Experimental Assessment of the Seismic Response of a Base-Isolated Building Through a Hybrid Simulation Technique

Hybrid shake table testing method: Theory, implementation and
application to midlevel isolation
Andreas H. Schellenberg1
, Tracy C. Becker2,
*,†
and Stephen A. Mahin1
1
Department of Civil and Environmental Engineering, University of California, Berkeley, CA, USA
2
Department of Civil Engineering, McMaster University, Hamilton, Canada
SUMMARY
A 2 m by 6 m unidirectional shake table was constructed at the University of California, Berkeley and combined
with a real-time hybrid simulation system creating a hybrid shake table. A series of tests were carried out to exam-
ine the viability of real-time hybrid simulation techniques to perform experimental simulations of buildings with
midlevel seismic isolation. The isolation system and superstructure were physically tested on the table while the
portion of the building below the isolation plane was numerically modeled. OpenFresco was used to interface
the numerical model with the control system. The isolated superstructure consisted of a two-story steel moment
frame on six triple friction pendulum bearings, which exhibit significant nonlinear velocity-dependent behavior,
necessitating real-time testing. Shear building models with a range of periods were used to represent the portion
of the building below the isolation plane. Increasing the number of degrees of freedom increased the control dif-
ficulty as higher modes were excited in the numerical model because of experimental errors caused predominantly
by feedback noise and table tracking. Nonetheless, the results illustrate that hybrid shake table tests are indeed an
economical and reasonably accurate method to assess the seismic behavior of midlevel isolation systems installed
in a range of building configurations. Results showed that midlevel isolation was beneficial for the superstructure
and, to a smaller extent, the substructure. However, to achieve maximum benefits, it is recommended that the ef-
fective period of the isolation system be sufficiently longer than the period of the substructure. Copyright © 2016
John Wiley & Sons, Ltd.
Received 23 October 2015; Revised 20 June 2016; Accepted 29 June 2016
KEY WORDS: hybrid shake table; midlevel isolation; experimental control; real-time hybrid; friction pendulum
1. INTRODUCTION
In many countries, seismic isolation technology is increasingly being used to improve the performance
of buildings and bridges and to avoid significant structural damage during ground shaking by concen-
trating the seismic deformations in the isolators while simultaneously providing supplemental energy
dissipation. For various reasons, the isolation plane is no longer always placed at the base of the build-
ing. Over the last decade, midlevel seismic isolation systems, where the isolation plane is placed higher
up in the building instead of at the base, have been studied [1–5], and several tall midlevel seismically
isolated building projects have been designed and constructed, especially in Japan [6,7]. Midlevel
isolation retains the basic principles of traditional seismic isolation; however, the seismic moat required
when the isolation layer is placed at the base of the building is no longer necessary, potentially decreas-
ing the cost of the overall structural system. In addition, midlevel seismic isolation systems can provide
more architectural flexibility, needed in multi-use applications where office space is combined with
condominium or hotel space, because transitions between different structural systems are common in
these types of applications. In this same vein, midlevel isolation can facilitate the addition of new
*Correspondence to: Tracy C. Becker, Department of Civil Engineering, McMaster University, Hamilton, Canada.
†
E-mail: tbecker@mcmaster.ca
STRUCTURAL CONTROL AND HEALTH MONITORING
Struct. Control Health Monit. (2016)
Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/stc.1915
Copyright © 2016 John Wiley & Sons, Ltd.

