You are on page 1of 9

Engineering Structures 32 (2010) 2931–2939

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Best-fit models for nonlinear seismic response of reinforced concrete frames


Andres Lepage a,∗ , Michael W. Hopper a , Sebastian A. Delgado b , Jeff J. Dragovich c
a
The Pennsylvania State University, 104 Engineering Unit A, University Park, PA, 16802, USA
b
Departamento de Estructuras, Facultad de Ingeniería, Universidad del Zulia, Av. Goajira, Ala D, Maracaibo, Zulia, Venezuela
c
National Institute of Standards and Technology, Building and Fire Research Laboratory (MS 8604), 100 Bureau Drive, Gaithersburg, MD 20899-8604, USA

article info abstract


Article history: This paper identifies the optimal combination of hysteresis-modeling and damping parameters for use in
Received 19 February 2010 practical nonlinear dynamic analysis to obtain satisfactory correlations in both amplitude and waveform
Received in revised form between the calculated and measured seismic response of reinforced concrete frames. In this study, frame
14 April 2010
members are characterized by five modeling parameters: initial stiffness, bond–slip rotations, post-yield
Accepted 24 May 2010
Available online 23 June 2010
stiffness, unloading stiffness, and viscous damping. The calculated response is compared with measured
data from three small-scale shake-table multistory test structures and from a seven-story instrumented
Keywords:
building. The three test specimens (structures MF1, MF2, and FNW) are each analyzed for two different
Frequency domain error base acceleration tests whereas the seven-story building (Holiday Inn at Van Nuys, CA) is analyzed using
Hysteresis a single recorded seismic event (1994 Northridge) in each of the two principal directions of the building
Nonlinear dynamic analysis (structures HNS and HEW). Analyses for all five structures are carried out using three different computer
Time–history analysis programs. The goodness-of-fit of the computed response to the recorded experimental data is measured
by the Frequency Domain Error (FDE) index. Simplified rules are presented to derive the best modeling
characterizations that give consistent low values of FDE for the various structures and structural analysis
programs considered.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction number and type of material nonlinearities often lead to erroneous


models and inadequate solutions. Simple rules are needed to iden-
Practicing engineers are increasingly using nonlinear static or tify and characterize assorted modeling parameters and to provide
dynamic analysis to evaluate the response of a structure subjected guidance for their implementation using standard structural anal-
to seismic events. When designing new buildings, structural en- ysis software. This paper stems from the search for practical com-
puter models capable of representing a realistic nonlinear seismic
gineers refer to the governing building code for loads and analy-
response of reinforced concrete frames with a satisfactory correla-
sis procedures. When designers choose a non-compliant system to
tion between calculated and measured response, not only in am-
resist lateral loads, sufficient analytical and test data must be sub-
plitude but also in waveform.
mitted to assure that the non-compliant system has adequate seis-
This study is limited to mid-rise reinforced concrete frames that
mic response comparable to that of a conventional code-compliant
are subjected to strong ground motions. The response-history anal-
system. Deviating from the building code is likely to trigger at
yses include geometric and material nonlinearities with hysteretic
least two actions: first, a traditional code-level design is performed response defined at a macroelement level. The primary aim is to
where engineers use a linear response analysis and relate it to identify the optimal values of the hysteresis modeling parameters
traditional lateral force-resisting systems; second, the building is after comparing calculated data with data recorded from the Uni-
evaluated using nonlinear dynamic analysis for its response to a versity of Illinois earthquake simulator [2] on three small-scale test
severe seismic event. These actions are commonplace in the seis- specimens and from the instrumented Holiday Inn building in Van
mic evaluation of existing buildings [1]. Nuys, California [3]. For this purpose, nonlinear dynamic analyses
Although a nonlinear dynamic analysis is an excellent way to were performed using five basic modeling parameters: initial stiff-
evaluate the performance of structures subjected to strong ground ness, bond–slip deformations, post-yield stiffness, unloading stiff-
motions, the modeling complexities involved in characterizing the ness, and type of damping.

2. Nonlinear dynamic analysis


∗Corresponding address: The Pennsylvania State University, Department of
Architectural Engineering, 104 Engineering Unit A, University Park, PA, 16802, USA. The seismic analysis in this study uses a numerical model which
Tel.: +1 814 865 3013; fax: +1 814 863 4789. directly incorporates the nonlinear load–deformation character-
E-mail address: lepage@psu.edu (A. Lepage). istics of individual frame members. The model is subjected to
0141-0296/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.05.012
2932 A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939

earthquake shaking represented by ground-motion acceleration 3. Modeling parameters


