Professional Documents
Culture Documents
2015 Guidelines-for-Nonlinear-Seismic-Analysis-of-Ordinary-Bridges-Version-2.0
2015 Guidelines-for-Nonlinear-Seismic-Analysis-of-Ordinary-Bridges-Version-2.0
2015 Guidelines-for-Nonlinear-Seismic-Analysis-of-Ordinary-Bridges-Version-2.0
December 2015
i
FUNDING AGENCY ACKNOWLEDGMENT
Support for this research was provided by the California Department of Transportation under
Research Contract No. 65A0454 (and amendments thereto), which is gratefully acknowledged.
We would like to acknowledge the valuable assistance and technical support of Caltrans staff in
this project, particularly Anoosh Shamsabadi. Caltrans Research Project Manager Peter Lee is
recognized for his assistance in contract administration.
DISCLAIMER STATEMENT
This document is disseminated in the interest of information exchange. The contents of this report
reflect the views of the authors who are responsible for the facts and accuracy of the data presented
herein. The contents do not necessarily reflect the official views or policies of the State of
California or the Federal Highway Administration. This publication does not constitute a standard,
specification or regulation. This report does not constitute an endorsement by the Department of
any product described herein.
For individuals with sensory disabilities, this document is available in Braille, large print,
audiocassette, or compact disk. To obtain a copy of this document in one of these alternate formats,
please contact: the Division of Research and Innovation, MS-83, California Department of
Transportation, P.O. Box 942873, Sacramento, CA 94273-0001.
ii
STATE OF CALIFORNIA DEPARTMENT OF TRANSPORTATION
CA15-65A0454
4. TITLE AND SUBTITLE 5. REPORT DATE
913
California Department of Transportation
Division of Research and Innovation, MS-83
1227 O Street
Sacramento CA 95814
15. SUPPLEMENTAL NOTES
Unclassified 167
Reproduction of completed page authorized
iii
iv
EXECUTIVE SUMMARY
This report presents Guidelines for Nonlinear Seismic Analysis of Ordinary Bridges: Version 2.0,
which was a joint effort by research teams from UC-Berkeley, UC-Irvine, and UCLA. The report
is based on the first generation of nonlinear analysis guidelines developed in 2008 through a PEER
Lifelines project. The main objective of this study was to extend further the first generation
guidelines by providing the following aspects: a) a comprehensive treatment of soil-structure
interaction effects, b) improved modeling of various bridge components (e.g., shear keys), c)
discussions on the identification of appropriate Ground Motion Selection and Modification
(GMSM) methods and their application, and d) improved solution strategies for nonlinear time
history analysis (NLTA). The focus of the guidelines is on Caltrans ordinary standard bridges. The
study was supplemented by extensive analyses conducted on three representative reinforced and
prestressed concrete bridges in California.
The study provided new macro-element models for ductile and brittle abutment shear keys that
were validated using available experimental data. These macro-elements were incorporated into
detailed structural models of selected typical ordinary bridges, and were utilized to quantify the
sensitivity of important Engineering Demand Parameters (EDPs) to various shear key and abutment
backfill model parameters. A physically parameterized hyperbolic force-deflection model was used
for the passive resistance of backfill normal to the backwall orientation. The effect of skew angle
in reducing passive resistance of the backfill soil was introduced in two different ways, namely a)
according to an assumed/heuristic backfill failure mechanism, and b) based on an empirical
relationship derived through available laboratory and field experiments. The lateral support that is
provided by the shear-keys in the transverse direction was modeled by considering two modes of
failure: brittle and ductile. Backbone curves of each type of response were developed based on the
dimensions and the recommended seismic reinforcement detailing of the shear-keys. It was
observed that brittle (i.e., sacrificial) shear keys increase the seismic demands (e.g., deck rotation)
which need to be controlled by the bridge columns. On the other hand, the ductile shear keys reduce
the seismic demands by controlling the deck movement, transfer large amount of forces to the
substructure, and eventually overload the abutments and piles. Accordingly, it is recommended to
provide a shear key with specific reinforcement details that behaves in a ductile manner to dissipate
energy with lower force capacity to avoid undesirable bridge damages and failures. It is also noted that
the existing provisions related to the sacrificial shear keys need to be revisited as the current values of
the coefficient of friction significantly underestimate the capacity of the sacrificial shear keys.
Regarding the backfills, the results of the study revealed that the backfill response of a skewed
abutment has a very significant effect on the seismic demands. Future research should focus on
investigating the backfill reaction against skew abutments so that existing models of this reaction are
validated or improved.
For modeling of the soil-structure interaction in deep foundation elements, foundation soil
was assumed to be composed of cohesive soil with upper weak strata underlaid by stiff clay. A
reduced-order modeling approach was adopted for the foundation where fiber-based beam
elements and nonlinear springs were used for modeling pile and foundation soil, respectively. The
findings indicated that the superstructure’s seismic responses are somewhat sensitive to the
features of the deep foundation elements. However, various simplifications can be made depending
on the soil conditions, foundation type, and the targeted performance level.
Solutions were sought to overcome convergence problems associated with the NLTA of
ordinary standard bridges. Two different analysis approaches were considered for this purpose.
v
The first approach was the use of integration methods alternative to the commonly (including the
bridge analyses conducted by Caltrans engineers) used Implicit Newmark (IN) integration, namely
the Explicit Newmark (IN), Operator-Splitting (OS), and the so-called TRBDF2 methods. In this
approach, the eigen properties of these alternative integrators were theoretically investigated for
the nonlinear response evaluation. Moreover, the accuracy of these integration methods was
investigated using the studied bridge models and a nonlinear test problem with an available closed
form solution. The second approach was the improvement of the robustness of IN, in the cases
where the considered alternative integrators are not applicable. The enhancement of the robustness
of IN was investigated with five considered parameters: nonlinear solvers, convergence tests,
convergence tolerances, integration time steps, and finally adaptive switching of integration
algorithms. The NLTA simulations conducted on the studied bridge models showed that the OS
and TRBDF2 algorithms provided very close results to the IN algorithm. The TRBDF2 algorithm
and adaptive switching of algorithms showed improved convergence performance as compared to
IN. Accordingly, it is concluded that the OS method, the TRBDF2 method and adaptive switching
of different integration methods can be considered as suitable alternatives to the IN method for
NLTA of highway bridges.
Accuracy and reliability evaluations of various GMSM methods, namely the first mode
spectral acceleration method, the Conditional Mean Spectrum (CMS) method, and the
Unconditional Selection (US) method, were conducted. The GMSM methods were evaluated using
the High End Prediction (HEP) approach. Three sample bridges with and without the consideration
of nonlinear soil structure interaction (SSI) effects were used in the evaluation process. Different
abutment skew angles as well as regular bridges, i.e. with a skew angle of zero, were considered.
The accuracy of the GMSM methods in predicting the median values of different EDPs was
evaluated. It was observed that none of the GMSM methods is particularly superior to the others.
Ordinary standard bridges play a crucial role in transportation and thus require short downtime
after severe earthquakes regarding both the emergency response and community resiliency aspects.
Therefore, it is suggested to use the slightly conservative US method for selection and modification
of ground motions for different intercept angles and to choose the most critical intercept angle that
yields the largest response.
It is recommended to use the improved modeling, analysis and ground motion application
methods produced in this study for performance-based earthquake engineering (PBEE) of Caltrans
ordinary bridges in order to develop related bridge PBEE guidelines. Moreover, it is suggested to
extend these guidelines in the future to address modeling and analysis issues of special bridges,
e.g. Ordinary Nonstandard and Important bridges. Such bridges may include numerous long and
curved spans that introduce severe geometric irregularities. Furthermore, for long-span and
irregular bridges, multi-support (spatially varying) excitation should be investigated.
vi
TABLE OF CONTENTS
1 Introduction ............................................................................................................................. 1
1.1 Overview .......................................................................................................................... 1
1.2 First Generation Nonlinear Analysis Guidelines ............................................................. 1
1.3 Reaserch Plan ................................................................................................................... 3
1.4 Report Outline .................................................................................................................. 4
2 An Overview of Performance-Based Earthquake Engineering Methodologies ...................... 5
2.1 Introduction ...................................................................................................................... 5
2.2 PEER PBEE Methodology ............................................................................................... 5
3 Bridge Structures and Analytical Models of Version 1.0 ....................................................... 9
3.1 Introduction ...................................................................................................................... 9
3.2 Selected Bridges ............................................................................................................... 9
3.3 Geometrical Characteristics ............................................................................................. 9
3.4 First Generation Guidelines for Modeling Ordinary Bridges ........................................ 10
4 Model Components ............................................................................................................... 14
4.1 Introduction .................................................................................................................... 14
4.2 Column Modeling .......................................................................................................... 14
4.3 Abutment Modeling ....................................................................................................... 15
4.4 Further Modeling Assumptions...................................................................................... 32
4.5 Modeling Soil-Structure Interaction in Shafts ............................................................... 33
4.6 In-Span Hinges ............................................................................................................... 40
5 Robust Integration and Solution Algorithms Selection ......................................................... 50
5.1 Introduction .................................................................................................................... 50
5.2 Integration Methods ....................................................................................................... 50
5.3 Eigen Property ................................................................................................................ 57
5.4 Integration Accuracy ...................................................................................................... 67
5.5 Applicability of the Alternative Integration Methods to The Bridge Models ................ 69
5.6 Improving Convergence of Implicit Newmark Integration............................................ 73
5.7 Vision for Implementation of Robust Integration and Solution Algorithms ................. 82
6 Ground Motion Selection and Modification ......................................................................... 83
6.1 Introduction .................................................................................................................... 83
6.2 Objective ........................................................................................................................ 83
6.3 Ground Motion Scenario ................................................................................................ 84
vii
6.4 Candidate GMSM Methods ........................................................................................... 85
6.5 Predominantly First Mode Engineering Demand Parameter (EDP) .............................. 91
6.6 Development of Point of Comparison: High End Prediction (HEP) ............................. 94
6.7 Evaluation of GMSM Methods ...................................................................................... 97
6.8 An Alternative Ground Motion Suite and Treatment................................................... 103
7 Case Studies ........................................................................................................................ 106
7.1 Introduction .................................................................................................................. 106
7.2 Ground Motion Selection and Application .................................................................. 107
7.3 Collapse Criteria........................................................................................................... 108
7.4 Probabilistic Approach for the Assessment of Demand and Fragility ......................... 108
7.5 Sensitivity to Various Modes of Shear-Key Failure .................................................... 109
7.6 Sensitivity to Backfill Geotechnical Properties and Epistemic Uncertainties in the Skew
Abutment Model.................................................................................................................. 117
7.7 Sensitivity to Foundation Model Characteristics ......................................................... 131
8 Guidelines and Recommendations for Modeling and Analysis .......................................... 147
8.1 Recommendations for the Abutment Backfill Model .................................................. 147
8.2 Recommendation for the Abutment Shear Key Model ................................................ 148
8.3 Recommendation for Soil-Structure Interaction in Deep Foundation Elements.......... 149
8.4 Recommendation for Robust Integration and Algorithms Selection ........................... 150
8.5 Recommendation for Ground Motion Selection and Modification (GMSM).............. 152
Appendix A Example Tcl File ............................................................................................. 166
viii
LIST OF FIGURES
Figure 2.1 Outline of the PEER PBEE Methodology. .................................................................... 7
Figure 3.1 Modeling of Bridge B [Kaviani et al. 2012]................................................................ 11
Figure 3.2 Column modeling scheme for Bridge B [Kaviani et al. 2012]. ................................... 12
Figure 3.3 Springs and gap elements used to model various components of the abutment: a) Type
I Modeling [Kaviani 2011], b) Type II Modeling [Aviram et al. 2008]. ...................................... 13
Figure 3.4 Shear key force-deformation backbone curve [Kaviani 2011].................................... 13
Figure 3.5 Configuration of a typical seat-type abutment [Kaviani 2011]. .................................. 13
Figure 4.1 beamWithHinges element: (a) Gauss-Radau Integration (b) modified Gauss-Radau
integration [McKenna et al. 2010]. ............................................................................................... 14
Figure 4.2 Column modeling: (a) Lumped plasticity column modeling [Berry and Eberhard
2008], (b) Moment deformation backbone for single column in OpenSees [McKenna et al. 2010]
using Beam with Hinges Element. ................................................................................................ 15
Figure 4.3 Effective abutment stiffness [Caltrans SDC 1.7]. ....................................................... 16
Figure 4.4 Shear key details based on Caltrans SDC 1.7: (a) brittle shear key, (b) ductile shear
key. ................................................................................................................................................ 19
Figure 4.5 Sliding shear failure mechanism and parameters. ....................................................... 19
Figure 4.6 Shear key models: (a) ductile, (b) brittle. .................................................................... 20
Figure 4.7 LSH intermediate and ultimate log-spiral failure surfaces for (a) Category I, (b)
Category II, (c) Category III, and (d) Category IV backfills (1 m = 39.37 in.). ........................... 27
Figure 4.8 Passive resistance of a unit width of a 1.9m-high abutment backfill due to SDC 1.7
bilinear backbones as well as LSH and GHFD models for typical backfills in California highway
bridges (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). ....................................................................... 28
Figure 4.9 Linear variation of backfill passive resistance from the obtuse to the acute corner of a
skewed abutment. .......................................................................................................................... 29
Figure 4.10 Reduction of backfill volume normal to backwall, from the obtuse to the acute
corner of a skewed abutment. ....................................................................................................... 30
Figure 4.11 Variation of passive resistance over the reduced-resistance region in the vicinity of
acute corner in skewed abutments. ............................................................................................... 31
Figure 4.12 Fiber-based element model of a circular reinforced concrete pile. ........................... 33
Figure 4.13 Macroelement for modeling passive pressure of soil in cyclic loading [Boulanger et
al., 1999]. ...................................................................................................................................... 34
Figure 4.14 Force-deformation relationship in OpenSees p-y macroelement. ............................. 35
Figure 4.15 Distribution of passive soil pressure on different cross-sections [Reese and Van
Impe, 2009]. .................................................................................................................................. 35
Figure 4.16 Pile-to-cap connection details: (a) fixed connection, (b) pinned connection [Silva and
Seible 2001]. ................................................................................................................................. 37
Figure 4.17 Typical details of a HP-steel pile-to-cap connection [Xiao et al. 2006]. .................. 38
Figure 4.18 Recommended details of HP-steel pile-to-cap connection [Xiao and Chen 2012]. .. 38
Figure 4.19 (a) Simplified contact model, (b) Impact material in OpenSees [Muthukumar &
DesRoches, 2006]. ........................................................................................................................ 40
Figure 4.20 Typical restrainers in ISH in Bridge [Muthukumar 2003]. ....................................... 41
Figure 4.21 Reinforcement details and in-span hinge geometry [Hube and Mosalam 2008]. ..... 43
ix
Figure 4.22 Idealized shear deformation relationship for ISH (based on Hube and Mosalam
2008). ............................................................................................................................................ 44
Figure 4.23 One-dimensional shear failure mechanism of in-span hinge. ................................... 45
Figure 4.24 Experimental Stress-Load relationship of specimen S1 and S2 [Hube and Mosalam
2008]. ............................................................................................................................................ 46
Figure 4.25 Comparison of load-displacement envelops at bearings of tested specimens [Hube
and Mosalam 2010]....................................................................................................................... 47
Figure 4.26 Load-deformation relationship for specimen S1: (a) experimental results, (b)
analytical model based on [Hube and Mosalam, 2008]. ............................................................... 47
Figure 4.27 Load-deformation relationship for specimen S2: (a) experimental results, (b)
analytical model based on [Hube and Mosalam 2008]. ................................................................ 48
Figure 4.28 Two Failure mechanisms in interior shear key based on [Megally et al., 2001]. ...... 48
Figure 4.29 Two Failure mechanisms in interior shear key based on [Megally et al., 2001]. ...... 49
Figure 4.30 Interior shear key model in OpenSees. ...................................................................... 49
Figure 5.1 A two DOF system used for eigen analysis of the OS integration. ............................. 63
Figure 5.2 Schematic of the nonlinear pendulum in a general deformed state. ............................ 68
Figure 5.3 Period elongation and amplitude decay for the pendulum problem with θ0=0.10π. ... 69
Figure 5.4 Period elongation and amplitude decay for the pendulum problem with θ0=0.50π. ... 69
Figure 5.5 Superstructure, column bent and abutment of Bridge B with identified nodes for
NLTA results [Kaviani 2011]. ...................................................................................................... 70
Figure 5.6 Maxerror of the OS algorithm for the three selected EDPs of Bridge B. .................. 71
Figure 5.7 Comparison of the IN and OS algorithm results for NLTA of Bridge B with Type II
abutment modeling (Ground motion #21 as an example). ............................................................ 72
Figure 5.8 Maxerror of the TRBDF2 algorithm for the three selected EDPs of Bridge B. .......... 73
Figure 5.9 Maxerror of each tolerance for zerolength element associated with node 100
(longitudinal displacement). ......................................................................................................... 80
Figure 5.10 Maxerror of each tolerance for node12 (longitudinal displacement). ....................... 80
Figure 6.1 Response spectra for a scenario earthquake for Bridge B site. ................................... 85
Figure 6.2 Campbell and Bozorgnia (CB) 2008 spectrum and CMS for 10% POE in 50 years for
Bridge B site. ................................................................................................................................ 92
Figure 6.3 Ratios of median EDPs for the two abutment modeling cases of Bridge A. ............... 93
Figure 6.4 Ratios of median EDPs for the two abutment modeling cases of Bridge B. ............... 94
Figure 6.5 Ratios of median EDPs for the two abutment modeling cases of Bridge C. ............... 94
Figure 6.6 Investigation of the intercept angle effect on EDP for the two abutment modeling
cases of Bridge B. ......................................................................................................................... 96
Figure 6.7 Investigation of the intercept angle effect on EDP for three abutment skew angles of
Bridge B with enhanced modeling. ............................................................................................... 97
Figure 6.8 Investigation of the adequacy of GMSMs in predicting the maximum column base
shear for the two abutment modeling cases of Bridge B. ............................................................. 98
Figure 6.9 Investigation of the adequacy of GMSMs in predicting the maximum column base
shear for three abutment skew angles of Bridge B with enhanced modeling. .............................. 99
Figure 6.10 Investigation of the adequacy of GMSMs in predicting the maximum column top
curvature for the two abutment modeling cases of Bridge B...................................................... 100
x
Figure 6.11 Investigation of the adequacy of GMSMs in predicting the maximum abutment
unseating displacement for the two abutment modeling cases of Bridge B. .............................. 100
Figure 6.12 Investigation of the adequacy of GMSMs in predicting the maximum column drift
ratio for the two abutment modeling cases of Bridge B. ............................................................ 101
Figure 6.13 Investigation of the adequacy of GMSMs in predicting the maximum column drift
ratio for three abutment skew angles of Bridge B with enhanced modeling. ............................. 101
Figure 6.14 Investigation of the adequacy of GMSMs in predicting the maximum abutment
unseating displacement for three abutment skew angles of Bridge B with enhanced modeling. 102
Figure 6.15 Investigation of the adequacy of GMSMs in predicting the maximum column top
curvature for three abutment skew angles of Bridge B with enhanced modeling. ..................... 103
Figure 6.16 Soil modulus degradation and damping curves cohesive and cohesionless soils
[EPRI, 1993]. .............................................................................................................................. 105
Figure 7.1 Definitions of seat length in skewed abutments. ....................................................... 108
Figure 7.2 Effect of shear key and backfill model on column ductility demand of Bridge A. ... 111
Figure 7.3 Deck displacement with respect to the abutments in the longitudinal and the
transverse directions of Bridge A. .............................................................................................. 112
Figure 7.4 Effect of shear key and backfill model on deck rotation of Bridge A. ...................... 113
Figure 7.5 Effect of backfill model on bridge response of bridge A: (a) non-uniform backfill, (b)
uniform backfill. ......................................................................................................................... 113
Figure 7.6 Effect of shear key and backfill model on column ductility demand of Bridge B. ... 114
Figure 7.7 Deck displacement with respect to the abutments in longitudinal and transverse
direction of Bridge B. ................................................................................................................. 114
Figure 7.8 The effects of shear key and backfill models on the deck rotations of Bridge B. ..... 115
Figure 7.9 Deck unseating fragility curve conditioned on seat-width for Bridge A................... 116
Figure 7.10 Deck unseating fragility curve conditioned on seat-width for Bridge B. ................ 117
Figure 7.11 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category I
GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). ............................................ 119
Figure 7.12 GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). ......................... 119
Figure 7.13 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category III
GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). ............................................ 120
Figure 7.14 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category IV
GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). ............................................ 120
Figure 7.15 Fragility curves for bridges with SDC 1.7 bilinear and Category I to IV GHFD
backfill model (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.). .......................................................... 122
Figure 7.16 Frequency of (a) column and (b) abutment collapses in bridges with SDC 1.7
bilinear and Category I to IV GHFD backfill models................................................................. 122
Figure 7.17 Schematic illustration of reaction forces on bridge deck for (a) longitudinal
translation, (b) transverse translation, (c) counter-clockwise rotation, and (d) clockwise rotation
of deck. ........................................................................................................................................ 123
Figure 7.18 Distribution of deck rotation in bridges with SDC 1.7 bilinear backfill model. ..... 124
Figure 7.19 Distribution of deck rotation in bridges with Category I GHFD backfill model. ... 124
Figure 7.20 Distribution of deck rotation in bridges with Category II GHFD backfill model. .. 125
Figure 7.21 Distribution of deck rotation in bridges with Category III GHFD backfill model. . 125
xi
Figure 7.22 Distribution of deck rotation in bridges with Category IV GHFD backfill model. 125
Figure 7.23 Fragility curves for acute- and obtuse-corner shear-keys in bridges with Category I
and IV GHFD backfill models. ................................................................................................... 126
Figure 7.24 Fragility curves for bridges with uniform and non-uniform reduction of passive
resistance in skew abutments with Category I and II GHFD backfill models. ........................... 127
Figure 7.25 Frequency of (a) column and (b) abutment collapse in bridges with Category I and II
GHFD backfill models and uniform reduction of passive resistance in skewed configurations. 128
Figure 7.26 Distribution of deck rotation in bridges with Category I GHFD backfill model and
uniform reduction of passive resistance in skewed configurations. ........................................... 128
Figure 7.27 Distribution of deck rotation in bridges with Category II GHFD backfill model and
uniform reduction of passive resistance in skewed configurations. ........................................... 129
Figure 7.28 Fragility curves for acute- and obtuse-corner shear-keys in bridges with Category I
and II GHFD backfill models and uniform reduction of passive resistance in skewed
configurations. ............................................................................................................................ 129
Figure 7.29 Fragility curves for bridges with SDC 1.7 bilinear and Category I to IV GHFD
backfill and ductile shear-key models......................................................................................... 131
Figure 7.30 Frequency of (a) column and (b) abutment collapse in bridges with SDC 1.7 bilinear
and Category I to IV GHFD backfill and ductile shear-key models........................................... 131
Figure 7.31 Moment-curvature relationship (under no axial force) of pile sections in the
specimen Bridges A and B. ......................................................................................................... 132
Figure 7.32 Variation of (1) peak ground acceleration, (2) spectral acceleration at 1 s, and (3)
maximum spectral acceleration of strike normal (top row) and strike parallel (bottom row)
components of ground motion, between ground surface and -40 ft below the surface. ............. 134
Figure 7.33 Variation of period corresponding to maximum spectral acceleration of a) strike
normal and b) strike parallel components of ground motion, between ground surface and -40 ft
below the surface. ....................................................................................................................... 135
Figure 7.34 Abutment longitudinal (top row) and transverse (bottom row) unseating in Bridges
(a) A and (b) B. ........................................................................................................................... 136
Figure 7.35 Column drift ratio in Bridges (a) A and (b) B. ........................................................ 138
Figure 7.36 Footing displacement in Bridges (a) A and (b) B.................................................... 139
Figure 7.37 Relative rotation between the top and bottom of column in Bridge B. ................... 140
Figure 7.38 Effect of axial load on the moment-curvature relationship of the RC pile in Bridge B.
..................................................................................................................................................... 140
Figure 7.39 Bridge A pile maximum longitudinal moment and curvature; (a) CIDH-RF-MSE, (b)
CIDH-RF-UE, and (c) HP-RF-MSE. (*Note the different scale on the curvature axis). ........... 142
Figure 7.40 Effect of axial load on the moment-curvature relationship of the RC pile section
(Bridge B). .................................................................................................................................. 143
Figure 7.41 Bridge B pile maximum longitudinal moment and curvature; a) CIDH-RF-MSE, b)
CIDH-HF-MSE, and c) CIDH-HF-UE. ...................................................................................... 144
Figure 7.42 Bridge B pile maximum transverse moment and curvature; (a) CIDH-RF-MSE, (b)
CIDH-HF-MSE, and (c) CIDH-HF-UE ...................................................................................... 145
Figure 7.43 Deformation of Bridge B longitudinal (top row) and transverse (bottom row) p-y
springs beyond yield point; (a) CIDH-RF-MSE, (b) CIDH-HF-MSE, and (c) CIDH-HF-UE. . 146
xii
Figure 7.44 Normalized backbone of Bridge B longitudinal p-y springs (shown only at the
location of peak “maximum to yield deformation ratio” of foundation soil). ............................ 146
xiii
LIST OF TABLES
Table 1.1 Components modeling [Aviram et al. 2008]. ................................................................. 2
Table 3.1 Characteristics of the selected bridges for this study. ................................................... 10
Table 4.1 Experimental results versus model predictions for a brittle shear key. ........................ 22
Table 4.2 Experimental results versus model predictions for a ductile shear key. ....................... 22
Table 4.3 Backfill soil type categories in California highway bridges. ........................................ 26
Table 4.4 Ratio of reduced-resistance width (Wr) to tributary width (Wt) of uniformly distributed
abutment springs in the specimen bridge model (Bridge B)......................................................... 30
Table 4.5 Concrete Properties [Hube and Mosalam 2008]. .......................................................... 42
Table 4.6 Average rebars properties [Hube and Mosalam 2008]. ................................................ 42
Table 4.7 Theoretical Strength Estimation [Hube and Mosalam 2008]. ...................................... 42
Table 4.8 Parameters for specimen S1 & S2. ............................................................................... 43
Table 5.1 Conditions for ρ(Ai)≤1 .................................................................................................. 62
Table 5.2 Coefficients for the OS method for a 2 DOF system using the tangential stiffness. .... 64
Table 5.3 Nonlinear pendulum ..................................................................................................... 68
Table 5.4 The convergence failure time [sec] of simulations for different initial nonlinear solvers
under GM31 for Bridge A with Type I abutment model. ............................................................. 78
Table 5.5 Total number of iterations for simulations with different convergence tests under
GM31. ........................................................................................................................................... 78
Table 5.6 Maxerror in long. deformation of node 100 for GM31 (different convergence tests). ... 79
Table 5.7 The convergence failure time [sec] of simulations for different integration time steps
under GM31 for Bridge A............................................................................................................. 81
Table 5.8 The convergence failure time [sec] of simulations for different implicit integration
methods. ........................................................................................................................................ 82
Table 7.1 Combination of various models of backfill and shear-key used in sensitivity analyses.
..................................................................................................................................................... 118
Table 7.2 Uniform reduction factors for the passive resistance of backfill at 15o to 60o skew
angles. ......................................................................................................................................... 127
Table 7.3 Geotechnical properties of foundation soil (composed of clay), used for deciding
properties of soil springs. ............................................................................................................ 132
Table 7.4 Input parameters for SHAKE91 deconvolution analysis. ........................................... 133
Table 7.5 Bridge matrix for sensitivity analysis. ........................................................................ 135
Table 7.6 Schematic analogy between simplified column models and cases of base-fixity and
column-to-pile-cap connection in Bridge B. ............................................................................... 138
xiv
1 Introduction
1.1 OVERVIEW
The outcome of the comprehensive nonlinear analysis guidelines developed for Caltrans ordinary
standard bridges are presented in this report. These new guidelines are intended to improve upon
the first generation nonlinear analysis guidelines previously developed through a project funded
through the Pacific Earthquake Engineering Research (PEER) Lifelines program [Aviram et al.
2008]. In the present study, the previous guidelines are extended by providing a comprehensive
treatment of soil-structure interaction effects, improved modeling of various bridge components
(e.g., shear keys), ground motion hazard modeling, and nonlinear dynamic analysis methodologies
for Caltrans ordinary standard bridges. Special attention is given to development of methodologies
and component models that will enable Caltrans engineers to develop sophisticated bridge models
at multiple tiers of complexity. Detailed sensitivity analyses are also provided to assist Caltrans
engineers in making informed decisions regarding the appropriate level of bridge model
complexity for different design and assessment tasks.
1.2.1 Outcomes
The Guidelines for Nonlinear Analysis for Bridge Structures in California [Aviram et al. 2008]
present a collection of general recommendations for the modeling and analysis of Caltrans standard
ordinary bridges subjected to earthquake ground motions, with the goal to evaluate the demands
imposed on the critical bridge components and systems. The contents of this report are summarized
in the following paragraphs.
The introductory chapters (i.e., Ch. 2 and 3) identify the relevance and importance of
nonlinear analysis procedures in bridge structures, including the advantages and drawbacks over
simpler linear analysis. Different types of nonlinearities to be incorporated in the analytical bridge
model are described briefly, with a list (Table 1.1) of the critical components of the structure that
require detailed inelastic modeling to guarantee a desired level of accuracy. The appropriate model
dimension (2D versus 3D) recommended for the application of nonlinear analysis procedures is
also justified in detail.
The bridge modeling chapter, Chapter 3, establishes a set of recommendations to simplify
the geometry of a real bridge structure, definition of elements and materials, and the assignment
1
of mass and boundary conditions, among others. A thorough explanation is presented that
addresses the minimum requirements in the modeling of the plastic hinge zone in bridge column
bents. The nonlinear behavior issues related to bridge abutments, foundations, and expansion joints
located along the superstructure are briefly discussed.
The focus of Chapters 5 and 6 is on bridge analysis. In these chapters, all the relevant
procedures and parameters to simulate the seismic demands on bridge structures in the form of
imposed static and dynamic forces or displacements are described in detail. These chapters provide
the necessary methodologies that will allow the design engineer to conduct modal, gravity load,
pushover, response spectra, and time history analyses, and to interpret the resulting response data
from those analyses. References are provided to other resources for the use of response spectrum
curves, selection and scaling of ground motions, and definition of additional parameters required
for the different types of nonlinear analyses approaches.
The report also presents ample recommendations for linear and nonlinear analysis of bridge
structures appropriate for any structural analysis program, as well as specific details on the use of
SAP2000 [Wilson and Habibullah, 1998] for such procedures. Additionally, a general review and
definitions related to structural dynamics, applicable to both linear and nonlinear analyses, are
presented throughout the document. It is noted that the emphasis is on the implementation of
nonlinear analysis procedures used primarily for the estimation of the demand on a bridge structure.
These first generation guidelines are intended for use on standard ordinary bridges designed by
2
and for the California Department of Transportation according to the 2004 Caltrans Seismic
Design Criteria (SDC), Section 1.1, reflecting the current state of practice at Caltrans.
1.2.2 Shortcomings
The aforementioned report [Aviram et al., 2008] addresses two types of nonlinearities: (1) material
nonlinear behavior in plastic hinge regions; and (2) geometric nonlinear behavior due to large
deformations (second-order effects). At the bridge system level, these guidelines address standard
ordinary bridges with no significant geometric irregularities (such as skewed or curved decks,
unequal or long spans, uneven mass distribution, numerous expansion joints).
Linear analysis of soil-structure interaction (SSI), linear foundation springs, and simplified
models for abutments and expansion joints are the principal shortcomings of these first generation
guidelines. The second-generation guidelines discussed in this report address the broad issue of
nonlinear SSI, and examine the modeling and analysis needs for columns, shear keys, abutments,
embankments, and piles in detail. Greater attention is paid to the nonlinear solvers and time
integration techniques to assure robustness of the nonlinear time history solution algorithms.
3
University of California, Berkeley
Development of practical recommendations regarding the use of various methods of
ground motion selection and scaling for various types of bridges addressed in this project
(Identified as Task 4).