stories on top of existing buildings while minimally increasing seismic demands on the existing build-
ing structure [8].
Because the dynamics of midlevel seismically isolated buildings are dependent on the interaction
between the substructure and superstructure as well as the isolation system, experimental validation
of a range of configurations becomes impractical by standard methods because of the large cost
associated with carrying out multiple large-scale experimental tests. In order to efficiently test the in-
teraction between an isolated superstructure with a range of substructure configurations, a specialized
real-time hybrid simulation (RTHS) method, called hybrid shake table testing, was adopted.
Hybrid simulation, sometimes referred to as pseudo dynamic testing, is an experimental method in
which the most critical portion of a system that exhibits uncertain nonlinear behavior is experimentally
tested in a laboratory. The remainder of the structure, which exhibits a behavior that can be predicted
numerically with confidence, is simultaneously analyzed in a finite element software application. In
this experimental testing technique, a simulation is executed based on a step-by-step numerical solu-
tion of the governing equations of motion for a model formulated considering both numerical and
physical components of a structural system. Originally, the hybrid simulation testing method was de-
veloped by Takanashi et al. [9] as an alternative experimental technique to shake table testing. Early
pseudo dynamic tests were executed at very slow speeds because of limitations in hydraulic actuator
control systems and computing performance related to the analysis of the numerical portions of a
structure. This meant that loading rates during a hybrid simulation were much slower than the com-
puted velocities, and rate effects and inertial effects in the experimental portion of the structure could
therefore not be captured. The first RTHS was achieved in 1992 [10], and research has since focused
on the development of delay compensation techniques [11–14] and real-time compatible integration
schemes [15–17] to reduce instability issues and improve testing accuracy. With these developments,
it became possible to test experimental specimens at loading rates that are equal to computed velocities
and accelerations. It is now possible to test at loading rates that are faster than real time, which is nec-
essary if experimental specimens are reduced-scale and similitude laws need to be considered during
testing.
Over the last decade, the introduction of high-performance devices to improve structural behavior of
components and/or systems under extreme loading conditions has increased notably. Most of these
devices, such as base isolation bearings, energy dissipation devices and high-performance materials,
exhibit intrinsic rate-dependent behavior. Thus, it is of utter importance that testing of structural sys-
tems incorporating such devices is performed at the correct loading rates. Real-time hybrid shake table
testing provides an attractive, versatile, and cost-effective alternative to traditional shake table testing
by applying RTHS principles to drive a shake table. Based on the same fundamental idea as in tradi-
tional hybrid simulation, the most critical portion of a structure that is not well understood can be
experimentally tested on a shake table, while the well understood remaining portions of the structural
system are modeled and analyzed numerically. In the case that the experimentally tested portion of the
structure is a non-destructive specimen, such as an isolation or viscous damper system, the hybrid
shake table testing method provides the unique ability to perform efficient parameter studies by chang-
ing the properties and behavior of the numerically modeled portion of the structural system.
Similar to RTHS, it is crucial in real-time hybrid shake table testing that systematic experimental
errors, such as time delays introduced by the control and data acquisition systems, are being corrected
for using delay compensation techniques to avoid instability and accuracy problems. However, typi-
cally, that task is even more formidable because shake table systems inherently have larger mass
and more frictional resistance, and it is thus more difficult to achieve good tracking performance with
minimal delays. Hybrid shake table tests, where the analytical portion of the hybrid model is placed
below the shake table platform, have been studied by several researchers [18–21], starting with
Igarashi et al. [22]. Research is also being conducted to develop methods that allow analytical subas-
semblies to be placed above the experimental specimen tested on the shake table [23]. Others have
even investigated RTHS methods where a shake table is being combined with additional dynamic ac-
tuators to conduct tests where analytical portions of the structure are placed above and below the ex-
perimental specimen tested on the shake table [24]. However, so far, many of these investigations
are proof of concept studies that have mostly used small and light shake table platforms with small-
scale experimental specimens.
A. H. SCHELLENBERG, T. C. BECKER AND S. A. MAHIN
Copyright © 2016 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2016)
DOI: 10.1002/stc

This paper applies the hybrid shake table testing method to a large experimental substructure with
fairly large specimen mass. It also demonstrates how hybrid shake table testing can be utilized to per-
form parameter studies efficiently and economically. To investigate the behavior of a midlevel seismi-
cally isolated structure, a unidirectional shake table was constructed and then connected with an
existing RTHS system that was modified to conduct hybrid shake table tests. For this study, friction
pendulum bearings were selected for the midlevel isolation system. These bearings exhibit a strong rate
dependency, especially at low velocities, and, thus, it was essential that the hybrid testing was per-
formed in real time to accurately capture the behavior of the entire structural system.
2. THEORY ON HYBRID SHAKE TABLE TESTING METHOD
Before discussing the modifications to the hybrid simulation testing approach that are necessary to per-
form real-time hybrid shake table tests, it is helpful to review the theory and the fundamental equations of
motion that are being solved during a hybrid simulation when tests are executed at different loading rates.
For slow tests, it is imperative that the physically tested portions of the structure do not exhibit any veloc-
ity or rate-dependent behavior, unless such dependency is compensated for in the analytical part of the
hybrid model. In addition, it is important that an uninterrupted execution, meaning a continuous actuator
motion, is maintained to avoid force relaxation and actuator stick-slip difficulties [25]. On the other hand,
for rapid hybrid simulations, it is crucial that the inertia and damping force contributions generated by the
physical portions of the hybrid model are correctly accounted and compensated for. Moreover, it is
important to recognize that for such tests high-force dynamic actuators with large accumulators and hy-
draulic pumping systems are required and that the accurate control of these actuators becomes much more
difficult due to inertia force feedbacks. This is especially true for hybrid shake table tests where simulator
platforms are typically quite heavy and large inertia forces are generated. Hence, advanced control strat-
egies and delay compensation techniques need to be employed when testing speeds are being increased.
2.1. Numerical formulation
The equations of motion to be solved during a hybrid simulation take on different forms depending on
the execution speed of the test [26]. It is important to note that in all cases, except for the hybrid shake
table case, the equations of motions are expressed in terms of response quantities relative to ground.
For slow hybrid tests, where negligible inertia forces are generated in the physical subassemblies, they
can be expressed in the following manner.
M €
Uiþ1 þ C _
Uiþ1 þ PA
r Uiþ1; _
Uiþ1