histories. The computed response includes load–deformation his-
tories for individual frame members as well as frame global re- Practicing engineers rarely test the assumptions built-in their
sponse parameters: base shear, base overturning moment, and computer models with measured data. With the intention of
story drifts. Analyses are carried out with three computer pro- helping engineers create meaningful models and computer output,
grams: LARZ [4], PERFORM 3D [5] and SAP 2000 [6]. this section describes the key modeling parameters considered and
The following general assumptions and simplifications were in- gives two relatively extreme values to each. Each value typifies
volved in the development of the numerical model for the frames a modeling assumption that is to be tested in combination with
the values assigned to the other parameters. The test is based on
considered: (1) the structure, loads, and response are defined in
determining the goodness-of-fit between the calculated nonlinear
one vertical plane; (2) the ground motion is considered in the
seismic response and the measured data for selected structures as
horizontal direction; (3) the foundation of the structure is rigid;
discussed later in Sections 5 and 6.
(4) masses are lumped at the floor levels and constrained by rigid It is assumed that the load–deformation characteristics of the
diaphragms; (5) each frame element is a line element with a linear- members are dominated by flexure. The effect of axial load was
elastic middle portion bounded by nonlinear springs connected taken into account when defining the moment–rotation relation-
to rigid beam–column joints; (6) linear-elastic response accounts ship for each column. Five basic modeling parameters are con-
for shear and flexural deformations; (7) axial deformations are as- sidered, four of which directly relate to the primary ‘‘backbone’’
sumed to be negligible; (8) material nonlinearity occurs only due moment–rotation relationship of frame members: initial stiffness,
to flexure; (9) slip of reinforcement is considered at beam–column bond–slip deformations, post-yield stiffness, and unloading stiff-
connections; (10) P–∆ effects are incorporated into the basic for- ness. The fifth parameter, viscous damping, also has a significant
mulation of the structural stiffness matrix using a geometric stiff- impact on the dynamic response. Table 1 shows a summary of the
ness correction that accounts for the effects of axial load on the parameters considered.
chord rotation of the columns. All of the three computer programs
considered allow the implementation of the above assumptions. 3.1. Initial stiffness
The modeling assumptions target ductile flexural response at
the member ends. Localized brittle failure modes such as shear fail- Moment–rotation primary curves are derived from moment–
ure and beam–column joint failure are not directly accounted for curvature relationships with the assumption that inflection points
in the analysis; the sudden strength loss typically associated with are at midspans. The initial slope of the moment–curvature dia-
the nonlinear modeling of a brittle failure may be computationally grams is determined using uncracked (U ) or cracked (C ) section
difficult and prone to solution convergence problems. The prob- properties. For U cases, the initial stiffness is based on gross section
properties and the cracking p point is calculated based on a mod-
ability that these types of failure may occur can be assessed by
ulus of rupture, fr , of 1/2 fc0 (MPa) 6 fc0 (psi) . For C cases, the
p
post-processing the analysis results to check the demands at the
initial stiffness is defined using the secant to yield. Both U and C
locations in question.
cases share the same yield point. These cases explore if satisfactory
In both LARZ and PERFORM 3D, the solution technique for inte- correlations may be obtained when the cracking point is ignored.
gration of the equation of motion uses the Newmark β method [7] The cracking and the yield point are derived from sectional analy-
with β = 1/4. In SAP 2000, the solution is based on the Fast Non- ses based on the concrete’s stress–strain relationship by Hognes-
linear Analysis (FNA) method by Wilson [8] with participation of tad [10] and the steel is taken as perfectly elastoplastic. Although
all translational modes of vibration (one per story). Stiffness char- the cracking and yield moments are determined for the initial axial
acteristics are assumed constant over the user-defined time step load due to gravity, both moments are assumed to be independent
and the programs divide the time step into smaller substeps if a of the change in axial load induced by the ground motion.
change of stiffness is detected. The computed response is calcu-
lated and reported at time steps no larger than the smaller of 1/20 3.2. Bond–slip rotation
of the fundamental period of the uncracked structure or 1/4 of the
time step used to define the ground acceleration. These limits are Bond–slip rotation accounts for the softening effect due to the
recommended to facilitate convergence of the solution technique elongation of the longitudinal reinforcement beyond the beam–
and to capture the high-frequency content of the response. column interface into the joint. Bond–slip rotation, θ 0 , is defined
The damping matrix, [C ], is constructed by a linear combination herein as a function of the development length expressed as a mul-
of the mass [M ] and stiffness [K ] matrix, both defined at global tiple of the bar diameter, λdb :
structural degrees of freedom:  2
1 M
θ0 = φy λdb (3)
[C ] = α [M] + β [K ] (1) 2 My
where α and β are defined using
f
where λ = 4uy , fy is the yield stress of reinforcement, u is the av-
erage bond stress, φy is the curvature at yielding, M is the applied
1 moment, and My is the yield moment.
ξi = α + βωi2