Development of recommendations for selection of robust solution algorithms and
integration methods from the existing OpenSees portfolio and developing procedures using
existing safeguarding methods (such as line searches, sub-incrementation, and generalized
displacement control) to ensure accurate and confident modeling of SSI for standard
ordinary bridges and the best practices aimed at managing the computational effort and
solution convergence checks involved in these potentially long-duration analyses.
(Identified as Task 5).
4
2 An Overview of Performance-Based
Earthquake Engineering Methodologies
2.1 INTRODUCTION
Traditional earthquake design philosophy is based on preventing structural damage in low-
intensity earthquakes, limiting the damage in to repairable levels in medium-intensity earthquakes,
and preventing the overall or partial collapses in high-intensity earthquakes. After 1994 Northridge
and 1995 Kobe earthquakes, the structural engineering community realized that the amount of
damage, the economic loss due to downtime, and repair cost of structures were unacceptably high,
even though those structures complied with available seismic codes based on traditional design
philosophy [Lee and Mosalam, 2006].
The concept of performance-based earthquake engineering (PBEE) has its roots from the
above-mentioned realization. Vision 2000 report [SEAOC, 1995] is one of the early documents of
the first generation PBEE in USA and addressed building structures. In this report, performance-
based earthquake design (PBED) is defined as a design framework, which results in the desired
system performances at various intensity levels of seismic hazard. The system performance levels
are classified as fully operational, operational, life safety, and near collapse, while hazard levels
are classified as frequent, occasional, rare, and very rare events. The designer and owner consult
to select the desired combination of performance and hazard levels to use as design criteria
(objectives). The intended performance levels corresponding to different hazard levels are either
determined based on the public resiliency requirements (e.g., hospital buildings), or by the private
property owners (e.g., residential or commercial buildings). Subsequent documents of the first
generation PBEE; namely ATC-40 [ATC, 1996], FEMA-273 [BSSC, 1997], and FEMA-356
[ASCE, 2000] express the design objectives using a similar framework, with slightly different
performance descriptions and hazard levels. The element deformation and force acceptability
criteria corresponding to the performance are specified for different structural and non-structural
elements for linear, nonlinear, static, and/or dynamic analyses. These criteria do not possess
probability distributions, i.e. the element performance evaluation is deterministic. The defined
relationships between engineering demands and component performance criteria are based
somewhat inconsistently on relationships measured in laboratory tests, calculated by analytical
models, or assumed on the basis of engineering judgment [Moehle, 2003]. In addition, the element
performance evaluation is not tied to a global system performance.
5
probabilistic manner where uncertainties in earthquake ground motion, structural response,
damage, and losses are explicitly considered. Main enhancements of the PEER PBEE
methodology with respect to the first generation methods are as follows:
Performance of a structure is determined in a rigorous probabilistic manner by considering
all sources of uncertainty that affect the performance.
Performance is defined with decision variables (DV), which reflect global system
performance.
Performance is defined with DVs in terms of the direct interest of various stakeholders.
PEER performance assessment methodology has been summarized in various publications
[Moehle 2003, Krawinkler 2002, Krawinkler and Miranda 2004, Moehle and Deierlein 2004,
Porter 2003] and various benchmark studies have been conducted [Comerio 2005, Goulet et al.
2006, Krawinkler 2005, Kunnath 2006, Mitrani-Reiser et al. 2006, Lee and Mosalam 2005].
Recently, Günay and Mosalam [2013] summarized and explained the methodology in a simplified
manner for its adoption by the broad engineering community. Below is a brief overview of the
methodology from this document.
PEER PBEE methodology consists of the four analysis stages shown in Figure 2.1, i.e.
hazard, structural, damage, and loss. The methodology focuses on the probabilistic calculation of
system performance measures meaningful to facility stakeholders by considering the four stages
of analysis in an integrated manner, where uncertainties are explicitly considered in all stages. As
shown in Figure 2.1, the outcome of each of the four stages is either a probability (or probability
of exceedance, POE) distribution. The probabilities determined in each stage are combined using
Equation 2.1.
The damageable parts of a facility are divided into damageable groups consisting of
components affected by the same EDP in a similar manner. Global collapse of a structure is treated
separately in this methodology since its probability does not change from one damageable group
to another. Equation 2.1, which resemble the well-known triple integration, referred to as the PEER
PBEE framework equation in [Moehle 2003, Krawinkler 2005], is applicable only for the case of
a single damageable group and no global collapse. A fourth summation is included to consider the
presence of different damageable groups in Equation 2.2. The most general format of the
formulation is given in Equation 2.3 for the case of multiple damageable groups and global
collapse.
6
Facility Definition: Location and Design
Hazard Analysis
P (IM) in t years
p(IM) in t years
Intensity measure (IM) Intensity measure (IM)
Structural Analysis
For each value (IMm) of the
PDFs
p( EDPj IMm)
intensity measure IM: m=1:
j=1:
Conduct nonlinear time
: # of IMs
history analyses with the
ground motions selected for : # of EDPs
IM=IMm Eng. demand param. (EDPj)
Damage Analysis
j= 1: # of damageable groups (= # of EDP’s) i= 1: # of data points for EDPj
P(DMEDPji)
p(DMEDPji)
p(DMEDPj)
&
k fragility functions
k=1: # of DM levels (n)
Eng. demand param. (EDPj) DM1 ... DMn DM1 DM2 ... DMn
Loss Analysis
(2.1)
(2.2)
(2.3)
7
where p(IMm) is the probability of the mth value of the earthquake intensity measure (IM),
determined as an outcome of hazard analysis, p(EDPji|IMm) is the probability of the ith value of the
EDP utilized for the jth damageable group, when the mth value of IM occurs (outcome of structural
analysis), p(DMk|EDPji) is the probability of the kth Damage Measure (DM) when subjected to the
ith value of the EDP utilized for the jth damageable group (outcome of damage analysis) and
P(DVjn|DMk) is the POE of the nth value of the DV for the jth damageable group when the kth DM
occurs (outcome of loss analysis). Moreover, p(C|IMm) and p(NC|IMm) are the probabilities of
having and not having global collapse, respectively, under ground motion intensity IMm. Finally,
P(DVn|C) is the POE of the nth value of DV in the case of global collapse. Index j is dropped in
Equation 2.1 since this equation represents the case of a single damageable group.
The POE of the nth value of a DV in Equation 2.2, P(DV n ) , is interpreted as a weighted
average of the POE of DVs from all possible cases, P(DVjn DM k ) , where each case is defined by
the probability of specific IM, EDP and DM—i.e., p(DM k EDPji ) p(EDPji IM m ) p(IM m ) .
Similarly, the weighted average of any other quantity, such as the expected value of DV, E(DV),
a valuable indicator, is calculated by replacing the POE with this quantity, as shown in Equation
2.4.
(2.4)
Equations 2.1-2.4 consider all possible scenarios of hazard, where each hazard level has a
specific probability of occurrence during a considered time span, e.g. 50 years, as calculated from
hazard analysis. However, in some situations, it may be useful to determine the loss in the case of
a specific hazard level certainly taking place during the considered time span. These situations,
which correspond to the case that the probability of the considered hazard level is equal to 1.0,
may arise if the considered structure is an important public facility such as a school or hospital or
an important bridge, or if the return period of the considered hazard level is likely to be completed
within the considered time span. In such cases, the POE and expected value of DV are represented
with Equations 2.5 and 2.6, respectively, where IMm is the considered intensity level.
(2.5)
(2.6)
Most commonly used DVs are fatalities, economic losses and downtime. PEER PBEE
methodology is particularly important and useful for Caltrans ordinary bridges, because it provides
a convenient venue to probabilistically determine the downtime for the highway bridges in
California. Accordingly, bridges with largest potential downtime can be determined and this
information, combined with other factors such as the importance of a particular bridge for post-
earthquake emergency response, bridges that require retrofit can be identified. The current study
mainly deals with the hazard and structural analysis stages of the PEER PBEE methodology and
as such targets to provide an accurate foundation for the complete application of the methodology
in consecutive projects.
8
3 Bridge Structures and Analytical Models of
Version 1.0
3.1 INTRODUCTION
It is stated in the Introduction that this study focuses on Caltrans ordinary standard bridges. For a
bridge to be considered as an ordinary standard bridge, it should satisfy the following conditions:
(1) the span length should be less than 300 feet; (2) the bridge should be constructed with normal-
weight concrete; (3) foundations must be supported on spread footings, pile caps with piles, or pile
shafts; (4) the soil is not susceptible to liquefaction, lateral spreading or scour; and (5) the
fundamental period of the transverse and longitudinal directions of the bridge should be greater
than or equal to 0.7 seconds [Caltrans 2010]. This chapter provides a brief description of selected
ordinary standard bridge structures and the related OpenSees [McKenna et al. 2010] modeling
technique based on the first generation of guidelines for modeling ordinary bridges [Aviram et al.
2008].
9
Table 3.1 Characteristics of the selected bridges for this study.
Bridge Designation A B C
Total Length 220.6 ft 300.0 ft 418.0 ft
3
2 2
Number of spans Span 1: 156.0 ft
Span 1: 110.3 ft Span 1: 150.0 ft
and Span Length Span 2: 144.0 ft
Span 2: 110.3 ft Span 2: 150.0 ft
Span 3: 118.0 ft
10
Figure 3.1 Modeling of Bridge B [Kaviani et al. 2012].
11
3.4.2 Column-bent modeling
To model the columns (Figure 3.2), nonlinear force-based beam-column elements are utilized with
fiber-discretized sections considering 10 integration points along the height, which is usually
deemed to provide adequate accuracy [Kaviani et al. 2012], to consider the progression of column
yielding and damage expected under strong ground motions. Three different constitutive rules are
used simultaneously within a fiber-discretized cross-section: (i) confined concrete for the core
concrete, (ii) unconfined concrete for the cover concrete, and (iii) steel rebar for the reinforcing
bars (Figure 3.2). OpenSees Concrete01 constitutive model is used for both the cover and core
concrete, and steel rebar is modeled by Steel02 material. A rigid element is attached to the top of
the nonlinear beam-column element to model the portion of the column-bent embedded in the
superstructure. The boundary condition of the column base proves to introduce significant impact
on the seismic responses obtained from NLTA [Kaviani et al. 2012]. In current models, the single-
column bridge (Bridge A) is modeled with a fixed base connection, while both pinned and fixed
base connection is assigned to the multi-column bridge (Bridge B and C).
Figure 3.2 Column modeling scheme for Bridge B [Kaviani et al. 2012].
12
Bearing pad
and stemwall
Expansion joint
Backfill Expansion joint
Backfill
Backfill
Bearing pad
and stemwall
a) Type I b) Type II
Figure 3.3 Springs and gap elements used to model various components of the abutment: a) Type
I Modeling [Kaviani 2011], b) Type II Modeling [Aviram et al. 2008].
13
4 Model Components
4.1 INTRODUCTION
The first generation guidelines [Aviram et al. 2008] address two types of nonlinearities: (1)
material nonlinear behavior in plastic hinge regions; and (2) geometric nonlinear behavior due to
large deformations (second-order effects). At the bridge system level, these guidelines address
standard ordinary bridges with no significant geometric irregularities (such as skewed or curved
decks, unequal or long spans, uneven mass distribution, numerous expansion joints).
Linear analysis of soil-structure interaction (SSI), linear foundation springs, and simplified
models for abutments and expansion joints are the principal shortcomings of these first generation
guidelines. This chapter addresses the improvements to the broad issue of nonlinear SSI, and
examines the modeling needs for columns, shear keys, abutments, in-span hinges, and piles.
(a) (b)
Figure 4.1 beamWithHinges element: (a) Gauss-Radau Integration (b) modified Gauss-Radau
integration [McKenna et al. 2010].
14
length. In the case of hardening section behavior, the computed element response will converge to
a unique solution by increasing the number of integration points. However, for softening section
behavior, the formation of localized deformations at the single integration point does not results in
a unique solution Therefore, the element response will change as a function of NP [Scott and
Fenves, 2006]. This problem has been addressed by Coleman and Spacone (2001) and modified
using plastic hinge integration method. The modified version of the forceBeamColumn is
implemented by Michael H. Scott in OpenSees [McKenna et al. 2010].
In this project we used the first method, beamWithHinges, with two plastic regions at two
ends of element and elastic linear region at the center of the element. Concrete01 and
ReinforcingSteel are used at the two ends (Figure 4.2). Plastic hinge length is calculated based on
SDC 1.7. Figure 4.2 shows lumped plasticity column modeling and the moment-column drift
relationship for a single column using the beamWithHinges element.
4
x 10
10
6
Moment(Kips.in)
-2
-4
-6
-8
-0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04
Displacement/Column length
(a) (b)
Figure 4.2 Column modeling: (a) Lumped plasticity column modeling [Berry and Eberhard 2008],
(b) Moment deformation backbone for single column in OpenSees [McKenna et al. 2010] using
Beam with Hinges Element.
15
4.3.1 SDC 1.7 Guidelines
Current Seismic Design Guidelines, SDC 1.7 [Caltrans, 2013] for ordinary highway
bridges assume simplified models for longitudinal and transverse response of bridge abutments.
The passive longitudinal response of backfill soil is modeled with an idealized bilinear elastic-
perfectly-plastic force-deformation relationship. The guidelines suggest minor adjustment to the
initial stiffness of the bilinear backbone curve when the backfill properties do not comply with
Caltrans Standard Specifications. Furthermore, per SDC 2010 guidelines, this initial stiffness
needs to be adjusted to an effective value to account for the expansion gap, in case of seat-type
abutments (Figure 4.3.)
16
4.3.2 Shear Keys
In bridge transverse direction, exterior and interior shear keys are designed to provide transverse
support to bridge superstructure and transmit the lateral shear forces—through vertical
reinforcement between the shear key and the stem wall—in minor-to-moderate earthquake events
and service loads. These elements are designed to break off once subject to strong motions in order
to save the abutment structure and piles. Shear-key design philosophy requires that these elements
behave similar to a fuse. In essence, shear-keys are designed to resist lateral service loads with
minor deformation and break under extreme loads to save the stem wall and the supporting piles.
In contrary, observations in previous earthquakes have shown multiple incidences of failure of
external shear keys that have propagated into abutment stem-walls [EERI, 1995]. In these cases,
diagonal cracks in the abutment stem-wall indicate a strut-and-tie behavior of the external shear
key that was not considered in the design process [Priestley et al., 1994; Moehle et al., 1995]. In
fact, the detailing of these shear-keys has led to engaging the stem walls’ reinforcements and
subsequently damaged the stem-wall. This is an undesirable behavior, and highlights the negative
effects of such shear key design approach [Yashinsky et al., 2010].
A number of studies have been directed towards quantifying abutment behavior in the
transverse direction. Two experimental studies were conducted at the University of California, San
Diego [Megally et al., 2001; Bozorgzadeh et al., 2004; Bozorgzadeh et al., 2006], which
investigated the behavior of shear keys and their failure mechanisms. The experimental program
conducted by Megally et al. [2001] comprised destructive testing of the exterior and interior shear
keys that are common to typical Caltrans bridges, and aimed at assessing the variation in shear key
responses under different loading protocols with respect to different detailing and reinforcement
ratios. Experimental results showed that exterior shear keys may fail in a sliding shear friction—a
brittle shear key behavior—or a strut-and-tie failure mechanism—a ductile shear key behavior.
The failure mechanism depends on the reinforcement detailing and the type of construction joint
between the shear key and the abutment stem wall [Megally et al., 2001].
Aviram et al. [2008] considered three different abutment modeling techniques. They
developed tri-linear backbone curves to capture the exterior shear key resistance levels when it
acts as a sacrificial element, and assumed the ultimate shear key strength to be 30% of the
superstructure dead load (cf. Eq. 7.47 of SDC 1.3). Zhang et al. [2008] and Elgamal et al. [2008]
developed 2D and 3D models of the Humboldt Bay Bridge with continuum soil models with an
aim to evaluate the effects of backfill soil and soil structure interaction on the overall bridge
response. In their bridge models, the behavior of the shear keys was modeled using an elastic-
perfectly-plastic backbone curve. Goel and Chopra [2008] evaluated the role of shear keys in
seismic behavior of bridges crossing fault-rupture zones. They concluded that seismic demands
for non-skewed bridges that cross fault-rupture zones are bounded between seismic demands
obtained from two simpler models: (i) a bridge model with elastic shear keys, and (ii) a bridge
model without shear keys.
Fell and Salveson [2013] assumed shear keys to be sacrificial elements (modeled as sliding
shear mechanisms) and compared the effects of two types of sacrificial shear key modeling
techniques on bridge responses: (i) the linear model suggested by California seismic provisions
SDC 1.7 (Caltrans, 2013), and (ii) a nonlinear model that included a gap distance and a bilinear
force-deformation backbone curve. In their nonlinear model, the shear key stiffness was obtained
based on a plane stress finite element model, and the ultimate strength was based on the Caltrans
friction shear method. Longitudinal springs were developed consistent with the requirements of
17
the SDC 1.7. Fell and Salveson [2013] suggested that the nonlinear shear key model is sufficiently
accurate for bridges with flexible substructures (T > 1 sec.), and the linear shear key model is
reasonable for short and stiff bridges. However, they ignored the ductile behavior of shear keys in
their non-linear model.
Kaviani et al. [2012] investigated the effect of different parameters such as column bent
height, symmetry of span arrangement, and abutment skew angle on various engineering demand
parameters—namely, column drift ratio, deck rotation, and abutment unseating. They showed that
the strength of a sacrificial shear key has a large influence on bridge deck rotations. Wasef [2013]
examined the effect of embankment modeling on displacement demand of shear keys; he assumed
that shear keys behave as sacrificial elements and fail in a sliding shear friction mechanism. Wasef
[2013] concluded that proper embankment modeling has a significant effect on the shear key
displacement demand.
The “Caltrans Seismic Design Criteria 1.7” document [Caltrans, 2013] provides two types
of detailing for exterior shear keys: (i) isolated shear keys based on the recommendations by
Megally et al. [2001], and (ii) non-isolated shear keys. According to these guidelines, lateral
stiffness of a bridge abutment in the transverse direction, which is mostly provided by the shear
keys, is 50% of the lateral stiffness of the adjacent bent in the global system. Both non-isolated
and isolated shear keys are assumed to fail based on the shear friction model. Experimental studies
[Megally et al., 2001] indicate, however, that shear keys with the reinforcement detailing of non-
isolated shear keys fail in a strut and tie failure mechanism rather than a shear-friction mechanism.
This form of failure can result in significant damage to the stem wall and change a bridge’s global
behavior under dynamic excitations. On the other hand, shear keys with reinforcement detailing
of isolated shear keys work as a fuse, and they fail based on the sliding shear friction mechanism.
The capacity of this type of shear key can be underestimated if the shear friction model presented
in Caltrans Bridge Design Specification [Caltrans, 1993] is used.
Shear key modeling techniques presented here are based on the experimental program
conducted by Megally et al. [2001]. In the said study, ten prototype abutments were built at 1:2.5
scale and tested. The experiments were classified into 5 categories based on the amount and
location of vertical reinforcement, existence of construction joints, and the type of horizontal
reinforcement at the top of the stem wall (i.e., either pre-stressing steel or regular hanger bars).
The test results demonstrated that an exterior shear key failure could occur in two different
mechanisms: (i) sliding shear friction, and (ii) diagonal tension. The type of shear key behavior
depends on reinforcement detailing and the construction joint between the shear key and the stem
wall (Figure 4.4). Typically, the maximum transmitted shear force along the horizontal crack at
the shear key interface with the stem wall can be estimated for isolated shear keys. This failure
mechanism does not engage the stem wall’s reinforcements and limits the damage at the interface
of the shear key and stem wall. In this form, the shear key behaves as a brittle element; after
reaching the maximum strength, the shear key resistance drops due to breakage of the bond and
yielding of the vertical reinforcements that connect the shear key to the stem wall at the
construction joint. Experimental results by Megally et al. [2001] further indicated the location and
the amount of the vertical reinforcement, as well as the type of the horizontal bar (prestressed vs.
regular reinforcement) at the top of the stem wall can also alter the behavior of shear keys.
18
External
shear key
Vertical
reinforcement
Bearing
Construction
joint
a) Stem wall b)
Figure 4.4 Shear key details based on Caltrans SDC 1.7: (a) brittle shear key, (b) ductile shear key.
Megally et al. [2001] and Bozorgzadeh et al. [2004, 2006] defined the nominal capacity of
brittle shear keys based on the sliding shear friction mechanism as shown in Eq. (4.1). This model
predicts the shear key capacity with a construction joint and lumped vertical reinforcement near
the centerline of the shear key (see Figure 4.5).
External
Back Wall shear key
Angle of kinking
Angle of inclined face of shear key
AVƒ f Kinematic coefficient of friction for
Construction
β Avf concrete
Area of vertical reinforcement
joint
α fsu Ultimate tensile strength of reinforcing
steel
Stem wall
19
Back Wall
Shear key Failure Shear Key
B’ Area=a x d
B
Shear key Failure
Vn
A’ d
C’ D’ b
A
Force
Vs k2
Concrete01 EPPG h
Force
ENT
k3
k1
Bilinear ENT Hysteretic C
Deformation
Deformation (a) (b)
Stem Wall K1 = (h/AG + h3/3EI)-1
K2 = f . k1
f = (0.5% - %2.5)
k3 = %2.5 . K1
h : Height of shear key
d : Depth of shear key
EPPG material
G : Shear modulus
Concrete 01 material
E : Modulus of elasticity
Vc I : Moment of inertia
Force
ENT material
Force
Force
Deformation Deformation
Force
Bilinear hysteretic
material
Vs Deformation
Force
Deformation
Figure 4.6 shows macroelement models for the two shear key types (ductile and brittle)
and their generic force-deformation backbone curves. In the present study, a tri linear backbone
curve is proposed to represent the constitutive model of a brittle shear key. The initial stiffness of
this backbone curve is the summation of the shear and flexural responses of a concrete cantilever
with the shear key dimensions. The stiffness of the hardening part is assumed to range between
0.5 to 2.5% of the initial stiffness, which depends on the type of construction joint (i.e., smooth or
rough), and the softening part of the backbone curve is assumed to be equal to 2.5% of the initial
stiffness [Aviram et al., 2008]. The maximum capacity of the shear key in sliding shear friction
mechanism can be estimated based on the shear key properties (see Eq. 4.1).
(4.1)
Experimental results suggest that the behavior of brittle shear keys can be summarized in
two steps, as illustrated in Figure 4.6. In the first step, a horizontal crack is initiated at the interface
of the stem wall and the shear key. This point is marked as point A in Figure 4.6. In step two, the
crack propagates horizontally, and the rebars start yielding until capacity is reached, which is
marked as point B in Figure 4.6. The breakage of the construction joint bond results in softening
after point B, the rupture of vertical reinforcements signifies point C.
20
The brittle shear key model’s prediction are compared in Table 4.1 with experimental data
from Megally et al. [2001] for a sacrificial shear key, which included a smooth construction joint
and lumped vertical reinforcements at the center. The tri-linear backbone curve for the brittle shear
key can be modeled using the hysteretic material in OpenSees [McKenna et al., 2000]. Since a
shear key can provide reaction in only one direction, a zero length element with elastic no-tension
material (denoted as ENT) is attached to the inelastic spring (see Figure 4.6).
The behavior of a ductile shear key is different due to lack of construction joint, large
amounts of vertical reinforcement, and the location of vertical rebars connecting the shear key to
the stem wall. Our proposed ductile shear key model backbone curve is quinque-linear and traces
the ductile failure mechanism, which initiates with diagonal tension failure cracks forming through
the shear key. Horizontal cracks then initiate at the interface of the shear key and the stem wall,
and propagate until they reach the first row of the vertical reinforcement of the shear key (point
A´ in Figure 4.6). As the lateral force increases and the shear key rebars yield, multiple inclined
cracks propagate toward the bottom of the stem wall, and the contribution of concrete in resisting
the lateral loads decreases. This leads to a capping point in the shear key force-deformation
backbone curve (point B´ in Figure 4.6). Once past the capping point, the capacity of the shear key
is solely due to the contribution of the rebars’ resistance (points C´ and D´ in Figure 4.6).
According to experimental results, the strain of rebars in the stem wall corresponds to the onset of
concrete degradation (i.e., 0.5 pct-strain, which is point C´ in Figure 4.6), and fracture of
reinforcement (i.e., 0. 7 pct-strain, which is point D´ in Figure 4.6). This form of failure mechanism
results in the development of a strut and tie mechanism {Megally et al., 2001; Bozorgzadeh et al.,
2004] for ductile shear keys.
According to Megally et al. [2001], the shear key capacity for ductile shear keys can be
obtained from the combination of two contributions as shown in Eqs. (2.4), where Vc and Vs are
the concrete and reinforcing steel contributions to shear resistance, respectively; Av is the area of
vertical reinforcement crossing the interface of shear key and stem wall; fy is the yield strength;
As1 is the area of horizontal reinforcement at the top of stem wall; As2 is the area of stem wall side
reinforcements; nh and nv are the numbers of side reinforcements in the stem wall (horizontal and
vertical) facing the inclined crack; d is the depth of shear key; b is the width of the stem wall; and
h is the stem wall height.
(4.2)
(4.3)
(4.4)
The macroelement model for a ductile shear key is obtained by incorporating the
contribution of reinforcing steel using the modified Ibarra-Medina-Krawinkler deterioration model
with bilinear hysteretic response (Bilin) material model in OpenSees; concrete contribution is
modeled using concrete02. Because concrete is activated after the initiation of horizontal cracks
at the shear key to stem wall interface, which results in the formation of tensile stresses in the first
row of vertical reinforcements, the gap between these two materials is simulated using the elastic-
21
perfectly-plastic gap material (EPPG) with high stiffness to eliminate the effect of gap on the
initial stiffness of concrete01 material (see Figure 4.6). The magnitudes of force and displacement
at each step were computed based on the equations proposed in Megally et al. [2001]. In order to
validate the macroelement, a typical Caltrans shear key with ductile behavior that had been tested
by Megally et al. [2001] was considered. Table 4.2 displays the measured responses and those
predicted by the macroelement model.
Table 4.1 Experimental results versus model predictions for a brittle shear key.
Proposed Model Experimental Data
Force kip Force kip
Point (kN) Disp. in (cm) (kN) Disp. in (cm)
Point A (Level I) 41.9(186.8) 0.007 (0.02) 27 (120.1) 0.011 (0.03)
Point B (Level II) 71 (315.8) 1.1 (2.8) 75 (333.6) 1.3 (3.3)
Table 4.2 Experimental results versus model predictions for a ductile shear key.
Experimental Data
Proposed Model [Megally et al., 2001]
Force kip
Point (kN) Disp. in (cm) Force kip (kN) Disp. in (cm)
Point A´ (Level I) 234 (1041) 0.2 (0.5) 230 (1023) 016 (0.4)
Point B´ (Level II) 316 (1406) 1.5 (3.8) 320 (1423) 1.22 (3.1)
Point C´ (Level III) 223 (992) 2.7 (6.8) 240 (1068) 2.4 (6.1)
22
parallel combination of soil and pile springs. The pounding element proposed by Muthukumar
[2003] is used to model effects of deck pounding on the abutment.
Aviram et al. [2008] proposed three types of abutment models with increasing levels of
complexity and studied the effects of abutment modeling on the output of modal, push-over, and
nonlinear time-history analyses of six bridges. In the most basic model, the interaction between
the backfill soil and abutment is neglected and the abutment is replaced with roller supports. In the
intermediate model, the original and modified SDC 1.7 bilinear springs are used to mimic the
response of backfill soil. The vertical responses of bearing pads are modeled with linear springs.
The most advanced model by Aviram et al. employs combinations of linear and bilinear springs,
rigid links, gap and contact elements, as well as lumped masses to mimic the interaction effects
between various structural elements (including bearing pads, backwall, stem wall and shear-keys),
abutment and backfill soil, and participating mass of embankment soil. The study concludes that
the roller-support boundary condition is an appropriate modeling option for long-span bridges,
where the interaction between the superstructure and the abutments does not dominate the overall
response. The intermediate model—although suitable for estimating the longitudinal response of
the bridge—does not incorporate adequate details for predicting the actual transverse responses of
the bridge.
In contemporary study, Zhang et al. [2008] created a detailed 2D finite element model of
the Humboldt Bay Bridge, which featured a large portion of the surrounding/supporting soil
domain in an attempt to capture the longitudinal response of the backfill soil with high accuracy.
Shear-key models—at the super-structure and abutment joints—are generated using a combination
of elastoplastic materials, gap and hook elements. A very similar continuum modeling approach
was later adopted in 3D modeling of the same bridge by Elgamal et al. [2008]. Similarly, Rahmani
et al. [2014] also created a 3D continuum model of a two-span bridge with integral abutments,
which were modeled using shell elements. Solid elements were employed to model the
embankment; and abutment-embankment contact was represented by constraining the common
degrees of freedom between the collocated nodes of the shell and solid elements.
Zhang and Makris [2002] developed an equivalent nonlinear approach for deriving the
impedance of an approach embankment for a given ground motion, from which they deduced
practical frequency-independent values for the embankment stiffness and radiation damping. Their
approach also provided the input motion amplification as the seismic waves travel from the base
of the approach embankment to its crest.
Huo [2011] used a 3D finite element abutment model to calibrate longitudinal and
transverse p-y springs for various abutment wall heights and material properties of sandy
embankments. In order to model the longitudinal gap and the friction between the backwall and
the backfill soil, the p-y springs were combined with a contact-friction element, which possesses
no-tension and pressure-dependent capacities in the longitudinal and transverse directions,
respectively. Abutment piles were incorporated into the model explicitly; yet, the computational
burden was reduced by replacing every seven or eight pile with a single super-pile. The equivalent
linear spring and viscous dashpots suggested by Zhang and Makris [2002] were later used by Huo
and Zhang [2012] along with gap elements, elastoplastic and contact linear springs to account for
the combined effects of embankment impedance, elastomeric bearings, and deck pounding.
Earthquake and bridge engineering communities have also directed significant attention
towards the seismic performance of bridges with skew abutments. These studies were arguably
23
prompted by the observed inferior seismic performance and higher vulnerability of skew bridges
[Buckle et al. 1994, Kawashima et al. 2011]. Consistent with field observations was the reduced
passive resistance of the backfill in a skew abutment configuration—a phenomenon that has been
demonstrated through three-dimensional finite element simulations [Shamsabadi et al. 2004], and
observed in laboratory [Rollins and Jessee, 2012] and large-scale field [Marsh et al. 2014] tests.
Shamsabadi and Kapuskar [2006] investigated the response of skew bridges to near-fault
ground motions considering various span and bent configurations. Deck rotation was shown to
build up in skew bridges shortly after the arrival of a velocity pulse (if any), resulting in residual
rotations and sometimes unseating of the bridge deck from the abutment. As expected, bridges
with single-column bents were found prone to larger deck rotations as a result of the lower global
torsional stiffness. These findings indicated the critical role of transverse shear-key failure in
amplifying deck rotations.
Kaviani et al. [2014] considered a linear trend for the variation of backfill passive
resistance between its acute and obtuse corners. This modeling approach was used along with the
simplified abutment model of Aviram et al. [2008] in three typical California bridges to study the
effects of skew angle on deck rotations, column drift ratios, and system-collapse fragilities.
Detailed sensitivity analysis revealed increases in seismic demand and collapse potential at higher
skew angles. The skew-abutment effects were amplified in case of bridges with single-column
bents due to amplified deck rotations, which, in turn, led to higher fragility of shear-keys. The
sensitivity analyses by Kaviani et al. also revealed that ordinary (skew or non-skew) bridges incur
higher seismic demands, in general, when subjected to near-fault ground motions that contain
energetic velocity-pulses.
