You are on page 1of 14

Implementation of Nonlinear Elements for Seismic Response

Analysis of Bridges
Kevin R. Mackie, F.ASCE1; and Michael H. Scott2

Abstract: General purpose finite-element software tools have put nonlinear analysis within an engineer’s reach for the assessment of bridge
response to seismic loading. Although these tools can capture strength, ductility, and nonlinear material and geometric effects more accurately
than response spectrum or linear methods, the response is extremely sensitive to modeling techniques, even when the same input parameters
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

for bridge geometry and material properties are used in different software packages. The resulting discrepancies in nonlinear response are due
to mathematical formulations of the element response, for example, concentrated or distributed plasticity, and software-dependent implemen-
tation of the formulations and their constitutive models. To illustrate the effects of modeling choices and the ability of two widely used soft-
ware packages, CSiBridge and OpenSees, to reproduce analytical solutions, concentrated and distributed plasticity models were applied to
cantilever bridge columns with simplified steel and concrete constitutive models. Discrepancies in stiffness and strength owing to the location
and length of plastic hinges can be resolved for these simple component models. The modeling strategies were extended to two ordinary stand-
ard bridges designed by Caltrans. Although modal analyses show the bridge models have approximately the same distribution of mass and
stiffness in the two software packages, only nominally consistent results can be achieved when using more realistic constitutive models for
nonlinear static analyses. DOI: 10.1061/(ASCE)SC.1943-5576.0000420. © 2019 American Society of Civil Engineers.
Author keywords: Uniaxial constitutive model; Reinforced concrete; Pushover analysis; Distributed plasticity; Concentrated plasticity.

Introduction response. To obtain an accurate prediction of the nonlinear response


of a bridge structure, it is necessary for the design engineer to under-
Compared with equivalent linear approaches to seismic design, sim- stand nonlinear analysis concepts and to have guidance on parame-
ulating the nonlinear response of a structural system gives more ter selections and the consequences of these choices.
accurate estimates of stresses, strains, deformations, forces, and dis- After the ATC-32 project (ATC 1996), numerous nonlinear
placements of critical components. Nonlinear static and dynamic analysis studies were conducted on bridges. Fenves and Ellery
analyses are able to incorporate realistic material behavior beyond (1998) investigated the failure mechanism of a bridge during the
the elastic limit, loss of stiffness due to nonlinear geometric (P-D) Northridge earthquake and made general modeling and analysis
effects, and contact nonlinearity. These analysis approaches have recommendations for highway bridges. Mackie and Stojadinovic
been widely studied for highway bridges in the framework of (2003) presented probabilistic seismic demand models generated
performance-based earthquake engineering (Cornell and Krawinkler for typical RC bridges in California that were parameterized in
2000; Kunnath et al. 2007; Mackie and Stojadinovic 2005). terms of common geometric and reinforcing parameters. Similar
Nonlinear analysis results can then be utilized to design and detail studies were conducted on bridges in the southeastern and central
bridge subsystems and to evaluate the global capacity and ductility United States by Nielson and DesRoches (2007). Hajihashemi et al.
of the system. In most cases, the additional effort of nonlinear analy- (2017) studied the nonlinear static response of common bridge
sis leads to cost savings compared with more conservative design types in the context of AASHTO design procedures. The bridge
approaches based on linear analysis. response was sensitive to the pushover analysis method, boundary
Although computing resources enable a large number of analy- conditions, and software used for the analysis.
ses to be carried out quickly, modeling a nonlinear bridge response Several recent studies have focused on best modeling practices
to seismic loading carries a large amount of uncertainty. Sources of and sensitivity to modeling choices. Aviram et al. (2008) generated
this uncertainty can be traced to modeling assumptions and theory, finite-element spine models of six typical California RC box girder
the need for several parameters to describe the nonlinear behavior bridges to develop practical modeling recommendations for nonlin-
of structural components, and the numerical formulations and ear analysis using SAP2000 and OpenSees. Omrani et al. (2015)
software implementation of the FEMs that simulate the bridge extended the study to abutments, expansion joints, and soil–structure
interaction (SSI) issues. The primary refinements were in the nonlin-
1
Professor, Dept. of Civil and Environmental Engineering, Univ. of ear backfill spring models of the abutments, SSI in pile shafts, shear
Central Florida, Orlando, FL 32816-2450. ORCID: https://orcid.org/0000 key models, and in-span hinge models. The study, which included
-0003-1287-6520. Email: kmackie@mail.ucf.edu several different skew angles, highlighted the importance of abut-
2
Professor, School of Civil and Construction Engineering, Oregon ment boundary properties on quantifying bridge responses.
State Univ., Corvallis, OR 97331 (corresponding author). ORCID: https:// The current state of analysis encompasses a wide range of mod-
orcid.org/0000-0001-5898-5090. Email: michael.scott@oregonstate.edu
eling approaches, element and constitutive model formulations, and
Note. This manuscript was submitted on October 15, 2018; approved
on December 14, 2018; published online on April 24, 2019. Discussion pe- solution strategies; however, significant knowledge gaps remain.
riod open until September 24, 2019; separate discussions must be submit- Although the differences in simulated results associated with the
ted for individual papers. This paper is part of the Practice Periodical on choice of nonlinear constitutive or element models and errors with
Structural Design and Construction, © ASCE, ISSN 1084-0680. respect to known benchmark solutions have been addressed,

© ASCE 04019011-1 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


differences in simulated response when using nominally identical
bridge models in separate software packages have not been enumer-
ated. This paper addresses these differences through nonlinear static
simulations of cantilever columns and two ordinary standard
bridges (OSBs) with common finite-element formulations of plas-
ticity. To focus the nonlinear response mechanisms on the column,
simple boundary conditions at the abutments and linear geometry
were assumed. The numerical models presented herein are based on
CSiBridge (version 17.3.0) and OpenSees (version 2.6.4).

Benchmark Columns

Bents are the primary lateral load carrying system for multi-span
bridges. Accordingly, the effect of nonlinear column models on the
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