(2)
2ωi Eq. (3) was derived based on four simplifying assumptions,
similar to those found in Refs. [4,11]: (i) pullout does not occur
and ξi is the fraction of critical damping for mode i of frequency ωi .
when the reinforcement develops fy ; (ii) the steel stress varies
It is convenient to define damping as a function of the first-mode
linearly from a maximum at the beam–column interface to zero
frequency, ω1 , based on uncracked section properties. For the case
inside the joint; (iii) the rotation due to bond slip is measured
of mass-proportional damping, α = 2ξ1 ω1 and β = 0, where with respect to the neutral axis; and (iv) the tensile stress in the
α affects the unchanging [M ] matrix. For stiffness-proportional reinforcement is proportional to the moment.
damping, α = 0 and β = 2ξ1 /ω1 , where β affects the continuously Two cases of bond–slip rotation are considered: a case of tight
updated [K ] matrix. (T ) bond where λ is taken as 20 bar diameter lengths, and a case
Additional details are presented in Ref. [9] including the input of loose (L) bond where λ is taken as 40 bar diameter lengths. The
data required for the definition of the nonlinear spring properties values of λ closely correspond to practical values of fy and u. For
for each frame member in all of the structural models considered instance, a case where fy = 410 MPa (60 ksi) and u = 5.2 MPa
herein. (0.75 ksi), gives λ = 20, and a case where fy = 410 MPa (60 ksi)
A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939 2933

Table 1
Parameter identification.
Parameter Characteristic value Symbol

fr = 1/2 (MPa)
p
Initial stiffness: Uncracked f0
c U
Cracked fr ≈ 0 C
Bond–slip: Tight λ = 20 T
Loose λ = 40 L
Post-yield stiffness: Hard Kp = 0.10 Ke H
Soft Kp = 0.02 Ke S
Unloading stiffness: Nonreducing γ =0 N
Reducing γ = 0.6 R
Viscous damping: 2% mass proportional ξ = 0.02 α2
5% mass proportional ξ = 0.05 α5
2% stiffness proportional ξ = 0.02 β2
5% stiffness proportional ξ = 0.05 β5

and u = 2.6 MPa (0.375 ksi), gives λ = 40. The large values of θ 0 when unloading from a positive direction depend on the values of
that correspond to the L cases may also be interpreted as including Ke , θe , and θm for the positive direction.
some joint flexibility. In relation to the representative moment–rotation curves
shown in Table 1, the yield point is identified with an open circle of
3.3. Post-yield stiffness coordinates (θe , My ) where My = Ke θe and Ke is the slope from the
origin to the yield point; Kr is the slope of the unloading segments
The moment–rotation post-yield stiffness, Kp , is expressed as a R or N; and θm is defined by the largest rotation along segments S
fraction of the secant-to-yield stiffness, Ke = My /θe where θe is the or H.
effective yield rotation which includes bond–slip effects. Two cases Two values of γ are considered: zero to represent a non-redu-
are considered: a soft case (S ) where Kp = 0.02Ke and a hard case cing (N ) unloading stiffness and 0.6 to represent a reducing (R)
(H ) where Kp = 0.10Ke . The values of Kp are directly defined in case. The case of γ = 0 is investigated because engineers often
programs PERFORM 3D and SAP 2000 via M–θ curves while LARZ choose oversimplified hysteresis models where the unloading stiff-
requires a special definition of the bond–slip beyond the yield point ness is assumed constant throughout the response history analysis.
to control Kp . The value of γ = 0.6 was selected over the original 0.4 recom-
The moment and rotation at ultimate depend on the assumed mended by Takeda [13] because pilot studies [9] showed an im-
material properties, the curvature distribution along the member proved correlation with the measured data considered.
length, and the plastic-hinge length. Instead of the precise calcula- A different hysteresis model is used by each of the programs
tion of the ultimate point, it is convenient to define the post-yield considered: the Takeda model [13] in program LARZ; the energy-
stiffness as a function of the secant to yield. A finite positive slope
ratio model [5] in PERFORM 3D; and the pivot model [14] in SAP
is assigned to the stiffness after yielding to incorporate the strain
2000. All of these models are stiffness-reducing models. In LARZ,
hardening characteristics of the longitudinal reinforcement. This
the value of γ is directly assigned as a Takeda hysteresis model
approach also gives the analyst the option of defining the primary
parameter but in PERFORM 3D and SAP 2000 the target value of γ
moment–rotation curve mainly as a function of the yield point.
has to be emulated through other hysteresis parameters.
Because the structures considered in this study predominantly
In PERFORM 3D, the N case is represented by an energy-
responded with rotation demands below the ultimate point, the
dissipation ratio of 0.5 and the R case by a ratio of 0.2. In SAP
suggested approach proved satisfactory without need for consid-
eration of strength degradation. 2000, the N case is represented with the pivot model using αpivot =
10, βpivot = 0.5, and ηpivot = 0; for the R case, the pivot model is
defined using αpivot = 1.0, βpivot = 0.3, and ηpivot = 10. These
3.4. Unloading stiffness
variables are described in detail in Refs. [5,6,14].
Measured load–deformation hysteresis data for reinforced
concrete members have shown that the unloading stiffness 3.5. Viscous damping
decreases as the maximum deformation increases [12]. The
unloading stiffness has been defined in relation to a stiffness- When inelastic behavior is explicitly considered, damping is
reducing exponent, γ , as implemented by the Takeda hysteresis meant to represent the energy dissipation associated with pre-
model [13]: yield stages of response and therefore is usually taken into ac-
γ count by means of a damping factor expressed as a fraction of
θe