þ PE
r Uiþ1
ð Þ ¼ Piþ1 P0;iþ1 (1)
where the terms with superscripts A are assembled from the analytical portions of the structure and the
terms with superscripts E are assembled from the experimental portions of the structure. As can be seen
from Equation (1), the mass and the viscous damping of the entire structure are treated numerically be-
cause the experimental resisting force vector PE
r does not include these contributions because of the
slow execution of the test. Hence, the mass matrix M and viscous damping matrix C can directly be
assembled from all the node and element contributions of both the analytical portions as well as the
experimental portions of a structure. On the right-hand side of the equations of motion, Pi + 1 are the
externally applied nodal loads and P0, i + 1 are element forces due to externally applied loads such as
body forces, boundary tractions, initial strains, and initial stresses.
Once the execution speed reaches real time, meaning that the experimental subassemblies are
loaded with the actual, calculated velocities and accelerations (advanced control and delay compensa-
tion strategies need to be employed here to guarantee this), an alternative, preferred form of the equa-
tions of motion can be employed.
MA €
Uiþ1 þ CA _
Uiþ1 þ PA
r Uiþ1; _
Uiþ1

þ PE
r Uiþ1; _
Uiþ1; €
Uiþ1

¼ Piþ1 P0;iþ1 (2)
where MA
and CA
are the mass and viscous damping matrices assembled from only the analytical sub-
assemblies and PE
r is the resisting force vector assembled from the experimental subassemblies, which
includes the inertia and damping forces of these physical portions of the structure. It is important to
HYBRID SHAKE TABLE TESTING FOR MIDLEVEL SEISMIC ISOLATION
Copyright © 2016 John Wiley Sons, Ltd. Struct. Control Health Monit. (2016)
DOI: 10.1002/stc

notice that the experimental resisting force vector is entirely assembled from forces measured by load
cells, which also capture the dynamic force contributions from the experimentally tested specimen.
However, it is crucial that inertia and energy dissipation forces generated by the testing apparatus
are not part of the forces fed back to the hybrid model, but only the dynamic forces from the test spec-
imen. If dynamic or friction forces from the testing apparatus contaminate the load cell measurements,
they need to be compensated for in real time using observer techniques.
Finally, in hybrid shake table testing where real-time hybrid simulation is utilized to drive a shake
table and the entire experimental subassembly is tested on the simulator platform, the experimental
resisting force vector PE
r is no longer a function of the relative response with respect to ground. Instead,
total (or absolute) response quantities (structural response plus ground response) need to be sent to the
control system driving the shake table so that inertia forces in the test specimen are correctly generated.
Therefore, in the case of a hybrid shake table test, the entire equations of motion are written using total
displacements, velocities, and accelerations.
MA €
Ut;iþ1 þ CA _
Ut;iþ1 þ PA
r Ut;iþ1; _
Ut;iþ1

þ PE
r Ut;iþ1; _
Ut;iþ1; €
Ut;iþ1

¼ Piþ1 P0;iþ1 (3)
where PE
r includes the inertia and damping forces of the test specimen excited by the total accelerations
€
UE
t iþ1 and total velocities _
UE
t iþ1. It is important to note that, unlike the previously presented equations
that are expressed relative to the ground, this modiﬁed equation (3) also includes the degrees of free-
dom at the supports of the structure. For earthquake excitations, this means that no external loads are
applied at any of the degrees of freedom so the external load vectors Pi + 1 and P0, i + 1 are zero. Instead,
seismic motions are directly imposed at the support degrees of freedom in terms of ground displace-
ments, ground velocities, and ground accelerations. In OpenSees, this approach is readily available
through the existing multi-support excitation load pattern capabilities.
2.2. Integration algorithm
To solve Equation (3) during a hybrid shake table test and advance the solution in time, direct integra-
tion methods based on Newmark’s method are employed. However, for real-time hybrid simulations,
the traditional, implicit Newmark method is not well suited. For real-time hybrid shake table testing,
direct integration methods should (1) make as few function calls, meaning resisting force acquisitions,
as possible to reduce computational cost; (2) be at least second-order accurate and possess favorable
experimental error propagation behavior to minimize error accumulation; (3) provide some adjustable
amount of algorithmic (numerical) energy dissipation to suppress the excitation of higher mode effects
because of experimental errors in the force feedbacks; and (4) execute a constant number of function
calls, which produce uniform displacement increments within each integration time step in order to
be real-time compatible. This ﬁnal requirement guarantees that not only displacements, but also veloc-
ities and accelerations that are used to drive the shake table are smooth and consistent.
Three direct integration methods, namely, the Explicit Newmark method, the Explicit Generalized
Alpha method, and the Generalized AlphaOS method, were employed for the hybrid shake table
tests described in this paper. All of these methods are real-time compatible and perform only one
function call per integration time step. In addition, all three methods use a constant Jacobian that
does not require any updates during the solution process. Lastly, the Explicit Generalized Alpha
method and the Generalized AlphaOS method provide adjustable algorithmic damping that is desir-
able in hybrid shake table testing. For the two explicit integration methods, stability was not a con-
cern as the integration time step size was dt= 5/1024 s, and the shortest natural period of the entire
structure was T5 = 0.059s, so testing occurred well below the stability limit of these explicit integra-
tion methods.
3. IMPLEMENTATION OF THE HYBRID SHAKE TABLE TESTING METHOD
The hybrid shake table test was composed of an experimentally tested, seismically isolated two-story
moment frame that represented the superstructure and a numerically simulated, lumped-mass
shear building that represented the substructure. A unidirectional shake table was constructed for the
DOI: 10.1002/stc