seismic response of bridges is a common theme in much of the afore-


mentioned research. For these column models, one-dimensional (1D) (a)
or macro FEMs are preferred over solid models for computational ef-
ficiency. These elements fall in to two categories: distributed plastic- Concrete section
ity (DP) and concentrated plasticity (CP).
In the DP approach, material yielding can initiate and spread
anywhere along the element based on either a displacement- or
10 -3
force-based finite-element formulation. Mixed formulations for DP
have also been developed (Alemdar and White 2005) but have not -5 5
been widely used to model bridge responses. In the displacement-
based formulation, the spread of plasticity is captured through mesh
refinement, that is, through using many elements along the column
length. On the other hand, for force-based elements, the spread of
plasticity is captured with additional integration points along the Steel section
element length (Neuenhofer and Filippou 1997). With the force-
based formulation it is also possible to limit the spread of plasticity
over analyst-specified plastic hinge lengths at the element ends
(Scott and Fenves 2006; Addessi and Ciampi 2007).
The CP approach represents plasticity using zero-length springs -0.02 -0.01 0.01 0.02
within a defined plastic hinge region, usually at the ends of the ele-
ment where the flexural demands are largest in the absence of mem-
ber loads. The CP elements are formed by placing the nonlinear
zero-length springs in series with an elastic frame element (Giberson (b)
1967). For an explicit consideration of the plastic hinge length, rigid
elements can be added to the series configuration with the zero- Fig. 1. Benchmark models of 3D cantilever steel and RC columns:
length spring located anywhere in the rigid zone. Macro-element (a) cantilever; and (b) sections and materials.
formulations of CP are typically benchmarked against analytical sol-
utions for DP, and calibrations of CP models are required in order to
achieve the desired element response (Scott and Ryan 2013). Nsteel = 4,450 kN (1,000 kip) and NRC = 600 kN (135 kip) were
applied to the steel and RC columns, respectively, along with
Methods and Models monotonically increasing lateral loads, l y and l z, of equal
magnitude.
Steel and RC columns with circular a cross section highlight the dif- The reinforcing steel was modeled using an EPP material with
ferences between these formulations and their implementations in yield stress Fy = 345 MPa (50 ksi) and elastic modulus Es =
CSiBridge and OpenSees. The columns and assumed material 200 GPa (29,000 ksi), as given in Fig. 1(b). An ENT material was
behaviors of concrete and steel are presented in Fig. 1. Simple con- used to model the normal-weight 27 MPa (4.0 ksi) concrete with
stitutive models—elastic-perfectly-plastic steel (EPP) and elastic- elastic modulus Ec = 24.8 GPa (3,605 ksi) ACI (2014). Although
no-tension (ENT) concrete—are used so that analytical solutions these uniaxial material models are not realistic for an RC section,
for the column response can be compared with formulations avail-
they ensure that both the CSiBridge and OpenSees sectional behav-
able in both software packages.
iors are identical and also make it possible to obtain analytical solu-
The coordinate axes for the 3D model are given in Fig. 1(a). All
tions for the section and element responses. The yield stress for the
column response quantities presented herein are the square-root-
steel cross section was also Fy = 345 MPa (50 ksi), but with 2%
sum-of-squares (SRSS) of the y and z responses. This bidirectional
strain hardening to illustrate the effect of different CP choices on
loading in a 3D model precludes the use of hinge types available in
the post-yield behavior.
commercial software that consider only an uncoupled unidirectional
response. The RC column diameter is D = 51 cm (20 in.) with eight
#8 longitudinal reinforcing steel bars and 2.5-cm (1-in.) clear cover. Sectional Analysis
The steel column has a homogeneous cross section, also with The cross sections of the steel and RC columns were discretized
D = 51 cm (20-in.) diameter. Constant compressive axial loads of with fibers. The RC section was discretized into 50 layers in the

© ASCE 04019011-2 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


tangential direction and 30 layers in the radial direction for the core D = 51 cm (20 in.). For comparison, the Caltrans seismic design cri-
and 50 layers in the tangential direction and 5 layers in the radial teria (SDC) (Caltrans 2013) plastic hinge length equation yielded
direction for the cover (although core and cover concrete properties an estimate of lph = 67.8 cm (26.7 in.). For the steel column, lph =
remained the same). Berry and Eberhard (2007) described other dis- D/2 = 25 cm (10 in.) was assumed.
cretizations of circular RC sections that yielded the same results There are three important CP modeling choices. The first choice
with more optimal placement of the fibers in the core. The steel is the length of the column that contributes to the elastic deforma-
cross section was discretized into 50 layers in the tangential direc- tions. In CP1, the full height H is assumed, which leads to an over-
tion and 40 layers in the radial direction. A relatively large number prediction of the elastic displacement because of the contribution of
of fibers was used in order to capture the initial stiffness of the col- the hinge. This is corrected in CP2 by using a rigid element (EI =
umn (Kostic and Filippou 2012). 1) over length lp. In the numerical implementations, it is not neces-
Sectional analysis was performed for the RC cross section under sary for the EI multiplier to be infinite, similar results can be
NRC axial load level to confirm the consistency of the constitutive obtained with values between 3 and 10. An alternative correction is
modeling. The closed-form moment–curvature response is pre- to make the hinge rigid before yield, which is CP5, although this
sented together with the numerical responses obtained from prevents the use of a fiber-discretized section. Finally, another
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

CSiBridge and OpenSees in Fig. 2. The transition from elastic to approach (CP6) to correcting the elastic deformation is to modify
plastic flexural responses of the section was due to the yielding of the properties of the elastic element but retain the hinge at the base.
the reinforcing steel bars as the neutral axis shifted. The moment– This approach is common in modeling the seismic response of
curvature responses for the steel section under Nsteel axial load level buildings (Zareian and Krawinkler 2009).
were also in agreement and are shown in Mackie et al. (2017). With In the second CP modeling choice (CP2), the location of the inte-
agreement of the section response established for the two software gration point that corresponds to the hinge is at the base of the col-
packages, attention now turned to the element response. umn (xh = 0). The implication is that the yield moment and therefore
the yield load are correct; however, the initial stiffness and post-
Element Formulations yielding stiffness are not correct. An appropriate correction can be
Using simplified constitutive models for the section response, it is made (i.e., that the location of the integration point is at some point,
possible to obtain objective analytical solutions for the element xh, above the base), with two different cases presented. CP3 locates
response according to either the DP or CP approach. Because plas- the hinge at the midpoint of lp, as assumed by Aviram et al. (2008),
ticity is allowed to spread along the entire element, obtaining an an- whereas CP4 locates the hinge at the top of the plastic hinge.
alytical solution for the DP approach is a straightforward applica- The third CP modeling choice (CP3) relates to the elastic ele-
tion of the principle of virtual forces (PVF) in conjunction with the ment properties, which are not uniquely defined due to axial–
moment–curvature response presented in Fig. 2. It was assumed moment interaction and shifting of the neutral axis in RC. As the
that shear deformations were negligible, that is, the section shear load level increases, the moment of inertia decreases from the gross
area was assumed to be large in the software implementations. moment of inertia, Ig, to the fully cracked moment of inertia, Icr. For
The CP approach has many variants, each of which can be repre- consistency of the CP1, CP5, and CP6 solutions, the properties for
sented by the schematic given in Fig. 3, where an elastic element is
in series with a hinge (at xh from the base) and, optionally, a rigid DP CP
element. The physical length of the plastic hinge is designated as lp,
whereas the analytical (or empirical) plastic hinge length is desig-
nated as lph. Variants of the CP approach are summarized in
Table 1. Regardless of the hinge properties, an analytical solution
for each CP variant can be obtained using the PVF where the curva-
ture is assumed constant over lph. The plastic hinge length for the
RC column was assumed to be equal to the column diameter, lph =
Inelastic Elastic
500 element element

400 Rigid
SRSS of moment (kN-m)

element
Hinge
300

Fig. 3. Element schematic for DP and CP configurations.


200

Table 1. Concentrated plasticity approaches used in benchmark columns


100 Exact
OpenSees Case lp xh Ieff =Ig Notes CSi Hinge
CSiBridge
0 CP1 0 0 0.987 Uncorrected Fiber
0 0.02 0.04 0.06 0.08 0.1 0.12 CP2 lph 0 0.6 Hinge at base Fiber
SRSS of curvature (1/m) CP3 lph lph =2 0.6 Hinge at midpoint Fiber
CP4 lph lph 0.6 Hinge at lph Fiber
Fig. 2. Analytical and numerical moment–curvature responses for the CP5 0 0 0.987 Rigid-plastic Interaction
RC cross section. CP6 0 0 0.987 Modified elastic portion Fiber