One of the most recent studies in this area is the work of Zakeri et al. [2013] who
investigated the fragilities of single-frame box girder skew bridges. Eight classes of such bridges
were studied, with categories based on the seismic detailing of the construction era (pre-1971 and
post-1994), number of columns in the bent (single or double), and abutment type (seat-type or
monolithic). Modeling details of abutment almost followed a combination of the approaches
adopted by Nilson [2005], Nilson and DesRoches [2006], Aviram et al. [2008]. Based on the
outcome of component and system fragility analyses for various damage states, the study inferred
that the seismic responses of post-1994 single-frame bridges with seat-type abutment are sensitive
to the abutment skew angle. Authors also suggested that the application of HAZUS-MH [FEMA
2005] skew modification factors for classes that were sensitive to the skew angle yields results that
are consistent with the direct fragility analyses of skewed bridges, while the application of the said
factors to insensitive classes leads to conservative estimate of fragilities.
At the present time there is no consensus model for inelastic responses of skew abutments,
and a study that provides the direct comparison of different approaches is lacking. The present
effort is intended to explore this area and (i) provide some guidelines for practicing engineers in
modeling skew abutments, and (ii) identify research needs. More specifically, this study aims to
assess current state-of-practice abutment modeling methodologies—adopted from Caltrans
Seismic Design Guidelines—as well as more advanced approaches, by quantifying their
predictions quantified through a set of Engineering Demand Parameters (EDPs) that are pertinent
for highway bridges.
In this study, emphasis is placed on adopting alternative models of backfill passive
resistance and abutment external shear-keys. The passive resistance of backfill soil is modeled
24
with either the SDC 1.7 bilinear force-deformation model or the generalized hyperbolic force-
deformation (GHFD) relationship proposed by Khalili-Tehrani et al. [2010]. The parameters of
the latter model are calibrated in terms of backwall height and physical material properties of
backfill soil that can be obtained through laboratory and/or field-testing. Additionally, two
approaches are employed to account for the effect of skew angle on the passive resistance of the
backfill. One of these is an approach based on an assumed soil failure mechanism, which predicts
the formation of intermediate and ultimate log-spiral failure surfaces [Shamsabadi et al. 2005,
2007]. The other model uses empirical coefficients derived through laboratory and full-scale
testing of skewed abutments [Rollins and Jessee, 2012, Marsh et al. 2014]. Improved shear-key
models—developed and calibrated using experimental data obtained by Megally et al. [2001]—
replicate brittle and ductile behavior of the shear-keys in the transverse direction.
25
1703 683.4tan 1.23 50 (4.6)
5.62tan 2 0.53 10.58tan 1.79 2.86 c, 0, c 0
1.06 5.62tan 0.53 ,
2
c0 (4.7)
0.50 2.63c, 0
2 c2
n 0.91tan 1.49
.2
(4.8)
0.90 c0
c
(4.9)
Number of d* W* * c*
Category* 50 Rf
Bridges* kip/ft3 (%) kip/ft3 (deg) (kip/ft2)
I 2 1.2 3-6 0.12 38 0 0.0035 0.96
II 9 0.12-1.2 5-14 0.11 33 0.5 0.0035 0.96
III 4 0.1-0.11 7-15 12 23 2 0.0035 0.96
IV 5 0.11-0.08 14-29 0.09 6 3.5 0.007 0.96
*Earth Mechanics Inc. [Earth Mechanics, 2005].
The values listed for and c are typical properties of compacted soil samples from each
backfill category. However, not all properties that are needed for LSH and GHFD methods are
provided in the report. Phase relationships are used to determine the total unit weight () from the
26
lower bounds of measured field dry density (d) and moisture content (w). The backwall-backfill
interface friction angle, , has been set to two-third of the soil internal friction angle per
recommendations of Shamsabadi et al. [2005, 2007]. Since 50 and Rf are also not available from
laboratory test results, their value are chosen according to recommendations of the same reference,
which is 0.96 for Rf, and 0.0035 and 0.007 for sand (Categories I to III) and clay (Category IV)
backfills, respectively.
Figure 4.7 displays the intermediate and ultimate failure surfaces generated by the LSH
method for each backfill category for a unit width of a 1.9m-high backwall (the specimen bridge
backwall). The corresponding LSH force-deformation curves are shown on Figure 4.8, where the
abscissa and ordinate of each data point correspond to the backwall deflection and lateral force
required for mobilizing one of the illustrated failure surfaces. GHFD backbone curves show
reasonable agreement with the LSH backbones as shown in Figure 4.7. The bilinear force-
deformation relationships recommended in SDC 1.7 are also illustrated along with the GHFD and
LSH backbones. Apparent from Figure 4.8, SDC 1.7 idealized bilinear backbone with unreduced
initial stiffness is in reasonable agreement with GHFD backbone of Category II backfill, which
constitutes 45% of the tested backfills. However, SDC 1.7 idealized backbones deviate
significantly from the GHFD backbones corresponding to the other three categories, which
constitute 55% of the tested backfills.
(a) (b)
(c) (d)
Figure 4.7 LSH intermediate and ultimate log-spiral failure surfaces for (a) Category I, (b) Category
II, (c) Category III, and (d) Category IV backfills (1 m = 39.37 in.).
27
Figure 4.8 Passive resistance of a unit width of a 1.9m-high abutment backfill due to SDC 1.7
bilinear backbones as well as LSH and GHFD models for typical backfills in California highway
bridges (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
The bilinear force-deformation recommended by SDC 1.7 and the nonlinear relationship given by
GHFD formulation are both adopted for modeling the longitudinal passive response of backfill. In
both cases, the gap between backwall and deck is integrated into the force-deformation models by
shifting the backbone with an amount equal to the gap.
A uniaxial compressive hyperbolic-gap material is available in OpenSees [McKenna et al.
2010. Backbone curve of the hyperbolic-gap material model is parameterized in terms of initial
stiffness, Kmax, failure ratio, Rf, and ultimate passive resistance, Fult, as follows,
(4.10)
A one-to-one comparison between the OpenSees [McKenna et al. 2010] and GHFD
formulation yields the following relationships between the input parameters of the OpenSees
[McKenna et al. 2010] material model and GHFD parameters,
(4.11)
4.3.3.2 Incorporating the effects of skew into the backfill’s passive response
A skew in the abutment alignment generally introduces a reduction in the passive lateral resistance
of the abutment backfill. This is primarily due to rotations induced by the eccentricity of the
resultant backfill reaction relative to the bridge’s center of stiffness and/or mass. Two heuristic
approaches are adopted in the present study to incorporate the said reduction.
28
4.3.3.2.1 Backfill with non-uniformly reduced GHFD passive resistance
This approach can be considered as an evolved version of the hypothetical skew abutment model
suggested by Kaviani et al. [25] in which the stiffness and ultimate strength of the backfill springs
were varied linearly over the entire width of the backwall between the obtuse (OBT) and acute
(ACT) corners. In order to achieve such effect, backfill passive resistance is modeled with five
hyperbolic springs that are evenly distributed over the skewed length of backwall as depicted in
Figure 4.9.
Bearing pad
and stemwall
Expansion joint
Backfill
Shear key
Figure 4.9 Linear variation of backfill passive resistance from the obtuse to the acute corner of a
skewed abutment.
The upper and lower bounds of the coefficient of linear variation are defined as 1± where,
for the skew angle , parameter is evaluated using Eq (4.12). Eq (4.12) postulates that a
maximum variation of 60% happens between the obtuse and acute corners at 60◦ skew angle. This
variation accounts (in a phenomenological sense) for the decrease in the available volume of soil
to be mobilized from the obtuse corner to the acute one, as schematically illustrated in Figure 4.10.
Nonetheless, it is possible that the variation of passive resistance along the backwall takes place
only over a fraction of the backwall. The width of the reduced-resistance region (Wr) can be
decided based on the formation of intermediate and final failure surfaces given by the LSH method
[Shamsabadi et al. 2005, 2007].
(4.12)
29
OBT
Wr
Lu a
ACT
Figure 4.10 Reduction of backfill volume normal to backwall, from the obtuse to the acute corner
of a skewed abutment.
In the extended model (of that proposed in Kaviani et al. [2012]) offered herein, the
nominal resistance provided by GHFD at a given coordinate over the width of the skewed backwall
is assumed to remain unchanged if the extent of backfill perpendicular to the backwall is long
enough to allow for the formation of the ultimate soil failure surface. However, if—as a result of
the skewed geometry of the abutment—the perpendicular length of backfill is less than the length
of the ultimate failure surface, the passive resistance of backfill needs to be reduced at that
coordinate. The reduced-resistance width (Wr) becomes then, a function of the skew angle and
becomes related to the length of ultimate failure surface (Lu) as given by,
(4.13)
It is observed in Figure 4.8 that the shape of the mobilized volume of soil predicted by LSH
for cohesive soil (Category IV) converges to the conventionally used Rankin failure wedge. The
length of the ultimate failure surface at the ground level decreases as characteristics of backfill
change gradually from granular to cohesive (i.e., form Category I to Category IV backfills). This
introduces larger reduction in passive resistance of backfill due to skewed abutment configurations
for granular backfills compared to cohesive backfills, since the premature termination of
intermediate failure surfaces before the mobilization of ultimate surface occurs over a larger width
of the backwall. The ratio of Wr to Wt, which is the tributary width of a hyperbolic spring, in the
specimen bridge and for an assumed arrangement of five backfill springs along the width of
backwall is provided for the four backfill Categories in Table 4.4.
Table 4.4 Ratio of reduced-resistance width (Wr) to tributary width (Wt) of uniformly distributed
abutment springs in the specimen bridge model (Bridge B).
For all backfill categories and at all skew angles (except Category I at the 60-degree skew
angle), the reduced-resistance width is contained within the tributary width of backfill for the first
hyperbolic spring located near the acute corner (i.e., Wr / Wt 1). For brevity, Figure 4.11 only
illustrates the general case where Wr extends to the tributary width of the second spring located
30
from the acute corner. Distribution of backfill passive resistance for the case of Wr / Wt 1 can be
easily deduced from the same diagram.
Wt Wt
Wr Wr
Wt Wt
F2
Pu
RPu F1
Figure 4.11 Variation of passive resistance over the reduced-resistance region in the vicinity of
acute corner in skewed abutments.
The force in the first acute corner spring of Category II, III, and IV backfills can be
approximated by Eq. (4.14) where Pu corresponds to the unreduced passive resistance provided by
unit width of backfill. For Category I backfill, Wr extends to the tributary width of the second
hyperbolic spring located near the acute corner. Therefore, the force in the first acute corner spring
is approximated with Eq. (4.15) where factor R is defined in Eq. (4.16) and Pint corresponds to the
force required for mobilizing an intermediate failure surface which fits within the available length
of backfill at the end of Wt of the first spring. This intermediate failure surface and the
corresponding point on the force-deformation backbone for Category I backfill are marked with a
dashed line in Figure 4.7.a, and with an asterisk in Figure 4.8, respectively. Factor R is
approximately 0.6 for the Category I backfill in the 60-degree skew geometry for the abutment in
the specimen bridge. The force in the second spring located from the acute corner of backfill is
similarly approximated as shown in Eq. (4.17). Eqs. (4.14) to (4.17) become clearer by examining
the force diagrams shown in Figure 4.
(4.14)
(4.15)
(4.16)
(4.17)
31
abutments. Rollins and Jessee [2012] and Marsh et al. [2013] suggest a passive resistance
reduction factor (R) as a quadratic function of skew angle () as given in
(4.18)
The passive resistance of backfill in a skewed abutment (Pskew) is equal to the product of the
reduction factor (R) and the passive resistance (P) of the same backfill for the projected (non-
skewed) length of backwall, as given by
(4.19)
The reduction factor given by Eq. (4.18) is applied uniformly to the distributed backfill
hyperbolic springs in Category I and II backfills, whose geotechnical properties comply with those
of the tested backfills. Per Marsh et al. [2013], P should be evaluated for the “projected” width of
the abutment. Therefore, the tributary width of backfill assigned to each spring is calculated based
on the non-skewed width of the deck. Reductions factors given by Eq. (4.18) for five representative
skew angles are listed in Table 4.4.
32
4.5 MODELING SOIL-STRUCTURE INTERACTION IN SHAFTS
The constitutive models for concrete and steel are OpenSees [McKenna et al. 2010]
Concrete01 and Steel02 uniaxial material models. The backbone curve of Concrete01 model is
based on the no-tension constitutive model of Kent-Scott-Park, and incorporates
unloading/reloading stiffness degradation under cyclic loading. Steel02 material is based on the
Giuffre-Menegotto-Pinto model, which is a bilinear model featuring strain hardening and a smooth
transition from the initial tangent stiffness to the post-yield tangent stiffness. Different stress-strain
pairs are chosen for characterizing the constitutive models of confined (core) and unconfined
(cover) concrete. In particular, zero ultimate (crushing) strength is assigned to the cover concrete.
The adoption of fiber-based modeling yields enough information for investigating the possibility
and location of plastic hinging in the pile, yet it involves less computational burden compared to
three-dimensional finite element modeling approach. The piles have been discretized to 1ft
sections and 10 Gauss integration points have been assigned along each element. Such level of
refinement should provide a robust insight into optimized element size and number of integration
points along the depth of the pile given the intensity of input motion.
33
The present study does not take into account the deteriorative effects of cyclic loading—as
described by API recommendations— on the lateral load bearing capacity of soil.
Figure 4.13 Macroelement for modeling passive pressure of soil in cyclic loading [Boulanger et al.,
1999].
Particular assemblages of the abovementioned nonlinear springs and a number of physical
and semi-physical components can incorporate additional near-field and far-field soil-pile
interaction effects—such as gapping, drag, and radiation damping—and are referred to as
macroelements [Taciroglu et al., 2006; Boulanger et al., 1999]. The macroelement described in
the work by Boulanger et al. (1999), hereinafter referred to as the p-y macroelement, is available
in OpenSees as (PySimple1 material) and has been adopted for the current study. The
macroelement (Figure 4.13) consists of a closure and a gap element placed in parallel with each
other that simulate loss of passive soil pressure throughout gapping, and friction forces incurred
as the pile moves in the gap. This combination, a plastic spring, and an elastic spring have been
placed in series. The backbone of the plastic and elastic spring together follows Matlock and API
p-y curves. The effects of radiation damping have been taken into account in the form of equivalent
viscous damping and by placing a dashpot parallel with the linear spring. Such configuration is
labeled as "series hysteretic/viscous damping" by Boulanger et al. (1999). It has been argued that
placing the dashpot in series with the nonlinear spring circumvents generation of unrealistically
large damping forces when the soil gets into the nonlinear deformation region. As recommended
in the work by Boulanger et al. (1999) the constant of equivalent viscous damping can be decided
based on the work by Gazetas and Dobry (1984).
Gazetas and Dobry (1984) in part of their study developed and evaluated a simple model
to obtain horizontal and vertical radiation damping coefficients of a circular piles resting on or
embedded in a homogeneous elastic space subjected to vertical and horizontal harmonic excitation.
An example of soil force-displacement relationship, which demonstrates the shape of
backbone curve as well as gapping and drag phenomena simulated by the PySimple1 material used
in the bridge model, is shown in Figure 4.14.
34
20
15
10
Force (kips)
0
-5
-10
-15
-20
-1 -0.5 0 0.5 1
Deformation (in)
Figure 4.15 Distribution of passive soil pressure on different cross-sections [Reese and Van Impe,
2009].
Obviously, if width (w) and depth (d) of the rectangular section are equal to the diameter
(b) of the circular section, then the rectangular section will experience larger resistance. Therefore,
the equivalent diameter should be chosen relevant to both width and depth and the passive and
35
frictional forces acting on the leading and side faces of the rectangular cross section. Based on this
intuitive concept, Reese and Van Impe (2009) suggest the following relationship for the equivalent
diameter of the rectangular section of Figure 4.15b,
(4.21)
in which Puc is the ultimate resistance of a circular section with a diameter equal to w; fz is the
shearing resistance along the sides of rectangular cross section at depth z below the surface given
as,
(4.22)
for cohesive soil and,
(4.23)
for cohesionless soil.
In the above equations, cu is the undrained shear strength and is the reduction factor,
which depends on the behavior of pile under axial loading. Kz is the lateral earth pressure
coefficient and its value depends on the pile installation method; z and z are the effective stress
at depth z and shear angle between soil and pile at the relevant shear strain level, respectively. The
value of z is often taken slightly smaller than the shear angle.
The equivalent diameter of other noncircular shapes can be decided based on the formula
suggested for the rectangular section and by considering the direction of loading. For example, in
the calculation of an equivalent diameter for the flat shape of Figure 4.15c, fz should be set to zero.
Reese and Van Impe (2009) recommend taking as unity and z is equal to shear angle for the H-
section embedded in cohesive and cohesionless soils, respectively, under the shown direction of
loading. In the orthogonal direction, the cross section can be treated as a rectangular section with
width equal to the height of cross-section web, and depth equal to the width of cross-section flange.
Per Reese and Van Impe (2009) recommendation, the depth-dependent equivalent diameter
can be calculated at several representative depths and be interpolated for the depths values in
between.
36
constraints pertinent to a rigid-body are imposed on the nodes of the pile-cap. The thickness of the
pile caps are decided based on the structural drawings of one of the seed models.
(a) (b)
Figure 4.16 Pile-to-cap connection details: (a) fixed connection, (b) pinned connection [Silva and
Seible 2001].
Pile to pile-cap connection modeling as an important phenomena has effects on transferring
moment and shear into the soil during the earthquake. Figure 4.16 illustrates standard pile to pile-
cap connection detailing for CIDH piles. Previous study (Richards et al., 2011) on the connection
of pile to reinforcement concrete cap shows that not only the steel reinforcement that extends into
the pile-cap can provide fixed connection (Figure 4.16a) but also a connection with adequate
embedment shows considerable flexural capacity. Richards et al. (2011) tested four full-scaled
reinforced concrete circular piles with different embedment lengths. They concluded that pile to
pile-cap connection that is usually assumed to behave as pinned (Figure 4.16b) connection show
considerable flexural capacity similar to a “fixed” connection.
In the case of steel pile (such as H Pile) and according to the current design standard, the
connection of steel pile-to-pile cap is designed as a pinned connection. However, previous studies
on the connection of steel pile to reinforcement concrete cap have shown that such a boundary
connection is invalid. Marcakis and Mitchell (1980) introduced an equation based on experiments
for the flexural capacity of steel-to-concrete connection in terms of embedment length.
Xiao et al. (2006, 2013) has performed experiments with five full-scale steel H-piles with
a shallow embedment and conventional V-bar connection details (Figure 4.17). They concluded
that V-bar connections cannot develop the design ultimate tensile strength required to develop
37
significant moment resistance at the pile to pile-cap connection, However, significant moment
resistance can be developed based on the embedment of pile in the pile cap.
Figure 4.17 Typical details of a HP-steel pile-to-cap connection [Xiao et al. 2006].
According to the current design standard, this type of connection is designed as a pinned
connection. Experimental results show that the connection can develop significant moment resistance
based on a flexural resisting mechanism and embedment mechanism. They have also concluded that V-
bars for the HP-pile connection cannot develop the design ultimate tensile strength. Therefore, they made
recommendations for this type of connections. They suggested welding anchor bars to the HP-steel piles
in the new design in this type of connection (Figure 4.18).
Figure 4.18 Recommended details of HP-steel pile-to-cap connection [Xiao and Chen 2012].
In this study based on details obtained from drawings and the calculated flexural capacity at the
connection of H pile-to-pile cap, fixed connections have been assumed in both selected bridges. At the
connection of CIDH pile-to-pile cap, both fixed (Bridge B) and pinned (Bridge A) connections has been
assumed.
38
4.5.4.3 Incorporation of Shadowing Effects
The overall resistance of a group of piles is less than the sum of the resistance of the individual
piles in the group. This is due to the disturbance of soil surrounding each pile by the motion of the
neighboring piles. The disturbance of the resisting wedge of one pile by another is referred to as
the "shadowing" effect. In general, shadowing effect is expected when the center-to-center spacing
of piles is around or less than 5 times the diameter of the piles in the group. Based on the modeling
and analysis approach, group effect can be taken into account either by reducing the overall
capacity of the pile group through the application of a "group efficiency" factor, or by reducing
the ultimate resistance in the py curve of each pile through the application of p-multipliers. The
second method is the approach of choice here due to the explicit modeling of each pile in the pile
groups and employment of py elements for modeling soil resistance. In general, the soil around
the leading row of piles experiences the least amount of disturbance and larger p-multipliers are
used for this row of piles. The shadowing effects become more prominent in the trailing rows of
piles. A survey on group effects has been conducted and based on the recommendation in the work
by Brown et al. (2001)—sponsored through a National Cooperative Highway Research Program
(NCHRP) project—it was decided to use a weighted average of the p-multipliers of the leading
and trailing rows. The p-multiplier for each row is chosen based on the empirical relationships
suggested by Rollins et al. (2006). Since the number of rows in the transverse and longitudinal
directions of seed bridges are not the same, different p-multiplier values are used in the two
orthogonal directions.
39
4.6 IN-SPAN HINGES
Figure 4.19 (a) Simplified contact model, (b) Impact material in OpenSees [Muthukumar &
DesRoches, 2006].
40
4.6.2 Restrainer stiffness in In-span Hinge
The 1971 San Fernando earthquake showed that narrow seat length of in-span hinges may result
in deck unseating. Therefore, Caltrans suggested using steel cables or bar restrainers in in-span
hinges to prevent excessive relative displacement and unseating of spans. Figure 4.20 shows a
typical detail of multi-frame strainers. Nevertheless, restrainer failure has been observed during
past earthquakes (e.g., 1995 Kobe earthquake, 1989 Loma Prieta earthquake, and 1994 Northridge
earthquake). At this time, typical restrainers used in California are ¾ inch (19 mm) diameter steel
cables with an area of 0.22 square inch (143 (mm)2 which are made of 6x19 strands (Scalzi and
McGrath, 1971; Section 83-2.02A, Standard Specifications).
41
Table 4.5 Concrete Properties [Hube and Mosalam 2008].
Hube and Mosalam (2008, 2010) concluded that ISHs failure can be described using two
failure mechanisms: beam shear, and punching shear. Table 4.7 shows estimation of strength for
two specimens.
Table 4.5 indicates that 1D shear strength at seat can be a good estimate for maximum
capacity of ISHs. We generate analytical force–deformation relationship using the aforementioned
background, and compared with the experimental backbone curve of specimens S1 & S2
introduced by Hube and Mosalam in 2007. Figure.4.21 shows 1D shear failure mechanism that
results in inclined crack in the seat, and effective reinforcements that develop this type of failure.
42
Figure 4.21 Reinforcement details and in-span hinge geometry [Hube and Mosalam 2008].
ACI 2008 defines total shear resistance as summation of two contributions, VS for
reinforcing steel and VC for concrete as illustrated in Equations (4.24-4.26). In these equations bw
is total diaphragm length, d is effective seat depth, AD1 is the total area of reinforcement D1, Av1 is
the total area of reinforcement V1, θ is the angle between D1 and horizontal line in longitudinal
direction of bridge, and β is the angle between vertical reinforcement, V1, and longitudinal
direction of bridge which is π/2 . D1 and V1 are two critical reinforcements in1D shear failure
mechanism in ISH.
(4.24)
(4.25)
Table 4.8 shows parameters and their corresponding values for specimen S1 & S2 which
are used in this study. Parametric study of reinforcing steel which was conducted by Hube and
Mosalam [2012] shows that the variation of the diagonal bars (D1), and the vertical bars of the seat
(V1) have significant effect on the vertical capacity of ISH.
43
Figure 4.22 represents the idealized shear-deformation relationship of ISH. This idealized
model is simulated using combination of Bilin material in OpenSees along with damage states at
each level observed by Hube and Mosalam [2008].
Figure 4.22 Idealized shear deformation relationship for ISH (based on Hube and Mosalam 2008).
In general, three damage states can be defined for 1D shear failure mechanism as described
below. These three damage states are marked on Figure 4.22 as I, II and III.
Damage State I
This level of damage corresponds to onset of yielding of diagonal bar D1 (Figure 4.21).
Deformation and force can be obtained as follows:
(4.28)
(4.29)
In this study d = 9.75”, f y = 69.5 ksi for the headed bar, = 1.0 for uncoated
reinforcement, and fC' = 63.0 ksi. la , which is defined as a crack region, is approximately equal
to h based on experimental results (Figure 4.23).
44
Figure 4.23 One-dimensional shear failure mechanism of in-span hinge.
Experimental results show that diagonal bar D1, yields approximately at the 50% of the
maximum applied load. As such, shear capacity of the ISH at this level can be obtain as bellow:
(4.30)
Damage State II
This level corresponds to the onset of 1D shear failure in the seat due to initiation of yielding of
vertical bars V1. At this damage level, cracks propagate toward point C as shown in Figure 4.23
since the crack width is measured at the location reinforcement near the compression toe (Point C
in Figure 4.23), the above equation for can be express as:
(4.32)
where s is the average reinforcement spacing in the seat and is approximately equal to 4” for both
specimens. And is implemented because of the angle between crack and reinforcements crossing
the crack which is equal to 45 degrees. Based on experimental results, strain of vertical bars, V1,
for both specimens at the maximum applied load reach the value of 0.04.
Damage State III
This level of damage corresponds to full reduction in contribution of concrete in shear resistance
mechanism (see Figure 4.26). Stress-force relationship of diagonal bars of the seat in experimental
results shows that strength of D1 increases after reaching the yielding point (Figure 4.24). Therefore,
the in-span hinge capacity at this level only relies on diagonal bars and hence Equation (4.33):
(4.33)
45
Figure 4.24 Experimental Stress-Load relationship of specimen S1 and S2 [Hube and Mosalam 2008].
Based on experimental force-deformation results, specimens S1 and S2 show significant
reduction in shear capacity when vertical displacements approximately reach 1.0 in and 2.0 in
respectively. The value of strain in steel rebars were back calculated from the assuming rigid body
deformation around point C and using the top displacement data.
(4.34)
(4.35)
(4.36)
Generally for all specimens, except specimen S3, significant reduction of ISH capacity
occurs at the displacement of 1.2 when strain of D1 reaches 0.04 (Figure 4.25).
46
Figure 4.25 Comparison of load-displacement envelops at bearings of tested specimens [Hube and
Mosalam 2010].
According to the three stages of damage, analytical backbone curve of a bridge in-span
hinge can be estimated (Figures 4.26 and 4.27).
500
400
Force(Kips)
300
200
100
0
0 0.2 0.4 0.6 0.8 1 1.2
Displacement(in)
(a) (b)
Figure 4.26 Load-deformation relationship for specimen S1: (a) experimental results, (b) analytical
model based on [Hube and Mosalam, 2008].
47
500
400
Force(Kips)
300
200
100
0
0 0.5 1 1.5 2 2.5
Displacement(in)
(a) (b)
Figure 4.27 Load-deformation relationship for specimen S2: (a) experimental results, (b) analytical
model based on [Hube and Mosalam 2008].
Figure 4.28 Two Failure mechanisms in interior shear key based on [Megally et al., 2001].
The experimental program by Megally et al., [2001] consists of two groups of tests. In the
first group, 3 shear keys with similar height-to-depth ratio (denoted with α), width-to-depth ratio
(denoted with β), and reinforcement ratio (denoted with ρ), were considered and subjected to three
different loading protocols: monotonic, quasi-static reversed cyclic, and dynamic reversed cyclic.
In the second group of tests, 4 specimens with different α and ρ were evaluated under dynamic
cyclic load. Figure 4.29 shows the force deformation of specimens in the first group. Based on
these 7 tests, Megally et al. [2001] recommended an analytical force-deformation relationship for
interior shear keys with the reinforcement ratios between 0.32 percent and 0.63 percent, aspect
ratios between 0.30 and 0.50, and width-to-depth ratio around 0.70. In this study, behavior of
48
interior shear keys simulated using a Hysteretic material in series with an Elastic-Perfectly Plastic
Gap Material (EPPG) in OpenSees (Figure 4.30). Since the interior shear keys work in both
tension and compression, two EPPG materials are defined. All materials are uniaxial material type
as defined in OpenSees platform.
Figure 4.29 Two Failure mechanisms in interior shear key based on [Megally et al., 2001].
49
5 Robust Integration and Solution Algorithms
Selection
5.1 INTRODUCTION
In order to ensure robustness of the NLTA of bridges with the enhanced modeling options, two
analysis approaches are considered. First approach is the use of integration methods alternative to
the commonly used (including the bridge analyses conducted by Caltrans engineers) Implicit
Newmark (IN) integration, while the second approach is the improvement of the robustness of IN,
in the cases where the considered alternative integrators are not applicable. This chapter covers
these two approaches.
Regarding the first approach, considered integrators are the Explicit Newmark (EN) and
Operator-Splitting (OS) methods, which do not require iterations. The eigen property of these
integrators is theoretically investigated for nonlinear response. Moreover, the accuracy of these
integration methods is investigated using a nonlinear test problem with an available exact solution.
In addition, eigen property and accuracy of the implicit TRBDF2 method, which is not as
commonly used as IN, but is observed to provide improved convergence, is investigated for
nonlinear response. Applicability of the mentioned alternative integration methods to the NLTA
of bridge models described in Chapter 3 is also explored.
The equations of motion of a multi-degree of freedom (MDOF) system under an external dynamic
force excitation can be defined as follows:
cu f p
mu (5.1)
where m is the mass matrix, c is the damping matrix, and u , u , f , and p are the acceleration,
velocity, restoring force, and external force vectors, respectively. The restoring force can generally
be defined as a function of displacement. Due to several factors, such as the random variation of
the external force with time, e.g. due to earthquake shaking, and the nonlinear variation of the
restoring force vector with displacement, closed form solution of Equation 5.1 is not possible
[Chopra 2006]. Therefore, numerical integration methods are used for the sought solution.
Differences between direct integration methods are mainly introduced by the way they
handle Equations 5.2-5.4, arranged as the Newmark difference equations for displacement and
velocity and the discretized dynamic equilibrium equation, respectively.
ui 1 ui t u i
t 2 1 2 u i 1
2 u
i (5.2)
2
u i 1 u i t 1 γ u i 1
i γ u (5.3)
50
i 1 cu i 1 fi 1 p i1
mu (5.4)
In the above equations, γ and parameters define the variation of accelerations over a
time step, t . For example, γ 1 2 and 1 4 represent constant average acceleration over one
time step, while γ 1 2 and 1 6 define linear variation of acceleration during the time step.
The following sub-sections describe the alternative integration methods, namely the
Explicit Newmark, Operator-Splitting, and TRBDF2 integrators and the commonly utilized
Implicit Newmark integration method, and the corresponding algorithms. This description starts
from the basic three Equations 5.2-5.4 defined above and emphasizes departure points and
differences between these methods.
Explicit Newmark (EN) integration [Newmark, 1959] is defined by setting 0.0 . Accordingly,
Equation 5.2 becomes
u i 1 u i t u i
t
2
u i (5.5)
2
Substituting Equations 5.3 and 5.5 in Equation 5.4, the linear system of equations defined
by Equations 5.6 is obtained, which can be solved to determine the acceleration. Subsequently,
Equation 5.3 is used to determine the velocity.
i 1 p eff
m e ffu
m e ff m t γ c (5.6)
p e ff p i 1 f i 1 c u i t 1 γ u
i
It is noted that setting 0.0 makes the method explicit. In this way, the implicit nature
of Equation 5.2 is eliminated by removal of the u i 1 term. The algorithm for an integration time
step of the EN method is summarized as follows:
1. Compute the displacement using Equation 5.5.
2. Obtain the restoring force, f , corresponding to the computed displacement from the
constitutive relationships of the defined materials and elements using a state determination
method [Spacone et al. 1996].