real-time loading. The shake table, shown in Figure 1, consists of a large steel platform isolated on low-
friction linear bearings. The friction in the rails of the linear bearings was assessed to be less than 10%.
The steel platform is 5.8m long by 2.0m wide. The platform is supported at six points, directly below the
locations of the six seismic isolators. The shake table is driven by a dynamic MTS actuator with +/0.5m
stroke and +/667kN force capacity. With this actuation, the table can realize velocities of up to 1.0m/s
and accelerations of up to 1.0 g. The maximum payload capacity at these accelerations is approximately
445 kN. The table is controlled using an MTS-493 real-time controller with the Structural Test System
software interface. The digital controller provides closed-loop proportional–integral–derivative (PID),
derivative feedforward (FF), and differential pressure (Delta-P) control capabilities.
To avoid instability and achieve good accuracy, the tracking performance over the frequency range
of interest needs to be excellent. The relevant frequency range for these tests was approximately 0.5 to
8 Hz. The frequency range remains fairly low due to the influence of the midlevel seismic isolation sys-
tem on the overall dynamic behavior of the structure. FF control was used to compensate for delays due
to the transfer system dynamics. In FF control, the derivative of the command signal is multiplied by a
user-defined gain value and then added to the servovalve command. In addition, Delta-P control was
utilized to suppress resonance phenomena around the oil column frequency that lead to inaccurate
tracking in that frequency band and introduce significant inertia forces, which are measured by the load
cells under the test specimen and then fed back into the hybrid model. Force oscillations due to exper-
imental errors can destabilize a hybrid shake table test. It is important to note that many other delay
compensation techniques and advanced control strategies exist in the literature, which can be beneficial
for hybrid shake table testing. For future testing, it is recommended that the methods developed by [27]
and [28] are investigated to improve tracking performance, especially in the high-frequency range. This
is particularly important for the hybrid shake table testing of non-isolated structures that may have
many relevant frequencies well above 8 Hz and potentially exhibit significant higher mode contribu-
tions to the overall response.
The basic outline of the hybrid shake table test is shown in Figure 2. Earthquake excitation, in terms
of imposed ground displacements, ground velocities, and ground accelerations, is input into the base of
the numerical substructure, modeled in OpenSees [29]. The absolute displacement at the top of the nu-
merical substructure is the target displacement, which is sent to OpenFresco [30]. OpenFresco serves as
the middleware that is used to interface the numerical substructure with the experimental superstructure
through the transfer system, consisting of control and data acquisition systems. A predictor–corrector
algorithm [26], running on a real-time digital signal processor (xPC-target), is then used to bridge the
difference between the analysis time step size (5/1024 s) and the smaller control system time step size
(1/1024 s). To synchronize the nondeterministic execution of the OpenSees/OpenFresco analysis with
the determinist execution of the control system, the predictor–corrector algorithm performs the follow-
ing tasks: (1) while the analysis software solves the equations of motion for the new target displace-
ment, the pc-algorithm generates command displacements based on polynomial forward prediction;
(2) once the new target displacement has been received, the pc-algorithm switches into the correction
mode where it generates command displacements driving the actuator response towards the new target
displacement; (3) if the new target displacement is not received within 60% of the simulation time step
(a) (b)
Figure 1. Unidirectional shake table.
DOI: 10.1002/stc