Kr = Ke (4) critical damping. Damping forces are assumed to be proportional
θm to the relative velocities of the points where the translational de-
where Kr is the unloading stiffness, Ke is the secant stiffness to the grees of freedom (at floor levels) are defined. The damping matrix
yield point, θe is the effective yield rotation including bond–slip is assumed to be a linear combination of the mass and stiffness
effects, and θm is the maximum rotation attained. The values of Kr matrices, as expressed in Eq. (1). This means that if damping is
2934 A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939

a b
Fig. 1. Test structures, (a) MF1 and MF2 and (b) FNW.

defined as stiffness proportional then the damping matrix is up- Generally, test runs of a given specimen included repetitions
dated every time an element cracks or yields, whereas if damping is of free vibration tests, to determine low-amplitude natural fre-
defined as mass proportional then it remains unchanged through- quencies, followed by earthquake simulation. The sequence was
out the analysis. In addition, stiffness-proportional damping im- repeated with the intensity of the earthquake simulation being in-
plies increased damping values on modes of vibration other than creased in successive runs. Only the first two runs are considered.
the fundamental mode whereas mass-proportional damping im- These runs targeted a roof displacement of approximately 1% the
plies reduced damping values on the higher modes. frame height for run 1 and 2% for run 2. The base motions were
Four cases for the amount and type of viscous damping are con- patterned after El Centro 1940 NS with the time scale compressed
sidered. Two cases use mass-proportional damping at 2% and 5% of by a factor of 2.5 to obtain realistic ratios between input frequency
critical damping (α2 , α5 ) and two cases use stiffness-proportional contents and natural frequencies of the specimens. Recorded data
damping at 2% and 5% of critical damping (β2 , β5 ). The constants α include displacements relative to the ground and absolute acceler-
and β (Eq. (1)) are determined using Eq. (2) with the fundamental ations, all measured at beam lines.
frequency of vibration always based on uncracked section proper- Specimens were cast using small-aggregate concrete with a
ties. nominal compressive strength of 35 MPa (5 ksi) and a modulus of
Table 1 presents a summary of the five modeling parameters elasticity of about 21,000 MPa (3000 ksi). All members were rein-
considered. Four parameters adopt two values and a fifth param- forced by steel wire gage No. 13 and 16, longitudinally and trans-
eter adopts four values, giving at most a total of 64 combinations versely, respectively. Longitudinal reinforcement with a nominal
per structure for a single ground motion and for each program of yield stress of 350 MPa (50 ksi) was continuous through all joints
analysis. and extended into exterior joint stubs for anchorage. Transverse re-
inforcement provided by rectangular spirals prevented shear fail-
4. Structures and ground motions ure before flexural failure. Beam–column joints were reinforced
with helical spirals to prevent joint failures. More details about the
Nonlinear dynamic analyses are performed on three small-scale fabrication of the specimens are found in Refs. [15–17]. Tabulated
test structures responding to two separate shake-table base mo-
data in Ref. [9] describe all member properties in relation to the
tions and a real building subjected to the ground motions recorded
modeling parameters presented in Section 3.
at its base during the 1994 Northridge earthquake. The analyses
incorporate the combinations of parameter values presented in
Section 3. 4.2. Instrumented building: HEW and HNS