© ASCE 04019011-3 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


the elastic element were taken from the CSiBridge model defaults are progressively overestimated. The optimal location of the hinge for
as the modulus of elasticity of concrete [24.8 GPa (3,605 ksi)] and purposes of initial stiffness therefore, by definition, will lead to an
moment of inertia I = 322,745 cm4 (7,754 in.4) [for comparison, overestimate of the yield force.
Ig = 326,908 cm4 (7,854 in.4)]. For CP2–CP4, a more appropriate CP5 partially corrects the response (Fig. 4). The initial elastic
softening of the cracked elastic properties was made using an aver- stiffness matches the exact solution and the yield load is preserved
age value of Ieff = 0.6 Ig. For comparison, ACI 318–14 (ACI 2014) (if the yield moment is properly computed). However, the post-
recommends a value of 0.7 Ig; Caltrans SDC (Caltrans 2013) speci- yielding slope is not correct and the CSiBridge implementation uti-
fies 0.53 Ig for the RC section considered; and Paulay and Priestly lizes the Interaction hinge, which removes the ability to study more
(1992) recommend 0.5–0.7 Ig for the range of axial loads considered explicit nonlinearities in the cross section enabled by the fiber dis-
in the benchmark problem. cretization. Finally, CP6 has the same model configuration as CP1;
however, the elastic element properties are modified so that the ini-
Benchmark Analytical Results tial stiffness is exact.
 The modified
 material stiffness for this col-
umn is Emod ¼ E 1 þ 3lph =H . The result of CP6 is algorithmically
The analytical solutions for the steel column using the DP and CP1 the same as with CP2–CP4, but with the hinge located at the xh solu-
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

implementations are given in Fig. 4. The difference between the two tion mentioned above, and therefore not shown.
solutions is due to the limitation of the formulations, not the numeri- The analytical solutions for the RC implementations are given in
cal (software) implementations. The results confirm the common ob- Fig. 5. The stiffness of CP1 is overpredicted after cracking (and
servation that a series representation of a hinge and column is more underpredicted before cracking) because it is based on the approxi-
flexible than considering the true response because the deformations mately uncracked stiffness of the column (0.987 Ig) and not a com-
accumulate simultaneously in both elements. However, because of posite or cracked stiffness. Although any selection of Ieff is approxi-
the series system, the forces are the same in the hinge and column, mate (because it does not change continuously with load level), it
and the yield moments (and, therefore, the yield load) is exact com- can be seen that the softening of Ieff in CP2–CP4 yields trends simi-
paring DP and CP1. lar to the DP solution.
As with any element formulation that attempts to account for the The location of the hinge at the base of the column (xh = 0) guar-
difference between rotation and curvature, the CP1 solution also antees that the yield loads are nominally equal for CP1 and CP2. As
depends on the selected value of lph. The larger the assumed lph, the xh increases (CP3 and CP4), so do the predicted yield and ultimate
larger the discrepancy between the DP and CP1 solutions in the loads. Due to the competing effects of the more flexible series sys-
yielding region. As with the elastic stiffness, the post-yielding stiff- tem, cracked Ieff in the elastic element, and the location of the hinge,
ness is not the same in the two approaches. The hardening stiffness CP2 and CP3 exhibit responses most similar to DP for the bench-
and ultimate load for the CP1 solution are larger than for the DP so- mark column. However, when strain hardening is added to the rein-
lution because there is no spread of yielding beyond the plastic forcing steel, the solutions will diverge at larger ductilities.
hinge length. The response of CP5 is highly sensitive to the chosen Ieff,
The rigid element over lp at the base (CP2–CP4) removes the addi- yield moment, and hardening slope. Here the stiffness is approx-
tional deformation in the series system arrangement of CP1. The CP2 imately equal to the initial elastic stiffness of DP because of the
initial stiffness is approximately correct, and generally there is a better high Ieff. Modifying the elastic element properties (CP6) com-
match between the CP2 and DP responses. The CP2 implementation pounds changes made to Ieff, leading to a response bounded by
has the added benefit of being able to use the fiber cross section, CP1 and CP5.
which is not possible with the rigid-plastic hinge (CP5). CP3 and CP4
also exhibit the approximately correct initial stiffness. It can be shown
that the exact elastic stiffness for this column can be derived by setting Benchmark Numerical Results
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
xh equal to H  H 2  lp H þ l2p =3. The moment arm decreases as The numerical implementations for the RC column are compared
with the analytical solutions in this section. The DP solutions are
xh increases (CP3 and CP4); therefore, the yield and ultimate loads

1500 DP
DP 80
CP1
CP1 CP2
CP2 70
CP3
CP3
SRSS of load (kN)

60 CP4
SRSS of load (kN)

CP4 75
1000 1400 CP5
CP5 50 70
1300
40 65
1200
30 60
500
1100 55
20
1000 50
10
900 45
0.14 0.16 0.18 0.2 0.22 0.08 0.1 0.12 0.14 0.16
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0 0.05 0.1 0.15 0.2 0.25 0.3
SRSS of displacement (m) SRSS of displacement (m)

Fig. 4. Analytical solutions for all implementations of steel bench- Fig. 5. Analytical solutions for all implementations of RC benchmark
mark columns with 2% hardening. columns.

© ASCE 04019011-4 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


presented in Fig. 6 for two modeling approaches in OpenSees. The The numerical implementations for CP1 are given in Fig. 7. The
DP approach is not an option in CSiBridge; however, the analytical minor differences in the CSiBridge solution are due to the differen-
DP solution is often used as a benchmark for calibrating the CP ces observed in the sectional response in Fig. 2, whereas differences
approaches available in CSiBridge. The first DP approach in in the OpenSees solution are due to the discrete moment–rotation
OpenSees uses a forceBeamColumn element with six Gauss- points that define the ElasticMultiLinear material. Finally, the CP2
Lobatto integration points. With this number of integration points, and CP5 numerical solutions from CSiBridge are given in Fig. 8.
the numerical solution captures the spread of plasticity along the The responses are nominally identical to the analytical solutions
column height and matches the analytical load–displacement rela- prior to yield, but the CP2 solution from CSiBridge exhibits much
tionship given in Fig. 6. stiffer post-yielding behavior.
The second OpenSees approach for DP uses the beamWithHinges
element in which plasticity is allowed to spread over plastic hinge Benchmark Problems Summary
regions at the element ends. Although it is not necessary, this element
uses an elastic section in the interior between the end hinges to limit The benchmark problems demonstrated that the inelastic column
the spread of plasticity. Therefore, the solution presented in Fig. 6 responses predicted by both CSiBridge and OpenSees can be under-
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

depends heavily on the hinge length specified and the cracked proper- stood relative to the element formulations employed in each soft-
ties assigned to the elastic portion of the element. Including the same ware package. A sequence of analyses was performed with 3D steel
fiber cross section in the element interior would lead to the same nu- (strain hardening) and RC columns (EPP steel and ENT concrete).
merical results as the forceBeamColumn implementation, but this is Numerous assumptions simplified the modeling. Specifically, the
not shown here. material models were selected to be simple to ensure that differen-
The permutations of CP approaches shown in Table 1 were imple- ces observed between the software implementations were not due to
mented in both CSiBridge and OpenSees. The CSiBridge models uti-
lized the Fiber P-M2-M3 hinge (CP1-CP4, CP6) and Interaction
P-M2-M3 hinge (CP5). These are the two most commonly used hinge 80
types in CSiBridge. The former is a discretized cross section that uses
individual uniaxial constitutive models for each fiber (as was utilized 70
in the moment–curvature analysis), whereas the latter has rigid-plastic
behavior, with the yield defined by an axial–moment interaction sur- SRSS of load (kN) 60
face. In CSiBridge, the Interaction hinge generates a moment–rotation 50
response based on the yield surface specified in FEMA (2000),
although the default moment–rotation backbone values are not 40
directly relevant for bridge analysis. The Interaction hinge is able to 30
capture the coupling of the yield surface due to bidirectional loading.
In OpenSees, the CP permutations were implemented using 20
CP1
elasticBeamColumn elements of different stiffness and a CP1 (CSiBridge)
10
CoupledZeroLength element at the hinge location. This arrange- CP1 (OpenSees)
ment is inconvenient because it requires a preliminary sectional 0
analysis to obtain the moment–curvature relationship that is con- 0 0.05 0.1 0.15 0.2 0.25 0.3
verted to a moment–rotation relationship through lph but is SRSS of displacement (m)
numerically equivalent for the case when the axial load is con-
stant. The moment–rotation is input using discrete points on the Fig. 7. Comparison of analytical and numerical implementations for
backbone curve into an ElasticMultiLinear uniaxial material. RC columns and model CP1.