3. Calculate the acceleration by solving the linear system of equations defined by Equations 5.6.
4. Determine the velocity using Equation 5.3.
5. Increment i and proceed with the next integration time step.
51
5.2.2 Operator-Splitting (OS) method
Similar to the EN method, the OS method [Hughes et al. 1979] eliminates the implicit nature of
the solution algorithm. However, instead of the direct elimination adopted by the EN, OS uses a
prediction-correction technique. The predicted displacement, u~ , is obtained by neglecting the
i 1
i 1 term in the bracketed part of Equation 5.2, i.e.
u
~ u t u t
2
u i 1 i i 1 2 u i (5.7)
2
After the prediction of the displacement, the method is defined by setting the restoring
force of the integration time step as the sum of the restoring force corresponding to the predicted
~
displacement, fi 1 , and the difference between the corrected and predicted displacements
multiplied by the tangential stiffness matrix, k T , i.e.
f i 1 fi 1 k T u i 1 u
~
~
i 1 (5.8)
The difference between the corrected and predicted displacements is defined with Equation 5.9
~
ui 1 ui 1 u (5.9)
i 1
f i 1 fi 1 k T u i 1
~
(5.10)
Subtracting Equation 5.7 from Equation 5.2 leads to the relationship between u i 1 and u
i 1 , i.e.
ui 1
i 1
u (5.11)
t 2
Substituting Equation 5.11 in the Newmark difference equation for velocity (Equation 5.3) gives,
u i 1 u i t 1 u
i ui 1 (5.12)
t
A linear system of Equations 5.16 is obtained by substitution of Equations 5.10, 5.11 and 5.12 in
Equation 5.4, which can be solved to determine the displacement along with Equation 5.14.
k e ff u i 1 p eff
1 γ
k e ff m c kT (5.13)
Δt β2
Δt β
p e ff p i 1 fi 1 c u i t 1 γ u
i
~
52
~ u
u i 1 u (5.14)
i 1 i 1
The algorithm for an integration time step of the OS method is summarized as follows:
1. Compute the predicted displacement using Equation 5.7.
~
2. Obtain the restoring force, f , corresponding to the predicted displacement from the
constitutive relationships of the defined materials and elements using a state determination
method [Spacone et al. 1996].
3. Determine the difference between the corrected and the predicted displacements by solving the
linear system of equations defined by Equations 5.13.
4. Determine the acceleration using Equation 5.11.
5. Calculate the corrected displacement using Equation 5.14.
6. Obtain the restoring force corresponding to the corrected displacement following the procedure
outlined in step 2 above.
7. Determine the velocity using Equation 5.12.
8. Increment i and proceed with the next integration time step.
It is noted that the initial stiffness matrix, k I , was used instead of k T in some of the studies
in literature. For example, Combescure and Pegon (1997) used k I in hybrid simulations due to
the difficulties in obtaining the tangent stiffness matrices of the test specimens.
The implicit nature of the solution algorithm is eliminated in the EN, OS, and OSG methods. In
contrast to these methods, the Implicit Newmark (IN) integrator [Newmark, 1959] treats the
governing equations (difference and dynamic equilibrium equations) directly without altering their
implicit nature. Rearranging the time-discrete equilibrium equations (Equation 5.4), one obtains,
pi 1 mu
i 1 cu i 1 fi 1 0 (5.15)
53
In order to solve for the displacement increments, the difference Equations 5.2 and 5.3 are
redefined such that the velocities and accelerations are expressed in terms of displacements.
u i 1 1
i 1
u u i 1u i (5.16)
t t
2
2
u i 1 u i 1 u i (5.17)
u i 1 u i 1 1 u i t 1 u
i (5.18)
t 2
g xk xk g xk (5.19)
where g and g represent the function and its derivative with respect to x , respectively, x is the
root of the function, k is the iteration number and x k is the difference between the value of x
in the current and previous iterations. If the left-hand side of Equation 5.15 is considered as the
function g in Equation 5.19, one can write the following in terms of displacement,
g uik1 uik1 g uik1 (5.20)
g u
ik1 mu
ik1 cu ik1 fik1 pi 1 (5.21)
g u ik1
1
m
γ
c kT (5.22)
t β
2
t β
where k T is the tangential stiffness matrix corresponding to the displacement vector uik1 , which
can be obtained as a result of a state determination method [Spacone et al. 1996]. Substitution of
Equations 5.21 and 5.22 in Equation 5.20 leads to the linear system of Equations 5.23 in the same
format as the other methods explained earlier.
At this point, it is beneficial to state that all three methods discussed above reduce the
nonlinear differential equations of motion to a system of linear algebraic equations. However,
54
depending on the way each method treats the three basic Equations 5.2-5.4, the resulting
coefficient matrix (or the Jacobian matrix, referring to the derivative on the left-hand side of
Equation 5.20), m e ff for EN, and k e ff for OS and IN, and the effective load vector, p e ff , differ
from one method to the other. Accordingly, these differences determine the adequacy and ease of
application of each method as explained later.
After the determination of the displacement increment for iteration k from Equations 5.23,
the method continues by the calculation of the displacement, velocity and acceleration for iteration
k 1 using Equations 5.24-5.26, respectively.
u ik11
t
u ik11 u i 1 u
t
1 u
i (5.25)
2
i
ik11
u
u ik11 u i
1
u i
1 u
i (5.26)
t
2
t 2
An iterative method requires an initial guess for the sought value, i.e. for u ik11 . For the
Newton-Raphson method, acceleration of the previous iteration can be used as the initial guess as
defined in Equation 5.27. Subsequent substitution of this equation into Equations 5.2 and 5.3 leads
to the corresponding velocity and displacement, i.e. u ik11 and uik11 vectors as follows,
uik11 ui (5.27)
u ik11 1 u i t 1 u
i (5.28)
2
1
ik11
u u i 1 u
i (5.29)
t 2
The algorithm for an integration time step of the IN method is summarized as follows:
1. Determine the initial guess, k 1 , for acceleration from Equation 5.27 and the corresponding
velocity and displacement from Equations 5.28 and 5.29, respectively.
For each iteration k =1: N, where N is the total number of iterations:
2. Obtain the restoring force, f , corresponding to the computed displacement from the
constitutive relationships of the defined materials and elements using a state determination
method [Spacone et al. 1996].
3. Determine the acceleration increment by solving the linear system of equations defined by
Equations 5.23.
4. Compute the acceleration, velocity and displacement using Equations 5.24-5.26.
55
5. Check convergence by comparing a calculated norm with a defined tolerance value. If the norm
is smaller than the tolerance, set N k , increment i and proceed to the next time integration
step; otherwise, increment k and go to step 2.
Regarding the above algorithm, two remarks are stated as follows:
The presence of the convergence check requires at least two iterations. Therefore,
unless a solution is separately coded for a linear case, a general nonlinear analysis
software requires at least two iterations for the IN integration, even for the case of a
linear problem.
Different norms can be used for the convergence check in step 5. Examples of these
norms are the displacement increment, unbalanced force and energy norms.
TRBDF2 method [Bank et al., 1985; Bathe and Baig, 2005; Bathe 2007] is a composite integration
method, that uses IN and three point backward Euler scheme, alternately in consecutive integration
time steps. The first step uses the IN method with constant average acceleration, i.e. with γ =1/2
and =1/4 in the difference Equations 5.2 and 5.3. The consequent step uses the equations of the
three-point Euler backward method (Equations 5.30 and 5.31) for the relation between the
displacement and velocity and acceleration, instead of the Newmark difference Equations 5.2 and
5.3.
u i 1 4u i 3u i-1
u i 1 (5.30)
t
Following the Newton-Raphson method as demonstrated for IN, the linear system of
Equations 5.32 is obtained.
k e ff u ik1 p e ff
9 3
k e ff m c kT (5.32)
t 2
t
9u i 1 12u i 3u i 1 4u i u i 1 3u ik1 4u i u i 1 k
k
p e ff p i 1 m c
f i 1
t 2
t t
After the determination of the displacement increment for iteration k from Equations 5.32,
the method continues by the calculation of the displacement, velocity and acceleration for iteration
k 1 using Equations 5.33-5.35, respectively.
56
k 1 u ik11 4u i 3u i-1
u i 1 (5.34)
t
It is noted that the consecutive steps are considered as the sub-steps of one step in [Bathe
and Baig, 2005; Bathe 2007], rather than considering them as consecutive steps, which is exactly
the same as the formulation presented above, but with a time step t 2 .
Bathe and Baig, 2005 and Bathe 2007 used the method in structural dynamics to conserve
energy and momentum at large deformations (not necessarily involving material nonlinearity)
which the IN method may fail to do so and become unstable. Herein, it is considered not because
of its superior stability performance than that of IN, but because of its better convergence behavior,
due to the numerical damping provided by the Euler backward method, as discussed in the next
chapter.
An unconditionally stable integration method requires that the solution for any initial conditions
does not grow without bound for any time step t . It also requires that any error of displacements,
velocities and accelerations at time t does not grow. On the other hand, conditionally stable means
that the same requirements only hold under certain conditions, for example when t Tmin , where
Tmin is the smallest natural period (corresponding to the highest mode of vibration) of interest of
the structure under consideration, is less than a certain value.
For each integration method, the following relationship, between the kinematic quantities
at time step i+1 and those at time step i, can be established for a single degree of freedom (SDOF)
system.
u i 1 u i
u A u Lp
i 1 i i 1 (5.36)
u i 1 u i
57
where λ i values are the eigenvalues of A. Recently, systematic and fundamental Lyapunov
stability and accuracy analyses of the above-mentioned four integration algorithms for both SDOF
and MDOF systems were conducted by Liang and Mosalam (2015, 2016a-c).
The following two sub-sections present the eigen properties of the four integration methods
described in Section 5.2. The discussion of the eigen property of these four methods is grouped
into three groups: OS with both initial and tangential stiffness covered in sub-section 5.3.1, EN
and IN covered in sub-section 5.3.2, and TRBDF2 covered in sub-section 5.3.3.
pi 1 fi 1 cu i t 1 ui
~
ui 1 (5.38)
m t c t k I
2
~
The tangential stiffness k T can be approximated by linearization as follows,
~
~ f f
k T ~i 1 i (5.39)
u i 1 u i
where u i and f i are the displacement and restoring force of the time step i, respectively. Using
the dynamic equilibrium equation at the time step i, the restoring force, f i , can be defined as
follows,
~
Substituting Equation 5.40 in Equation 5.39 gives the following approximation of fi 1 ,
fi 1 p i mu i cu i k T ~
u i 1 u i
~ ~
(5.41)
Substitution of Equation 5.41 in Equation 5.38 leads to the following explicit expression of u i 1
based on the displacement, velocity and acceleration of the previous step,
p i 1 p i mu i k T ~
u i 1 u i ct 1 u i
~
u i 1 (5.42)
m t c t k T
2 ~
~ t 2
p i 1 p i m ct 1 k T
1 2 ui ~kT t u i
u i 1 2 (5.43)
m t c t k T
2 ~
58
Substitution of Equation 5.37 in Equations 5.3 and 5.11 results in the explicit expressions relating
u i 1 and u i 1 to the displacement, velocity and acceleration of the previous step.
Bathe and Wilson (1976) state that (1) The stability of an integration method is determined
by examining the behavior of the numerical solution for arbitrary initial conditions, (2) In general,
small values of viscous damping do not change the overall stability characteristics of an integration
scheme with respect to an evaluation conducted with zero damping. Accordingly, the following
stability analyses are conducted for the case of p (external force) and c (damping) being equal to
zero. For γ 1 2 and 1 4 (average acceleration method), the approximation operator A
becomes
1 B 4B
B 1 t B 1 0
t 1 B
A 0 (5.44)
B 1 B 1
t 2 t
1
2B 1 B 1
B
t k T
2~
1 B 2 B
λ1,2
B 1 (5.45)
λ3 1
~
For positive k T , using the fact that B is imaginary, Equation 5.45 can be modified as
λ1,2
1 B2 2B B 12 11 (5.46)
B 12 B 12
For the zero damping case, Newmark’s method is conditionally stable if [Newmark 1959]:
γ 1/2
β 1/2 (5.47)
Δt 1
Tmin 2π γ/2 β
59
β0
Δt 1 (5.48)
Tmin 2π γ/2
2β 1/2 (5.49)
However, if γ 1 2 , errors are introduced. Accordingly, the common use of the IN integration
method assuming average acceleration, i.e. γ 1 2 and 1 4 , is unconditionally stable.
Equations 5.47-5.49 are only valid for linear problems. In order to investigate the case
when nonlinearity is present, the SDOF nonlinear system discussed in Section 5.3.1 is considered.
Based on Equation 5.4, the acceleration can be determined as follows,
pi 1 cu i 1 fi 1
ui 1 (5.50)
m
f i 1 f i
kT (5.51)
u i 1 u i
The restoring force at the time step i, f i , can be determined with Equation 5.40. Thus f i 1
is approximated as follows,
Substitution of Equation 5.52 in Equation 5.50 along with the Newmark difference equations lead
to Equation 5.53. From this equation, the explicit expression relating u i 1 to the displacement,
velocity and acceleration of the previous step can be stated.
Using Equation 5.53 and the Newmark difference Equations 5.2 and 5.3, the explicit
expression relating u i 1 to the displacement, velocity and acceleration of the previous step can be
obtained. For the case of average acceleration, i.e. γ 1 2 and 1 4 , the relationship between
u i 1 and the kinematic quantities of the previous step can be represented as follows,
2C 4(1 C)
u i 1 u i u i (5.54)
C Ct
60
k T t
2
where C 1 . Substitution of Equation 5.54 in Equations 5.3 and 5.11 results in the
4m
explicit expressions relating u i 1 and u i 1 to the displacement, velocity and acceleration of the
previous step, where the approximation operator A is expressed as follows,
2 C 4(1 C)
C Ct
0
t 2C
A 0 (5.55)
C C
t 2 t
1
2C C
2 C 2 1 C
λ1,2
C (5.56)
λ3 1
It is obvious that C is always larger than one for positive tangent stiffness, and thus
λ1,2
2 C2 4C 4 C2
11 (5.57)
C2 C2
2(D 1)
ui 1 Dui u i (5.58)
t
k T t
2
where D 1 . Substitution of Equation 5.58 in Equations 5.3 and 5.11 results in the
2m
explicit expressions relating u i 1 and u i 1 to the displacement, velocity and acceleration of the
previous step, where the approximation operator A is expressed as follows,
2(1 D)
D 0
t
t (1 D)
A D 0 (5.59)
2
t
2
t 1
2
61
λ1,2 D D 2 1
(5.60)
λ3 1
1 D 1 (5.61)
i.e.
t 1
(5.62)
TT
. The conditions for A 1 are summarized in Table 5.1 for the case of zero
m
where TT 2
kT
viscous damping ( 0 ), which is the most critical case for the stability analysis of direct
integration algorithms.
62
m2 u2
k2
m 0 c c c2 k k 2 k2
m 1 , c 1 2 , k 1
k 2
m1 u1
0 m2 c2 c2 k2
k1
Figure 5.1 A two DOF system used for eigen analysis of the OS integration.
The approximation operator A for this two DOF system is a 6×6 matrix relating the
response quantities of the previous time step and those of the current time step
u1i 1 u 1i
1 1
u i 1 u i
u i 1
1 u1i
2 A 2 (5.63)
u i 1 u i
u 2 u 2
i21 i2
i 1
u u i
where the superscripts denote the first and second DOF. For the case of c1=c2=0, A is as follows,
1
1 1 t 2 t 2
4
0 3 4
2 2 t 2 2
0 1 1 2 0 3 4
t t 2 t t
0 4
1
4
2 0
4
3
4
4
t 2
t 2 t 2 t 2
A (5.64)
1
0 3 4 1 1 t 2 t 2
4
0 2 2 2 2 t
3 4 0 1 1 2
t t t t 2
4 4 4 4
0 3 4 0 1 2
t 2 t 2 t 2 t 2
63
Table 5.2 Coefficients for the OS method for a 2 DOF system using the tangential stiffness.
Coefficients Coefficients
~ ~
k1T k 2T t
~ ~
4m 2 k1T k 2T ~
4 m 2 k 2T
1 t 3
tR
R
2
1~ ~
k1T k 2T t 2
4
4 m 1
t
m
2
2 ~ ~
~
k 2T m1 k1T k 2T m 2 4
~
2 m 2 k 2T
R R
~
~ ~ 4m1k 2T ~
1 k1T k 2T t 3 4 m 1 k 2T
t
tR
R
2
1~ ~
k1T k 2T t 2
4
4 m
t
1 m
2
2
~ ~
~
k1T k 2T m 2 - k 2T m1
4
~
2 m 1 k 2T
R R
R
~ ~
k1T k 2T
16m1m 2
4
~
~ ~
k 2T m1 k1T k 2T m 2
t 4 t
~ ~ ~
For the case of m1 m2 m , and k1T k 2T kT 0 , eigenvalues of the approximation
operator A are obtained as follows,
λ1, 2 1
m
where a ~ . The following inequality can be obtained in a straightforward manner:
k T t 2
Using Equation 5.66, Equation 5.67 is obtained, which proves that the OS method is
unconditionally stable explicit method for the two DOF system of Figure 5.2 with m1 m2 m ,
~ ~ ~
and k1T k 2T kT 0 .
64
λ 3 ,4
16a 2
2
4 5a 1 8 5a 128 5a 3 24a 128a 2 384a 3
16a 2
12a 1
2
λ 5 ,6
16a 2
2
4 5a 1 128 5a 8 5a 24a 128a 384a
3 2 3
16a 2
12a 1
2
As mentioned before, two consecutive steps are considered as sub-steps of one integration time
step in the TRBDF2 method. However, these two sub-steps are explained as two consecutive
integration time steps in Section 5.2.4 in order to compare the TRBDF2 method to the IN method.
However, stability analysis requires relating the response quantities in the previous integration
time step to those in the current. Therefore, the stability analysis in this section is conducted by
considering two sub-steps, defined as follows,
t
u i 0.5 u i u i u i 0.5 (5.68)
4
t
u i 0.5 u i ui ui 0.5 (5.69)
4
1 4 3
u i 1 ui u i 0.5 u i 1 (5.71)
t t t
1 4 3
ui 1 u i u i 0.5 u i 1 (5.72)
t t t
For the SDOF system with p (external force) and c (damping) equal to zero, Equations 5.70
and 5.73 become
65
f i 1 f i 0.5 f f f f
kT i 0.5 i i 1 i (5.75)
u i 1 u i 0.5 u i 0.5 u i u i 1 u i
t
1 E u i ui
u i 0.5 2 (5.76)
1 E
k T t 2
where E . Substitution of Equation 5.76 in Equation 5.68 gives
16m
t t 2
u i 0.5 u i ui ui (5.77)
2 1 E 8 1 E
4 3 3
u i u i 0.5 u i u i u i 0.5
u i 1 t 16E 4E (5.78)
9
1
16E
Substitution of Equations 5.76 and 5.77 in Equation 5.78 gives the explicit expression that relates
u i 1 to the displacement, velocity and acceleration of the previous time step as follows,
u i 1
9 47E u i 9 5E t u i (5.79)
16E 9 1 E
Substitution of Equations 5.76 and 5.79 in Equation 5.72, and substitution of Equations 5.72 and
5.75 in Equation 5.74 gives the explicit expression that relate ui 1 and u i 1 to the displacement,
velocity and acceleration of the previous time step as follows,
80E 2
144E u 9 47E u
t
i i
u i 1 (5.80)
16E 9 1 E
t 2
u i t 9 5E u i 2E 9 u i
u i 1 2 (5.81)
16E 9 1 E
Thus, the approximation operator A is expressed as follows,
66
80E 2 144E
9 47E 0
t
1
A t 9 5E 9 47E 0 (5.82)
16E 9 1 E 2
t
2E 9 t 9 5E 1
2
9 - 47E 36 - 20E -E
λ1,2 (5.83)
16E 2 25E 9
λ3 1
For positive k T
9 - 47E 36 - 20E E
2 2
λ1,2
16E
2
2
25E 9 (5.84)
400E 3 796E 2 450E 81
1
256E 4 800E 3 913E 2 450E 81
The accuracy of the numerical integration algorithms depends on several factors, e.g. the loading,
the time-step size, and the physical parameters of the system. In order to develop an understanding
of this accuracy, a nonlinear test problem with an available closed-form exact solution is analyzed
in this section.
Consider a simple pendulum (Figure 2) of length l , forming a time-dependent angle (t )
with the vertical axis and undergoing time-dependent angular acceleration (t ) . The governing
equation, initial conditions, exact solution, and period of vibration are summarized in Table 4
where g is the gravitational acceleration, n K r 0t , 0 g l , K r is the complete
elliptical integral of the first kind, and snn; r is the Jacobi elliptic function (Abramowitz and
Stegun 1972).
67
Table 5.3 Nonlinear pendulum
Exact solution (Beléndez et al. 2007) t 2 arcsin sin 0 2snn; r , r sin 2 0 2
Period T 4 K r 0
Figures 3 and 4 present the period elongation and the amplitude decay of the investigated
integration algorithms for θ0 0.10 π and θ0 0.50 π , respectively. The period is shortened using
explicit Newmark (Chopra 2006), and elongated by the other algorithms. It is observed that
OStangent and implicit Newmark present similar period elongations. The TRBDF2 has the smallest
period change while it is about twice computationally expensive compared to the other algorithms.
Considering roughly the same computational efforts, e.g. t T 0.08 for TRBDF2 and
t T 0.04 for the others, the accuracy becomes comparable. Moreover, the accuracy of all
algorithms is indifferent for the integration time steps required for accuracy, i.e. t T 0.01
(Bathe 2006). All algorithms do not result in any significant amplitude decay except TRBDF2,
which presents some amplitude decay due to introduced numerical damping. Up to t T 0.1 ,
period elongation (< ±3%) and amplitude decay (< 1%) are acceptable.
68
8 8
Percentage period elongation OS OS
0 0
-2 -2
-4 -4
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
t / T t / T
Figure 5.3 Period elongation and amplitude decay for the pendulum problem with θ0 0.10 π .
8 8
OS OS
Percentage period elongation
0 0
-2 -2
-4 -4
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
t / T t / T
Figure 5.4 Period elongation and amplitude decay for the pendulum problem with θ0 0.50 π .
Applicability of the utilized alternative integrators for NLTA of the mentioned bridges is
investigated in this section.
t 1
(5.85)
Tn
69
where Tn is the period of the highest mode of vibration of the analyzed structure. Equation 5.85
restricts the use of the EN method for structures with massless degrees of freedom (DOF), since
the presence of such DOF results in a singular mass matrix that yields zero-period modes.
Accordingly, the EN method is not applicable to building and bridge structures which are modeled
with massless rotational DOF, unless these DOF are condensed out. It is within the future
objectives of this study to implement a condensation algorithm in OpenSees [McKenna et al. 2010]
to facilitate broader use of the EN algorithm. Therefore, the EN algorithm is not pursued further
for the NLTA of the investigated bridge systems in this report.
As mentioned before, the version of the OS algorithm considered herein (i.e. the one using the
tangential stiffness matrix), possesses similar stability properties and accuracies to those of the IN
integration. Moreover, the TRBDF2 algorithm is considered because of its superior stability and
better convergence behavior due to numerical damping.
Figure 5.5 Superstructure, column bent and abutment of Bridge B with identified nodes for NLTA
results [Kaviani 2011].
The NLTA is conducted for both abutment modeling approaches described in Chapter 3 of
Bridges A, B, and C using the IN, OS, and TRBDF2 algorithms under 40 pulse-like three-
component ground motions described in [Jayaram et al. 2010]. These ground motions, which are
not selected to match any target response spectrum, contain strong velocity pulses of varying
periods ranging from 1.0 sec. to 12.9 sec., in their strike-normal direction. These velocity pulses
are expected to occur near fault ruptures in some ground motions. The pulse-like ground motions
tend to introduce highly nonlinear responses. Therefore, they are selected in this study to assess
the validity of the discussed integration algorithms for NLTA of bridges. The results from the IN
70
algorithm, which is considered as the common algorithm used for NLTA, are considered as the
reference results. It is to be noted that the use of a numerical solution as the reference is attributed
to the lack of an available closed-form exact solution or reliable experimental data for the analyzed
bridges under the selected earthquake excitations. Although analyses are conducted for all 40
ground motions, detailed results are presented herein for selected ground motions which result in
the highest level of nonlinearity.
Figure 5.5 illustrates the nodal and element designation of Bridge B with Type I abutment
modeling. Two-digit numbers are used to represent the nodes in the columns (e.g. node12), while
the nodes of the superstructure are represented by three-digit numbers (e.g. node105). Five-digit
numbers are assigned to the zerolength elements in the abutments (e.g. 10100 associated with node
100). For comparison of the results, three EDPs are selected, namely the abutment unseating
displacement, column drift ratio, and column base shear. As identified in Figure 5.5 for Bridge B,
these three EDPs correspond to the longitudinal displacement of Node 100, column drift ratio of
Node 12, and column base shear of Node 11. Similar nodes are also utilized for Bridges A and C
to investigate the NLTA results. The comparison is conducted quantitatively by using the error
measure Maxerror , defined as follows:
Maxalt Maxref
Maxerror (5.86)
Maxref
where Maxref and Maxalt are the maximum absolute response of the considered EDP provided
by the IN and by the other algorithms, respectively. The NLTA is conducted for all three bridges
with detailed results presented for Bridge B and brief ones given for Bridges A and C. Figure 5.6
shows the Maxerror of the OS algorithm for the three selected EDPs of Bridge B, analyzed with
both abutment modeling approaches I and II (Figure 3.3), from all 40 ground motions.
0.4 0.4
Unseating displcement Unseating displcement
0.2 0.2
0 0
-0.2 -0.2
0 10 20 30 40 0 10 20 30 40
Maxerror (%)
Maxerror (%)
0.4 0.4
Column drift ratio Column drift ratio
0.2 0.2
0 0
0 10 20 30 40 0 10 20 30 40
0.4 0.4
Column base shear Column base shear
0.2 0.2
0 0
0 10 20 30 40 0 10 20 30 40
GM # GM #
Figure 5.6 Maxerror of the OS algorithm for the three selected EDPs of Bridge B.
It is observed that all obtained errors are insignificant (< 0.2%). The abutment unseating
displacement from IN and OS with Type II modeling is plotted in Figure 5.7a for ground motion
#21 (Earthquake: Northridge – 01; Station: Sylmar – Olive View Med FF), which yields the largest
71
Maxerror as shown in Figure 5.6b. Figure 5.7b represents the corresponding moment-curvature
plot ( M ) from the IN to reflect the obtained high level of nonlinearity in this case.
5
5 x 10
1.2
IN
Longitudinal Displacement (in)
4
OS 0.8
3
2 0.4
M
0
0
-0.4
-1
-2 -0.8
-3
0 10 20 30 40 -1.2 -0.8 -0.4 0 0.4 0.8 1.2
Time (s) -3
x 10
a) Abutment unseating displacement b) Moment-curvature response
Figure 5.7 Comparison of the IN and OS algorithm results for NLTA of Bridge B with Type II
abutment modeling (Ground motion #21 as an example).
The good match of the time history responses for the highly nonlinear case in Figure 5.8a
and the small values of the error measure in Figure 5.6 indicate that the explicit OS algorithm is
suitable for the intended objective of overcoming the convergence problems in the NLTA of RC
highway bridges, while maintaining the accuracy of the results provided by the IN method. Similar
good match and small error levels are also obtained for Bridges A and C.
Figure 5.8 shows the Maxerror of the TRBDF2 algorithm for the three selected EDPs of Bridge B,
analyzed with both abutment modeling approaches I and II (Figure 3.3), from all 40 ground
motions. Analogous to the OS results in Figure 5.6, small errors in Figure 5.8 imply that the
accuracy of the implicit TRBDF2 and IN algorithms are comparable. Furthermore, the TRBDF2
algorithm exhibits superior convergence features as discussed in the next section.
72
0.4 0.4
0.2 Unseating displcement 0.2 Unseating displcement
0 0
-0.2 -0.2
0 10 20 30 40 0 10 20 30 40
Maxerror (%)
Maxerror (%)
0.4 0.4
Column drift ratio Column drift ratio
0.2 0.2
0 0
0 10 20 30 40 0 10 20 30 40
0.4 0.4
Column base shear Column base shear
0.2 0.2
0 0
0 10 20 30 40 0 10 20 30 40
GM # GM #
Figure 5.8 Maxerror of the TRBDF2 algorithm for the three selected EDPs of Bridge B.
As demonstrated in the previous chapter, the explicit methods are suitable alternatives of the IN
method to avoid convergence problems. However, there exist some conditions where these
methods are not applicable, e.g. the case of significant stiffening response due to closing of the
gap elements in the above mentioned bridge models. Moreover, aside from the chosen integration
method, formulation of some of the elements, e.g. the force-based beam-column elements or the
materials, e.g. Bouc-Wen type, in OpenSees [McKenna et al. 2010] are iterative. Accordingly, the
IN method may be the only option for NLTA of models containing such elements and materials if
the convergence problems at the element and material levels cannot be eliminated while using the
alternative integrators.
The following sections investigate the effect of different parameters in improving the
convergence properties of the IN method while preserving its accuracy for the bridge models
described in Chapter 3. It is noted that this investigation is based on the displacement formulation
of the method, which is observed to result in an improved convergence performance compared to
the acceleration formulation. The considered parameters are categorized in five groups as follows:
1. Type and sequence of nonlinear equation solvers
2. Convergence test type
3. Convergence tolerance
4. Integration time step
5. Adaptive switching of integration algorithms
In order to achieve convergence, OpenSees [McKenna et al. 2010] allows trying various nonlinear
solvers consecutively for any iteration of an integration time step. Therefore, one of the potential
items that can be considered to improve convergence is the type and sequence of the nonlinear
73
equation solvers. This investigation is further divided into two sub-groups, namely, (a)
determination of the most suitable initial solver and (b) sequence of other solvers after the initial
one. Before presenting the results of this investigation, the following nonlinear solvers: Newton–
Raphson, Broyden, Newton with Line Search (NLS), Broyden–Fletcher–Goldfarb–Shanno
(BFGS), and Krylov-Newton, all available in OpenSees [McKenna et al. 2010], are briefly
described below.
Equation 5.4 can be written in a residual form as follows,
(5.87)
The statement of equilibrium requires that the residual forces are zeros, i.e.
(5.88)
The following sub-sections describe how each of the discussed nonlinear solvers attempt
to satisfy Equation 5.87. It is noted that the subscript i representing the number of the integration
time step is dropped to simplify the expressions in the following sub-sections.
The Newton-Raphson (NR) algorithm is based on linear approximation of the residual vector as
follows,
(5.89)
The superscript k denotes the iteration number within one time step and the matrix
is called the system Jacobian matrix, which can be denoted as follows,
(5.90)
where k T is the tangential stiffness matrix. The algorithm starts with an initial guess and iterates
with the following equations until a certain convergence criterion is met.
(5.91)
In NR algorithm, the Jacobian matrix is computed at every iteration, which is a complicated and
expensive operation. The idea behind Broyden method is to compute the whole Jacobian only at
the first iteration, and to do a rank-one update at the other iterations [Broyden 1965], i.e.
74
(5.92)
(5.93)
Then the algorithm proceeds with Equations 5.91 as in the Newton algorithm. It is to be noted that
the Modified NR is a special case where this rank-one update is ignored.
The direction of determined by the NR method is often correct. However, the same is not
always true for the step size . Furthermore, it is computationally less expensive to compute
the residual for several points along rather than forming and factorizing a new Jacobian
[Crisfield 1991].
In Newton with Line Search (NLS) algorithm, the regular NR method is used to compute
the . However, only a certain portion of the calculated is used to determine the
displacement in the next iteration as follows,
(5.94)
Four types of line search algorithm are available in OpenSees [McKenna et al. 2010]:
Bisection, Secant, RegulaFalsi and Interpolated. The different line search algorithms embrace
different root finding methods to determine the factor . A root to the function is defined as
follows,
(5.95)
(5.96)
75
5.6.1.4 Broyden–Fletcher–Goldfarb–Shanno (BFGS) Algorithm
From an initial guess u0 and an approximate Hessian matrix2 B 0 , the following steps are repeated
until convergence [Bathe and Cimento 1980]:
1) A direction is obtained by solving the following equation,
(5.97)
2) A line search is performed to find the step size for the kth iteration and update the
displacement as follows,
(5.98)
3) Set the Hessian matrix, which is updated for the next iteration as follows,
(5.99)
where
(5.100)
(5.101)
At each time step, instead of Equation 5.88, the Krylov-Newton algorithm seeks the solution to
the system of preconditioned residual equations [Scott and Fenves 2010]
(5.102)
where k 0 is the tangential stiffness at the first iteration of the time step. The solution to Equation
5.102 is equivalent to that of Equation 5.88 as long as k 0 is nonsingular. Thus, Equation 5.89
becomes
2
Hessian matrix is a square matrix of second-order partial derivatives of a function.
