or rocking to

category control the in-plane behavior of walls.
All buildings in Evaluate the flexibility of the roof diaphragm in establishing the
category. period of vibration of the building and in creating the NSP
curve. The evaluation shall consider appropriate base conditions,
such as pinned or flexible, for the out-of-plane walls.
Re-entrant corner Evaluate collectors at re-entrant corners for the maximum force
per UBC (1997) that can be delivered to them by the system, and include

38
Note that if during inspection the engineer finds serious deficiencies in the lateral load resisting system
due to missing or damaged elements (e.g., in prefabricated metal buildings, to allow equipment installation
or door openings, rod braces are sometimes cut or removed, or baseplate bolts are missing) that were
included in the design drawings, he/she should require these elements be installed or replaced before using
Tables 2a to 2e to estimate epistemic uncertainty on the building response.

93
Table 16-M is collector effects in the pushover analysis.
present.
Metal-deck roof Evaluate roof diaphragm behavior according to test data on
metal deck diaphragms and an assessment of governing behavior
modes.
In-plane wall Evaluate wall in-plane behavior according to behavior modes
behavior affects the identified in FEMA 306 (Chapter 5 for Concrete and Chapter 6
pushover curve. for Block) and FEMA 307 (Chapter 2 for Concrete and Chapter
3 for Block).

Mill-type buildings
The evaluation uncertainty is Improved if all the required evaluations that the engineers
deems important for estimating the structural response are carried out when the
conditions identified below are present. Otherwise the evaluation uncertainty is Baseline.
Note that some of the evaluations listed in the table may have to be considered in a
Baseline evaluation as well to ensure reasonably accurate response estimates.

The βU values for this building category are reported in Table 2c. The βU values given for
Baseline assume that some but not all of the evaluations have been carried out. . In
intermediate cases the engineer may use his/her judgment whether interpolation between
Baseline and Improved βU values are justified for the case considered.

Evaluations for MILL-TYPE BUILDINGS

Condition Required evaluation
All buildings in category Evaluate the contribution of original steel braces, original
concrete walls, and any added seismic elements according
to their interaction in both the elastic and inelastic ranges of
behavior.
All buildings in category Evaluate in-plane composite action between steel framing
and boundary members and concrete walls or diaphragms.
In-plane wall behavior Evaluate in-plane behavior of concrete walls according to
affects the pushover behavior modes identified in FEMA 306 Chapter 5 and
curve. FEMA 307, Chapter 2.
Concrete roof diaphragm Evaluate in-plane behavior of roof diaphragm according to
behavior affects the behavior modes identified for walls in FEMA 306, Chapter
pushover curve 5 and FEMA 307, Chapter 2.

Prefabricated metal buildings

If the building contains any of the vertical structural irregularities of UBC (1997) Table
16-L or any of the horizontal structural irregularities of UBC (1997) Table 16-M, the
evaluation uncertainty is Baseline.

94
If there are no structural irregularities, the evaluation uncertainty is Improved if all the
required evaluations that the engineers deems important for estimating the structural
response are carried out when the conditions identified below are present. Otherwise the
evaluation uncertainty is Baseline. Note that some of the evaluations listed in the table
may have to be considered in a Baseline evaluation as well to ensure reasonably accurate
response estimates.

The βU values for this building category are reported in Table 2d. The βU values given for
Baseline assume that some but not all of the evaluations have been carried out. . In
intermediate cases the engineer may use his/her judgment whether interpolation between
Baseline and Improved βU values are justified for the case considered.

Evaluations for PREFABRICATED BUILDINGS

Condition Required evaluation
All buildings in category Field inspect the condition of lateral bracing and the bolting
of column base plates, and include in the nonlinear static
analysis the effects of any deficiencies observed.
All buildings in category Include the contribution of cladding in the nonlinear static
analysis, based on its behavior under in-plane lateral forces
and deformations.

All other buildings

The evaluation uncertainty is Baseline if either horizontal irregularities 3 or 4 from UBC
(1997) Table 16-M is present, or if horizontal irregularity 1 of UBC (1997) Table 16-M is
present (torsional irregularity), or if system in classified as torsionally unrestrained
according to Paulay (2000). The evaluation uncertainty is also Baseline if vertical
irregularities 1, 2, or 3 of UBC (1997) Table 16-L are present (soft story, weight
irregularity, or geometric irregularity). If none of these irregularities is present, the
evaluation uncertainty is as determined below.

The evaluation uncertainty is Improved if all the required evaluations that the engineers
deems important for estimating the structural response are carried out when the
conditions identified below are present. Otherwise the evaluation uncertainty is Baseline.
Note that some of the evaluations listed in the table may have to be considered in a
Baseline evaluation as well to ensure reasonably accurate response estimates.