80 80

70 70
SRSS of load (kN)
SRSS of load (kN)

60 60

50 50

40 40

30 30 DP
DP CP1
20 20
DP (OpenSees ForceBeam) CP2
DP (OpenSees BeamWithHinges El.) 10 CP2 (CSiBridge)
10
DP (OpenSees BeamWithHinges In.) CP5 (CSiBridge)
0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
SRSS of displacement (m) SRSS of displacement (m)

Fig. 6. Comparison of analytical and numerical implementations for Fig. 8. Comparison of analytical and numerical implementations for
RC columns and model DP. RC columns and models DP, CP1, CP2, and CP5.

© ASCE 04019011-5 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


the constitutive behavior. Likewise, no attempt was made to include to limit modeling uncertainties for comparison, the abutments in
bond-slip springs, shear deformations, damage effects, and so on. both models were made to be simple rollers (the roller model has a
These assumptions and the presented results are not meant to imply single vertical constraint as well as a torsional restraint about the
that proper constitutive models and modes of deformation should axis of the deck). A schematic of both bridges is presented in Fig. 9,
be ignored in more complex bridge models. and the major geometric, material, and modeling parameters are
Discrepancies in stiffness and strength are due to the series given here. More detailed descriptions of the bridges can be found
arrangement of hinge and frame elements, the location of the plastic in Mackie et al. (2017).
hinge along the column height, and the cracked moment of inertia The constitutive models for the column concrete and steel stress–
in the elastic element. Therefore, the same responses can be gener- strain behavior were subjected to monotonic tension and compres-
ated in both software packages as long as consistent modeling sion to illustrate the differences in the CSiBridge and OpenSees real-
choices are made (i.e., selection of properties that govern the CP izations. The comparisons were then extended to sectional analysis
elements). of the columns, pushover of only the hinge macroelements at the top
Several approximate corrections for the lumped plasticity of the columns, pushover analysis of the bents, and finally modal
approaches to better predict the DP response were demonstrated. analysis of the entire bridge system.
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

The introduction of a rigid element over the plastic hinge region,


while retaining the original position of the hinge itself (CP2), led Ordinary Standard Bridge 1
to the best approximation to stiffness and strength. For the RC
column, CP3 with a modifier of 0.6 Ig also provided similar As presented in Fig. 9, OSB1 was a two-span bridge with a single
results. The fiber hinges enabled the modeler to retain fiber cross two-column bent at the center. The clear column height was 5.6 m
sections, which is important for enabling both 2D and 3D (18.5 ft) and each span was 45.7 m (150 ft) long. The bent contained
responses to capture changes in stiffness due to axial–moment two 1.7-m (5.5-ft) diameter circular RC columns at 7.3-m (24-ft)
interaction. (center-to-center) spacing. The circular cross sections contained
36 #11 Grade 60 longitudinal bars and #8 transverse spirals spaced
15 cm (6 in.) on center. The columns were founded on pile caps
Benchmark Bridges with a pipe pin for the pile cap-to-column connection. The specified
material properties were fc0 ¼ 24:8 MPa (3.6 ksi) for the bent pile
To ensure that the commonalities identified by correcting the col- cap and fc0 ¼ 27:6 MPa (4.0 ksi) for the bridge structural concrete.
umn responses extend to the bridge level, two OSB models devel- The superstructure was a continuous cast-in-place (CIP) post-
oped by Caltrans were studied. The OpenSees models were imple- tensioned concrete box girder with five cells and it was integral with
mented to match the response from the CSiBridge models provided the bent cap. The standard seat-type abutments had a 5-cm (2-in.)
by Caltrans. No changes or corrections were made to these movement rating on elastomeric bearings. The end diaphragm was
CSiBridge models in developing the OpenSees models. However, also integral with the superstructure.

Fig. 9. Schematic of OSB1 and OSB2 showing geometry and column cross sections.

© ASCE 04019011-6 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


CSiBridge account for the circular yield surface, and the portions above and
The superstructure center of mass in the CSiBridge model was below the hinge were made rigid. The second approach used a
6.1 m (20 ft) above the column bases. A 0.85-m (2.8-ft) long rigid forceBeamColumn with a finite hinge length and a single integration
element was introduced at the top of each column by multiplying point. To accommodate an off-center hinge location for OSB1 in
the moments of inertia of the rigid element by three. A frame hinge OpenSees, it was necessary to use the FixedLocation integration
type (Fiber P-M2-M3) was introduced at the middle of the rigid ele- rule at the location 40 cm (15.6 in.) above the elastic portion of the
ment. The rigid offset [0.46 m (1.5 ft)] between the column top and column. The benefit of the latter approach is that there is a fiber
the deck was modeled with a rigid zone factor of 0.5. This rigid off- cross section at the single integration point that accounts for axial–
set within the top frame element containing the hinge forced the moment interaction, that is, a prior moment–curvature analysis is
hinge to move 0.46 m (18 in.) below the nodes of the deck centerline not necessary. The second approach was adopted for the OSB1
rather than 0.43 m (16.8 in.) below, which would have been the case results presented herein.
without the rigid offset. The moment of inertia for the remaining The columns were modeled with stiff elasticBeamColumn ele-
length of the column was reduced by a modifier of 0.35 in both ments to represent the rigid portions and with the same property
directions and no frame hinge was introduced in this part of the col- modifiers as CSiBridge for the interior portions. The fiber cross sec-
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

umn. The bottom end joint of the column at the pile cap was pinned. tion discretization matched CSiBridge, with individual core con-
The concrete constitutive models employed were Mander- crete, cover concrete, and longitudinal reinforcing steel constitutive
unconfined and Mander-confined with input parameters as given in models. Concrete04 was used for both the core and cover concrete
Tables 2 and 3, respectively. The constitutive model for the longitu- and its properties are given in Table 4. Concrete04 does not model
dinal steel was a table-based input. The important parameters of the the linear descending branch of the unconfined concrete indicated
longitudinal steel model were Young’s modulus Es = 200 GPa
in the CSiBridge manual (CSI 2011) and constitutive model dialog
(29,000 ksi), yield stress 420 MPa (60 ksi), ultimate stress 630 MPa
boxes. Instead, the stress immediately drops to zero at the specified
(90 ksi), strain at onset of strain hardening 0.01, ultimate strain
ultimate strain. The values for the ultimate strain and crushing
capacity 0.09, and final slope 0.1 Es.
strength of the confined concrete were taken directly from
The superstructure material was based on the elastic modulus of
CSiBridge for consistency. In addition, although the CSiBridge
24.8-MPa (3.6-ksi) normal-weight concrete and the section proper-
ties were selected to be non-prismatic, with variations due to girder manual indicates there is tension behavior in the confined concrete,
flare and changes near the supports. A rigid elastic material defined it was not observed during the constitutive model tests presented in
the bent cap and end diaphragms of the bridge. The frame elements the next subsection. Therefore, the tensile strength of the confined
for the bent cap and end diaphragms utilized rectangular prismatic concrete in the OpenSees model was set to zero. The longitudinal
sections, and the additional weight and mass due to the bent cap and reinforcing steel used the same table data to identify the tension and
end diaphragms were included. Due to the assumption of roller compression backbone points (symmetric) and the MultiLinear ma-
abutments, the end diaphragm elements do not appear in the model. terial was wrapped in a MinMaxMaterial to ensure that the stress
dropped to zero.
OpenSees
The elastic material and section properties used for the columns,
superstructure, bent cap, and end diaphragm were duplicated Ordinary Standard Bridge 2
directly from CSiBridge. The elements assumed to remain elastic
OSB2 was a two-span bridge with a single column bent at the center.
were modeled using the elasticBeamColumn. Two-column hinge
Each span was 45.7 m (150 ft) long and the clear column height was
modeling strategies were considered. The first approach was similar
5.7 m (18.5 ft). The column had a 1.7-m (5.5-ft) diameter circular
to that employed in the benchmark problems, where sectional anal-
RC section. The column section contained 22 bundles of #11 Grade
ysis with an assumed axial load was performed prior to the bridge
60 longitudinal bars, totaling 44 bars, and #8 transverse spirals
analysis. The curvature was converted to a rotation and lumped
40 cm (15.6 in.) from the top of the elastic portion of the column. spaced at 15 cm (6 in.) on center. The columns were founded on pile
The spring was a CoupledZeroLength element that was able to caps. The specified material properties were fc0 = (24.8 MPa) 3.6 ksi
for the bent cap and fc0 = 27.6 MPa (4.0 ksi) for the bridge structural
concrete. The superstructure was a continuous CIP post-tensioned
Table 2. Mander-unconfined concrete model parameters for CSiBridge
concrete box girder with three cells. The total depth and width of the
Parameter OSB1 OSB2 superstructure were 1.8 m (6 ft) and 11.4 m (37.5 ft), respectively.
The bent cap was integral with the superstructure. The abutments
Compressive strength (MPa) 24.8 27.6
were standard seat-type abutments with 5-cm (2-in.) movement rat-
Strain at compressive strength 0.00222 0.002
Ultimate strain capacity 0.005 0.005
ing on elastomeric bearings. The end diaphragm was also integral
with the superstructure.