size (3/1024 s), the pc-algorithm gradually slows down the command displacements until the new tar-
get displacement is received. So in order to achieve a real-time execution of the hybrid shake table test
without any slowdowns, the analysis software needs to compute a new target displacement in less than
3/1024 s. Predictor–corrector states were monitored and recorded for all hybrid tests. No slowdowns
were encountered, and on average, the pc-algorithm performed 1/1024s of prediction and 4/1024 s
of correction meaning that real-time execution was achieved for all the tests. The resulting displace-
ment of the table and the shear force under the physical specimen, recorded using load cells under each
of the isolators, are measured. The measured displacement is fed back into the controller and predictor–
corrector algorithm while the measured shear force is fed back into the predictor–corrector algorithm
and then the numerical OpenSees model for the next analysis time step. To minimize communication
delays among the three machines (OpenSees/OpenFresco analysis machine, xPC-target digital signal
processor, and MTS-493 controller), a SCRAMNet (Shared Common RAM Network) ring is
employed.
4. MIDLEVEL ISOLATION SET-UP
To investigate a variety of midlevel isolation configurations using hybrid simulation, a single isolated
superstructure was designed to be physically tested while the substructure was varied numerically. This
approach provides a very cost-effective experimental testing method. The physical superstructure is a
1/3rd scale steel moment frame isolated on six triple friction pendulum bearings, shown in Figure 3
with properties given in Table I. The ultimate displacement capacity of the model-scale isolators is
163mm. The effective period of the bearings is 1.32s at 100 mm, and the post-yield second sliding
stage period is 1.87s. At full scale, the effective period of the isolator would be 2.29s, and the second
sliding stage period would be 3.24s.
The superstructure frame is two stories above the isolators. The first and second story heights are 1.7
and 1.5 m, respectively. The frame has two bays in the direction of loading with a span of 2.44m. The
frame was constructed using the NEES Reconfigurable Platform for Earthquake Testing (REPEAT
frame), which uses clevises with replaceable steel coupons at locations of expected plastic hinges.
The frame was loaded with additional concrete blocks to reach a total weight of 380kN so that the
pressure on the sliding surfaces in the isolators would be large enough (~10N/mm2
on the outer sliding
surface of a corner isolator) to ensure stable friction behavior. The frame was designed to yield; at the
same time, the bearings reached their ultimate displacement capacity, at roughly 30% g. The fixed base
periods of the superstructure are approximately 0.43 and 0.14s, which were found by matching a
numerical model to the experimental data.
Figure 2. Data flow in the hybrid shake table test.
DOI: 10.1002/stc

Multiple substructure configurations were used to examine the ability of the hybrid model to be
tested in real time with various properties for the numerical portion. Two main substructure configura-
tions were used: a one-story and a three-story building, shown in Figure 4. Both configurations used
simple numerical shear building and lumped-mass modeling assumptions, the properties of which
are listed in Table II. The one-story substructure, or Model A, was assigned a weight slightly larger
than the total weight of the superstructure. The period of the Model A substructure was changed from
0.125 to 1.0 s in the tests. For the three-story substructure, or Model B, each floor had a weight approx-
imately equal to the bottom floors of the superstructure. The period of the Model B substructure was
changed from 0.25 to 1.0 s in the tests. Both models were assigned 3% equivalent viscous damping
and were assumed to remain linear elastic. For Model A, damping was assigned as initial stiffness pro-
portional damping anchored at the first mode period. For Model B, damping was assigned as Rayleigh
damping anchored at the first and third mode periods.
(a)
(b)
(c)
Figure 3. (a) Isolated frame superstructure installed on the shake table (b, c) triple friction pendulum bearing and
backbone curve.
Figure 4. Hybrid model configurations, above the isolation layer is the physical specimen, below is the numerical
substructure: (a) Model A; (b) Model B.
Table I. Triple friction pendulum properties.
Surfaces 1, 2 Surfaces 3, 4
R 76 mm 473 mm
Din 44 mm 76 mm
Dout 66 mm 229 mm
μ 0.03 0.13
DOI: 10.1002/stc

Two ground motions were used in this study: the fault normal components of Loma Prieta, Gilroy
Array #4 and Superstition Hills, Westmoreland Fire Station. A length scale of 3 was used to match the
scale of the physical specimen. The ground motions were additionally amplitude scaled for use in later
experimental tests when the input displacement would be applied directly to the isolated frame, without
a numerical substructure. The motions were scaled so that in those tests the expected displacement of
the bearings was just within the maximum displacement capacity of the bearings under the maximum
considered earthquake level, or 2% probability of exceedance in 50 years. For this study, scale factors
corresponding to the 50% and 10% in 50 years seismicity levels were used, which were 0.64 and 1.70
for the Loma Prieta motion, and 0.65 and 1.75 for the Superstition Hills motion. The response spectra
for the scaled 10% in 50-year motions are shown in Figure 5.
5. EXPERIMENTAL RESULTS
5.1. Experimental control
Before discussing the behavior of the midlevel isolated buildings, it is important to look at the ability of
the hybrid simulation transfer system to accurately link the numerical and physical portions of the hy-
brid test. Figure 6 shows the tracking indicator histories, which give a measure of the enclosed area in a
synchronization subspace plot where the measured displacement is plotted against the command dis-
placement [31]. A decreasing value indicates a lag in the control resulting in energy being added be-
cause of tracking errors. Tracking indicators for the different substructure periods, input motions, or
hazard levels should not be compared against each other in magnitude; for this, the normalized root-
mean-square (RMS) tracking error at the end of the test is given in Table III. The RMS tracking error,
which provides a measure of accuracy that was achieved in the synchronization of the numerical and
physical portions, is normalized by the full amplitude range of the measured displacement feedbacks.
This allows tracking performance to be compared between tests with varying substructure properties
and ground motion inputs.
The tracking indicators showed lag for all testing, which is a direct result from the controller. The
RMS tracking errors reveal that, in general, the shorter the period of the substructure, the larger the
error between the target and measured displacements. For the Loma Prieta 10% in 50-year hazard level
motion, the 1.0-s period substructures resulted in displacement errors of less than 2% of the maximum
Table II. Numerical substructure parameters.
Floor weight (kN) Story stiffness (kN/m) Period (s) Damping ratio
1 Story — Model A 445 1751 1.0 0.03
445 7005 0.5 0.03
445 28,020 0.25 0.03
445 112,081 0.125 0.03
3 Story — Model B 142 2802 1.0 (0.36, 0.25) 0.03 (Rayleigh)
142 11,208 0.5 (0.18, 0.13) 0.03 (Rayleigh)
142 44,832 0.25 (0.09, 006) 0.03 (Rayleigh)
Figure 5. Response spectra of scaled 10% in 50-year input motions, 5% damped.
DOI: 10.1002/stc