4.1. Test structures: MF1, MF2, and FNW The seven-story Holiday Inn building in Van Nuys (California)
was built in 1966 on the northeast side of the Los Angeles basin
The tests were performed using the University of Illinois earth- at ±6 km (4 miles) from the epicenter of the 1994 Northridge
quake simulator [2]. Geometries of the specimens are shown in earthquake. The structure is essentially symmetrical in elevation
Fig. 1. Structures MF1 [15], MF2 [16] and FNW [17] were composed and in plan about both axes (Fig. 2), except for light framing
of two planar frames working in parallel. The ten-story structures, members supporting the stairway and elevator openings on the
MF1 and MF2, had identical geometries except for the discontin- southwest end. The nonlinear dynamic analysis of the building
ued beam at the first story in MF2. The nine-story structure, FNW, is performed using two-dimensional models for each direction
had a tall first story. Nominal cross sectional dimensions of beams of analysis: the East–West frames, or structure HEW, and the
and columns were identical for all three frames, with all columns North–South frames, or structure HNS. Given the symmetrical
stiffer and stronger than the beams. The heavy base girders and layout and the presence of perimeter moment frames, the merits
the specimen-to-simulator connections were designed to simulate of a 2D model have been explored and proven satisfactory [3].
a fixed-base condition. The story weights were nominally 4.45 kN The floor system is a reinforced concrete flat plate that is
(1 kip) with the exception of the first story in MF2 where nearly 220 mm (8.5 in.) thick at typical floors, 250 mm (10 in.) at the first
1/3 of the weight was removed. The weights were transferred di- floor, and 200 mm (8 in.) at the roof. Spandrel beams, 410 mm
rectly to the column centerlines such that each column carried an (16 in.) wide in the EW direction and 360 mm (14 in.) in the
equal fraction of the story weight. NS direction, are typically 570 mm (22.5 in.) deep, except at the
A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939 2935

Imaginary
to top of slab (Roof)

(Rc, Ic)

(Base) Err
Elevation EW Elevation NS or
(Section A–A) (Section B–B)

d
Calculate
(Rm, Im)

d
sure
NORTH Mea

Plan
Real

Fig. 2. Holiday Inn building in Van Nuys, California. Fig. 3. FDE representation.

first floor and roof with depths of 760 mm (30 in.) and 560 mm domain. For each frequency, the real and imaginary components
(22 in.), respectively. Seismic design of the building was based on of the complex number can be thought of as x–y coordinates
the assumption that the code-specified lateral forces were resisted (Argand diagram) defining the vectors shown in Fig. 3. The FDE in-
by the combined action of the interior slab–column frames and dex is based on the error vector (Fig. 3) normalized by the sum of
exterior beam–column frames. The interior column cross sections the vector magnitudes of the measured and calculated signals:
are 510 mm (20 in.) square in the first story and 460 mm (18 in.) f2
square above. Exterior columns are 360 × 510 mm (14 × 20 in.) for
q 2 2
P
RMi − RCi + IMi − IC i
the full building height with the 510 mm (20 in.) dimension along i=f1
the NS direction. Although the column longitudinal reinforcement FDE = (5)
f2 q  q 
R2Mi + IM
2
R2Ci + IC2i
P
ratios range between one and three percent of the column gross +
i
area, the transverse reinforcement, spaced at 300 mm (12 in.), is i=f1
noncompliant with modern special moment frame requirements. where RMi and IMi are the real and imaginary components of the
All concrete is normal-weight with the first-story columns measured signal at frequency i; RCi and ICi are the real and imag-
having a specified compressive strength of 35 MPa (5 ksi). The inary components of the calculated signal at frequency i; f1 and
first-level floor slab and columns between levels 1 and 2 were f2 are the starting and ending frequencies for summation. In this
built with 28 MPa (4 ksi) concrete. All other concrete above grade study, frequencies f1 and f2 are defined as a function of the fun-
had a specified concrete strength of 21 MPa (3 ksi). Columns are damental period, T1 , of the uncracked structure, using 1/(4T1 )
reinforced with steel having a specified yield of 410 MPa (60 ksi), and 1/(0.1T1 ), respectively. This range is expected to capture the
while beams and slabs used 280 MPa (40 ksi) steel. The foundation most relevant frequency content of the signals representing inelas-
system consists of groups of cast-in-place reinforced concrete tic structural response. The duration of the response history over
friction piles with pile caps connected by grade beams. which the index is calculated is set to approximately 30T1 .
The structure damage due to the 1994 Northridge earthquake
was serious. The perimeter frames had extensive cracking of 6. Analysis cases and results
concrete related to shear and bond stresses, especially at columns.
A total of 16 accelerometers located at the roof, fifth, second, first, To identify the combinations of modeling parameters that
and ground levels recorded motions in the EW, NS, and vertical lead to a realistic response, the FDE index is determined for the
directions. The recorded horizontal peak ground accelerations roof displacement history obtained from each of the nonlinear
were 0.45 g (EW) and 0.42 g (NS). The building experienced roof dynamic analysis cases considered. First, trends are identified in
displacements of about 1.3% the building height. the data corresponding to the test structures; and second, the
Tabulated data are included in Ref. [9] describing all member Holiday Inn building is analyzed to verify if the combinations that
properties in relation to the modeling parameters presented in give satisfactory calculated-to-measured correlations for the test
Section 3. structures also work for the full-scale building.
Program LARZ was used in all three test structures (MF1, MF2,
5. FDE index and FNW) and in the Holiday Inn building (HEW and HNS) for all
64 possible parameter value combinations (see Table 2). Programs
The FDE index [18] measures the correlation between two PERFORM 3D and SAP 2000 were also used to model the same
waveforms. The index is used herein to help identify the combi- structures but with a reduced number of cases due to program lim-
nations of modeling parameters that lead to the best correlations itations. When using discrete nonlinear springs at member ends,
between measured and calculated roof displacement response for both PERFORM 3D and SAP 2000 cannot appropriately represent
the structures considered. Roof displacement is chosen as the key the U models using a breakpoint before the yield point.
response parameter because it has proven satisfactory for charac- For each analysis program, the FDE index data were sorted
terizing the global and local response of frames having individual for each type of structure and damping and presented in the
elements without abrupt changes in stiffness and with columns form of FDE clocks (Figs. 4–9). An FDE clock [18] is a graphical
proportioned to develop limited yielding [4]. representation resembling a dart board, where the center of the
The FDE index quantifies amplitude and phase deviations be- chart (FDE = 0) indicates a perfect correlation between the
tween two signals and gives a value between 0 and 1, where 0 is measured and calculated response histories. A point near the
for a perfect correlation and 1 is for 180° out of phase. The index periphery (FDE = 0.75) represents a very poor correlation. The
uses the Fourier transform of both the measured and calculated circle is divided into 16 sectors, where each sector represents one
signals to represent them as complex numbers in the frequency of the 16 models after the combination of U /C , H /S, N /R, and T /L.
2936 A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939