Table 3. Mander-confined concrete model parameters for CSiBridge Table 4. Concrete04 material properties for OpenSees

Parameter OSB1 OSB2 OSB1 OSB1 OSB2 OSB2


Parameter Core Cover Core Cover
Tangent modulus of elasticity (GPa) 23.6 24.8
Secant modulus of elasticity (GPa) 5.20 6.63 Compressive strength (MPa) 35.1 24.8 38.0 27.6
Compressive strength of unconfined concrete (MPa) 24.8 27.6 Strain at compressive strength 0.0069 0.0022 0.0056 0.002
Compressive strength of confined concrete (MPa) 35.1 38.0 Crushing strength (MPa) 32.1 20.4 33.3 20.7
Strain at strength of unconfined concrete 0.00222 0.002 Strain at crushing strength 0.0163 0.0044 0.0153 0.004
Ultimate strain capacity of unconfined concrete 0.005 0.005 Tensile strength (MPa) 0 3.1 0 3.3
Strain at compressive strength of confined concrete 0.0069 0.0056 Elastic modulus (GPa) 28.0 23.6 29.2 24.9

© ASCE 04019011-7 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


CSiBridge Column Constitutive Models
Similar to OSB1, no changes were made to the CSiBridge model
The column constitutive models were studied independently under
obtained from Caltrans. An inadvertent modeling choice was made monotonically increasing strains. Comparative results are presented
for the column hinges that resulted in a much stiffer and more brittle here for CSiBridge and OpenSees for each of the longitudinal steel,
bridge response than would otherwise have been anticipated. This cover concrete, and core concrete models. The procedure for obtain-
original model was retained and a matched OpenSees model gener- ing the stress–strain relationships in OpenSees is direct (strain his-
ated, with the understanding that the model as intended would result tory desired is prescribed directly on a single fiber). However, in
in more ductile behavior. CSiBridge the stress–strain behavior was obtained using frame ele-
The center of mass in the CSiBridge model was located 6.1 m ments with modified section and material properties (and was there-
(20 ft) above the column bases. Rigid elements were introduced fore subject to any strength loss or negative stiffness limitations
at the top and bottom of the column. Each rigid element was made internally). For the steel materials and unconfined concrete
0.85 m (2.8 ft) long and used a multiplier of three for its moments models, a single fiber hinge of unit length and area was used to cap-
of inertia. A frame hinge type (Fiber P-M2-M3) was introduced at ture plastic behavior. For the confined concrete model, the frame
the middle of the rigid element in each hinge. Unlike OSB1, no was split into two elements: the rigid element and an element of unit
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

rigid offset was used in the OSB2 model, so the effective length length with the cross section of the column’s plastic hinge. The
of the columns was longer in OSB2. The inadvertent choice cross section of the column’s plastic hinge was modified to have no
occurred due to the introduction of a separate element to model cover concrete and zero strength longitudinal bars, thus leaving
the rigid end regions with hinges. The default hinge length of only the confined core concrete. The forces obtained from this
10% of the element length was retained, resulting in a plastic model were normalized by the area of the core to obtain stress.
hinge length of 8.5 cm (3.36 in.). The steel monotonic stress–strain responses are presented in
The moment of inertia for the remaining length of the column Fig. 10. Due to the table-based inputs associated with the steel mod-
was reduced by using a modifier of 0.35 in both directions and no els, the responses were identical between the two software pack-
frame hinge was introduced in this part of the column. The bottom ages. The only distinction purposely introduced was that the
of the column was fixed in all degrees of freedom. The longitudinal OpenSees model unloaded to zero before being removed from the
steel constitutive model was identical to that shown for OSB1. The analysis (through the MinMaxMaterial) to eliminate spurious nega-
Mander model was used for the concrete stress–strain behavior; tive strength when unloading from the softening backbone.
however, the parameters differed slightly from OSB1 due to the The cover concrete (unconfined) monotonic responses from
27.6-MPa (4.0-ksi) strength, as shown in Tables 2 and 3 for uncon- CSiBridge and OpenSees are presented in Fig. 11(a), together with
fined and confined concrete, respectively. the predictions from the equations provided in the CSiBridge user
A rigid elastic material defined the bent cap and end diaphragms manual. The predictions for the compression response of cover con-
of the bridge. The frame elements for the bent cap and end dia- crete are consistent for the monotonic case; however, the CSiBridge
phragms utilized rectangular prismatic sections, but mass and solution only matches the backbone in a piecewise linear way. It
weight contributions were not added for consistency with the input was verified that this piecewise linear behavior was not due to the
files provided by Caltrans. convergence tolerance specified or the number of strain increments
utilized in the history. The OpenSees compressive stress drops rap-
idly to zero after twice the strain at the peak compressive stress, a
OpenSees documented behavior of the Concrete04 material. However, the
The OpenSees OSB2 model followed the approach taken for CSiBridge results continued to decrease linearly to zero, accommo-
OSB1 in imitating the CSiBridge models provided by Caltrans. dating very large strains. The unconfined behavior was not consist-
The elastic material and section properties used for the columns, ent with either the linear drop predicted by the equations from the
superstructure, bent cap, and end diaphragm were duplicated CSiBridge user manual or the figure shown in the constitutive
directly. The elements assumed to remain elastic were modeled model dialog boxes. The OpenSees model used an exponential
using the elasticBeamColumn. The same two plastic hinge mod-
eling strategies mentioned with OSB1 were considered. The
implementation for OSB2 was simpler due to the absence of the
rigid offset, so the hinge element of length 8.5 cm (3.36 in.) was 800
used directly in the center of the 0.85 m (2.8 ft) rigid region at the 600
top and bottom hinges for the first approach. The second
approach used a forceBeamColumn with a finite hinge length, 400
but with only a single integration point at the center of the ele-
Stress (MPa)

ment. The second approach was again adopted for the results pre- 200
sented here. 0
The rigid portions of the columns were modeled with stiff
elasticBeamColumn elements. The interior of the columns were -200
modeled as elasticBeamColumn elements with the same property
-400
modifiers as implemented in the CSiBridge models. The bottom OpenSees
node of the column was fixed. The fiber cross section discretization -600 CSiBridge
was created to match CSiBridge, with individual core concrete, CSiBridge input table
cover concrete, and longitudinal reinforcing steel constitutive mod- -800
-0.15 -0.1 -0.05 0 0.05 0.1 0.15
els. Concrete04 was again used for both the core and cover concrete
Strain
and the properties are given in Table 4. The ultimate strain and
crushing strength of the confined concrete were taken directly from Fig. 10. Steel monotonic stress–strain response for OSB1 and OSB2.
CSiBridge for consistency.