target displacements. Comparatively, for the 0.25-s period substructures, the displacement error
reached 10% of the maximum target displacements. This is most likely due to the use of a
displacement-only control system using linear non-adaptive proportional–integral–derivative tuning
parameters. This type of control delivers good tracking at low frequencies, but is less accurate at higher
frequencies. The FFTs of the acceleration histories for the 10% in 50-year motions are shown in
Figure 7. The systems with larger requested peak input accelerations and higher requested frequency
content (at the top of the substructure) resulted in larger tracking errors. Using more advanced control
strategies such as multi-variable control, acceleration trajectory tracking control, H∞ loop shaping con-
trol algorithms, or sliding mode control techniques may reduce these errors further.
The short-period substructures exhibit similar tracking delays for both Model A and Model B, with
a small increase in lag for Model B. However, for the 1.0-s substructure, there is a significantly larger
lag with Model B. This change in tracking can be seen in the displacement demand histories at the top
of the numerical substructure (which are also the command displacements into the shake table), shown
in Figure 8. For the short-period substructures, the demand is similar for both models. For the 1.0-s
substructure, the demand is increased significantly with Model B, which is most likely a result of
the energy imparted to the system from the larger tracking errors.
Figure 6. Control tracking indicators for the various substructure configurations and ground motion inputs
Table III. Normalized RMS errors at the end of the tests.
Norm RMS error (%) Model A, Model B
Ground motion T = 0.125 s T = 0.25 s T = 0.5 s T = 1 s
Loma Prieta 50/50 years 0.51, n/a 0.57, 0.70 0.34, 0.35 0.16, 0.15
Loma Prieta 10/50 years 0.45, n/a 0.56, 0.59 0.25, 0.27 0.13, 0.12
Superstition Hills 50/50 years 0.32, n/a 0.34, 0.49 0.25, 0.28 0.12, 0.13
Superstition Hills 10/50 years 0.30, n/a 0.27, 0.29 0.19, 0.19 0.09, 0.10
RMS, root-mean-square.
DOI: 10.1002/stc

Examining the FFTs of the acceleration histories (Figure 7), Model B exhibits conspicuous peaks at
high frequencies (14 and 17 Hz) that are not present in Model A. These high-frequency peaks corre-
spond to the fifth and sixth modes of Model B with a 0.25-s substructure and the sixth mode of Model
B with a 1.0-s substructure (Figure 9). It can be observed that modal displacement amplitudes at the
isolation level are large for these mode shapes when the isolators respond in their initial stiffness range
during the unloading and reloading phases. In a real-world application, these modes would not be
excited significantly because of their low mass participation factors; however, in these hybrid shake
table tests, such modes are excited due to the high-frequency nature of the experimental errors, espe-
cially the oil column frequency errors of the shake table actuator, which were not entirely suppressed
by the differential Delta-P control. As a result of the erroneous higher mode effects, the peak floor
accelerations at the top of the substructure were consistently and sometimes significantly larger for
Model B. These higher modes are also apparent in the FFTs of the tracking errors, shown in
Figure 10 for the Loma Prieta 10% in 50-year hazard level motion, which is representative of all
the inputs. However, even when considering these peaks at the substructure frequencies, the FFTs
of the errors are broadband. Thus, while the lag was significant in some cases, the test system was
able to reproduce displacements for the full range of desired frequencies in nearly all the hybrid shake
table tests.
Figure 7. FFTs of the acceleration histories for 10% in 50-year motions.
DOI: 10.1002/stc