3
where i, j and k are unit vectors in the directions x, y and z , respectively.
76
(5.103)
where A is the identity matrix when modified NR algorithm is used, while when using
the NR algorithm. The tangent at the initial guess is used for the modified NR algorithm in the
iterations, instead of the current tangent.
The Krylov-Newton algorithm decomposes into two components:
(5.104)
where is the acceleration component and is the standard modified Newton component.
is further represented as a linear combination of the vectors from the subspace of
displacement increments with size m, i.e.
(5.105)
To satisfy Equation 5.103, the first step is to minimize the norm of vector ,
which represents an over-determined system of equations for the unknown coefficients
in Equation 5.105, by least-squares analysis [Golub and Van Loan 1996]. The second step for
satisfying Equation 6.17 is to solve the following,
(5.106)
where , and A is the identity matrix using the Modified N-R. Thus,
(5.107)
The displacement increment is then calculated by summing the two components determined with
Equations 5.105 and 5.107.
5.6.1.6 Results
Based on the convergence situations from the simulations with several relatively strong ground
motions (GM1, GM11, GM18, GM19, GM28, and GM31) with scale factors ranging from 1.0 to
2.0, NLS proves to be the most suitable initial solver. Table 6.1 shows the convergence condition
for different scales of GM31, where it can be observed that all the simulations which used the NLS
solver as the initial solver are completed, whereas the simulations with other initial solvers fail to
converge for some of the scales.
Using the NLS as the initial solver, no difference was found when different subsequent
orders of other solvers (NR, Broyden, BFGS, and Krylov-Newton) were investigated. It may be
concluded that as long as a suitable initial solver is determined, the order of subsequent other
solvers do not have significant impact on the convergence. It is beneficial to note that this finding
may be specific to the investigated structure since this investigation is not repeated for other
structures. However, it is still a useful conclusion because of two reasons. First, this finding sets
the NLS to be a suitable initial solver as the first simulation trial of future nonlinear models. Second,
77
it shows to the analyst the importance of proper selection of the initial solver compared to trying a
variety of solver combinations afterwards. The same observations were found for the two other
Bridges B and C.
Table 5.4 The convergence failure time [sec] of simulations for different initial nonlinear solvers
under GM31 for Bridge A with Type I abutment model.
Scale Factor NR Krylov-Newton Broyden NLS BFGS
1.0 Completed 21.820 Completed Completed Completed
1.1 35.645 21.820 35.660 Completed 35.820
1.2 35.650 21.820 Completed Completed 41.010
1.3 35.655 6.115 35.655 Completed Completed
1.4 Completed 6.115 35.260 Completed 27.985
1.5 Completed 6.115 77.505 Completed Completed
1.6 Completed 6.115 Completed Completed 42.600
1.7 Completed 6.115 Completed Completed 36.155
1.8 35.710 6.115 Completed Completed 37.270
1.9 Completed 6.115 35.265 Completed 35.915
2.0 35.730 6.115 52.540 Completed 24.675
Similar to the previous section, Bridge A with Type I abutment modeling is used to evaluate the
effect of the convergence test type. The following five convergence tests: Energy Increment, Norm
Displacement Increment, Relative Norm Displacement Increment, Total Relative Displacement
Increment, and Relative Energy Increment, are compared for the 4 ground motions (GM18, GM19,
GM28 and GM31) with scale factors ranging from 1.0 to 2.0. The comparisons are based on
counting the total number of iterations for each simulation, see Table 5.5. It is observed that the
Energy Increment test leads to significantly less number of iterations compared to the other test
types. It is to be noted that very small response differences are obtained from the simulations with
different convergence tests, as indicated by Maxerror for the longitudinal direction displacements
of one of the nodes of the abutment, i.e. node 100 (Figure 5.5) in Table 5.6. Here, Maxerror is
calculated for the simulations with different convergence tests considering the simulations with
the Energy Increment test as the reference.
Table 5.5 Total number of iterations for simulations with different convergence tests under GM31.
78
Table 5.6 Maxerror in long. deformation of node 100 for GM31 (different convergence tests).
Scale RelativeNorm RelativeTotal Relative
NormDisIncr
Factor DispIncr NormDisIncr EnergyIncr
1.0 1.82×10-6 1.82×10-6 1.82×10-6 1.82×10-6
1.1 0.0 0.0 0.0 0.0
1.2 1.30×10-4 0.0 0.0 0.0
1.3 7.67×10-6 7.67×10-6 7.67×10-6 0.0
1.4 0.0 0.0 0.0 0.0
1.5 0.0 0.0 0.0 0.0
1.6 0.0 0.0 0.0 0.0
1.7 0.0 0.0 0.0 0.0
1.8 0.0 0.0 0.0 2.21×10-6
1.9 0.0 0.0 0.0 0.0
2.0 Failed 1.91×10-4 2.38×10-3 2.70×10-6
It is beneficial to note that the same tolerance value of 108 is used for all the convergence
tests, which is the main factor that makes the Energy Increment test to have the least amount of
iterations. Multiplication of the displacement increment with the unbalanced force, both less than
1.0, results in a norm value smaller than the other norms. However, aside from this explanation,
observing that the simulations with the less stringent Energy Increment leads to practically the
same results with significantly less number of iterations compared to the other tests, it is concluded
that the Energy Increment test is the most suitable convergence test for the analyzed model. Similar
to the investigation in previous sections, this finding may be specific to the investigated model.
However, for similar reasons as those mentioned above, it is still a useful conclusion by providing
an initial trial suggestion for other models and simulations.
For Bridge B with Type I abutment modeling, the effect of different tolerances of 1.0, 0.1, 103
and 105 on the convergence and accuracy of the obtained results is investigated using the Energy
Increment test. Although analyses are conducted for all 40 ground motions, detailed results are
presented herein for four selected ground motions (GM13, GM16, GM18 and GM19) and the mean
results of all 40 ground motions with scale factors ranging from 1.0 to 2.0. The longitudinal
direction deformation of one of the nodes of the abutment, corresponding to node 100 (refer to
Figure 5.5), and the longitudinal displacement of a column node, node12, for selected ground
motions and the mean of all ground motions are presented in Figures 5.9-5.10 for the simulations
with different tolerances. In the calculation of these error quantities, simulations with the tolerance
of 108 are considered as the reference.
Selected tolerances may have considerable effect on the nonlinear response. A large
tolerance may result in a premature convergence and corresponding deviation from the true
response. The small errors between the simulations with tolerances of 105 and 108 under the
effect of the considered ground motions resulting in highly nonlinear response indicate that the
increase of the tolerance can be used as a reasonable option to overcome convergence issues, while
preserving accuracy. It is to be noted that the above tolerance values are used for all the integration
time steps of a particular simulation. Considering that a common application is the increase of the
convergence tolerance only at the integration time steps with convergence problems, the errors in
the obtained results in such cases of selective adoption of tolerance values will be even less than
the errors plotted in Figures 5.9-5.10.
79
0
10
GM13
-1
GM16
10 GM18
GM19
10
-2 Mean
Maxerror
-3
10
-4
10
-5
10 0 -1 -3 -5
10 10 10 10
Tolerance
Figure 5.9 Maxerror of each tolerance for zerolength element associated with node 100 (longitudinal
displacement).
0
10
-1
10
-2
10
Maxerror
-3
10 GM13
GM16
-4 GM18
10
GM19
Mean
-5
10 0 -1 -3 -5
10 10 10 10
Tolerance
Use of a smaller integration time step during the simulation does not necessarily improve the
convergence behavior as seen from Table 5.7, which compares the convergence condition for the
simulations with different scales of GM31 using Newton-Raphson as the initial solver. Based on
the simulations conducted with the seed Bridges A, B and C in this study, it is observed that
reduction of the integration time step, only when needed, is useful to overcome the convergence
problems. However, this requires attention to be paid to preventing the simulation from being
completed before the duration of the external excitation, where the integration time step should be
reset to its original value after completion of all of the reduced time steps that represent the original
step size, e.g. using automatic adaptive time increments [DIANA 2005].
80
Table 5.7 The convergence failure time [sec] of simulations for different integration time steps
under GM31 for Bridge A.
Scale
t = 0.01 t = 0.005 t = 0.0025 t = 0.001
Factor
1.0 Completed Completed Completed Completed
1.1 35.6500 35.6450 35.6375 35.6360
1.2 35.6600 35.6500 35.6450 35.6430
1.3 35.6700 35.6550 35.6525 35.6510
1.4 Completed Completed Completed Completed
1.5 Completed Completed Completed Completed
1.6 Completed Completed Completed Completed
1.7 Completed Completed Completed Completed
1.8 35.7200 35.7100 35.7050 35.7040
1.9 Completed Completed 41.7225 62.6450
2.0 35.7400 35.7300 36.3475 36.3460
As mentioned and demonstrated in the previous sections, the explicit OS integration algorithm is
a suitable alternative of the IN to avoid the problems of convergence. However, there exist some
conditions where this algorithm is not applicable. For example, aside from the chosen integration
method, formulations of some of the elements and materials, such as the force-based beam-column
elements or the Bouc-Wen material in OpenSees [McKenna et al., 2010], are iterative. Accordingly,
the implicit algorithms may be the only option for NLTA of models containing such elements and
materials if problems of convergence at the element and material levels cannot be eliminated while
using the explicit OS integrator. This does not, however, prevent taking advantage of the explicit
algorithms in certain time steps where the implicit algorithm fails to converge. In OpenSees (2010),
the adaptive switching of integration algorithms, i.e. from IN to OS, is triggered when IN fails to
converge, say at time step i 1 . Then, the simulation automatically returns to the previously
converged time step, i.e. time step i , and is rerun from time step i to i 1 using the OS algorithm.
Subsequently, the integration algorithm is switched back to IN. Therefore, IN is reused for the
simulation starting from time step i 1 until another convergence difficulty is encountered. As
demonstrated in the previous sections, the OS algorithm possesses similar stability and accuracy
properties to those of the IN integration. Therefore, adaptive switching from IN to OS at
problematic time steps, where convergence issues are bypassed, does not affect the stability and
accuracy of simulations [Liang and Mosalam 2015a, b].
NLTA results in Table 5.8 are selected examples from simulations for the seismic response
of bridges investigated in more detail in the next chapter. It is observed from Table 5.8 that the IN
algorithm fails to converge at the indicated times in the 4th column, where the responses are at high
levels of nonlinearity. It is noted that the ground motions in this table are identified by its sequence
number in [NGA database 2011]. On the other hand, the simulations with the same ground motions
are completed using adaptive switching between integration algorithms, i.e. from IN to OS, at the
time steps when IN fails to converge. Therefore, the adaptive switching of algorithms is considered
81
to be a viable and readily available option, e.g. in OpenSees (2010), to overcome the problems of
convergence. Moreover, this statement is supported by the previously conducted theoretical
investigation [Liang and Mosalam 2015a, b] and successful completion of the simulations
discussed in the next chapter.
Table 5.8 The convergence failure time [sec] of simulations for different implicit integration
methods.
From Table 5.8, it is also observed that simulations that experienced problems of
convergence are all completed with the use of TRBDF2. This superior convergence performance
of TRBDF2 is attributed to the numerical damping introduced by the three point backward Euler
scheme. It is noted that several simulations in Table 5.8 fails to converge at as early as 0.495 sec
when IN is used. This early stage convergence issues can be attributed to several possible reasons,
such as near fault ground motions that cause nonlinear responses very early or the abrupt stiffness
change due to the opening and closing of the gap elements used in the modeling of abutments.
Previously indicated adaptive switching of integration algorithms and nonlinear solution methods
have already been implemented for the conducted analyses of the three considered bridges in Tcl
[Ousterhout, 1994] files, which are used as input to OpenSees. An example of such robust
algorithm is provided in Appendix A. Interested OpenSees users can access these algorithms and
the corresponding Tcl files upon communicating with the authors. These files provide the best
possible algorithm options for the investigated bridges. However, OpenSees users can
conveniently modify these files for the analyses of other bridges and structures, if needed.
OpenSees is commonly used in the academia as an analysis tool, while it is not as common
among the practicing engineer community including the Caltrans engineers. In order to make the
robust algorithms accessible to the practicing engineers, these algorithms are planned to be
implemented soon in more user friendly tools that use OpenSees, such as the Graphical User’s
Interface (GUI) developed by Ahmed Elgamal in UC San Diego for modeling bridges in OpenSees.
The implementation of these algorithms in the mentioned GUI based application is envisioned to
result in a valuable tool that can efficiently be used by Caltrans and other practicing engineers.
82
6 Ground Motion Selection and Modification
6.1 INTRODUCTION
Nonlinear structural response of bridges, similar to most other structures, is so complicated that it
is often highly sensitive to the selection and modification of the input ground motions.
The goal in this chapter is to evaluate current ground motion selection and modification
methodologies and to develop new methods for response assessment of Caltrans standard ordinary
bridges. Ultimately, the goal of this chapter is to provide practical recommendations regarding use
of various methods of ground motion selection and scaling for various types of bridges addressed
in this project. The methodology presented herein follows the general framework used in a
previous and extensive research project performed on a similar topic (but for buildings), by the
Ground Motion Selection and Modification (GMSM) Working Group [Haselton et al. 2009],
which is summarized as follows:
Definition of the Objective: Define the objective of the structural analysis, i.e.
determine median response, the variability in response, etc., such that the reference
response can be clearly determined.
Selection of the Ground Motion Scenario: Define the ground motion scenario for
which the responses of the bridge are to be evaluated.
Selection of Candidate GMSM Methods and Development of Ground Motion
Sets: Determine the short-list of candidate GMSM methods to be evaluated. Then,
select a ground motion set for each GMSM method, corresponding to the defined
ground motion scenario, and predict the statistical parameters (in relation to the
defined objective) of the bridge response from NLTA using the selected ground
motion set.
Development of Point of Comparison using High End Prediction (HEP) to
evaluate the GMSM Methods: Create a high-end prediction (HEP) based on the
NLTA using a large number of ground motions. Then, compare the various bridge
response predictions to this HEP in order to evaluate each GMSM method in terms
of accuracy and precision of the bridge response predictions.
6.2 OBJECTIVE
Several objectives may be of interest for nonlinear dynamic analysis of a bridge structure. The
following lists the four possible objectives articulated in the previously mentioned PEER GMSM
report [Haselton et al. 2009].
83
Objective 1: Predict the probability distribution of structural response for an
earthquake of a given magnitude (M) and source-to-site distance (R), i.e., a given
earthquake scenario.
Objective 2: Predict the median structural response for an earthquake of a given M
and R.
Objective 3: Predict the probability distribution of the structural response for a
ground motion scenario described by the spectral acceleration value at the
fundamental period of the structure Sa(T1), M and R. Therefore, it is necessary to
have certain knowledge of the structure, i.e. its fundamental period T1.
Objective 4: Predict the median structural response for a ground motion scenario
defined by a given Sa(T1), M and R.
The fourth objective is the main focus of this study to maintain consistency with the common
practice of NLTA of bridge structures by Caltrans engineers and the requirements in current
seismic codes and provisions.
84
1
10
Median response spectrum
Median+1.5 response spectrum
Conditonal mean spectrum
Target Sa(T1)
0
10
S (g)
a
-1
10
-2
10
-2 -1 0 1
10 10 10 10
T (s)
Figure 6.1 Response spectra for a scenario earthquake for Bridge B site.
85
6.4.1 Sa(T1) selection and scaling method
The Sa(T1) scaling method selects ground motions from earthquakes with magnitude M and type
of faulting F, recorded at sites with distance R, and soil classification S as close as possible to those
of the scenario of interest. After applying the selection criteria, the ground motions are selected
randomly from the candidate set of motions if the number of eligible ground motions is larger than
the target number; otherwise, the selection criteria would need to be relaxed.
Once the ground motions are selected, each of them is scaled in amplitude such that their
Sa(T1) value is equal to the target Sa(T1) value of the ground motion scenario. This procedure to
select and scale ground motions does not take into account the shape or the variability of the target
response spectrum, as it considers only the target Sa(T1) value.
The selection procedure of the Sa(T1) scaling method is summarized below:
Step 1: Select the ground motion based on an M-R-S-F (magnitude, source-
to-site distance, site classification, and type of faulting) bin that is consistent
with the given scenario. The criteria utilized in this study are:
1) 6.6 < M < 7.4
2) 5 km < R < 15 km
3) Vs,30 ≥ 183 m/s
4) Faulting Type F: Not constrained
In addition to the above, to ensure that all ground motions have frequency
content appropriate for evaluating the bridges of interest, ground motions
with lowest usable frequency of 0.25 Hz are selected.
Step 2: Scale each component of record to the target ground motion level
based on their geometric mean (Equation 7.1). The target Sa(T1) is the
median+1.5ε value predicted by the attenuation model from [Campbell and
Bozorgnia 2008] for the given M, R, S and F scenario.
(6.1)
where SFi is the scaling factor, the subscript i denotes the ith GM
components pair, and the superscripts C1 and C2 refer to components 1 and
2. The scaling factor is the same for both components.
Step 3: Select desired number of records from the bin. In this study, 40
records are selected. Two algorithms are available.
1. Based on the equation for the proportion of pulse-like records [Hayden
et al. 2012],
(6.2)
86
substitution of R = 10 km and ε = 1.5 in Equation 6.2 results in a Proportion
value = 0.7. Thus, 28 of the 40 motions should be pulse-type. Select 28
records from the pulse-type bin and 12 records from the no pulse-type bin
with smallest scaling factors from step 2.
2. Another algorithm is to select 40 records solely based on the scaling
factors. Therefore, 40 records with smallest scaling factors are selected
from the bin with both pulse-type and no pulse-type ground motions.
(6.3)
(6.4)
87
The mean and the standard deviation referred to in this step are the
quantities presented in Step 1, i.e. and ,
respectively. In this step, these values are computed at periods included in
the range of periods of interest. Existing ground motion models, such as the
attenuation model from [Campbell and Bozorgnia 2008], can be used to
compute these terms.
Step 3: Compute ε at other periods, given ε(T*)
This step consists of computing conditional mean ε-values for the
other periods Ti, which can be calculated as the product of ε(T*) and the
correlation coefficient between the ε-values at the two periods , i.e.
(6.5)
(6.6)
(6.7)
88
Step 1: Scale ground motions to match the target Sa(T*)
Before selecting the ground motions, they are scaled to match the target
Sa(T*), or Sa(T1) in the present study. The scaling factor for each
bidirectional ground motion is the same as that defined in Equations 6.2.
Step 2: Parameterize the multivariate normal distribution of lnSa at
multiple periods
It has been shown that the set of logarithmic spectral accelerations at various
periods is a random vector that follows a multivariate normal distribution
[Jayaram and Baker 2008]. The parameters of the multivariate normal
distribution are the means and variances of lnSa at all periods and the
correlations between these values at all pairs of periods. These parameters
should be set equal to their target values, which depend on the considered
T*.
Step 3: Perform Monte Carlo simulations to probabilistically generate
response spectra from the multivariate normal distribution
The number of simulated and generated spectra is equal to the desired
number of ground motions to be selected.
Step 4: For each simulated response spectrum, select a ground motion
which best matches the simulated spectrum
The sum of squared errors (SSE) is used to select the best matching ground
motion as follows,
(6.8)
89
have slight deviations from the target values. The magnitude of these deviations can be estimated
from the sum of the squared differences between the target and the sample means and variances
(SSEs) over the period range of interest according to the following expression.
(6.9)
where m̂ln Sa(T ) is the mean value of , is the target mean value of ,
i
w is the weighting factor indicating the relative importance of the errors in the standard deviation
and the mean, and p is the number of periods considered to compute the error.
The mean and standard deviation of each set can be calculated as follows:
(6.10)
(6.11)
where is the of jth record of the set and n is the total number of records in the
set.
SSEs is the parameter to be minimized using an optimization technique as proposed in step
5 above for selecting and modifying sets of records to match the target CMS and its variance, which
consists of replacing each ground motion selected previously. This replacement is performed one
at a time, with a ground motion from the database that results in the best improvement in the match
between the target and the sample means and variances, i.e. results in the major reduction of the
SSEs value. Similar to the Sa(T1) scaling method, 40 records are considered.
91
CB 2008, 10% in 50 years CB 2008, 10% in 50 years
1
Conditional mean spectrum 10 2.5 and 97.5 percentile response spectra
Target Sa
0
10
0
10
Sa (g)
Sa (g)
-1
10
-1
10
-2
10
-2 -1 0 -2 -1 0
10 10 10 10 10 10
T (s) T (s)
Figure 6.2 Campbell and Bozorgnia (CB) 2008 spectrum and CMS for 10% POE in 50 years for
Bridge B site.
Figure 6.2a shows the response spectrum by the attenuation model from Campbell and
Bozorgnia (2008) at hazard level of 10% POE in 50 years for Bridge B site, i.e. the target spectrum
for the second “reference” group. Also shown in Figure 8a is the CMS [Baker 2011] anchored at
the Bridge B fundamental period of 1.1 sec, which is the target spectrum for the first group. As
mentioned before, both groups of ground motions are selected using a method proposed by
Jayaram et al. (2011) that seeks to match the mean and variance of the target spectrum (Figure
6.2b). A detailed explanation of this method refers to [Jayaram et al. 2011].
As discussed before, two approaches for abutment modeling, namely Type I (Figure 3.3a)
and Type II (Figure 3.3b), are considered. For each abutment modeling (2) of each scenario (3),
40 ground motion records are selected for each ground motion group (2). Therefore,
40×2×3×2×3=1440 NLTA simulations in total are conducted for the considered bridge systems
(3). A large amount of problems of convergence, e.g. those indicated in Table 5.8, are encountered.
Most of these problems of convergence are overcome by the proposed solutions, i.e. OS, TRBDF2,
and adaptive switching of integration algorithms. Those NLTA simulations that still fail to
converge are primarily due to the large ground motions, e.g. several ones in the “reference” group
at hazard level of 2% POE in 50 years, which lead to significantly large nonlinear responses in the
range of collapse limit state and probably corresponding to physical partial or complete bridge
collapse.
92
1.4 1.4
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
50% 10% 2% 50% 10% 2%
POE in 50 years POE in 50 years
Figure 6.3 Ratios of median EDPs for the two abutment modeling cases of Bridge A.
Figures 6.3-6.5 present the ratios of the median EDPs obtained from the ground motions
of the first group (CMS), i.e. first-mode dominant, to those obtained from the ground motions of
the second “reference” group, i.e. higher mode response. Therefore, the smaller the ratio, the more
the considered EDP is affected by the higher modes. It is observed that the ratio for the column
base shear is close to 1.0 and almost invariant for both modeling cases and all three scenarios. On
the other hand, the ratios for the column drift and deck acceleration are always less than 1.0 and
generally reduce as the hazard level and the corresponding nonlinearity level increase. Accordingly,
it is concluded that the higher mode effects are more pronounced on column displacements and
deck accelerations than on column shear forces. Moreover, the effects of higher modes on the
column drift and deck acceleration increase with increasing hazard level and nonlinearity. The
column base shear is likely to be a first-mode dominant EDP, irrespective of the hazard level.
Accordingly, an investigation that uses the base shear as the EDP of interest may make use of
computationally less demanding single mode nonlinear static analyses in PBEE computations. On
the contrary, NLTA must be used if the column drift and deck acceleration are the important EDPs.
It is noted that the comparison is not recorded for Bridge C with Type I abutment modeling for the
hazard level of 2% POE in 50 years. This is due to the fact that 7 out of 40 ground motion records
in the second group failed to converge, despite significant efforts to improve convergence [Liang
et al. 2014, 2016a,b]. The results in Figures 6.3-6.5 are in agreement with the statement that the
higher modes affect the response of bridges to a greater extent than that of buildings in [Kappos et
al. 2013].
93
1.4 1.4
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
50% 10% 2% 50% 10% 2%
POE in 50 years POE in 50 years
Figure 6.4 Ratios of median EDPs for the two abutment modeling cases of Bridge B.
1.4 1.4
Ratio of EDP median of two groups
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
50% 10% 50% 10% 2%
POE in 50 years POE in 50 years
Figure 6.5 Ratios of median EDPs for the two abutment modeling cases of Bridge C.
94
1) 6.5 ≤ Mw ≤ 7.5 (note: , where M0 is the seismic
moment in dyne-cm.)
2) 0 km < R ≤ 20.0 km
3) VS30 ≥ 183 m/s
4) Lowest usable frequency of 0.25 Hz
5) Faulting F: Not constrained.
99 pairs of bidirectional horizontal ground motions are selected from the PEER
Next Generation Attenuation (NGA) Project ground motion library [Chiou et al.
2008] based on the criteria above.
Scale the 99 pairs of ground motions based on peak ground velocities (PGVs) of
one standard deviation above and below the PGV of the target earthquake scenario,
i.e.
(6.12)
where SFi max and SFi min are the maximum and minimum scaling factors for the ith
GM components pair. The scaling factor is the same for both components.
Furthermore, two rules are applied:
1) Exclude the GM components pair whose maximum scaling factor is larger than
5;
2) Exclude the GM components pair if any peak ground acceleration (PGA) of its
components after scaling is larger than 1.6g.
60 pairs of bidirectional horizontal ground motions are left after these two rules
applied. Interpolate between SFi max and SFi min to get 10 scaling factors in total for
each GM pair. This yields 600 pairs of bidirectional horizontal ground motions
totally.
Perform nonlinear time history analyses (NLTA) using the ground motions selected
in the previous step applied with intercept angles varying at 30 degrees increments
from 0 to 150 degrees, and record the response of interest (e.g. the EDP in terms of
maximum bridge column drift ratio) from each analysis.
Investigate response differences of different intercept angles of ground motions.
The median responses from the simulations in the last step are the HEPs for EDPs.
Compare predicted responses and the probability of collapse from each GMSM
method and the HEP and evaluate the GMSM Methods.
Bridges A, B and C described in Chapter 3 are used in the GMSM evaluation process in
addition to a fourth case that models Bridge B with the enhanced modeling options described in
Chapter 4. As mentioned in Chapter 4, the enhanced modeling options consist of mainly the
consideration of nonlinear soil structure interaction (SSI) effects. Three different abutment skew
95
angles are considered for the fourth case, while the first three cases are considered as regular
bridges, i.e. with a skew angle of 0. It is noted that the skew angle is defined as the angle between
the centerline of an abutment/pier and a line normal to the roadway [Kaviani, 2011].
The effects of the intercept angle are investigated on the bridge models mentioned above.
The intercept angle is the angle between the ground motion’s strike-normal direction and the
longitudinal direction of the bridge (Kaviani, 2011). Four Engineering Demand Parameters (EDPs)
are selected to demonstrate he effects. These four EDPs are the maximum column base shear,
maximum abutment unseating displacement, maximum column drift ratio, and maximum
curvature at the column top. Although simulations are done for all four bridges, only results of
Bridge B with and without SSI effects are presented for brevity (Figures 6.6 and 6.7).
1.6 1.6
Column base shear Column base shear
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
30 60 90 120 150 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.6 Investigation of the intercept angle effect on EDP for the two abutment modeling cases
of Bridge B.
Normalized ratio of median EDP, N ij (vertical axis of Figures 6.6 and 6.7) is defined as
follows:
HEPji
N ij (6.13)
HEPj0
where HEPj0 represents the median of the EDP j mentioned above obtained from the HEP
analyses with 0 degree intercept angle, and HEPji represents the median of an EDP j obtained
from the HEP analyses with different intercept angles i , ranging from 30 to 150 degrees.
96
1.6
Column base shear
0.8
0.6
0.4
0.2
0
30 60 90 120 150
Intercept angle ()
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
30 60 90 120 150 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.7 Investigation of the intercept angle effect on EDP for three abutment skew angles of
Bridge B with enhanced modeling.
It is observed that the ratios for all four EDPs are around 1.0 for all the considered bridge
models (Figures 6.6 and 6.7) indicating that the intercept angle plays an insignificant role in
evaluating HEPs. It is noted that this is a useful conclusion regarding the investigation of seismic
response of typical California bridges, because it is computationally expensive and not always
possible to determine the largest response that is generated by several rotated versions of a ground
motion. It is further useful to note that this conclusion is not effected by the presence of the SSI
modeling as the ratios are around 1.0 for both modeling cases of Bridge B i.e. with and without
the enhanced modeling options in Figures 6.6 and 6.7a.
97
are presented. Ratio of median EDP by different GMSM methods to the HEPs, R ij, k (vertical axis
of Figures 3 to 10) is defined as follows
RES ij, k
R ij, k (6.14)
HEPji
where RES ij, k stands for median response of EDP j with intercept angle of i by the considered
GMSM method k . In order not to produce any bias due to the use of different number of ground
motions, cases where more than 10% of the ground motion records failed to converge, despite the
use of all the improved convergence methods described in Chapter 5 (e.g. US method for Bridge
B), are not presented in these figures. Following remarks can be stated regarding Figures 6.8-6.15.
It is observed that the ratio for the column base shear is close to 1.0 and almost invariant for all
bridge models (Figures 3 and 4) and all GMSM methods, i.e. all the investigated GMSM methods
provide accurate estimates of column base shear. This result is mainly because of reaching the
column force capacity in both HEP analyses and in those conducted using the ground motions
selected by the GMSM methods. Accordingly, it can be mentioned that all the GMSM methods
are successful at least in the prediction of damage at a global level.
1.6 1.6
CMS CMS
1.4 US 1.4 US
Ratio of EDP median to HEP
SaT11 SaT11
1.2 SaT 2 1.2 SaT 2
1 1
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.8 Investigation of the adequacy of GMSMs in predicting the maximum column base shear
for the two abutment modeling cases of Bridge B.
98
1.6
CMS
1.4 US
0.8
0.6
0.4
0.2
0
0 30 60 90 120 150
Intercept angle ()
SaT11 SaT11
1.2 SaT 2 1.2 SaT 2
1 1
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.9 Investigation of the adequacy of GMSMs in predicting the maximum column base shear
for three abutment skew angles of Bridge B with enhanced modeling.
6.7.2 Evaluation of the GMSM Methods regarding Column Drift Ratio, Abutment
Unseating Displacement and Column Top Curvature:
Regarding the other three EDPs, the responses obtained from the US method are almost always on
the conservative side and are usually the most conservative of all GMSM methods (Figures 6.10
and 6.11). The CMS method underestimates the response of Bridge A (not shown here),
underestimates maximum column top curvature of Bridge B (Figure 6.10), while overestimating
the other EDPs and overestimates the response of Bridge C (not shown here). The SaT1 method
with different scaling algorithms provides predictions similar to those of the CMS method (Figure
6.11).
99
2 2
CMS CMS
1.8 US 1.8 US
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.10 Investigation of the adequacy of GMSMs in predicting the maximum column top
curvature for the two abutment modeling cases of Bridge B.
2.2 2.2
CMS CMS
2 2
US US
Ratio of EDP median to HEP
1.4 1.4
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.11 Investigation of the adequacy of GMSMs in predicting the maximum abutment
unseating displacement for the two abutment modeling cases of Bridge B.
Comparison of Bridge B with and without SSI modeling (e.g. Figures 6.12 and 6.13a) indicates
that modeling of the SSI effects play an insignificant role in evaluating GMSM methods. The
accuracy of some simulated GMSM methods could be investigated in the future. For example, the
accuracy of the simulated ground motion method by Rezaeian and Der Kiureghian (2010a, b, 2011)
can be investigated by selecting an earthquake scenario without considering SSI effects
( Vs 30 600m / s ).
100
2.2 2.2
CMS CMS
2 2
US US
Ratio of EDP median to HEP
1.4 1.4
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.12 Investigation of the adequacy of GMSMs in predicting the maximum column drift ratio
for the two abutment modeling cases of Bridge B.
2.2
CMS
2
US
Ratio of EDP median to HEP
1.8 SaT11
1.6 SaT 2
1
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 30 60 90 120 150
Intercept angle ()
1.4 1.4
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.13 Investigation of the adequacy of GMSMs in predicting the maximum column drift ratio
for three abutment skew angles of Bridge B with enhanced modeling.