© ASCE 04019011-8 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


5 0

-5
0
-10
-5

Stress (MPa)
Stress (MPa)

-15

-10 -20

-25
-15
-30
OpenSees OpenSees
-20 CSiBridge CSiBridge
-35
Mander Mander
-25 -40
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

-6 -4 -2 0 2 -20 -15 -10 -5 0


Strain -3
Strain 10-3 10
(a) (a)

5 0

0 -5

-5 -10
Stress (MPa)

Stress (MPa)
-15
-10
-20
-15
-25
-20
OpenSees -30
OpenSees
-25 CSiBridge CSiBridge
Mander -35
Mander
-30 -40
-6 -4 -2 0 2 -20 -15 -10 -5 0
Strain 10-3 Strain 10
-3

(b) (b)

Fig. 11. Cover concrete monotonic stress–strain response: (a) OSB1; Fig. 12. Core concrete monotonic stress–strain response: (a) OSB1;
and (b) OSB2. and (b) OSB2.

curve for the tension softening behavior; therefore, they were not
identical as modeled here. The standardization of the column constitutive models was
The core concrete (confined) monotonic stress–strain responses extended to the section level, where moment–curvature analyses
and the equations from the CSiBridge user manual are presented in were performed on the fiber cross sections of OSB1 and OSB2.
Fig. 12(a). Convergence issues prevented the softening portion of Here it was assumed that the axial load was zero. Due to the sim-
the CSiBridge model from being obtained in the monotonic case, ilarity of the monotonic stress–strain backbone curves of the
the results shown are based on a slightly relaxed convergence crite- constituent materials, it was anticipated that the sectional
rion. As with the unconfined case, there appears to be a linear soft- response curves between CSiBridge and OpenSees would be
ening portion that extended well beyond the crushing strain of the similar. The differences obtained were, therefore, due to the dis-
confined concrete. cretization of the cross section and minor constitutive model
The OSB2 cover concrete (unconfined) monotonic stress–strain differences.
responses are presented in Fig. 11(b). The cover compressive pre- The moment–curvature responses are presented in Fig. 13.
dictions deviated slightly beyond the peak compressive stress, in The results began to diverge beyond the peak moment, even
addition to the previously observed match to the backbone in a though the steel softening backbones were identical between the
piecewise linear way. The OpenSees model used an exponential two software packages. By setting the maximum steel strain in
curve for the tension softening behavior, therefore, they were not OpenSees to 0.09 (at the peak strength), it is possible to make
identical as modeled here. The compressive stress dropped rapidly the moment–curvature points of rapid descent align. It is not
to zero after twice the strain at the peak compressive stress. clear what caused the sudden drop in the CSiBridge results that
However, the CSiBridge results did not converge for the cover con- then stabilized; however, it was likely due to the residual
crete case beyond the last point shown in the plot. The core concrete capacity in the confined concrete observed in Figs. 12. The cur-
(confined) monotonic stress–strain responses are presented in vature at peak moment and peak moment for OSB2 are slightly
Fig. 12(b). Conclusions similar to those for the OSB1 case can be sensitive (<2%) to the orientation of the local axes for the sec-
drawn from the responses. tion due to the asymmetric bar arrangement.

© ASCE 04019011-9 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


Hinge Force–Deformation Analysis hinge formulation, as well as to the underlying material (stress–
strain) and section (moment–curvature) inputs that defined the
The nonlinear static response of the bridge systems depends on
flexural properties of the hinges.
the performance of the hinge elements. The differences in the
The hinge at the top of the column as shown in Fig. 14, was
responses between software packages are potentially due to the
extracted from the overall bridge models in CSiBridge and
OpenSees. This hinge was fully restrained at the base and then sub-
jected to a monotonically increasing displacement at the top in each
14000 lateral direction independently. The results are shown with an axial
load of zero. The resulting force–displacement curve incorporated
12000
the modeling assumptions specific to the hinge, such as the rigid
10000
zone factor, rigid offset, and location of the integration point. Results
Moment (kN-m)

from OSB1 are presented in Fig. 15(a). The two CSiBridge cases
8000 shown are for the model as originally created, followed by a modi-
fied version where the rigid offset was removed.
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

6000 Although the moment–curvature response was slightly more


ductile in OpenSees, the hinge pushover indicated a degradation of
4000 OpenSees OSB1 the strength near the peak force, as well as more brittle behavior
CSiBridge OSB1 than CSiBridge. This is due to the loss of concrete stress in
2000 OpenSees OSB2 Concrete04 once the ultimate strain is reached in compression. The
CSiBridge OSB2
location of the integration point also contributes to the hardening
0
0 0.02 0.04 0.06 0.08 0.1 slope and peak strength; however, without a physical basis for
Curvature (1/m)
4
10
Fig. 13. Moment–curvature responses for OSB1 and OSB2. 3
OpenSees
2.5 CSiBridge
CSiBridge no offset

Total base shear (kN)


2

1.5

0.5

0
0 0.02 0.04 0.06 0.08
Displacement (m)
(a)

104
3.5

3
Total base shear (kN)

2.5

1.5

1
OpenSees
0.5 CSiBridge 0.1
CSiBridge 2.8 ft
0
0 0.01 0.02 0.03
Displacement (m)
(b)

Fig. 14. Elements extracted at top of column for hinge force–deforma- Fig. 15. Column top hinge load–displacement response: (a) OSB1;
tion analysis. c.o.m. = center of mass. and (b) OSB2.

© ASCE 04019011-10 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


moving it, the value reported in CSiBridge for the absolute hinge during the hinge pushover can be attributed solely to the confined
location was utilized as described in the modeling section. The soft- concrete model. The resulting force–displacement curves are pre-
ening behavior prior to the peak shear force in OpenSees is also evi- sented in Fig. 15(b). The two CSiBridge cases are for the model as
dent in the analyses presented in the “Bent Pushovers” section. It is originally created (labeled as 0.1 to indicate 10% of the rigid zone
possible to get the hinge pushover and bent pushovers to agree; length) and what would have been achieved if the full plastic hinge
however, it requires removing the ultimate compressive strain limit length of 0.85 m (2.8 ft) was used. All responses between CSiBridge
(drop to zero stress) in concrete, which leads to an increase in the and OpenSees agreed except the peak strength, as was mentioned
predicted moment capacity and ductility capacity. As a result, the previously for the OSB1 hinge pushover analysis. The stiffer and
standardization of constitutive models to match the section level substantially more brittle response of the short hinges is what creates
responses is maintained throughout. the unusual nonlinear static pushover of the bridge bent. Therefore,
For OSB2, the hinge was more clearly defined with no rigid off- care should be exercised when viewing those results relative to the
sets or rigid zone factors, and therefore the differences observed modeling assumptions.

Bent Pushovers
6000
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

To isolate the effect of the bent and column models in the nonlinear
5000 static (pushover) response, the roller abutment boundary case was
Total base shear (kN)

adopted. Two independent pushover analyses were performed.