5.2. Midlevel isolation
The peak accelerations and story drifts for the entire structure under the 10% in 50-year motions are
plotted in Figures 11 and 12, respectively. Response trends differ for the substructures and superstruc-
tures. As expected, the story drifts in the substructure decreased as its period decreased (stiffness in-
creased). The drift of the top of the substructure relative to the ground of Model B were on the same
order as Model A for the substructures with the same periods. As the effective isolation period ranged
from roughly 1.0 to 1.5 s, it might be assumed that having a long-period substructure would cause large
isolator displacements as the input motion to the superstructure has larger low-frequency content
(Figure 7). However, the peak superstructure responses, including isolator response, are the lowest
in the case of the 1.0-s substructure. In fact, the peak isolation drift is less than the peak drift at the
top of the substructure for the 1.0-s substructure case.
The displacements at the instant of maximum isolation drift are presented in Figure 13. The dis-
placements at the instant of maximum roof displacement are nearly identical. The figure shows typical
isolation performance with deformations concentrated in the isolators. From the mass participation fac-
tors in Figure 9, it is seen that this behavior is the combination of two modes, both of which activate the
isolation layer. For the 1.0-s substructure, both modes incorporate significantly more displacement in
the substructure, which is exhibited in the earthquake displacements.
To show the change in demands if an isolated superstructure is added over an existing building, the
performance of the midlevel isolation systems is compared against the substructure alone in Figure 14.
Of course, before adding extra floors on top of an existing building, the engineer should check axial
load capacities of the existing structure. The existing building is assumed to remain elastic with the
same properties shown in Table II. Adding the isolated superstructure reduces the accelerations for
all values of substructure flexibility and lateral displacements for the majority of cases. However, for
Figure 8. Numerical substructure top displacement histories for the various substructure configurations.
DOI: 10.1002/stc

the most ﬂexible substructure, with Tsub =1.0 s, the story drifts increased under the Superstition Hills
motion at the level just below the added isolated superstructure. The increase was more signiﬁcant
for Model B, where it was on the order of 20%. It is likely that this local increase in drift was exacer-
bated by the increased tracking errors seen for the 1.0-s substructure with Model B (Figure 6),
which, as discussed previously, contributed to the increase in top of substructure displacement
Figure 9. Mode shapes and frequencies with mass participation factors in parentheses for Model B with a 0.25-s
and 1.0-s substructure.
Figure 10. FFT of the error between the commanded and measured displacements input into the physical super-
structure for Loma Prieta 10% in 50 years.
DOI: 10.1002/stc

(shake table inputs) (Figure 8). As this displacement increase is seen in Superstition Hills and not
Loma Prieta, it is ground motion dependent.
Figures 15 and 16 compare the peak responses of the building with the midlevel isolation system
with those of a counterpart elastic fixed base building having the same total number of stories. The
fixed base buildings have the same mass and stiffness of the isolated buildings, but the isolation layer
is removed and 3% Rayleigh damping is assigned to the building as a whole. The building is modeled
as elastic to compare analogous demands. For this comparison, only the drifts of the individual stories
are presented as there is no isolation layer in the fixed base counterparts. With the exception of story
drift for the 1.0-s substructure, having midlevel isolation significantly improves the overall drift
response of the building. While there is reduction in demands up the height of the building, the benefit
of the midlevel isolation is clearly more pronounced for the floors above the isolation layer (Floors 2
and 3 for Model A, and Floors 4 and 5 for Model B).
The floor response spectra, often used for estimating response of nonstructural components, are
shown in Figure 17. Comparing ground motion spectra with the spectra at the top of the substructure
shows that, especially for stiffer substructures, the motion at the top of the substructure is filtered so
Figure 11. Peak story accelerations for the 10% in 50-year hazard level motions.
Figure 12. Peak relative story drifts for the 10% in 50-year hazard level motions.
DOI: 10.1002/stc

Figure 13. Instantaneous displacements at the moment of peak isolator drift
Figure 14. Percent change of the peak responses in the midlevel isolated substructure with respect to the substruc-
ture alone.
Figure 15. Percent change of the peak responses in the midlevel isolated structure compared with when the isola-
tion system is removed for Model A (note: story drifts not shown for the isolation layer).
DOI: 10.1002/stc

that the predominant period aligns closely to the period of the substructure and accelerations in this
range are amplified. These spectra at the top of the substructure show the frequency characteristics
of the motion experienced by the isolated superstructure. Accelerations are lower for all levels for
the long-period substructure systems. However, for the long-period substructure, the isolation only
maintains similar levels of acceleration for the superstructure, while for the short-period substructure,
the isolation system decreases the accelerations. The spectra for the top of the substructure have larger
response for higher frequencies for Model B because of the inclusion of higher modes in the model.
However, this trend does not transfer to the spectra of the roof, which do not change significantly in
shape between the two models.
In Figure 18, the roof response spectrum of the existing building without any added superstructure
is compared with the response spectrum of the midlevel isolated structure below the isolation level and
the response spectrum of the building with fixed superstructure at the floor level equivalent to the
Figure 16. Percent change of the peak responses in the midlevel isolated structure compared with when the isola-
tion system is removed for Model B (note: story drifts not shown for the isolation layer).
Figure 17. Floor response spectra for the various substructure configurations.
DOI: 10.1002/stc