101
6.7.4 Effects of Skew Angle
Predictions of GMSM methods do not vary significantly with the skew angle (Figures 6.14 and
6.15).
2.2
CMS
2
US
1.6 SaT 2
1
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 30 60 90 120 150
Intercept angle ()
1.4 1.4
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.14 Investigation of the adequacy of GMSMs in predicting the maximum abutment
unseating displacement for three abutment skew angles of Bridge B with enhanced modeling.
102
2.2
CMS
2
US
1.6 SaT 2
1
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 30 60 90 120 150
Intercept angle ()
1.4 1.4
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Intercept angle () Intercept angle ()
Figure 6.15 Investigation of the adequacy of GMSMs in predicting the maximum column top
curvature for three abutment skew angles of Bridge B with enhanced modeling.
Based on simulation results, the US method are almost always on the conservative side and
are usually the most conservative of all GMSM methods. CMS and SaT1 methods present similar
predictions of responses, which underestimate responses sometimes. Standard ordinary bridges
play a crucial role in transportation and thus require short downtime after severe earthquakes
regarding both the emergency response and community resiliency aspects. Therefore and given
that none of the GMSM methods is particularly superior to the others, it is suggested to use the US
method for selection and modification of ground motions for different intercept angles and to
choose the most critical intercept angle that yields the largest responses.
104
2. Use the initial values to derive time history of shear strain for each layer.
3. Compute effective shear strain for each layer based on the maximum shear strain
in the computed shear strain time history.
4. Compute new values of G and ξ for each layer based on using shear modulus and
damping curves for different soil layers (Figure 7.6).
Repeat step 2-4 until get the similar values with respect to the previous step. Generally,
strain compatible properties for each layer can be derived after 5-8 iterations [Shake 91’s manual,
1992].
35.0
Soil Damping Ratio
Clay (PI=15)
Clay (PI=50)
30.0 Sand at 0-20 ft
Sand at 21-50 ft
Sand at 51-120 ft
25.0
Sand at 250-500 ft
Damping Ratio (%)
Mean Gravel
20.0
15.0
10.0
5.0
0.0
0.0001 0.001 0.01 0.1 1 10
Shear Strain (g)
1.20
Soil Modulus Degradation
Clay (PI=15)
Clay (PI=50)
1.00 Sand at 0-20 ft
Sand at 21-50 ft
Sand at 51-120 ft
0.80 Sand at 250-500 ft
Mean Gravel
G/Gmax
0.60
0.40
0.20
0.00
0.0001 0.001 0.01 0.1 1 10
Shear Strain (g)
Figure 6.16 Soil modulus degradation and damping curves cohesive and cohesionless soils
[EPRI, 1993].
105
7 Case Studies
7.1 INTRODUCTION
The sensitivity of seismic response of the specimen bridges to variation in modeling options has
been investigated within three major focus areas:
Sensitivity to various modes of shear-key failure
Sensitivity to backfill geotechnical properties and epistemic uncertainties in skew abutment
model
Sensitivity to soil-structure interaction in deep foundation elements
The first focus area investigates the effect of shear-key ductility on the seismic performance of the
specimen bridges. The properties of backfill springs are chosen pertinent to the geotechnical
properties of Category II backfills, which comply well with Caltrans Specifications. The skew
abutment model is kept relatively simple and the same as the model proposed by Kaviani et al.,
(2012) i.e. the initial stiffness and ultimate strength of backfill hyperbolic springs are varied
linearly between the obtuse and acute corners of the abutment. Considering the role of bent
torsional and transverse rigidity—which depend on the number of columns in the bent—on the
transverse response of superstructure, both specimen bridges (i.e. those with single and double
column bents) are included in this phase.
The second focus area addresses the effect of geotechnical properties of backfill on the
seismic response of bridge superstructure, considering the documented deviation of the properties
of a noticeable percentage of existing backfills from Caltrans specifications. The more evolved
skew abutment models developed based on a well-established soil failure mechanism and the other
based on an empirical relationship that was derived from past experimental data are used here. The
outcomes of the sensitivity studies relevant to the first focus area reveal that, in general, the lower
torsional and transverse stiffness of the single-column specimen bridge (compared to the double-
column bridge) amplify deck rotation, particularly in configurations with skewed abutments. The
sensitivity studies in this phase are limited to the response of the specimen bridge with double-
column bent (i.e., Bridge B). The additional torsional and transverse rigidity of the double-column
bent allows to better track the effect of backfill geotechnical properties and skew models—which
might be masked by the lower torsional and transverse rigidity of the bent in a single-column
bridge—on superstructure rotation and transverse response.
The third focus area involves identifying the appropriate level of Soil-Structure Interaction
(SSI) modeling for seismic response assessment of ordinary bridges. To this end, effect of soil-
structure interaction on the seismic response of two typical concrete overcrossing bridges is
investigated. The interaction between deep foundation elements and supporting soil is simulated
through reduced-order modeling of individual piles and supporting soil with nonlinear fiber-based
106
beam elements and series of nonlinear springs, respectively. Through an extensive sensitivity study,
it is demonstrated that the compound effect of two parameters may significantly affect the seismic
response of specimen bridges in the presence of explicit accounts of soil-structure interaction
effects: (1) the degree of moment fixity at the column-to-pile-cap connection, and (2) lateral
flexibility and yield mechanism of the deformable foundation. The study also reveals that
accounting for the variation of input motion at various depths of foundation vs. a uniform
excitation pattern yields larger uncertainty in the predicted response of the superstructure demand,
as well as more complex deformation patterns in the foundation. The findings underline the
importance of analyst’s judgment in adopting appropriate level of modeling and analysis intricacy
on a project-specific basis and by considering the scope of simulation.
107
7.3 COLLAPSE CRITERIA
Survival or collapse of the specimen bridges is decided based on two failure criteria: column
excessive rotation, and deck unseating; whichever happens first. The limit state corresponding to
column excessive rotation is defined as exceeding 8% drift ratio [Hutchinson et al. 2004]. Deck
unseating is assumed to happen when the center of the deck (shown with a diamond mark on Figure
7.1) is displaced to a coordinate outside the seat area of the abutment—i.e., anywhere in the hashed
area of Figure 7.1. The seat-length of the specimen bridge is approximately 33.85 in (86 cm),
which include the 1 in-gap (2.54cm-gap) between the deck and the backwall. The normal distance
between the backwall and the seat edge for all skewed configurations of the abutment is kept equal
to the seat length of the non-skewed configuration.
S S/cos()
)
Figure 7.1 Definitions of seat length in skewed abutments.
(7.1)
108
Assuming that all EDPs tend to infinity in a collapsed bridge, the first term on the right-
hand-side of Eq. (7.1) is equal to 1. Furthermore, since bridge “survival” and “collapse” are
complementary events, the probability of one can be expressed solely in terms of the probability
of the other,
(7.2)
(7.3)
(7.4)
Eq. (7.4) can be used to decide on the conditional cumulative probability of the qth percentile as.
(7.5)
where P(collapse|IM) can be evaluated from the observed data. If an analytical expression is
available for the conditional cumulative distribution function, FEDP, the qth-percentile can be
backcalculated subsequently.
The fragility curve or the conditional probability of collapse—i.e., P(collapse|IM)—is
evaluated here by fitting a binomial distribution to the number of observed collapses versus the
logarithm of IMs. A simple EDP-IM functional relationship is assumed by fitting a straight line to
the IM-EDP pairs in the logarithmic scale: the mean (μ) and standard deviation (σ) of a presumed
conditional log-normal probability distribution for the column drift ratio are equal to the value of
the linear fit at each IM, and the standard deviation of fit residuals, respectively.
7.5.1 Effects of Shear Key Behavior on Column Ductility Demand and Deck
Rotation
Column ductility demand is defined as the ratio of the maximum column top displacement to the
yielding displacement of the column. Based on static pushover analyses, the columns of both
bridges yield when the column drift ratio reaches 1%. Results show that shear key behavior
significantly affects column ductility demand and deck rotation, specifically under high-intensity
ground motions. Figures 7.2 to 7.8 show the effects of shear key behavior and backfill model on
column demand ductilities and deck rotations of the case-study bridges for 4 different skew angles.
The solid colored lines correspond to the variation of the median of EDP as a function of the
abutment skew angle and the two types of backfill models for a bridge with brittle shear keys. The
solid black lines show similar statistics for each bridge with ductile shear keys. The shaded areas
correspond to the EDPs’ ± one standard deviation from the median.
The aforementioned computed results are best interpreted by considering the equilibrium
of dynamic forces, which comprises backfill, column, and shear key reactions. Single column
bridges (i.e., bridges similar to Bridge A) with brittle shear keys experience large lateral
displacements at the top of the column (Figure 7.2), and large deck rotations (Figure 7.4). This is
mainly due to exhaustion of the brittle shear keys’ strength—which happens at relatively small
deformations—and the availability of only small rotational and translational capacities at the single
column bent. This trend can be easily traced in Figure 7.3 wherein the maximum deck
displacement in the longitudinal and the transverse directions for Bridge A are plotted for two
critical skew angles, 30o and 45o (note that only those cases that did not lead to collapse are plotted
in this figure). Different shades represent different ground motion intensities. Migration of data
points to positions above the 1:1 line at these two critical skew angles reveals that Bridge A with
brittle shear keys experience larger transverse and longitudinal deck movements compared to
Bridge A with ductile shear keys due to lower shear key ductility. Also, larger deck movements in
the transverse direction are correlated to larger deck rotations (Figure 7.4). Results reveal that
because of lower torsional resistance of the single column bent; the behavior of this type of bridge
mostly depends on the capacity of shear keys and backfill resistance.
The mode of failure is different for a non-uniform backfill compared to a uniform backfill
model. The role of the backfill in the transformation of mode of failure is more prominent in a bridge
with less shear key resistance (i.e., brittle shear key). The passive resistance of a non-uniform backfill
based on its definition [Kaviani et al., 2012] reduces linearly from the obtuse to acute corner of the
abutment as the skew angle increases (Figure 7.5), which drives the bridge to have larger deck rotation
in larger skew angles (see Figure 7.4). This fact increases the number of abutment unseating rates rather
than column failure (Figure 7.5). In contrast, in a bridge with uniform backfill, increasing the skew
angle results in significant reduction in backfill resistance [Rollins and Jessee, 2012], which causes the
110
translational movement of the deck in both transverse and longitudinal directions and a larger number
of column failures. This matter has been comprehensively addressed in the following sections through
evaluating the effects of backfill models on the seismic performance of bridges with various skewed
abutment models.
The behavior of Bridge B is slightly different from Bridge A, which is primarily due to the
number of columns present in each bridge. Figures 7.6 to 7.8 display the effects of shear key and
backfill models on column ductility demand and deck rotation for Bridge B that has a two-column
bent. Similar to the single column bridge, as the skew angle increases, the bridge with the smaller
shear-key ductility experiences larger deck rotations. However, the backfill model has a significant
effect on altering column ductility demand of Bridge B that has a two-column bent.
Uniform backfill(Bridge A)
PGV (m/s)
Non-uniform backfill(Bridge A)
PGV(m/s)
Figure 7.2 Effect of shear key and backfill model on column ductility demand of Bridge A.
111
Figure 7.3 Deck displacement with respect to the abutments in the longitudinal and the transverse
directions of Bridge A.
Response of Bridge B to seismic excitation depends on the equilibrium of three sets of reaction
forces acting on the deck—namely, those from the backfill, the shear-keys, and the bent columns.
Results shown in Figure 7.6 indicate that the column ductility demand of a bridge with a uniform
backfill model is not sensitive to the type of shear keys. This behavior can be understood if one
considers the large stiffness of columns in the transverse direction and the small backfill resistance. As
mentioned above, the uniform backfill model exhibits less resistance against deck movements in both
the transverse and the longitudinal directions, especially at larger skew angles. Additionally, the large
stiffness of the two-column bent in the transverse direction tends to resist the deck movements. Results
further reveal that brittle shear keys cause the columns to experience larger longitudinal top
displacements than transverse; and for the same ground motion and incident angle, this trend is
reversed for a ductile shear key. Therefore, the total displacements of the column in Bridge B with two
different shear keys are similar. This is evident from the increase in both medians (brittle and ductile,
and particularly ductile) shown in the upper row compared to the lower row in Figure 7.6.
Figures 7.7 and 7.8 reveal the trends in deck rotations for Bridge B. Generally, bridges with brittle
shear keys subjected to ground motions with larger intensity exhibit larger deck rotations. In contrast,
the larger initial stiffness of brittle shear keys results in smaller deck rotations in bridges with brittle
shear keys subjected to ground motions with smaller intensity. For non-uniform backfill similar to
Bridge A with a single column, lateral and longitudinal deck displacements in the case of brittle shear
keys is absolutely larger than the bridge with ductile shear keys, which results in larger column ductility
demand in the bridge with brittle shear keys (Figure 7.6 and 7.7) and larger deck rotation (Figure 7.8).
112
Uniform backfill(Bridge A)
PGV(m/s)
Non-uniform backfill(Bridge A)
PGV(m/s)
Deck rotation
Figure 7.4 Effect of shear key and backfill model on deck rotation of Bridge A.
a)
Column
Resultant Force
Non-uniform
backfill resistance
b)
Column
Resultant Force
uniform backfill
resistance
Figure 7.5 Effect of backfill model on bridge response of bridge A: (a) non-uniform backfill, (b)
uniform backfill.
113
PGV (m/s) Uniform backfill(Bridge B)
Non-uniform backfill(Bridge B)
PGV (m/s)
Figure 7.6 Effect of shear key and backfill model on column ductility demand of Bridge B.
Figure 7.7 Deck displacement with respect to the abutments in longitudinal and transverse
direction of Bridge B.
114
Uniform backfill(Bridge B)
PGV(m/s)
Med. Brittle
Med. ±σ Brittle
Med. Ductile
Med. ±σ Ductile
Non-uniform backfill(Bridge B)
PGV(m/s)
Deck Rotation
Figure 7.8 The effects of shear key and backfill models on the deck rotations of Bridge B.
7.5.2 Sensitivity of Abutment Unseating to Shear Key Behavior and Seat Width
Abutment seat width has a significant effect on the occurrence of longitudinal abutment unseating and
failure of bridges. As per Caltrans SDC 1.7, the seat width is required to be at least 30 in, and is
computed by summing different variables such as the displacements attributed to pre-stress shortening,
creep/shrinkage, thermal expansion/contraction, and the displacement demand from the adjacent
frame. Here, the collapse fragility curves conditioned on abutment seat widths are generated in order
to investigate the effect of the two alternative shear key modeling techniques on abutment unseating
(see Figures 7.9 and 7.10). In all prior simulations, abutment unseating, as a mode of failure, has been
prevalent in bridges with non-uniform backfill models only; as such, abutment unseating is considered
only for the non-uniform backfill in this section.
Three different seat-widths were considered—namely, 0.5, 1.0, and 2.0 times the design
seat-width, which are 30 in (0.76 m) and 34 in (0.86 m) for Bridges A and B, respectively. A logistic
regression was fitted to 105 data points per bridge per skew angle. Results indicate that in both bridges
with brittle shear keys, the probability of collapse conditioned on the seat width is larger than that for
the bridge with a ductile shear key for all seat widths (especially for larger skew angles). Therefore, it
appears that the shear key behavior has a discernible effect on the probability of collapse due to
abutment unseating, even when abutment seat width exceeds the Caltrans SDC 1.7 requirement by a
factor of 2.0. This observation can be explained when one considers the interplay of all three kinematic
modes of the bridge deck—i.e., translational and longitudinal displacements, and rotation. It is clear
that only rotations and transverse movements of a deck will engage the shear keys and these are
orthogonal modes to longitudinal displacements. However, brittle shear keys fail earlier than their
ductile counterparts and once that happens the ability of the shear key to dissipate the deck’s kinetic
115
energy gets diminished. This leads to larger deck movements overall including those in the longitudinal
direction, which eventually lead to unseating.
Non-uniform backfill(Bridge A)
o
0 30o
0.8 0.8
Ductile SK-0.5*Original Seat
P(C|Seat width)
0 0
0 1 2 0 1 2
o o
45 60
0.8 0.8
P(C|Seat width)
0.4 0.4
0 0
0 1 2 0 1 2
PGV (m/s) PGV (m/s)
Figure 7.9 Deck unseating fragility curve conditioned on seat-width for Bridge A.
116
Non-uniform backfill(Bridge B)
0o 30o
0.8 0.8
Brittle SK-0.5*Original Seat
P(C|Seat width)
0 0
0 1 2 0 1 2
45o 60o
0.8 0.8
P(C|Seat width)
0.4 0.4
0 0
0 1 2 0 1 2
Figure 7.10 Deck unseating fragility curve conditioned on seat-width for Bridge B.
117
of abutment modeling that are used in the subsequent sections. It should be noted that the
investigated backfill soil types include clay (GHFD III and IV in Table 4.3) where the use of clay
is typically not pursued by standard Caltrans practice.
Table 7.1 Combination of various models of backfill and shear-key used in sensitivity analyses.
118
Figure 7.11 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category I GHFD
backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
Figure 7.12 GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
119
Figure 7.13 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category III
GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
Figure 7.14 Comparison of column drift ratio in bridges with SDC 1.7 bilinear and Category IV
GHFD backfill models (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
120
the fitted binomial distributions at the five representative skew angle values for all cases of backfill
modeling are shown in Figure 7.15.
The probability of collapse is relatively low at 0 skew angle and rises to at most ~10% for
spectral acceleration equal to 1.5g. An inverse correlation is observed between the probability of
collapse and passive resistance of backfill soil at IMs larger than 0.75g: the fragility increases as
the passive resistance of soil decreases. Furthermore, the performance of bridges with Categories
I and II, hyperbolic and SDC 1.7 bilinear backfill models are similar at the 0o skew angle. The
probability of collapse drops at 15o skew angle compared to the non-skew case. No particular trend
is observed at the 15o skew angle. For skew angles larger than 15o, the probability of collapse rises
with the increase in skew angle, reaching a maximum of 80% for Category IV backfill at the
extreme case of 60o skew angle. Surprisingly, bridges with the least-resistant type of backfill (i.e.,
Category I) demonstrate the lowest fragility, which does not exceed 40% at 60 o skew. This
probability is increased to approximately 60%, 70%, and 80% for bridges with Category II, III,
and IV backfills, respectively.
The present investigation shows that a strong backfill soil can lead highly skewed bridges
to collapse due to unseating. Figure 7.16 illustrates the proportion of the two collapse mechanisms
(i.e., exceeding 8% column drift and abutment unseating) at each skew angle in bridges with
various models of backfill response and material. Column failure is the dominant failure
mechanism at 0o and 15o skew angles. In non-skew bridges, a very clear correlation is observed
between the number of collapses due to column failure and the variation of backfill passive
resistance—a trend identical to the correlation that was previously observed between the
deformation demands exerted on the columns and backfill resistance as well as in collapse fragility
curves. However, regardless of the degree of abutment skew, the trend in the number of column
failures from one case to another is lost in the skewed configurations.
Abutment unseating does not become significant until the skew angle exceeds 15 o, after
which a rise is observed in the number of both column failure and abutment unseating of all cases
with increasing skew angle. At 30o, 45o, and 60o skew angles, column failure and abutment
unseating cases contribute at comparable levels to the overall collapse cases. Furthermore, the
number of abutment unseating cases demonstrates a consistent pattern with the variation of backfill
model at these skew angles: in general, higher number of unseating cases are observed in models
with higher passive resistance of backfill soil (with the exception observed between Category III
and IV GHFD backfills). Such an observation is consistent with the fragility curves of Figure 7.15.
121
Figure 7.15 Fragility curves for bridges with SDC 1.7 bilinear and Category I to IV GHFD backfill
model (1 cm = 0.3937 in., 1 kN/m = 5.71 lb/in.).
(a) (b)
Figure 7.16 Frequency of (a) column and (b) abutment collapses in bridges with SDC 1.7 bilinear
and Category I to IV GHFD backfill models.
The higher fragility and abutment unseating in bridges whose backfills consist of stronger
materials (i.e., Categories III and IV backfills) can be explained by exploring the forces acting on
the deck of a skew bridge, as are schematically shown in Figure 7.17 for various modes of deck
displacement. These forces include the resultants of the backfill’s passive resistance, the column
shears, and the shear-key reactions that are marked on the figure with Fp, Fc, and Fsk, respectively.
At zero or small skew angles, deck rotation is mainly due to the randomness of the resultant
direction of ground’s horizontal motions at each instant of time, as well as response nonlinearities,
which may result in loss of configuration symmetry, and trigger rotational motions of the deck.
Therefore, the deck motion is mainly in the translation mode and collapse occurs as a result of
122
columns reaching their limit state. Consequently, bridges that receive less lateral support from the
backfill soil exhibit a higher chance of column failures and probabilities of collapse.
(a) (b)
Fsk Fsk
Fc Fp
Fc Fc
Fp Fc
Fsk
(c) (d)
Fsk Fc
Fp Fc
Fc Fp
Fp Fsk
Fc
Fsk
Figure 7.17 Schematic illustration of reaction forces on bridge deck for (a) longitudinal translation,
(b) transverse translation, (c) counter-clockwise rotation, and (d) clockwise rotation of deck.
At large skew angles, the passive resistance of the backfill soil triggers deck rotation
(clockwise for the shown configuration of skewed abutments), as illustrated in Figure 7.19a and b,
for pure longitudinal and transverse modes of deck displacement. In case of an accidental counter-
clockwise rotation, all three pairs of reaction forces act in a restoring mode, to reverse the direction
of rotation (Figure 7.19c). Finally, in case of an accidental clockwise rotation (Figure 7.19d),
backfill passive resistance tends to amplify the clockwise rotation, while the force-couples
generated by the reaction of shear-keys and the shear resistance of columns resist against such
clock-wise rotation. Eventually, the relative magnitudes of the forces acting on the deck determine
the extent and direction of its rotation. Backfills composed of stronger materials develop larger
passive pressures on a bridge deck, and subsequently, larger moments, which lead to further
clockwise deck rotations. This explains higher probability of abutment unseating and collapse
potential that is observed at high skew angles in bridges with Category III and IV backfills
compared to bridges with backfills that possess lower resistance (i.e., Category I, II, and SCD 1.7
bilinear backfills). The same explanation is valid when comparing the instances of abutment
unseating and probability of collapse between the bridges with Category I and Category II backfills.
In addition to increasing the probability of unseating, large deck rotations translate to additional
column drifts, as observed in the second rows of Figures 7.11 to 7.13.
However, between bridges with Category III and Category IV backfills, despite the larger
forces, which can be developed by the passive resistance of the Category III backfill, the former
shows a lower probability of collapse compared to the latter. Indeed, the competing role of deck
rotation and lateral translation is best comprehended here: the higher lateral support provided by
the stronger Category III backfill, as compared to Category IV backfill, limits deck translation.
Restraint of deck translation significantly lowers column fragility, but reduces the probability of
deck unseating, only marginally.
123
The intuitive interpretation of force diagram is backed by the distribution of deck rotations
shown in Figure 7.18 and 7.19 for cases that the bridge survives to the end of ground motion
excitation. The maximum deck rotations generated by the 40 ground motions applied at 21
intercept angles are distributed among 40 bins in between −0.002 and +0.002 radians.
Subsequently, all rotations with absolute values larger than 0.002 rad are lumped into the extreme
bins centered at −0.002 and +0.002 radians. The adopted sign convention considers a clockwise
rotation (as shown on Figure 7.23) as negative. In the non-skew configurations the maximum
rotations have a high frequency of occurrence at very small values, and are evenly distributed in
the positive and negative regions. At 15o skew, an obvious accumulation is observed in the first
bin corresponding to negative rotations. As the skew angle increases, deck rotations reach larger
values, and the distribution extends to the bins away from 0 rad. Deck rotations with absolute
magnitudes larger than 0.002 radians tend to be in the negative (clockwise) direction, as evident
by higher population of the bin centered at −0.002 radians. Dominance of clockwise rotation of
deck is more pronounced in bridges with backfills possessing higher passive resistance.
Figure 7.18 Distribution of deck rotation in bridges with SDC 1.7 bilinear backfill model.
Figure 7.19 Distribution of deck rotation in bridges with Category I GHFD backfill model.
124
Figure 7.20 Distribution of deck rotation in bridges with Category II GHFD backfill model.
Figure 7.21 Distribution of deck rotation in bridges with Category III GHFD backfill model.
Figure 7.22 Distribution of deck rotation in bridges with Category IV GHFD backfill model.
Shear-key fragility curves for bridges with the lowest and highest probabilities of collapse
(i.e., bridges with Category I and IV backfills, respectively) are shown in Figure 7.23. The limit
state that corresponds to shear-key failure is defined as reaching the onset of plastic deformation,
125
at which the shear-key essentially loses its load bearing capacity. The explanation conjectured
earlier by examining the free-body diagram of the deck and distribution of the maximum deck
rotation is verifiable against the increased probability of failure in the two obtuse-corner shear-
keys compared to the two acute-corner ones, as demonstrated in Figure 7.23. Furthermore, the
fragility curves often demonstrate an inverse correlation between the numbers of failed obtuse-
and acute-corner shear-keys—i.e., a higher probability of obtuse-corner shear-key failure is
associated with a lower probability of acute-corner shear-key failure. This implies that failure of
the obtuse-corner shear-keys triggers further clockwise rotation of the deck (for the given
configuration of the skewed abutments), hence, a higher probability of survival of the acute-corner
shear-keys. It should be emphasized that although strong correlation exists between shear-key
failure and bridge collapse, the former is neither necessary nor sufficient for occurrence of the
latter. Therefore, bridge and shear-key fragility curves are not identical.
Another significant observation is the reversal of probability of shear-key failure between
bridges with Category I and IV backfills at 60o skew angle. In a skewed configuration, 1 unit lateral
displacement of deck normal to the backwall translates into 1sin() unit displacement in the
transverse direction—i.e., ~0.71 and ~0.87 units of transverse displacement at 45o and 60o skew
angles, respectively. Transformation of approximately 87% of the relatively large lateral
displacements—allowed by the very soft Category I backfill soil—into transverse displacement at
60o skew angle results in high probability of shear-key failure, although it may not necessarily
result in the abutment unseating. In other words, at large skew angles, abutment unseating is
subsequent to, and in direct correlation with, deck rotation, whereas shear-key failure is caused by
a combination of deck rotation and transverse displacement. Amplification of the latter at the 60o-
skew configuration of bridges with Category I backfill increase the probability of shear-key failure
in such bridges.
OBT
ACT
ACT OBT
Figure 7.23 Fragility curves for acute- and obtuse-corner shear-keys in bridges with Category I
and IV GHFD backfill models.
126
7.6.3 Influence of passive resistance reduction method on collapse mechanism
The two approaches for incorporating skew effects in reducing the passive resistance of backfill
(i.e., uniform and non-uniform reduction methods) are compared by investigating bridge collapse
fragility curves with Category I and II backfills (Table 7.2, Figure 7.24). With the uniform passive
resistance reduction coefficients, the collapse potential consistently increases with an increase in
skew angle, since the reduction factor is a hyperbolic function of the skew angle. The maximum
discrepancy in the output of the two models is about 15%, which corresponds to the probability of
collapse at the 30o-skew configuration of bridges with Category I backfill.
Despite the similarities of collapse probability in the two models, the collapse mechanism
is completely altered when the uniform reduction factors are used, instead of reducing the passive
resistance over a limited width of backfill. As shown in Figure 7.25, column failure turns out to be
the dominant collapse mechanism, regardless of skew angle. Application of uniform reduction
factors results in two- to three-fold increase in the frequency of column failure, while the cases of
abutment unseating are almost halved in quantity, compared to the cases with non-uniform
reduction of backfill passive resistance.
Table 7.2 Uniform reduction factors for the passive resistance of backfill at 15 o to 60o skew angles.
Skew Angle 15o 30o 45o 60o
R 0.7465 0.5290 0.3475 0.2020
Figure 7.24 Fragility curves for bridges with uniform and non-uniform reduction of passive
resistance in skew abutments with Category I and II GHFD backfill models.
Distributions of deck rotation and shear-key fragility curves are shown in Figures 7.26 and
7.27. As the skew angle increases, the maximum deck rotations gradually migrate towards positive
values, which correspond to counter-clockwise rotation of the deck for the skewed configuration
of Figure 7.17. Reversal of the direction of deck rotation alters the shear-key failure pattern: acute-
127
corner shear-keys demonstrate higher failure probability compared to the obtuse-corner shear-keys
at 45o and 60o skew angles. As discussed earlier, relative magnitudes of the moments generated by
the force-couples shown on the deck force diagram dictates the direction of rotation. The very
large amount of reduction in the passive resistance due to the hyperbolic relationship of Eq. 4-15
for 45o and 60o skew angles (Table 7.2) substantially reduces the magnitude of the clockwise
moment that is generated by such forces. Consequently, the counter-clockwise moment formed by
the compression-only reaction forces—which are developed in the obtuse corner shear-keys as a
result of small amounts of clockwise deck rotations—can alter the direction of rotation to
counter-clockwise and trigger subsequent failures of the acute-corner shear-keys.
Figure 7.25 Frequency of (a) column and (b) abutment collapse in bridges with Category I and II
GHFD backfill models and uniform reduction of passive resistance in skewed configurations.
Figure 7.26 Distribution of deck rotation in bridges with Category I GHFD backfill model and
uniform reduction of passive resistance in skewed configurations.
128
Figure 7.27 Distribution of deck rotation in bridges with Category II GHFD backfill model and
uniform reduction of passive resistance in skewed configurations.
OBT
ACT
ACT OBT
Figure 7.28 Fragility curves for acute- and obtuse-corner shear-keys in bridges with Category I
and II GHFD backfill models and uniform reduction of passive resistance in skewed
configurations.
129
backfill, shear-keys, and bent columns on the deck. Throughout the previous sections, the response
of transverse shear-keys was modeled with the tri-linear backbone, which corresponds to a brittle
mode of failure. In the subsequent section, the ductile response is used along with the non-uniform
GHFD backfill models. The outcomes will further the understanding of coupled response of
backfill and shear-keys in skew abutment bridges.
Bridge fragility curves (Figure 7.29) and the number of collapsed cases due to either
column failure or abutment unseating (Figure 7.30) do not exhibit any specific trends with the
passive resistance of the backfill soil or the skew angle. Apparently shear-key ductility is
responsible for lowering the fragility. Furthermore, the cases of column failure outnumber the
cases of abutment unseating. The maximum difference observed in the probability of collapse
among all abutment configurations and materials is ~10% for the same backfill Category (i.e.,
Category III backfill at 30o and 45o skew angles) and ~15% for different type backfills (i.e.,
Category II and III backfills at 30o and 45o skew angles). It seems that the residual load bearing
capacity provided by the ductile shear-key over a relatively large extent of plastic deformations
plays a major role in limiting and/or reversals of deck rotation, and thus, in preventing abutment
unseating. The shear-key compression-only reaction and backfill passive resistance compete in
deciding the direction of deck rotation, and based on their relative magnitudes and the skew angle
value the bridge may become prone to rotate in either of the directions. This assumption can be
verified through the examination of distribution of deck rotation for the least and most resistant
backfill Categories (i.e., Categories I and III). The force-couple formed by the reaction of shear-
keys control the direction of rotation to be predominantly counter-clockwise, when the passive
resistance is provided by Category I backfill. With Category III backfill, the moment due to the
backfill passive resistance is still large enough to surpass the tendency for counter-clockwise
rotation due to shear-key reactions. However, the maximum rotations are more evenly distributed
between 0.001 to −0.002 rad, rather than piling up in the bin centered at −0.002rad. This indicates
the capability of ductile shear-key in restricting the deck rotation to smaller angles.
The significant role of shear-key ductility in controlling the seismic response of skewed
bridges cannot be overstated. This matter has been comprehensively addressed in the second focus
area.
130
Figure 7.29 Fragility curves for bridges with SDC 1.7 bilinear and Category I to IV GHFD backfill
and ductile shear-key models.
Figure 7.30 Frequency of (a) column and (b) abutment collapse in bridges with SDC 1.7 bilinear
and Category I to IV GHFD backfill and ductile shear-key models.