4000 Both utilized a single monitoring point in the center of the bent at
the level of the superstructure mass. Due to each of the OSBs having
a single bent and roller abutment boundaries, a single 445 kN
3000
(100 kip) reference load was applied at the monitoring point for the
longitudinal pushover analysis. For the transverse pushover analy-
2000 sis, a point load of the same magnitude was applied at the monitor-
Longitudinal CSi
Transverse CSi ing point as well as at each end of the spans (at the location of the
1000 Longitudinal OpenSees roller abutment). Nonlinear analysis with displacement control was
Transverse OpenSees adopted (in 100 steps).
0 The total pushover curves (summation of the base shears for each
0 0.1 0.2 0.3 0.4 0.5 column in the case of OSB1) are shown on the same plot for
Displacement (m) CSiBridge and OpenSees. The longitudinal and transverse pushovers
(a) from OSB1 are presented in Fig. 16(a), and the longitudinal and trans-
verse pushovers from OSB2 are presented in Fig. 16(b). For OSB1,
7000 although the sectional responses and hinge pushovers exhibited
nearly identical initial stiffness and yield points, the pushover curves
6000 showed stiffer behavior and a larger yield force in the OpenSees mod-
els. These nominal differences were likely due to the combined
5000 effects of the complex choice of hinge parameters in CSiBridge (rigid
Base shear (kN)

offset plus rigid zone factor) that were not easily duplicated in
4000
OpenSees, plus the use of the non-prismatic superstructure, whereas
the OpenSees model was assumed prismatic for modeling simplicity.
3000
These two items created a slightly different boundary condition at the
2000
top of the bent columns between the two software packages.
Longitudinal CSi
Transverse CSi The other obvious difference for OSB1 was the extended ductility
1000 Longitudinal OpenSees of the CSiBridge model, which causes a nearly 20% difference in the
Transverse OpenSees estimated peak base shear. The only way to recover this ductile behav-
0 ior in the OpenSees model is to remove the limitation on the crushing
0 0.02 0.04 0.06 strain of the confined concrete in compression. If the OpenSees consti-
Displacement (m) tutive model is allowed to continue along the compressive backbone
without dropping to zero stress, the same peak base shear and corre-
(b)
sponding displacement can be recovered. The subsequent softening
Fig. 16. Bent load–displacement pushover comparison: (a) OSB1;
that occurred at 5,340 kN (1,200 kip) was triggered by the steel reach-
and (b) OSB2.
ing the softening branch of the constitutive model. However, to
achieve the ultimate steel strain, the corresponding compressive strain

Table 5. Modal periods and frequencies obtained from CSiBridge and OpenSees models of OSB1 with roller abutments

CSiBridge OpenSees
Mode Period (sec) Frequency (Hz) Mode Shape Period (sec) Frequency (Hz) Mode Shape
1 6.033 0.166 Torsion 5.934 0.169 Torsion
2 1.012 0.988 Longitudinal 1.002 0.998 Longitudinal
3 0.915 1.092 Transverse 0.863 1.159 Transverse
4 0.567 1.762 Vertical (asymmetric) 0.533 1.876 Vertical (asymmetric)
5 0.405 2.463 Vertical (symmetric) 0.390 2.564 Vertical (symmetric)

© ASCE 04019011-11 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


in the core concrete must exceed 0.03. Note the extended ductility
prior to softening is also in conflict with the moment–curvature results
obtained from CSiBridge (it appears the moment–curvature is based
on the theoretical constitutive models, not the actual ones used in the
frame hinges).
OSB2 was a simpler model of both the superstructure and the
hinges, and therefore easier to replicate. The difficulty with OSB2
was more in the modeling choice (short hinge length) that led to
brittle results, which is obvious when comparing the displacement
capacity of the bridges in Figs. 16(a and b). The effect of the ulti-
(a) mate concrete strain limit in OpenSees is also evident for OSB2, as
with OSB1, although the change in ductility is not nearly as pro-
nounced given the short hinge length.
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

Modal Analysis
Modal analysis was performed to ensure the distributions of mass
and stiffness matched for the CSiBridge and OpenSees OSB mod-
els. The first five modes for OSB1 are presented in Table 5 for the
CSiBridge and OpenSees implementations. The first four mode
shapes are presented in Fig. 17 for roller abutments. The vertical
mode shapes are listed as either antisymmetric or symmetric rela-
(b) tive to the sense of vertical deformation on either side of the bent.
The OSB1 mode shapes matched exactly for the roller abutment
models between the two software packages. The periods were also
very close, with the OpenSees being slightly stiffer, as evidenced in
the nonlinear static responses presented previously. The vertical
eigenvalues were not identical in this case due to the different dis-
cretization of superstructure mass and use of non-prismatic mem-
bers in CSiBridge. The frame action in the transverse direction is
sensitive to the modeling assumptions of the hinge.
OSB2 exhibits similar natural periods of vibration for the roller
abutment condition in the CSiBridge and OpenSees models, as pre-
sented in Table 6. The mode shapes given in Fig. 18 also match
(c) between the two software packages. Overall, the periods of OSB2
are shorter than those of OSB1 due to the fixed column base and the
shorter hinge length. The transverse mode is again sensitive to the
modeling assumptions of the hinge. However, unlike OSB1, both
the second and fifth modes contain a longitudinal deformation of
the bent in addition to asymmetric bending of the superstructure.

Benchmark Bridge Summary


The constitutive models, section behavior, modal properties, and
nonlinear static behaviors of two Caltrans OSBs were presented.
The bridge descriptions, geometry, properties, and modeling param-
eters were used to ensure consistent inputs for both CSiBridge and
(d)
OpenSees models. The CSiBridge models were utilized directly
Fig. 17. Mode shapes obtained from CSiBridge and OpenSees models from Caltrans, with the only modification being to consider the sim-
of OSB1 with roller abutments: (a) Mode 1, torsion; (b) Mode 2, longi- pler case of roller abutments. The OpenSees models were developed
tudinal; (c) Mode 3, transverse; and (d) Mode 4, vertical (asymmetric). with the goals of matching the CSiBridge models and understanding
the limitations of the underlying material and element models. It was

Table 6. Modal periods and frequencies obtained from CSiBridge and OpenSees models of OSB2 with roller abutments

CSiBridge OpenSees
Mode Period (sec) Frequency (Hz) Mode Shape Period (sec) Frequency (Hz) Mode Shape
1 5.09 0.196 Torsion 6.66 0.158 Torsion
2 0.686 1.457 Longitudinal 0.698 1.492 Longitudinal
3 0.629 1.587 Transverse 0.685 1.562 Transverse
4 0.427 2.339 Vertical (symmetric) 0.428 2.340 Vertical (symmetric)
5 0.391 2.553 Vertical (asymmetric) 0.418 2.525 Vertical (asymmetric)

© ASCE 04019011-12 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


demonstrated that the steel models could be specified to match
exactly, whereas both the unconfined and confined concrete models
exhibited unanticipated differences. It was not possible to directly
match all aspects of the concrete models, specifically the softening,
ultimate compressive strain, and tension softening.
The second phase of calibration was on the section response of
the column cross sections and nonlinear static pushover response of
the column hinge elements. The differences in the softening behav-
iors of the concrete models yielded slightly different curvature
capacity estimates, as well as peak load and ductility of the column
(a) hinges during nonlinear static pushover analysis. Finally, modal
analysis was performed to demonstrate that the mass distribution
was consistent in the two software packages, and that the modifica-
tions to achieve consistent initial elastic stiffness in the benchmark
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

problems extended to full bridge models.