Table 2
Analysis cases.
Program Structure Base motion Parameters Cases

LARZ MF1/MF2/FNW run 1/run 2 U /C , H /S , N /R, T /L, α2 /α5 /β2 /β5 3 × 2 × 64 = 384
HEW/HNS 1994 Northridge U /C , H /S , N /R, T /L, α2 /α5 /β2 /β5 2 × 1 × 64 = 128
PERFORM 3D MF1/MF2/FNW run 1/run 2 C , H /S , N /R, T /L, α2 /α5 /β2 /β5 3 × 2 × 32 = 192
HEW/HNS 1994 Northridge C , H /S , N /R, T /L, α2 2 × 1 × 8 = 16
SAP 2000 MF1/MF2/FNW run 1/run 2 C , H /S , N /R, T /L, α2 /α5 /β2 /β5 3 × 2 × 32 = 192
HEW/HNS 1994 Northridge C , H /S , N /R, T /L, α2 2 × 1 × 8 = 16

Fig. 4. FDE clocks, structure MF1, program LARZ. Fig. 6. FDE clocks, structure FNW, program LARZ.

calculated roof displacement is 80% or less than the maximum


measured roof displacement. Note that the α5 models are more
likely to underestimate the roof displacement. For run 1, the low-
est FDE index averages corresponding to the data in Figs. 4–6
include: CHRT-α2 , CHRL-α5 , UHRT-β2 , and UHRL-β5 , with FDE in-
dex averages of 20%, 26%, 19% and 19%, respectively. For run 2, the
only model with FDE indexes below 25% for all test structures was
CHRL-α2 .
FDE clocks corresponding to the calculated response using
PERFORM 3D and SAP 2000 for α2 models are shown in Figs. 7–9.
The trends identified with LARZ were corroborated by data from
PERFORM 3D and SAP 2000: for run 1, model CHRT-α2 was the top
performer, whereas for run 2, model CHRL-α2 averaged the lowest
FDE index.
Roof displacement histories (Figs. 10 and 11), base shear histo-
ries (Fig. 12), and story drift ratios (Fig. 13) for a few of the models
representative of the lowest FDE index averages, indicate a satis-
factory correlation between the calculated and measured response.
There is a tendency for the analytical models to underestimate the
maximum story drift ratios in some stories, especially for the run 2
cases, as shown in Fig. 13 for the first and top stories. This suggests
that the analytical models deviate from the idealized fixed-base
condition and do not properly account for the high-mode effects
Fig. 5. FDE clocks, structure MF2, program LARZ.
in cases where the structures had previously yielded.