131
(transverse) rows of 39 ft CIDH circular RC piles (2 ft diameter with 8#9 rebars) in Bridge B
provide support to the columns. Considering the different mechanisms of plastic hinging in RC
and steel sections, the simulations were repeated with RC piles (2 ft diameter with 6#9 rebars) for
Bridge A, in addition to the simulations which were done with its as-built steel HP piles. The RC
section chosen to replace the H section in Bridge A is selected from Caltrans standard plans for
RC piles such that the moment-curvature response of the substitute RC pile is comparable to that
of the as-built steel HP pile about its strong axis, as displayed in the moment-curvature plots of
Figure 7.31.
The backfills are assumed to be composed of “medium dense silty sand”, which is the soil
type identified in the backfill of about 50% of bridges that were investigated in a field study to
classify existing backfills throughout California. The soil profile under the column footing has
been chosen based on the findings of the work by Mackie et al. [2012] in evaluating the effects of
interaction between the ground and bridge foundation on the loss and repair time of a two-span
bridge with a single column bent damaged during seismic excitation. Four different foundation
conditions were adopted in the aforementioned study, including fixed base condition
(corresponding to rigid soil), two gradually varying clay soil profiles, and a clay soil profile with
“weak upper strata.” The last scenario has been adopted in the current study, as it has been shown,
in general, to yield more pronounced effects on the seismic demand of deep foundation elements
and variation of superstructure demand from the fixed base condition. In both specimen bridges,
the weak layer is assumed to extend to 30 ft below the surface. Table 7.3 lists the geotechnical
properties of the upper and lower strata.
Figure 7.31 Moment-curvature relationship (under no axial force) of pile sections in the specimen
Bridges A and B.
Table 7.3 Geotechnical properties of foundation soil (composed of clay), used for deciding
properties of soil springs.
Layer (lb/ft3) Su (psi)
Upper (surface to −30 ft) 95 4.5
Lower 125 21
132
The suite of 40 near-fault seismic records are employed to study the seismic performance
of the two specimen bridges. The orthogonal directions for applying the two horizontal
components of ground motion have been rotated at 9 degree increments in order to eliminate any
bias that may result from a particular incident angle. In order to assess the seismic effects of vertical
variation of ground motion along the embedded length of pile, the input motion is applied once as
uniform-excitation and once as multiple-support-excitation. For the latter, horizontal components
of motion of soil at various depths are generated through deconvolution of the surface motion
(using SHAKE91, [Idriss 1993]) to the depth of pile tip, and are imposed at the free ends of the
orthogonal p-y macro-elements. The input parameters for the deconvolution analysis are listed in
Table 7.4. The surface motion (applied at the pile heads) is imposed at the abutments, as well.
Effects of vertical ground excitation have been excluded from the current study.
Peak ground acceleration, spectral acceleration at 1 second (as a representative period), and
maximum spectral acceleration at various depths (between the surface and −40 ft, which is the
vicinity of pile tips in the specimen bridges) for the strike normal and strike parallel components
of the 40 earthquake events are shown in Figure 7.32. The line-up of the events for each plot is
based on the surface value of the quantity shown on the vertical axis, e.g., PGA0 in Figure7.32a.1
corresponds to the peak ground acceleration of the strike normal component of the events at surface
(elevation 0). Therefore, the line-up of the events is not necessarily the same amongst all plots.
While the vertical variation of the abovementioned characteristics seem to be negligible for smaller
events, it can reach extreme values as large as 1.0 g at higher peak or spectral accelerations. For
such cases, both peak ground acceleration and maximum spectral acceleration show significant
fluctuations below the depth of 20 ft. On the other hand, the spectral acceleration at 1 second
becomes smaller from the surface to the bottom of the weak upper layer, and remains constant in
the lower strong layer. The period that yields maximum spectral acceleration at each elevation is
illustrated in Figure 7.33. Some period elongation is observed between the depths of -10 and -30
ft at smaller spectral accelerations. However, such trend is lost as the spectral acceleration becomes
larger.
The additional modeling burden associated with the application of multi-support excitation
is often a concern in routine engineering tasks, particularly in cases where the number of supports
with different imposed motion is relatively large. The response of the bridge models with deep
foundation elements to uniform (surface) excitation were also simulated and compared to the
133
response under multi-support excitation, to assess the extent of discrepancy and similarity of the
response under the two excitation scenarios at various intensities of shaking. Table 7.5 shows the
cases that were included in the sensitivity analysis, given various modeling options and methods
of input motion application.
In order to gain insight into the sensitivity of seismic response to variation of modeling
parameters listed in Table 7.5, variation of response parameters against a measure of ground
motion intensity have been investigated. The intensity measure chosen here is the spectral
acceleration GMRotI50 [Baker et al. 2011] at 1 sec. It is assumed that maximum response at each
intensity measure (i.e. EDP|IM) is a random variable with a lognormal distribution. Variation of
the median and mean of the response parameters given the level of ground motion intensity
measure are examined. Looking simultaneously at the median and mean of the response parameter
allows one to investigate how the distribution is changing from one case to another. The difference
between the medians of two cases indicates whether the distribution’s central tendency is changing
from one case to another. Subsequently, the difference between the means of two cases indicates
whether the variance of the lognormal distribution (i.e., dispersion of the response variable), in
addition to its median, is changing from one case to another.
Figure 7.32 Variation of (1) peak ground acceleration, (2) spectral acceleration at 1 s, and (3)
maximum spectral acceleration of strike normal (top row) and strike parallel (bottom row)
components of ground motion, between ground surface and -40 ft below the surface.
134
(a) (b)
135
Another noteworthy observation is the role of degree of rotational fixity at the column
footing or in the column-to-pile-cap connection in controlling deck displacement in Bridge B. As
expected, deck displacements are smaller in the model with fixed column footings. However, as
evident from the plots, constraining rotational degrees of freedom at the column-to-pile-cap
connection can also surpass the effect of additional lateral flexibility that is introduced by the deep
foundation elements; i.e., median of maximum deck displacements are smaller in the model with
rigid column-to-pile-cap connection than in the model without the deep foundation elements but
hinged column footing.
(a1) (b1)
(a2) (b2)
Figure 7.34 Abutment longitudinal (top row) and transverse (bottom row) unseating in Bridges (a)
A and (b) B.
136
7.7.2 Sensitivity of column drift ratio to foundation model characteristics
Figure 7.35 shows the variation of maximum column drift ratios in bridges A and B. The deviation
of mean from median slightly increases in case of hinged column-to-pile cap connection in bridge
B. However, the difference between mean and median is not as profound as what was observed
for the transverse displacement of deck.
In both bridges, the column drift ratio in models with rigid column-to-pile-cap connections
is smaller compared to the models with fixed column footing. Smaller lateral stiffness at the
column base, due to the presence of deformable foundation and weak layers of soil, allows the
column footing to move along with the superstructure, as such, reduce the relative displacement
between the top and bottom of the column.
Footing displacements (as shown in Figure 7.36) are the largest in bridge A model with
steel H piles. Formation of plastic hinges and accelerated increase of curvature in HP piles
compared to CIDH piles after reaching yield curvature (as will be discussed later) is the main
reason for such observation. Subsequently, reduction of column drift ratio in Bridge A model with
HP piles is the most significant. Evident from Figure7.36.b, column footings undergo larger
displacements in bridge B model with rigid column-to-pile-cap connection compared to the model
with hinged connections. This is due to the transfer of moments from rigid column-to-pile-cap
connection to the piles, which results in larger pile deformation and subsequently larger
displacement of pile cap.
Despite the added lateral flexibility of the base in the presence of deep foundation elements,
the column drift ratios in Bridge B models with hinged column-to-pile-cap connection increase
relative to the hinged base case. In order to explain this trend, one may consider an analogy
between the simplified column models shown in Table 7.6 and the models of bridge with various
footing fixity conditions. The deformation of the extension of the column below the ‘base’ yields
additional rotation between the top and the ‘base’, particularly when the moments are released at
the ‘base’ (similar to the bridge model with hinged column-to-pile-cap connection). The relative
rotation between column top and base, about Bridge B transverse direction, is shown in Figure
7.37. While the relative rotation between the top and bottom of the column is very similar between
the fixed and hinged bases cases, it increases—as explained earlier—in the presence of deep
foundation elements. Although the increased rotation-induced column drift is offset by the effects
of added lateral flexibility at the base to some extents, it still elevates the drift ratios in the cases
of the specimen bridge B with hinged column-to-pile-cap connections.
In order to better understand the effects of soil-foundation-structure interaction on the
seismic response of superstructure observed so far, force-deformation behavior of piles are studied
next.
137
Table 7.6 Schematic analogy between simplified column models and cases of base-fixity and
column-to-pile-cap connection in Bridge B.
Base Condition Fixed Base Rigid Footing Hinged Base Hinged Footing
Simplified
Column
Analogy
(b)
(a)
138
(a) (b)
139
Figure 7.37 Relative rotation between the top and bottom of column in Bridge B.
Variation of median of maximum moment and curvature with the ground motion intensity
measure along the height of a single pile for the specimen bridges are shown in Figures 7.39
through 7.42. Presence of a few outliers in the curvature of the elements close to the tip of the piles
in some of the cases results in fictitious large skewness of median in the vicinity of the pile tip.
Aside from that, it was observed that the median and mean of both quantities are similarly
comparable for all cases. Therefore, in order to circumvent the effect of fictitious outliers, normal
distributions were adopted instead for the probabilistic assessment of foundation response.
Assigning normal distributions to the moment and curvature at the elevations away from the tip
along the piles yields median (mean) values, which are similar to the medians deduced from
lognormal distributions.
Figure 7.38 Effect of axial load on the moment-curvature relationship of the RC pile in Bridge B.
140
In general, the maximum curvature is larger in the longitudinal direction compared to the
transverse direction. Such observation is consistent with the larger longitudinal displacements,
which were observed at the location of the abutments. Transfer of column footing moments to the
piles in cases with rigid column-to-pile-cap connections also leads to increased pile curvature,
compared to the cases with hinged connections.
Peak curvatures occur at the pile head and around the depth of -15 ft below the pile head.
The curvatures monotonically decrease at depths below 20 ft from the surface in Bridge A piles
and in bridge B under uniform-excitation. However, a different trend is observed in Bridge B under
multiple-support excitation: at higher intensities of ground shaking the curvature rise again in the
vicinity of the boundary between the week upper strata and the stiffer lower layer. Apparently, two
factors lead to the non-monotonic variation of curvature at lower depths in Bridge B: non-uniform
support excitation, and the additional fixity of the pile tip due to the extended length of Bridge B
piles (compared to the shorter Bridge A piles) in the stiff clay layer at the bottom. This observation
underlines the importance of vertical variation of soil profile and input motion in defining the
deflected shape of piles.
The head curvature in some cases for Bridge B is above the curvature, which corresponds
to the inelastic deformation of the RC pile section (about 1.510-4). However, both transverse and
longitudinal curvatures in all cases of Bridge B are well below the curvatures corresponding to the
onset of softening and unbounded deformation of the RC section, i.e., about 510-4 and 1.510-3,
respectively. Furthermore, variation of moment at all elevations along the pile depth is almost
proportional to the variation of curvature in all three cases as demonstrated in Figures 12 and 13
for Bridge B. This indicates that no plastic hinge is forming in the piles of Bridge B.
On the other hand, the curvature increases abruptly at the pile head relative to the lower
elevations in Bridge A, while the head moment does not increase with the same rate as the
curvature. Therefore, formation of plastic hinges at the pile head is evident in Bridge A models.
The increase of curvature at the pile head is even more acute in the model with steel HP piles, as
a result of HP section moment-curvature diagram almost reaching a plateau upon the initiation of
inelastic deformation in the section. Plastic hinging is also expected to happen about -15 ft below
the head in steel HP pile at high intensities of shaking, considering the magnitude of curvature at
this location, which is beyond the yield curvature of 2.510-4. Consistent with this speculation, at
high intensity measures the magnitude of longitudinal moment at this depth is almost the same as
the moment at the location of head plastic hinge (Figure 7.39.c.2).
These observations are in agreement with the previously observed response of columns: an
inverse relationship exists between the column drift ratio and the lateral deformation of the
foundation. Formation of plastic hinges in Bridge A piles yield large lateral deformations, thus, the
foundation acts similar to a ‘base isolation layer.’ Accordingly, the very large pile head curvatures
in Bridge A model with steel HP piles explain the drastic reduction of column drift ratios in this
model, compared to all other cases.
In general, the computational burden associated with nonlinear time-history analysis rise
as the number of nonlinear elements in the structural model increase. Therefore, it would be
beneficial to gain insight about the depth and extent of soil nonlinearity, in the specimen bridges.
Median of the ratio between maximum and yield deformation (i.e., /y) of the p-y springs along
the height of one of the corner piles is examined, as a measure of soil nonlinearity. Figure 7.43
shows variation of this ratio with ground motion intensity measure, at various depths measured
141
from the pile head for three cases of Bridge B modeling. As a general trend, inelastic deformation
increase at shallower depths and under higher ground motion intensity levels. Smooth variations
are observed with respect to the depth and intensity of shaking under uniform-excitation (Figure
7.43.c). However, the transition is not as smooth under non-uniform input motion, i.e., from the
tip of the pile to the depth of -20 ft, the plastic deformations of the springs do not necessarily
increase. Noticeably, loss of a uniform transition trend is not restricted to the vicinity of the
interface between loose and stiff soil (at -30 ft) and extends beyond this region. Consistent with
the larger pile deformations that were observed in the longitudinal direction, longitudinal springs
undergo larger amount of inelastic deformation compared to the transverse springs. Such
difference is most pronounced in Bridge B with rigid connection between column base and pile
cap (Figure 7.43.a).
Normalized backbones of longitudinal p-y springs of Bridge B at the location of pile head,
as well as 24 and 30 below surface, which are the location of local /y peaks, are illustrated in
Figure 7.44 in order to achieve a better understanding of the magnitude of /y. For example, the
/y ratio at 24 ft below pile head becomes as large as 5 at high intensity measures. Apparently,
the p-y spring corresponding to this depth experiences significant amount of softening for such
/y ratio, thus, replacing it with a linear spring with the purpose of reducing computational and
model complexity may not be a valid choice for the specimen bridge.
Figure 7.39 Bridge A pile maximum longitudinal moment and curvature; (a) CIDH-RF-MSE, (b) CIDH-
RF-UE, and (c) HP-RF-MSE. (*Note the different scale on the curvature axis).
142
(a1) (b1) (c1*)
Figure 7.40 Effect of axial load on the moment-curvature relationship of the RC pile section (Bridge
B).
143
(a1) (b1) (c1*)
Figure 7.41 Bridge B pile maximum longitudinal moment and curvature; a) CIDH-RF-MSE, b) CIDH-
HF-MSE, and c) CIDH-HF-UE.
144
(a1) (b1) (c1*)
Figure 7.42 Bridge B pile maximum transverse moment and curvature; (a) CIDH-RF-MSE, (b) CIDH-
HF-MSE, and (c) CIDH-HF-UE
145
(a1) (b1) (c1*)
Figure 7.43 Deformation of Bridge B longitudinal (top row) and transverse (bottom row) p-y springs
beyond yield point; (a) CIDH-RF-MSE, (b) CIDH-HF-MSE, and (c) CIDH-HF-UE.
Figure 7.44 Normalized backbone of Bridge B longitudinal p-y springs (shown only at the location
of peak “maximum to yield deformation ratio” of foundation soil).
146
8 Guidelines and Recommendations for
Modeling and Analysis
This report presented the second-generation comprehensive guidelines for performance-based
seismic analysis (PBSA) of ordinary bridges. Chapter 2 presented an overview of Performance-
Based Earthquake Engineering (PBEE) methodologies with an emphasis on the methodology
developed at the Pacific Earthquake Engineering (PEER) Center. Chapter 3 discussed the bridge
structures used in the present study and provided an overview of the bridge analysis models of the
first-generation guidelines. Chapter 4 investigated the enhanced modeling approaches used to
overcome the shortcomings of the first-generation guidelines. Chapters 5 provided different
approaches to eliminate potential convergence problems associated with nonlinear time-history
analyses (NLTA) required for PBEE. Ground motion selection and scaling was the core of Chapter
6. The sensitivity analyses were carried out in Chapter 7, which illuminated those features of bridge
models that control the predicted seismic responses.
Key findings from these previous chapters are organized and offered in the present chapter
in the form of recommendations on modeling and analysis procedures for PBSA of ordinary
bridges.
147
For the modeling case of reduced passive backfill resistance over only a limited width of
the skewed backwall, the forces developed by the passive backfill resistance tend to rotate
the deck in a direction that results in the shearing of the shear-keys located at the obtuse
corners of the abutment. Total loss of residual load bearing capacity in brittle shear-keys
after reaching the yield deformation amplifies the deck rotation, which also translates to
larger column drifts. Thus, in general, an increase in the magnitude of backfill passive
resistance leads to higher probability of collapse.
Application of the large empirical reduction factors to the total passive resistance of a
skewed backfill completely alters the collapse mechanism, and unseating becomes a highly
rare event, even at large skew angles.
Dependence of the bridge fragility on the passive resistance of backfill at large skew angles
vanishes as the transverse ductility is increased. The residual capacity of ductile shear-
keys provides lateral support against rotations and unseating, and thus, significantly lowers
the probability of collapse.
These findings underline the significance of the coupling between the passive backfill resistance
and shear-key responses in defining the fragilities and collapse mechanisms of typical two-span
two-column-bent skew bridges. The outcomes also highlighted the sensitivity of the seismic
performance of the specimen bridge to the force-deformation models of the abutment backfill and
the shear-keys. Therefore, incorporation of reliable constitutive models that exhibit the real-life
characteristics of the backfill and the shear-keys is essential for achieving robust
prediction/assessment of the seismic performance of ordinary highway bridges and for the
development of effective retrofit measures.
148
Both shear key and backfill soil properties play important and coupled roles on bridge
responses. Shear key behavior is more influential on seismic demands when the non-uniform
backfill pressure assumption is utilized.
Bridges with brittle shear keys exhibit a higher probability of unseating compared to bridges
with ductile shear keys, even for large seat-width values. Therefore, realistic modeling of shear
keys is a critical issue for quantifying the probability of failure due to abutment unseating.
The following recommendations can be made based on the results of the sensitivity analyses:
Brittle (i.e., sacrificial) shear keys increase seismic demands (e.g., deck rotation) that mostly
need to be controlled by bridge columns. On the contrary, ductile shear keys reduce seismic
demands by controlling the deck movement, and transfer large amounts of forces to the
substructure. Consequently, this behavior leads to overloading of the abutments and piles.
Therefore, providing a shear key with specific reinforcement details that behaves in a ductile
manner to dissipate energy with lower force capacity may be a solution to avoid undesirable
damages and bridge failures.
Based on the Caltrans guideline (SDC, 1.7), the capacity of a ductile (i.e., non-isolated) shear
key should be obtained using the shear friction method given in Caltrans Bridge Design
specifications (BDS Caltrans, 2003). However, experiments have shown that use of the shear
friction model with the current values of the coefficient of friction significantly underestimates
the capacity of the sacrificial shear keys. As such, these existing provisions (Megally et al.,
2001) should be revised.
The results reveal that the backfill response of a skewed abutment has a very significant effect
on seismic demands. More research is needed in this area to investigate the backfill reaction
against skew abutments so that existing models of this reaction are validated (or improved, as
needed).
149
Slight increase in the median longitudinal displacement of the deck is observed in the
single-column bridge. The effect of including deep foundation elements on the deck
longitudinal response is less pronounced in the bridge with multiple columns. Response
dispersion remains relatively the same for all cases of modeling.
Column drift ratios are reduced in models with rigid column-to-pile-cap connections,
particularly in cases where plastic hinges are formed in the piles compared to the cases
with fixed column footings. This observation is explained by the transfer of moments from
the superstructure to the foundation and subsequent large lateral deformation of the
foundation. However, the effect of foundation lateral deformation on column drift ratio is
masked by the increased rotation at the top of the columns relative to their base in the
model with hinged column-to-pile-cap connection. Therefore, the column drift ratio
increased in such models.
Releasing the moments at the column-to-pile-cap connection can significantly lower the
moment demands that are developed in the piles, as well as the amount of nonlinear
deformations of the foundation soil. Furthermore, no tensile forces are developed in the
piles of models with hinged column-to-pile-cap connections.
The additional moment capacity that an RC section can sustain compared to HP pile section
after reaching the onset of inelastic deformation can significantly limit the amount of pile
curvature, and subsequently the lateral deformation of the foundation layer.
Exceedance of the yield curvature and the formation of plastic hinges are not necessarily
restricted to the vicinity of rigid pile-to-pile-cap connections; they should also be expected
at the bottom of the upper-half of steel HP piles.
Application of multiple-support ground motion together with the extended length of piles
in the lower stiff layer yields different patterns in the variation of the pile curvature,
compared to the case of uniform-excitation. Mainly, the pile curvature may show local
increase at, and in the vicinity of, the interface between the two layers of soil foundation.
Application of multiple-support-excitation also alters the pattern observed in the
deformation of soil springs along the foundation depth. Under uniform-excitation, soil
deformations increase monotonically from the bottom to the top—similar to the trend
observed in the pushover of a free-head embedded pile. On the other hand, under multiple-
support-excitations, the monotonic pattern is lost at depths below the mid-elevation of the
piles in the specimen bridge; significant fluctuations and increases are observed in the
vicinity and beyond the interface of the two layers of foundation soil.
These findings underline the superstructure’s seismic responses are not necessarily insensitive to
the features of the deep foundation elements. However, various simplifications can be made
depending on the soil conditions, foundation type, and the targeted performance level.
150
the analysis procedures and integration methods, are represented below in the form of
recommendations to the Caltrans bridge engineers:
The explicit Operator-Splitting (OS) algorithm with tangent stiffness formulation
possesses similar stability and accuracy properties as those of Implicit Newmark (IN)
algorithm.
All of the investigated numerical integration algorithms provide very close results to the
exact theoretical solution for the nonlinear SDOF pendulum problem.
The simulations of the investigated three bridges show that the OS and TRBDF2 algorithms
provide very close results to the IN algorithm. Accordingly, it is concluded that the OS and
TRBDF2 method can serve as a suitable alternative to the IN method for NLTA of highway
bridges.
The TRBDF2 algorithm and adaptive switching of algorithms showed improved
convergence performance as compared to IN.
An analyst should be aware of the presence of the alternative methods such as the EN and
OS methods and the option of adaptive switching among these methods and should use
these methods whenever they are applicable in order to avoid the possible convergence
problems associated with the IN method. It is beneficial to verify the accuracy of these
alternative methods by comparison with the IN method as demonstrated in Chapter 5 for
highly nonlinear cases.
For the implicit integration methods, Newton with Line Search is observed to be the most
suitable initial nonlinear solver in terms of convergence. Accordingly, an analyst can start
with this method as the initial solver in the first simulation trial of an analytical model.
The sequence of the nonlinear solvers after a proper selection of the initial solver is
observed to be insignificant. Accordingly, an analyst should pay more attention to the
determination of the initial solver than the determination of the sequence of the subsequent
solvers in the NLTA.
Simulations with the convergence test based on the Energy Increment lead to the same
solution with significantly less number of iterations compared to other convergence tests.
Accordingly, an analyst can consider the Energy Increment test in the first simulation trial
of an analytical model subjected to NLTA.
Simulations conducted with tolerances of 10-5 and 10-8 selected for the Energy Increment
test for all the integration time steps (of the simulation) are observed to result in very similar
response calculations. Accordingly, the increase of the convergence tolerance for the
integration time steps with convergence problems is a valid option to achieve convergence
improvement.
Use of a smaller integration time step during the simulation does not necessarily improve
the convergence behavior. However, selective reduction of the integration time step, i.e.
only when needed, is useful to overcome the convergence problems, as long as the
integration time step is reset to its original value after completion of the reduced time steps
151
representing the original size of a time step. This resetting process is essential to prevent
the simulation from being completed before the duration of the ground motion.
Using explicit integrators only at the numerically problematic steps is a viable and effective
option to overcome the convergence problems of IN integration.
Higher mode effects are more pronounced on column displacements and deck accelerations
than the column shear forces. The column base shear is likely a first mode predominant
Engineering Demand Parameter (EDP), irrespective of the hazard level. The effect of
higher modes is observed to increase with hazard level and nonlinearity.
The intercept angle plays an insignificant role in evaluating HEPs. This is a useful
conclusion regarding the investigation of seismic response of typical California bridges,
because it is computationally expensive and not always possible to determine the largest
response that is generated by several rotated versions of a ground motion. It is further useful
to note that this conclusion is not effected by the presence of the SSI modeling.
All the investigated GMSM methods provide accurate estimates of column base shear. This
result is mainly because of reaching the column force capacity in both HEP analyses and
in those conducted using the ground motions selected by the GMSM methods. Accordingly,
it can be mentioned that all the GMSM methods are successful at least in the prediction of
damage at a global level.
Regarding the other three EDPs, the responses obtained from the Unconditional Selection
(US) method are almost always on the conservative side and are usually the most
conservative of all the investigated GMSM methods.
Modeling of the SSI effects play an insignificant role in evaluating GMSM methods.
Accordingly, the accuracy of some simulated GMSM methods could be investigated in the
future. For example, the accuracy of the simulated ground motion method by Rezaeian and
Der Kiureghian (2010a, b, 2011) can be investigated by selecting an earthquake scenario
without considering SSI effects ( Vs 30 600m / s ).
Predictions of GMSM methods do not vary significantly with the skew angle.
Standard ordinary bridges play a crucial role in transportation and thus require short
152
downtime after severe earthquakes regarding both the emergency response and community
resiliency aspects. Therefore and given that none of the GMSM methods is particularly
superior to the others, it is suggested to use the US method for selection and modification
of ground motions for different intercept angles and to choose the most critical intercept
angle that yields the largest response.
153
REFERENCES
AASHTO LRFD Bridge Design Specifications, American Association of State Highway and
Transportation Officials, 5th edition, Washington, DC, 2010.
AASHTO “Standard Specifications for Highway Bridges,” 16th edition, American Association of
State Highway and Transportation Officials, Washington, D.C., 1995.
American Society of Civil Engineers (ASCE). “Prestandard and Commentary for the Seismic
Rehabilitation of Buildings,” Report No. FEMA-356, Washington, D.C., 2000.
Applied Technology Council (ATC). “Seismic Evaluation and Retrofit of Concrete Buildings,”
Report No. ATC-40, Volume 1-2, Redwood City, CA, 1996.
Assimaki, D., Kausel, E., and Whittle, A. J. "Model for dynamic shear modulus and damping for
granular soils," Journal of Geotechnical & Geoenvironmental Engineering, 126(10), 859-869,
2000.
Arici, Y. and Mosalam K.M., "System Identification of Instrumented Bridge Systems,"
Earthquake Engineering and Structural Dynamics, 32(7), 999-1020, 2003.
Arici, Y. and Mosalam K.M., "Modal Identification of Bridge Systems Using State-Space
Methods," Journal of Structural Control and Health Monitoring, 12(3-4), 381-404, 2005a.
Arici, Y. and Mosalam K.M., "Statistical Significance of Modal Parameters of Bridge Systems
Identified from Strong Motion Data," Earthquake Engineering and Structural Dynamics,
34(10), 1323-1341, 2005b.
Arici, Y. and Mosalam K.M., “Modal Identification and Health Monitoring of Instrumented
Bridges Using Seismic Acceleration Records,” EERC Technical Report, June 2006.
Aviram A., Mackie K.R., and Stojadinović B., “Guidelines for Nonlinear Analysis of Bridge
Structures in California” Technical Report 2008/03, Pacific Earthquake Engineering Research
Center, University of California, Berkeley, CA, 2007.
Aviram, A., Mackie, K. R., and Stojadinovic, B. “Effect of abutment modeling on the seismic
response of bridge structures.” Earthquake Engineering and Engineering Vibration, 7(4), 395–
402, 2008b.
Aydinoglu M.N., “An incremental response spectrum analysis procedure based on inelastic
spectral displacements for multi-mode seismic performance evaluation,” Bulletin of
Earthquake Engineering, 1, 3–36, 2003.
Baker, J.W., “Conditional Mean Spectrum: Tool for Ground Motion Selection” ASCE Journal of
Structural Engineering, 137(3), 322-331, 2011.
Baker J W “Quantitative Classification of Near-Fault Ground Motions Using Wavelet Analysis,”
Bull. Seism. Soc. Am, 97(5): 1486-1501, 2007.
Baker, J. W., Lin, T., Shahi, S. K., and Jayaram, N. (2011). “New ground motion selection
procedures and selected motions for the PEER transportation research program.” PEER Report
2011/03, Pacific Earthquake Engineering Research Center, University of California, Berkeley,
CA.
154
Bank R.E., Coughran W.M., Fichter W., Grosse E.H., Rose D.J., Smith R.K. “Transient
simulations of silicon devices and circuits” IEEE TransCAD, 4, 436–451, 1985
Bathe K.J., “Conserving Energy and Momentum in Nonlinear Dynamics: A Simple Implicit Time
Integration Scheme” Computers and Structures, 85, 437-445, 2007.
Bathe K.J. and Baig M.M.I., “On a composite implicit time integration procedure for nonlinear
dynamics” Computers and Structures 83, 2513–2524, 2005.
Bathe K.J. and Wilson E.L., “Stability and accuracy analysis of direct integration methods”
Earthquake Engineering and Structural Dynamics, 1, 283-291, 1973.
Bathe, K.J. and Wilson E.L., Numerical Methods in Finite Element Analysis, Prentice Hall,
Englewood Cliffs, N.J., 1976.
Bathe K.J. and Cimento A.P., "Some Practical Procedures for the Solution of Nonlinear Finite
Element Equations", Computer Methods in Applied Mechanics and Engineering, 22, 59-85,
1980.
Beléndez A., Pascual C., Méndez D.I., Beléndez T. and Neipp C., “Exact solution for the nonlinear
pendulum” Revista Brasileira de Ensino de Física, 29(4), 645-648, 2007.
Berry M. P., and Eberhard M. O. “Performance Modeling Strategies for Modern Reinforced
Concrete Bridge Columns” PEER 2007/07, April 2007.
Boore D.M., Atkinson G.M., “ Ground motion prediction equations for the average horizontal
component of PGA, PGV and 5% damped PSA at spectral periods between 0.01s and 10.0s,”
Earthquake Spectra, 24, 1, pp. 99-138, 2007.
Boore, D. M., Jennie, Watson-Lamprey, and Norman, A. Abrahamson. "Orientation-independent
measures of ground motion." Bulletin of the Seismological Society of America 96.4A: 1502-
1511, 2006.
Boulanger R.W., Curras C. J., Kutter B.L, Wilson D.W., Abghari A., "Seismic soil-pile-structure
interaction experiments and analysis," Journal of Geotechnical and Geoenvironmental
Engineering, 125(9): 750-759, 1999.
Bozorgzadeh A, Megally S, Restrepo J, and Ashford S A “Capacity evaluation of exterior
sacrificial shear keys of bridge abutments,” Bridge Engineering 555–565, 2006.
Bozorgzadeh A, Megally S, and Restrepo J “Seismic Response of Sacrificial Exterior Shear Keys
in Bridge Abutments: Recommended Design and Construction Details,” Rep. to CALTRANS,
Contract No. 59A0337, Dept. of Structural Engineering, UC San Diego, San Diego, 2004.
Brown, D. A., O'Neill, M. W., Hoit, M., McVay, M., El Naggar, M. H., & Chakraborty, S. “Static
and dynamic lateral loading of pile groups,” No. Project: E24-9 FY'97, 2001.
Building Seismic Safety Council. “NEHRP Guidelines for the Seismic Rehabilitation of
Buildings,” Report No. FEMA-273, Federal Emergency Management Agency, Washington,
D.C., 1997.
Buckle, I. G. (1994). “The Northridge, California earthquake of January 17, 1994: Performance of
highway bridges.” Technical Report NCEER-94-0008, National Center for Earthquake
Engineering Research, State University of New York, Buffalo, NY.
155
Caltrans SDC (ver. 1.7), Caltrans Seismic Design Criteria, Version 1.7, California Department of
Transportation, Sacramento, 2013.