The primary challenges encountered in producing OpenSees
models to closely match the CSiBridge counterparts were 2-fold.
The first related to the differences in the achieved responses of the
underlying constitutive models than would have been expected
based on the inputs. The second related to modeling choices made
for OSB1 and OSB2. The OSB1 hinge at the top of the bent col-
umns was complicated by the choice of rigid offset and rigid zone
(b)
factor that were not easily reproduced. The OSB2 hinge was inad-
vertently assigned a plastic hinge length 10% of the expected value,
which led to much more brittle behavior, and therefore sensitive to
changes in the inputs and constitutive models. Knowledge of the
software implementations helps to ensure that the differences in
computed response are understood. Further modeling details and
recommendations are summarized in Mackie et al. (2017).

Acknowledgments

This study was sponsored by the California Department of


Transportation under contract 65A0559. The views and findings
(c) reported here are those of the authors alone. The contents do not
necessarily reflect the official views or policies of the State of
California or the Federal Highway Administration. This study does
not constitute a standard, specification, or regulation. The authors
would like to acknowledge the contributions to the modeling by
Munaf Al-Ramahee, Karryn Johnsohn, and Michael Steijlen.

References

ACI (American Concrete Institute). 2014. Building code requirements for


structural concrete (ACI 318-14) and commentary. Rep. No. ACI 318-
14. Farmington Hills, MI: ACI.
(d) Addessi, D., and V. Ciampi. 2007. “A regularized force-based beam ele-
ment with a damage-plastic section constitutive law.” Int. J. Numer.
Fig. 18. Mode shapes obtained from CSiBridge and OpenSees models Methods Eng. 70 (5): 610–629. https://doi.org/10.1002/nme.1911.
Alemdar, B. N., and D. W. White. 2005. “Displacement, flexibility, and
of OSB2 with roller abutments: (a) Mode 1, torsion; (b) Mode 2, verti-
mixed beam–column finite element formulations for distributed plastic-
cal (asymmetric); (c) Mode 3, transverse; and (d) Mode 4, vertical ity analysis.” J. Struct. Eng. 131 (12): 1811–1819. https://doi.org/10
(symmetric). .1061/(ASCE)0733-9445(2005)131:12(1811).
ATC (Applied Technology Council). 1996. Improved seismic design crite-
ria for California bridges: provisional recommendations. Rep. No.
demonstrated that a standardized response could be nominally ATC-32. Redwood City, CA: ATC.
achieved for nonlinear static analysis; however, this does not imply Aviram, A., K. R. Mackie, and B. Stojadinovic. 2008. Guidelines for non-
linear analysis of bridge structures in California. Rep. No. PEER-2008/
nonlinear dynamic results are comparable due to additional discrep-
03. Berkeley, CA: Pacific Earthquake Engineering Research Center,
ancies in the hysteretic behavior of the constitutive models in Univ. of California Berkeley.
CSiBridge and OpenSees (Mackie et al. 2017). Berry, M. P., and M. O. Eberhard. 2007. Performance modeling strategies
The first step of calibration was standardizing the stress–strain for modern reinforced concrete bridge columns. Rep. No. PEER-2007/
responses of the column constitutive models: longitudinal reinforcing 07. Berkeley, CA: Pacific Earthquake Engineering Research Center,
steel, unconfined cover concrete, and confined core concrete. It was Univ. of California Berkeley.

© ASCE 04019011-13 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011


Caltrans. 2013. Caltrans seismic design criteria. Version 1.7. Sacramento, Mackie, K., and B. Stojadinovic. 2003. Seismic demands for performance-
CA: California Dept. of Transportation, Sacramento. based design of bridges. Rep. No. PEER 2003/16. Berkeley, CA: Pacific
Cornell, C. A., and H. Krawinkler. 2000. “Progress and challenges in seis- Earthquake Engineering Research Center, Univ. of California Berkeley.
mic performance assessment.” PEER Center News 3 (2): 1–2. Mackie, K., and B. Stojadinovic. 2005. Fragility basis for California high-
CSI (Computers & Structures, Inc.). 2011. CSI analysis reference manual. way overpass bridge seismic decision making. Rep. No. PEER 2005/02.
Berkeley, CA: CSI. Berkeley, CA: Pacific Earthquake Engineering Research Center, Univ.
FEMA. 2000. Prestandard and commentary for the seismic rehabilitation of California Berkeley.
of buildings. Report No. FEMA-356. Washington, DC: FEMA. Neuenhofer, A., and F. C. Filippou. 1997. “Evaluation of nonlinear frame
Fenves, G. L., and M. Ellery. 1998. Behavior and failure analysis of a finite-element models.” J. Struct. Eng. 123 (7): 958–966. https://doi.org
multiple-frame highway bridge in the 1994 Northridge earthquake. /10.1061/(ASCE)0733-9445(1997)123:7(958).
Rep. No. PEER 98/08. Berkeley, CA: Pacific Earthquake Engineering Nielson, B. G., and R. DesRoches. 2007. “Analytical seismic fragility curves
Research Center, Univ. of California Berkeley. for typical bridges in the central and southeastern United States.”
Giberson, M. F. 1967. “The response of nonlinear multistory structures sub- Earthquake Spectra 23 (3): 615–633. https://doi.org/10.1193/1.2756815.
jected to earthquake excitation.” Ph.D. thesis, California Institute of Omrani, R., B. Mobasher, X. Liang, S. Günay, K. M. Mosalam, F. Zareian,
Technology. https://thesis.library.caltech.edu/3604/. and E. Taciroglu. 2015. Guidelines for nonlinear seismic analysis of or-
Hajihashemi, A., S. Pezeshk, and T. Huff. 2017. “Comparison of nonlinear dinary bridges: Version 2.0. Caltrans Final Rep. No. 15-64A0454.
Downloaded from ascelibrary.org by Auburn University on 04/28/19. Copyright ASCE. For personal use only; all rights reserved.

static procedures and modeling assumptions for seismic design of ordi- Berkeley, CA: Univ. of California Berkeley.
nary bridges.” Pract. Period. Struct. Des. Constr. 22 (2): 04016022. Paulay, T., and M. J. N. Priestly. 1992. Seismic design of reinforced con-
https://doi.org/10.1061/(ASCE)SC.1943-5576.0000309. crete and masonry buildings. New York: Wiley.
Kostic, S. M., and F. C. Filippou. 2012. “Section discretization of fiber beam– Scott, M. H., and G. L. Fenves. 2006. “Plastic hinge integration methods for
column elements for cyclic inelastic response.” J. Struct. Eng. 138 (5): force-based beam–column elements.” J. Struct. Eng. 132 (2): 244–252.
592–601. https://doi.org/10.1061/(ASCE)ST.1943-541X.0000501. https://doi.org/10.1061/(ASCE)0733-9445(2006)132: 2(244).
Kunnath, S. ed. 2007. Application of the PEER PBEE methodology to the I- Scott, M. H., and K. L. Ryan. 2013. “Moment-rotation behavior of force-
880 viaduct. Rep. No. PEER 2006/10. Berkeley, CA: Pacific Earthquake based plastic hinge elements.” Earthquake Spectra 29 (2): 597–607.
Engineering Research Center, Univ. of California Berkeley. https://doi.org/10.1193/1.4000136.
Mackie, K., M. Scott, K. Johnsohn, M. Al-Ramahee, and M. Steijlen. 2017. Zareian, F., and H. Krawinkler. 2009. Simplified performance-based earth-
Nonlinear time history analysis of ordinary standard bridges. Caltrans quake engineering. Rep. No. 169. Stanford, CA: Stanford Univ., John
Final Rep. No. 17-65A0559. Corvallis, OR: Oregon State Univ. A. Blume Earthquake Engineering Research Center.

© ASCE 04019011-14 Pract. Period. Struct. Des. Constr.

Pract. Period. Struct. Des. Constr., 2019, 24(3): 04019011

You might also like