6.1. Test structures: MF1, MF2, and FNW 6.2. Instrumented building: HEW and HNS

The goodness-of-fit data for the calculated roof displacement The goodness-of-fit data for the calculated roof displacement
response for all 384 LARZ cases (see Table 2) are summarized in response for all 128 LARZ cases (see Table 2) are summarized in
the FDE clocks shown in Figs. 4–6. The plotted data include a cross the FDE clocks shown in Fig. 14. The FDE clock data suggest that
marker to identify cases where the amplitude of the maximum the U models in LARZ have a better correlation with the measured
A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939 2937

a b
Fig. 7. FDE clocks, structure MF1, (a) PERFORM 3D and (b) SAP 2000.

Fig. 10. Roof displacement histories, MF2 run 1, program LARZ.

a b
Fig. 8. FDE clocks, structure MF2, (a) PERFORM 3D and (b) SAP 2000.

a b
Fig. 9. FDE clocks, structure FNW, (a) PERFORM 3D and (b) SAP 2000.
Fig. 11. Roof displacement histories, test structures run 2, program LARZ.

response when compared to the C models (except for the α5 7. Conclusions


models). The FDE clocks obtained for α2 damping (Fig. 15) with
programs PERFORM 3D and SAP 2000 also support the relatively To identify the modeling assumptions that lead to the best cor-
poor correlation of the C models. This may be an indication that the relation between calculated and measured seismic responses, a
nonstructural components of the building affected the response by series of nonlinear dynamic analyses was performed on three
providing an initial stiffening effect and/or additional damping. small-scale shake-table test structures (MF1, MF2, and FNW) and
The Holiday Inn building had suffered minor damage during the on the orthogonal structural systems (HNS and HEW) of an instru-
1971 San Fernando earthquake but was subsequently repaired to mented seven-story building located in Van Nuys, California. The
a nearly uncracked state. Therefore, a comparison with the test structures represent multibay multistory reinforced concrete mo-
structures subjected to run 1 is appropriate. When comparing the ment frames. The nonlinear dynamic analyses were implemented
data from the analysis of the test structures (run 1) with that of the using three programs (LARZ [4], PERFORM 3D [5] and SAP 2000 [6]).
Holiday Inn building, only two models give an FDE index below 25% Each program was used to represent the influence of five mod-
for all structures: models UHRT-β2 and UHRL-β5 . Roof displace- eling parameters: initial stiffness, bond–slip rotations, post-yield
ment response of these models when applied to HEW and HNS are stiffness, unloading stiffness, and viscous damping. Each of the
shown in Fig. 16, indicating a satisfactory correlation between the modeling parameters is assigned two extreme but plausible val-
measured and calculated data, with only a slight overestimation of ues to help identify the direction of the parameter value that
the maximum displacement. most favorably represents a realistic response. In all, for a single
2938 A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939

a b
Fig. 15. FDE clocks, structures HEW and HNS, (a) PERFORM 3D and (b) SAP 2000.

Fig. 12. Base shear histories, test structures run 2, program LARZ.

Fig. 13. Displacement response envelopes, MF2 run 2, model CHRL-α2 .

Fig. 16. Roof displacement histories, structures HEW and HNS, program LARZ.

consistent correlation with the measured data. The calculated re-


sponse wasprocessed using the FDE clocks, which is an effective
tool for visualizing the influence of multiple parameter values on
the correlation between the calculated response and the measured
response histories.
From the limited number of structures, modeling assumptions,
modeling parameter range of values, and computer programs con-
sidered, the following are concluded (refer to Table 1 for parameter
identification):
(1) For the test structures (MF1, MF2, and FNW) subjected to
run 1, the best models for each type of damping were: CHRT-α2 ,
CHRL-α5 , UHRT-β2 , and UHRL-β5 .
These models are derived by the following three rules: (i) use
10% post-yield stiffness ratio and an unloading stiffness parameter
γ of 0.6 (i.e. use models with H and R); (ii) for models using un-
cracked initial stiffness select stiffness-proportional damping and
for models using cracked initial stiffness select mass-proportional
damping (i.e. use U models with β or C models with α ); and
Fig. 14. FDE clocks, structures HEW and HNS, program LARZ. (iii) for models where bond–slip deformations are based on devel-
opment lengths of 20 bar diameters use 2% damping and for de-
structure and ground motion, up to 64 analysis cases were consid- velopment lengths of 40 bar diameters use 5% damping (i.e. use T
ered per program. The Frequency Domain Error (FDE) index [18] models with 2% damping or L models with 5% damping). Note that
helped identify the analytical models having the best and most constant damping (mass proportional) works well with the softer
A. Lepage et al. / Engineering Structures 32 (2010) 2931–2939 2939