Caltrans SDC (ver. 1.6), Caltrans Seismic Design Criteria, Version 1.6, California Department of
Transportation, Sacramento, 2010.
Campbell, K.W. and Bozorgnia, Y., “NGA Ground Motion Model for the Geometric Mean
Horizontal Component of PGA, PGV, PGD and 5% Damped Linear Elastic Response Spectra
for Periods Ranging from 0.01 to 10s” Earthquake Spectra, 24, 139–171, 2007.
Chiou, B., Darragh, R., Gregor, N., and Silva, W. (2008). NGA Project Strong-Motion Database,
Earthquake Spectra, 24(1), 23-44.
Choi, E. (2002). “Seismic analysis and retrofit of mid-America bridges.” Ph.D. Dissertation,
Georgia Institute of Technology, Atlanta, GA.
Chopra, A.K., Dynamics of Structures: Theory and Applications to Earthquake Engineering,
Pearson Prentice Hall, 3rd Edition, Upper Saddle River, NJ, 2006.
Chung, J., Hulbert, G.M., “A Time Integration Algorithm for Structural Dynamics with Improved
Numerical Dissipation: The Generalized-alpha method” Journal of Applied Mechanics, 60,
371-375, 1993.
Coleman, J., and Spacone, E. “Localization issues in force-based frame elements.” Journal of
Structural Engineering, 12711, 1257–1265, 2001.
Combescure, D. and Pegon, P., “-Operator Splitting Time Integration Technique for Pseudo-
Dynamic Testing Error Propagation Analysis” Soil Dynamics Earthquake Engineering, 16,
427–443, 1997.
Comerio M.C. Editor, PEER Testbed Study on a Laboratory Building: Exercising Seismic
Performance Assessment. Report No: PEER 2005/12, Pacific Earthquake Engineering
Research Center, PEER, 2005.
Crisfield, M.A., Non-linear Finite Element Analysis of Solids and Structures, Volume 1: Essentials,
John Wiley & Sons Ltd., Baffins Lane, Chichester, 1991.
DesRoches, R., and Fenves, G.L. “Simplified Restrainer Design Procedure for Multiple-Frame
Bridges,” Earthquake Spectra, 17(4), 551-567, 2001.
DIANA, User’s Manual: Analysis Procedures, de Witte, F.C. and Kikstra, W.P. (Editors), TNO
DIANA, 2005.
Earth Mechanics, INC., Filed Investigation Report for Abutment Backfill Characterization, Report
No. SSRP-05/02, Final Report Submitted to the California Department of Transportation
(Caltrans), under Contract No. 59A0287.
Duncan, J.M., and Mokwa, R.L. “Passive earth pressures: theories and tests.” Journal of
Geotechnical and Geoenvironmental Engineering, 127(3), 248-257, 2001.
Earth Mechanics, INC. (2005). “Field investigation report for abutment backfill characterization.”
Report No. SSRP-05/02, University of California, San Diego, CA.
EERI—Earthquake Engineering Research Institute “Northridge Earthquake Reconnaissance
Report.” Earthquake Spectra, Special Suppl. to vol. 11, 523 pp., 1995.
156
Electric power research institute (EPRI), “guidelines for determining design basis ground motion,”
final rep. No. TR-102293, Palo Alto, CA, 1993.
Elgamal, A., Yan, L., Yang, Z., and Conte, J.P. “Three-dimensional seismic response of Humboldt
Bay bridge-foundation-ground system.” ASCE Journal of Structural Engineering, 134(7),
1165–1176, 2007.
FEMA. (2005). HAZUS-MH software, Washington, DC.
Fell, B. V., and Salveson, M. W. “Parametric Simulation of Nonlinear Shear Key Behavior on the
seismic response on Bridge Abutments,” Bridge Engineering, January 5, doi: 10.1061/ (ASCE)
BE.1943-5592.0000465, 2013.
Gazetas, G., Dobry, R., "Simple radiation damping model for piles and footings," Journal of
Engineering Mechanics, 110(6): 937-956, 1984.
Goel R. K., and Chopra A. K. “Role of Shear Keys in Seismic Behavior of Bridges Crossing Fault-
Rupture Zones,” Journal of Bridge Engineering 13.4 398-408, 2007.
Golub, G.H., and Van Loan, C.F., Matrix computations, Johns Hopkins University Press, 3rd
Edition, Baltimore, Md.
Goulet C., Haselton C.B., Mitrani-Reiser J., Deierlein G.G., Stewart J.P., Taciroglu E. Evaluation
of the seismic performance of a code-conforming reinforced-concrete frame building – part I:
ground motion selection and structural collapse simulation. 8th National Conference on
Earthquake Engineering (100th Anniversary Earthquake Conference), 2006; San Francisco,
CA.
Günay S., Mosalam K.M. PEER performance-based earthquake engineering methodology,
revisited. Journal of Earthquake Engineering, 17(6): 829-858, 2013.
Haselton, C.B., Baker, J.W., Bozorgnia, Y., Goulet, C.A., Kalkan, E., Luco, N., Shantz, T.J.,
Shome, N., Stewart, J.P., Tothong, P., Watson-Lamprey, J.A. and Zareian, F., “Evaluation of
Ground Motion Selection and Modification Methods: Predicting Median Interstory Drift
Response of Buildings” Technical Report 2009/01, Pacific Earthquake Engineering Research
Center, University of California, Berkeley, CA, 2009.
Hashash, Y. M. A “DEEPSOIL V 3.7, Tutorial and User Manual” University of Illinois at Urbana-
Champaign, Urbana, Illinois, 2009.
Hashash, Y. M. A., and Park, D. “Non-linear one dimensional seismic ground motion propagation
in the Mississippi embayment,” Engineering Geology, 62(1-3), 185-206, 2001.
Hashash, Y.M.A, Groholski, D.R., Phillips, C. A., Park, D, Musgrove, M., DEEPSOIL5.1, User
Manual and Tutorial. 107 pp., 1991.
Hayden, C., Bray, J., Abrahamson, N. and Acevedo-Cabrera, A.L., “Selection of Near-Fault Pulse
Motions for Use in Design” 15th International World Conference on Earthquake Engineering,
Lisboa, 2012.
Hilbert, H.M., Hughes, T.J.R. and Taylor, R.L., “Improved Numerical Dissipation for Time
Integration Algorithms in Structural Dynamics” Earthquake Engineering Structural Dynamics,
5, 283-292, 1977.
157
Hube M. A. and Mosalam K.M. “Experimental and Computational Evaluation of Current and
Innovative In-Span Hinge Details in Reinforced Concrete Box-Girder Bridges, Part 2: Post-
Test Analysis and Design Recommendations” PEER 2009/107, University of California,
Berkeley, 2009.
Hube M.A. and Mosalam K.M. “Experimental and Computational Evaluation of Current and
Innovative In-Span Hinge Details in Reinforced Concrete Box-Girder Bridges, Part 1:
Experimental Findings and Pre-Test Analysis” PEER 2008/103, University of California,
Berkeley, 2007.
Hughes, T.J.R., Pister, K.S. and Taylor, R.L., “Implicit-Explicit Finite Elements in Nonlinear
Transient Analysis,” Computer Methods in Applied Mechanics and Engineering, 17/18, 159-
182, 1979.
Huo, Y. (2011). “Seismic response Assessment and improvement of highway bridges using
fragility function method.” Ph.D. Dissertation, University of California, Los Angeles, CA.
Huo, Y., and Zhang, J. “Effects of pounding and skewness on seismic responses of typical
multispan highway bridges using the fragility function method.” ASCE Journal of Bridge
Engineering, 18(6), 499–515, 2012.
Hutchinson T.C., Chai Y.H., and Boulanger R.W. “Inelastic Seismic Response of Extended Pile-
shaft-supported Bridge Structures,” Earthquake Spectra, 20(4): 1057-1080, 2004.
Idriss, I.M., and Sun, J.I. "SHAKE91: A computer program for conducting equivalent linear
seismic response analyses of horizontally layered soil deposits." Department of Civil and
Environmental Engineering, University of California Davis, 1992.
Jayaram, N. and Baker, J.W. “Statistical tests of the joint distribution of spectral acceleration
values” Bulletin of the Seismological Society of America, 98, 2231-2243, 2007.
Jayaram, N., Lin, T. and Baker, J.W. “A Computationally Efficient Ground-Motion Selection
Algorithm for Matching a Target Response Spectrum Mean and Variance” Earthquake
Spectra, 27(3), 797-815, 2011.
Katsanos, E.I., Sextos, A.G. and Manolis, G.D., “Selection of Earthquake Ground Motion Records:
A State-of-the-Art Review from a Structural Engineering Perspective” Soil Dynamics and
Earthquake Engineering, 30: 157-169, 2010.
Kaviani, P., Performance-Based Seismic Assessment of Skewed Bridges, Ph.D. Dissertation,
University of California, Irvine, 2011.
Kaviani, P., Zareian, F. and Taciroglu, E., “Seismic Behavior of Reinforced Concrete Bridges with
Skew-Angled Seat-Type Abutments” Engineering Structures, 45, 137–150, 2012.
Kaviani, P., Zareian, F., and Taciroglu E. “Performance-based seismic assessment of skewed
bridges,” PEER Report 2014/1, Pacific Earthquake Engineering Research Center, University
of California, Berkeley, CA, 2014.
Kawashima, K., Unjoh, S., Hoshikuma, J.I., and Kosa, K. (2011). “Damage of bridges due to the
2010 Maule, Chile, earthquake.” Journal of Earthquake Engineering, 15(7), 1036–1067.
Khalili-Tehrani P., Taciroglu E., Shamsabadi A. “Backbone curves for passive lateral response of
walls with homogeneous backfills,” Proc. of 2009 Soil-Foundation-Structure Interaction
158
Workshop (Orense, Chouw, and Pender, eds.) University of Auckland, New Zealand, 2, 149-
154, 2010.
Kondner, R.L., and Zelasko, J.S. "Hyperbolic stress-strain formulation of sands." Second pan
American Conference on Soil Mechanics and Foundation Engineering, Sao Paulo, Brazil, 289-
324, 1963.
Kornkasem, W, “Seismic Behavior of Pile-Supported Bridges,” Ph.D. Dissertation, University of
Illinois at Urbana-Champaign, 2001.
Krawinkler H. A general approach to seismic performance assessment. Proceedings of
International Conference on Advances and New Challenges in Earthquake Engineering
Research, ICANCEER. 2002; Hong Kong.
Krawinkler H. Editor, Van Nuys Hotel Building Testbed Report: Exercising Seismic Performance
Assessment. Report No: PEER 2005/11, Pacific Earthquake Engineering Research Center,
PEER, 2005.
Krawinkler H., Miranda E. Performance-based Earthquake Engineering. Chapter 9 of Earthquake
Engineering: from Engineering Seismology to Performance-based Engineering. Bozorgnia, Y.
and Bertero, V.V., Editors, CRC Press, 2004.
Kunnath S.K. Application of the PEER PBEE Methodology to the I-880 Viaduct. Report No:
2006/10, Pacific Earthquake Engineering Research Center, PEER, 2006.
Kwok, A.O.L., Stewart, J. P., Hashash, Y.M.A., Matasovic, N., Pyke, R., Wang, Z., and Yang, Z.
"Use of exact solutions of wave propagation problems to guide implementation of nonlinear
time-domain ground response analysis routines." ASCE Journal of Geotechnical and
Geoenvironmental Engineering, 133(11), 1337-1481, 2007.
LaFave J., Fahnestock L., Foutch D., Steelman J., Revell J., Filipov E., and Hajjar J. “Experimental
Investigation of the Seismic Response of Bridge Bearings” A report of the findings of ICT-
R27-070, Calibration and Refinement of Illinois’ Earthquake Resisting System Bridge Design
Methodology, Illinois Center for Transportation, Research Report No. FHWA-ICT-13-002,
University of Illinois at Urbana-Champaign and Northeastern University, 2013.
Lee T.H., Mosalam KM. Seismic demand sensitivity of reinforced concrete shear-wall building
using FOSM method. Earthquake Engineering and Structural Dynamics 2005; 34(14): 1719-
1736.
Lee T.H., Mosalam KM. Probabilistic Seismic Evaluation of Reinforced Concrete Structural
Components and Systems. Report No: PEER 2006/04, Pacific Earthquake Engineering
Research Center, PEER, 2006.
Lee, H. and Mosalam K.M., “Seismic Evaluation of the Shear Behavior in Reinforced Concrete
Bridge Columns Including Effect of Vertical Accelerations,” Earthquake Engineering and
Structural Dynamics, Vol. 43, No. 3, pp. 317-337, 2014.
Lee, H., Kumar P., Günay M.S., Mosalam K.M., Kunnath S.K., “Effect of Vertical Ground Motion
on Shear Demand and Capacity in Bridge Columns,” Caltrans Technical Report, March 2012.
Liang, X. and Mosalam, K. M. (2015). Lyapunov Stability and Accuracy of Direct Integration
Algorithms in Nonlinear Dynamic Problems and Considering the Strictly Positive Real
Lemma, SEMM Technical Report UCB/SEMM-2015/01, April.
159
Liang, X. and Mosalam, K. M. (2016a). “Lyapunov Stability and Accuracy of Direct Integration
Algorithms Applied to Nonlinear Dynamic Problems,” Journal Engineering Mechanics,
142(5), 0401622.
Liang, X. and Mosalam, K. M. (2016b). “Lyapunov Stability Analysis of Explicit Direct
Integration Algorithms Considering Strictly Positive Real Lemma,” Journal Engineering
Mechanics, 142(10), 04016079.
Liang, X. and Mosalam, K. M. (2016c). “Lyapunov Stability Analysis of Explicit Direct
Integration Algorithms Applied to Multi-Degree-of-Freedom Nonlinear Dynamic Problems,”
Journal Engineering Mechanics, 1061/(ASCE)EM.1943-7889.0001162, 04016098.
Liang, X., Günay, S. and Mosalam, K. M. (2014). “Integrators for Nonlinear Response History
Analysis: Revisited,” Proc., Istanbul Bridge Conf., Istanbul, Turkey.
Liang, X., Mosalam, K. M. and Günay, S. (2016a). “Direct Integration Algorithms for Efficient
Nonlinear Seismic Response of Reinforced Concrete Highway Bridges,” Journal of Bridge
Engineering, 21(7), 04016041.
Liang, X., Günay, S. and Mosalam, K.M. (2016b). Chapter 12: Seismic Response of Bridges
Considering Different Ground Motion Selection Methods, in Developments in International
Bridge Engineering, Springer Tracts on Transportation and Traffic 9, Springer Int. Publishing,
Switzerland.
Marcakis, K., and Mitchell, D. “Precast concrete connections with embedded steel members.” PCI
J., 254, 88–116. Precast/Prestressed Concrete Institute PCI, 1980.
Marsh, A. K., Rollins, K. M., Franke, B., Smith, J., and Palmer, K. “Passive force-deflection
behavior for 0° and 30° skewed abutments.” Transportation Research Record: Journal of the
Transportation Research Board, 2363(1), 12–20, 2013.
Martin, G. R., Yan, L. (1995). “Modeling passive earth pressure for bridge abutments.” ASCE
Geotech. Special Publ., 55, Earthquake-induced Movements and Seismic Remediation of
Existing Foundations and Abutments, ASCE 1995 Annual National Convention, San Diego,
October, 1995.
Mattock, A. H.,” Shear transfer in concrete having reinforcement at an angle to the shear plane,
Shear in Reinforced Concrete,” ACI Special Publication 42, 17–42, 1974.
McKenna, F., Fenves, G. L., and Scott, M. H. “Open System for Earthquake Engineering
Simulation University of California Berkeley”, California, 2000.
Mackie, K. R., Lu, J., & Elgamal, A. “Performance-based earthquake assessment of bridge systems
including ground-foundation interaction.” Soil Dynamics and Earthquake Engineering, 42,
184-196, 2012.
Matasovic, N., and Vucetic, M. "Cyclic Characterization of Liquefiable Sands." ASCE Journal of
Geotechnical and Geoenvironmental Engineering, 119(11), 1805-1822, 1993.
Megally, S., Silva, P. F., and Seible, F. “Seismic Response of Sacrificial Shear Keys in Bridge
Abutments,” Report No. SSRP–2001/23, University of California, San Diego, CA, 2001.
Mitrani-Reiser J., Haselton C.B., Goulet C., Porter K.A., Beck J., Deierlein G.G. Evaluation of the
seismic performance of a code-conforming reinforced-concrete frame building - part II: loss
160
estimation 8th National Conference on Earthquake Engineering (100th Anniversary
Earthquake Conference). 2006; San Francisco, CA.
Moehle, J., Fenves, G., Mayes, R., Priestley, M. J. N., Seible, F., Uang, C. M., Warner, S., and
Aschheim, M. “Northridge earthquake of January 17, 1994.” reconnaissance report,
Earthquake Spectra, Chapter 6, vol. 1, 1995.
Moehle J.P. “A framework for performance-based earthquake engineering.” Proceedings of ATC-
15-9 Workshop on the Improvement of Building Structural Design and Construction Practices,
2003; Maui, HI.
Moehle J.P., Deierlein G.G. A framework for performance-based earthquake engineering.
Proceedings of 13th World Conference on Earthquake Engineering. Paper No 679. 2004;
Vancouver, Canada.
Mosalam, K.M., Liang X., Günay M.S. and Schellenberg A., "Alternative Integrators and Parallel
Computing for Efficient Nonlinear Response History Analyses," COMPDYN 2013, 4th
ECCOMAS Thematic Conference on Computational Methods in Structural Dynamics and
Earthquake Engineering, Kos Island, 2013.
Muthukumar S. “Contact Element Approach with Hysteresis Damping For the Analysis and
Design of Pounding in Bridges,” Ph.D. dissertation, Georgia Institute of Technology,
November, 2003.
Naeim, F., and Kelly, J.M., “Design of Seismic Isolated Structures: From Theory to Practice,”
John Wiley & Sons, Inc., New York, 1999.
Newmark, N.M., “A Method of Computation for Structural Dynamics” ASCE Journal of the
Engineering Mechanics Division, 85(EM3), 67-94, 1959.
Nielson, B.G. “Analytical fragility curves for highway bridges in moderate seismic zones.” Ph.D.
Dissertation, Georgina Institute of Technology, Atlanta, GA, 2005.
Nielson, B. G., and DesRoches, R. “Influence of modeling assumptions on the seismic response
of multi-span simply supported steel girder bridges in moderate seismic zones.” Engineering
Structures, 28(8), 1083–1092, 2006.
Ousterhout, J. Tcl and the Tk Toolkit. Addison-Wesley, 1994.
Park, D., and Hashash, Y.M.A. "Rate-dependent soil behavior in seismic site response analysis."
Canadian Geotechnical Journal, 45(4), 454-446, 2007.
Priestley, M.J.N., Seible, F., and Uang, C.M. “The Northridge earthquake of January 17, 1994:
Damage analysis of selected freeway bridges,” Structural Systems Research Rep. SSRP-94/06,
Dept. of Structural Engineering, Univ. of California San Diego, La Jolla, California, 1994.
Porter K.A. An overview of PEER’s Performance-based earthquake engineering methodology.
Conference on Applications of Statistics and Probability in Civil Engineering (ICASP9). Civil
Engineering Risk and Reliability Association (CERRA), 2003; San Francisco, CA.
Rahmani, A., Taiebat, M., and Liam Finn, W.D. “Nonlinear dynamic analysis of Meloland Road
Overpass using three-dimensional continuum modeling approach.” Soil Dynamics and
Earthquake Engineering, 57, 121–132, 2014.
161
Rollins, K. M., and Jessee, S.J. “Passive force-deflection curves for skewed abutments.” ASCE
Journal of Bridge Engineering, 18(10), 1086–1094, 2012.
Rayleigh, J. W. S., and Lindsay, R. B. “The theory of sound, “ Dover Publications, New York,
1945.
Rezaeian, S. and Der Kiureghian, A., “Simulation of synthetic ground motions for specified
earthquake and site characteristics” Earthquake Engineering and Structural Dynamics, 39:
1155-1180, 2010a.
Rezaeian, S. and Der Kiureghian, A., Stochastic Modeling and Simulation of Ground Motions for
Performance-Based Earthquake Engineering. Report No: 2010/06, Pacific Earthquake
Engineering Research Center, PEER, 2010b.
Rezaeian, S. and Der Kiureghian, A., “Simulation of Orthogonal Horizontal Ground Motion
Components for Specified Earthquake and Site Characteristics” Earthquake Engineering and
Structural Dynamics, 41(2): 335-353, 2011.
Richards P.W., Rollins K.M., and Stenlund T.E. “Experimental Testing of Pile-to-Cap
Connections for Embedded Pipe Piles” Journal of Bridge Engineering, Vol. 16, No. 2, March
1, 2011.
Roeder, C.W., Stanton, J.F, “State-of-the-Art Elastomeric Bridge Bearing Design,” ACI Structural
Journal, 88(1), 31-41, 1991.
Roeder C.W., Stanton C.F., Taylor A.W. “Performance of Elastomeric Bearings” National
Cooperative Highway Research Program Report 298, Transportation Research Board,
National Research Council, Washington D.C., 1987.
Rollins, K.M., Olsen, R.J., Egbert, J.J., Jensen, D.H., Olsen, K.G., and Garrett, B.H. “Pile spacing
effects on lateral pile group behavior: load tests.” Journal of Geotechnical and
Geoenvironmental engineering, 132(10), 1262-1271, 2006a.
Rollins, K. M., Olsen, K. G., Jensen, D. H., Garrett, B. H., Olsen, R. J., & Egbert, J. J “Pile spacing
effects on lateral pile group behavior: Analysis.” Journal of Geotechnical and
Geoenvironmental Engineering, 132(10), 1272-1283, 2006b.
Saiidi, M., Arede, A., Cardone, D., Delgado, P., Dolce, M., Fischinger, M., Isakovic, T.,
Pantazopoulou , S., Pekcan, G., Pinho, R., and Sextos, A. “Modelling of Bridges for Inelastic
Analysis” Chapter 2: from Seismic Design and Assessment of Bridges, A. J. Kappos , M. S.
Saiidi, M. N. Aydınoglu, and T. Isakovic´, Editors, Springer Science+Business Media
Dordrecht, ISBN 978-94-007-3942-0, 2012.
Saiidi, M., Maragakis, E., Abdel-Ghaffar, S., Feng, S., and O’Connor, D. “Response of bridge
hinge restrainers during earthquakes-field performance, analysis, and design” Report No.
CCEER 93/06, Center for Civil Engineering Earthquake Research, University of Nevada, Reno,
1993.
Seed, H. B., and Idriss, I. M. “Influence of soil conditions on ground motions during earthquakes.”
ASCE J. Soil Mech. Found Div., 95, 99-137, 1969.
Scalzi, J., and McGrath, W. “Mechanical Properties of Structural Cables,” ASCE Journal of
Structural Engineering, 97(12), 2837-2844, 1971.
162
Schnabel, P. B., Lysmer, J. L., and Seed, H. B. "SHAKE: A computer program for earthquake
response analysis of horizontally layered sites." EERC-72/12, Earthquake Engineering
Research Center, Berkeley, CA, 1972.
Schellenberg, A.H., Mahin, S.A. and Fenves, G.L., “Advanced Implementation of Hybrid
Simulation” PEER Technical Report 2009/104, 2009.
Scott, M.H. "Evaluation of Force-Based Frame Element Response Sensitivity Formulations."
ASCE Journal of Structural Engineering, 138(1): 72-80, 2012.
Scott, M.H. and Fenves, G.L., “Krylov Subspace Accelerated Newton Algorithm: Application to
Dynamic Progressive Collapse Simulation of Frames” ASCE Journal of Structural
Engineering, 136:473-480, 2010.
Scott, M.H. and G.L. Fenves, "Plastic Hinge Integration Methods for Force-Based Beam-Column
Elements," ASCE Journal of Structural Engineering, 132(2): 244-252, 2006.
SEAOC Vision 2000 Committee. “Performance-based Seismic Engineering,” Structural Engineers
Association of California, Sacramento, CA, 1995.
Shahi, S. K., and Baker, J. W. “Regression models for predicting the probability of near-fault
earthquake ground motion pulses, and their period,” 11th International Conference on
Applications of Statistics and Probability in Civil Engineering, Zurich, Switzerland, 8p, 2011.
Shamsabadi, A., Yan, L., and Martin, G.R. “Three dimensional nonlinear seismic soil-foundation-
structure interaction analysis of a skewed bridge considering near fault effects.” Proc., US-
Turkey Soil-Structural Interaction Workshop, Turkey, 2004.
Shamsabadi, A., Khalili-Tehrani P., Stewart, J.P., and Taciroglu, E. “Validated simulation models
for lateral response of bridge abutments with typical backfills,” ASCE Journal of Bridge
Engineering, 15(3), 302-311, 2010.
Shamsabadi, A., Rollins, K.M., and Kapuskar, M. "Nonlinear soil–abutment–bridge structure
interaction for seismic performance-based design," Journal of Geotechnical and
Geoenvironmental Engineering, 133(6): 707-720, 2007.
Shamsabadi, A., Kapuskar, M. “Nonlinear seismic soil-abutment-structure interaction analysis of
skewed bridges.” Proc., 5th National Seismic conference on bridges and highways, San
Francisco, CA, 2006.
Shamsabadi, A., Ashour, M., and Norris, G. “Bridge Abutment Nonlinear Force-Displacement-
Capacity Prediction for Seismic Design,” Geotechnical and Geoenvironmental Engineering,
Vol. 131, No. 2, pp.1-9, 2005.
Shamsabadi, A., Yan, L., and Martin, G. R. “Three dimensional nonlinear seismic soil-foundation-
structure interaction analysis of a skewed bridge considering near fault effects.” Proc., US-
Turkey Soil-Structural Interaction Workshop, Turkey, 2004.
Silva, P. F., Seible, F., and Priestley, M. J. N. “Influence of Standard Development Length on
Formation of Plastic Hinges in Prestressed Piles.” PCI Journal, 46(3), 76-89, 2001.
Stewart, Jonathan P., Ertugrul Taciroglu, John W. Wallace, Eric R. Ahlberg, Anne Lemnitzer,
Changsoon Rha, Payman Tehrani, Steve Keowen, Robert L. Nigbor, and Alberto Salamanca.
163
"Full scale cyclic testing of foundation support systems for highway bridges. Part II: Abutment
backwalls." University of California, Los Angeles, UCLA-SGEL Report 2007/02.
Sugito, M., Goda, H., and Masuda, T. "Frequency dependent equi-linearized technique for seismic
response analysis of multi-layered ground." Doboku Gakkai Rombun-Hokokushu/Proceedings
of the Japan Society of Civil Engineers, 493(3-2), 49-58, 1994.
Taciroglu, E., Rha, C., Wallace, J.W. “A robust macroelement model for soil-pile interaction under
cyclic loads," Journal of Geotechnical and Geoenvironmental Engineering, 132(10): 1304-
1314, 2006.
Terzaghi, K. Theoretical soil mechanics, John Wiley and Sons, Inc., New York, NY, 1943.
Terzaghi, K., Peck, R. B., and Mesri, G. “Soil mechanics in engineering practice.” 3rd Ed., John
Wiley and Sons, Inc., New York, NY, 1996.
The George E. Brown Network for Earthquake Engineering Simulation (NEES), NEEShub:
Cyber-infrastructure of the Network for Earthquake Engineering Simulation, 2013,
http://nees.org.
Trochalakis, P., Eberhard, M. and Stanton, J “Evaluation and Design of Seismic Restrainers for
In-Span Hinges,” Journal of Structural Engineering, ASCE, Vol. 123(No.4), pp. 103 – 113,
1997.
Vucetic, M. and Dobry, R. (1991). Effect of soil plasticity on cyclic response. ASCE Journal of
Geotechnical Engineering, 117(1), 89-117.
Wasef, N.M. “An Analysis of Bridge Abutment Shear Key Behavior Due to Embankment
Modeling and Earthquake Intensity.” Ph.D. Dissertation, Univ. of California, San Diego, 2013.
Wilson E.L., Static & Dynamic Analysis of Structures, Computers and Structures, Inc., 4th Edition,
2004.
Wilson, E.L. and Habibullah, A., SAP2000-Structural Analysis User’s Manual, Computers and
Structures, Inc., 1997.
Xiao Y., Wu H., Yaprak T.T., “Experimental studies on seismic behavior of steel pile-to-pile-cap
connections,” ASCE Journal of Bridge Engineering, 11(2):151-9, 2006.
Yashinsky, M., Oviedo, R., Ashford, S., Fargier-Gbaldon, L., Huob, M. (2010). “Performance of
Highway and Railway Structures during the February 27, 2010 Maule Chile Earthquake.”
http://eqclearinghouse.org/co/20100227-chile/reports-from-the-field/report-from-the-bridge-
team.
Zakeri, B., Padgett, J.E., and Amiri, G.G. “Fragility analysis of skewed single frame concrete box
girder bridges.” ASCE Journal of Performance of Constructed Facilities, 28(3), 571–582, 2013.
Zhang, J. and Makris, N. "Kinematic response functions and dynamic stiffnesses of bridge
embankments," Earthquake Engineering and Structural Dynamics, 31(11): 1933–1966, 2002.
Zhang J, Makris N. “Seismic response analysis of highway overcrossings including soil–structure
interaction,” PEER-2001/02, Pacific Earthquake Engineering Research Center, University of
California, Berkeley, 2001.
164
Zhang, Y., Conte, J.P., Yang, Z., Elgamal, A., Bielak, J., and Acero, G. “Two-dimensional
nonlinear earthquake response analysis of a bridge-foundation-ground system.” Earthquake
Spectra, 24(2), 343–386, 2007.
165
Appendix A Example Tcl File
for {set ik 1} {$ik <= $Nsteps} {incr ik 1} {
set ok [analyze 1 $DtAnalysis];
# Convergence
if {$ok != 0} {
puts "Trying Newton with line search ...";
# change the nonlinear solver to NewtonLineSearch
algorithm NewtonLineSearch;
# do the anlysis with NewtonLineSearch
set ok [analyze 1 $DtAnalysis]
# change back to the initial nonlinear solver
algorithm $algorithmType;
};
if {$ok != 0} {
puts "Trying KrylovNewton ...";
algorithm KrylovNewton;
set ok [analyze 1 $DtAnalysis]
algorithm $algorithmType;
};
if {$ok != 0} {
puts "Trying BFGS ...";
algorithm BFGS;
set ok [analyze 1 $DtAnalysis]
algorithm $algorithmType;
};
if {$ok != 0} {
166
puts "Trying Broyden ...";
algorithm Broyden;
set ok [analyze 1 $DtAnalysis]
algorithm $algorithmType;
};
if {$ok != 0} {
puts "Trying OS ...";
# change the integrator to AlphaOS
integrator AlphaOS 1.00;
# change the solver to Linear
algorithm Linear;
# do the analysis with AlphaOS
set ok [analyze 1 $DtAnalysis]
# change back to the initial integrator implicit Newmark
integrator Newmark $NewmarkGamma $NewmarkBeta;
# change back to the initial nonlinear solver
algorithm $algorithmType;
};
if {$ok != 0} {
puts "Trying BackwardEuler ...";
integrator BackwardEuler;
set ok [analyze 1 $DtAnalysis]
integrator Newmark $NewmarkGamma $NewmarkBeta;
};
# Trying more iterations
if {$ok != 0} {
puts "Trying more iterations...";
# change the number of iterations to 1000
167
test $TestType $Tol 1000 $printFlag;
# do the analysis with 1000 iterations
set ok [analyze 1 $DtAnalysis]
# change back to original number of iterations
test $TestType $Tol $maxNumIter $printFlag;
};
if {$ok != 0} {
puts "Trying tolerance 1.0e-7 ...";
# change the tolerance to 10e-7
test $TestType 1.0e-7 $maxNumIter 0;
# do the analysis with the tolerance 10e-7
set ok [analyze 1 $DtAnalysis]
# change back to the original tolerance
test $TestType $Tol $maxNumIter 0;
};
if {$ok != 0} {
set Nstepsmax [expr $ik-1];
break;
};
};
168