The βU values for SMRF buildings are reported in Table 2e. The βU values for other
types of buildings belonging to this “all other”category can be estimated based on the βU
values for tilt-up, mill-type, and prefabricated metal buildings provided in Tables 2a to
2d. Interpolation may sometimes be in order. The βU values for mill-type buildings
could be considered as an upper bound for any structure in this category. Note that the βU
values given for Baseline assume that some but not all of the evaluations have been
carried out. . In intermediate cases the engineer may use his/her judgment whether

95
interpolation between Baseline and Improved βU values are justified for the case
considered.
Evaluations for ALL OTHER BUILDINGS
Condition Required evaluation
All buildings in category Consider in the evaluation building elements such as
partitions, infill walls, cladding, or gravity framing that
affect the seismic response of the structure. Such
elements are included in the evaluation whether their
effect is beneficial or detrimental and irrespective of
whether they have traditionally been identified as non-
structural elements or elements not part of the seismic-
force-resisting system.
Seismic-force-resisting Explicitly consider the potential behavior modes, using
elements are discontinuous, laboratory test results, of the elements that support the
or vertical irregularity 4 of discontinuous elements, in comparison to the maximum
UBC (1997) Table 16-L is force and displacement demands that can be imposed on
present (in-plane offset). the elements by the system.
Vertical irregularity 5 of Run the nonlinear static analyses with a range of vertical
UBC (1997) Table 16-L is distributions of lateral force, or run nonlinear time-
present (weak story). history analyses, to assess the potential for developing a
story mechanism and to evaluate the expected story
deformation as a function of global deformation.
Horizontal irregularity 2 of Evaluate collectors at re-entrant corners for the
UBC (1997) Table 16-M is maximum force that can be delivered to them by the
present (re-entrant corner). system, and include collector effects in the pushover
analysis.
Horizontal irregularity 5 of Carry out a three-dimensional nonlinear static analysis of
UBC (1997) Table 16-M is the structure considering a number of possible loading
present (nonparallel directions.
systems).

Elicitation of βU values

The βU values elicitation exercise was conducted for different cases that ranged from
relatively simple buildings analyzed using considerable modeling effort (e.g., a
prefabricated metal building at Improved level) to complicated buildings and limited
modeling effort (e.g., a mill type building at Baseline level). Values were elicited for the
Onset of Damage, Yellow, Red, and Collapse limit states.

To avoid any type of “anchoring” due to previous experience with quantification of

epistemic uncertainty not consistent with these guidelines (e.g., in HAZUS), the experts
were not asked directly for βU values but rather for estimates of structural deformation
values (e.g., roof drift) corresponding to incipient OD, Y, R, and Collapse limit states. In
the case of tilt-up buildings the experts preferred to provide their estimates in terms of the

96
spectral acceleration at incipient limit state rather than in terms of drift. The experts were
asked to provide estimates of median values and of 25% and 75% percentile values of
such deformations or spectral accelerations The experts were told to assess the 25% and
75% percentile values in such a way that they felt that the “true” value of the parameter
was equally likely to be inside or outside of that range. The experts provided their
estimates with the understanding that they previously had conducted thorough analyses of
the building as described by these guidelines. By design, the estimates did not include
engineer-to-engineer tagging variability. As per PG&E request, such variability will be
dealt with outside these guidelines by thoroughly training the PG&E building inspectors.

The βU values for each expert were then computed assuming that such deformations
followed a lognormal distribution. The βU values included in Tables 2a to 2e consist of
the weighted average values where each expert was given equal weight. The values
included in the tables were obtained using the weighted average of the βU values
computed by considering the drift ratios (or spectral acceleration ratios in the case of tilt-
ups) of the 25%-ile to the 50%-ile values, the 75%-ile to the 50%-ile values, and the
25%-ile to the 75%-ile values. The βU values computed using the three ratios are
somewhat different because the data elicited from the experts do not exactly follow a
lognormal distribution as hypothesized during the calculation.

97
 Appendix C

Copy of Most Relevant References

For convenience, a copy of the following papers is included:

• Bazzurro, P., Cornell, C.A., Menun, C., and M. Motahari, 2004. “Guidelines for Seismic
Assessment of Damaged Buildings", Proceedings of 13th WCEE, Paper 1708, Vancouver,
B.C., Canada, August 1-6.

• Luco, N., Bazzurro, P., and C.A. Cornell, 2004. “Dynamic Versus Static Computation of
the Residual Capacity of a Mainshock-Damaged Building to Withstand an Aftershock”,
Proceedings of 13th WCEE, Paper 2405, Vancouver, B.C., Canada, August 1-6.

• Yeo, G.-L., and C.A. Cornell, 2004. “Building Tagging Criteria Based on Aftershock
PSHA", Proceedings of 13th WCEE, Paper 3283, Vancouver, B.C., Canada, August 1-6.

98