C models and that higher values of damping work well with the References
softer L models.
(2) For the test structures (MF1, MF2, and FNW) subjected to [1] American Society of Civil Engineers/Structural Engineering Institute. Seismic
run 2, model CHRL-α2 outperformed all others. This model may be evaluation of existing buildings. ASCE/SEI 31-03. Reston (VA); 2003.
[2] Sozen MA, Otani S, Gulkan P, Nielsen NN. The University of Illinois earthquake
viewed as a logical outcome from the rules given for run 1 because simulator. In: Proc. of the fourth world conference on earthquake engineering.
before run 2 the structures had already yielded and experienced vol. 3. 1969. p. 136–50.
some damage. The change from T to L is justified because the [3] Browning JP, Lepage A. Discussion of nonlinear analyses of an instrumented
structure damaged in the 1994 Northridge earthquake (by Li, Y, Jirsa, JO).
structure is effectively softer.
Earthquake Spectra 1999;15(1):175–9.
(3) For the full-scale Holiday Inn building (structures HNS and [4] Saiidi M, Sozen MA. Simple and complex models for nonlinear seismic
HEW) two of the best models were consistent with the rules given response of reinforced concrete structures. Struct. research series, no. 465.
above in conclusion (1): UHRT-β2 and UHRL-β5 . Univ. of Illinois at Urbana-Champaign; 1979.
In general, the models derived with the first two rules of [5] Computers and Structures, Inc. PERFORM 3D: nonlinear analysis and per-
formance assessment for 3D structures. Version 4. Berkeley (CA); 2006.
conclusion (1) had satisfactory accuracy when representing the [6] Computers and Structures, Inc. SAP 2000: static and dynamic finite element
global measured response of the test structures for both run 1 and analysis of structures. Nonlinear Version 14.0.0. Berkeley (CA); 2009.
run 2. Calculated roof displacement, base shear, and overturning [7] Newmark NM. A method of computation for structural dynamics. J Eng Mech
moment histories successfully tracked the measured response. Div, ASCE 1959;85(3):69–86.
[8] Wilson EL. A new method of dynamic analysis for linear and nonlinear systems.
Although the models for the test structures in run 1 were able to Finite Elem Anal Des 1985;1(1):21–3.
satisfactorily represent the local measured response, indicated by [9] Hopper MW. Analytical models for the nonlinear seismic response of
story drift ratios, the models for run 2 were not as accurate. This reinforced concrete frames. M.S. thesis. University Park (PA): The Pennsylvania
is possibly due to limitations in the analytical models to properly State Univ.; 2009.
[10] Park R, Paulay T. Reinforced concrete structures. John Wiley and Sons, Inc.;
account for high-mode effects.
1975. p. 11–5.
The findings from this study suggest that a valuable contribu- [11] Sozen MA. Seismic behavior of reinforced concrete buildings. In: Bozorgnia Y,
tion to practicing engineers would be to have developers of struc- Bertero VV, editors. Earthquake engineering: from engineering seismology to
tural software incorporate moment frame model templates with performance-based engineering. Boca Raton (FL): CRC Press; 2004 [chapter
13].
pre-assigned nonlinear springs. The default spring properties may
[12] Otani S. Hysteresis models of reinforced concrete for earthquake response
be based on conclusion (1) for an undamaged structure and on analysis. J Fac Eng Univ Tokyo 1981;36(2):125–59.
conclusion (2) for a previously damage structure. These proper- [13] Takeda T, Sozen MA, Nielsen NN. Reinforced concrete response to simulated
ties would set the stage for a more reliable evaluation of the rel- earthquakes. J Struct Div, ASCE 1970;96(12):2557–73.
ative merits of several preliminary designs before the final design [14] Dowell RK, Seible F, Wilson EL. Pivot hysteresis model for reinforced concrete
members. ACI Struct J 1998;95(5):607–17.
is reached. [15] Healey TJ, Sozen MA. Experimental study of the dynamic response of a ten-
Research is needed to develop more definite recommendations story reinforced concrete frame with a tall first story. Struct. research series,
of parameter values following statistical analysis of a larger num- no. 450. Univ. of Illinois at Urbana-Champaign; 1978.
ber of structures and a more refined grid of parameter values. [16] Moehle JP, Sozen MA. Earthquake-simulation tests of a ten-story reinforced
concrete frame with a discontinued first-level beam. Struct. research series,
no. 451. Univ. of Illinois at Urbana-Champaign; 1978.
Acknowledgements [17] Moehle JP, Sozen MA. Experiments to study earthquake response of R/C
structures with stiffness interruptions. Struct. research series, no. 482. Univ.
The writers are grateful to Prof. Mete Sozen (Purdue University) of Illinois at Urbana-Champaign; 1980.
for sharing the data from the University of Illinois earthquake [18] Dragovich JJ, Lepage A. FDE index for goodness-of-fit between measured and
simulator. calculated response signals. Earthq Eng Struct Dyn 2009;38(15):1751–8.