existing building’s roof. In addition, the roof response spectra of the buildings with superstructures
added either as fixed or isolated are compared. Figure 18 shows these spectra for Model A; Model B
shows similar trends, however, with amplifications at higher frequencies for all building models.
The spectra show that adding an isolated superstructure to an existing building reduced floor accel-
erations for all but the stiffest building where acceleration levels are maintained. As this building is
practically rigid, it would make sense that the behavior below the isolation level would not be altered
significantly by any superstructure addition. For the most flexible substructure, the reduction in the ac-
celerations at the first floor level are on the same order as if the superstructure is added as fixed, sug-
gesting the benefit comes from slightly elongating the period of the building. This is true for the roof as
well, where adding the superstructure using midlevel isolation, rather than a fixed condition, signifi-
cantly reduces roof accelerations for all but the 1.0-s substructure system, for which the accelerations
are of similar magnitudes. These results, coupled with the comparison of story drifts, show that for
midlevel isolation to be effective, there should be a distinct separation between the substructure and
isolation periods, with Tiso/Tsub at least greater than two. However, if a benefit is desired to both the
substructure and the superstructure, the substructure should not behave rigidly. This is in agreement
with results found by Wang et al. [5] in which response spectrum analysis was used to investigate a
three degree of freedom midlevel isolation model. They found a decrease in substructure inertia forces
and story shears when Tiso/Tsub decreased (larger substructure period) and a decrease in superstructure
forces when Tiso/Tsub increased (larger separation between periods).
6. CONCLUSIONS
In order to conduct efficient experimental tests of a midlevel isolated building, which must be tested in
real time because of the velocity dependence of the friction pendulum bearings selected for the isola-
tion layer, a shake table was constructed and interfaced with a real-time hybrid control system. The iso-
lated superstructure was experimentally tested on the shake table, while the substructure was
numerically modeled. In this way, it was possible to examine the behavior of the isolated superstruc-
ture with multiple substructure configurations.
Figure 18. Response spectra from midlevel isolation (Mid Iso), the building without the added superstructure (No
Sup) and the building where the superstructure has no isolation (No Iso), Model A.
DOI: 10.1002/stc

The hybrid testing method was able to reproduce the full range of input frequencies for nearly all
test cases. However, when more degrees of freedom were introduced for the numerical substructure,
increased response was observed in high-frequency modes because of high-frequency experimental
errors, which added energy into the system. This had the greatest effect on the longest period substruc-
ture, which saw significantly increased tracking errors, resulting in an increase in the displacements at
the top of the substructure. These errors are believed to be associated with the test setup (large table
mass, friction in the linear bearings, noise in the load cell readings, oil column resonance, etc.) and
the use of a single variable (displacement) non-adaptive control algorithm for the shake table. Greater
fidelity can be achieved by further refinement of the setup and control system.
For the midlevel isolation, the short-period substructures resulted in higher accelerations input to the
superstructure. As a result, the displacement of the isolators was largest for the shorter period substruc-
tures, and the peak floor responses, especial accelerations, were smaller for the long-period structure.
When comparing against other design options, including a building without the addition of extra stories
or a building in which additional stories are added on without the use of isolation, midlevel isolation
exhibited larger reductions in peak responses (both story drifts and floor accelerations) for the systems
with substructure periods of 0.5 s and less. For the 1.0-s substructure period, adding an isolated super-
structure increased the story drifts compared with the other design options for some cases. Thus, it is
suggested that the effective period of the isolated superstructure be at least twice that of the fixed base
substructure. However, for the floor response spectra below the isolation plane, midlevel isolation re-
sulted in the greatest reductions when the substructure was more flexible. No reduction was seen for the
substructure floor response spectra when Tsub = 0.125 s. This suggests that if reduction in floor spectra
is desired below the isolation plane, the substructure should have some flexibility.
Overall, midlevel isolation delivered improved performance against other design options, reducing
both accelerations and story drifts, and the real-time hybrid shake table testing approach proved to be a
reliable experimental method to rapidly assess the behavior of midlevel isolation systems with a variety
of substructure conditions. Nevertheless, future research is needed for the control and compensation of
spurious experimental errors. This research will allow for the testing of more complex substructure
types, for example, finite element building models with nonlinear behavior or nonlinear soil for soil-
structure interaction.
ACKNOWLEDGEMENTS
Funding for this work was provided in part by the National Science Foundation through Grant No. CMMI-0724208.
The authors appreciate the assistance of Dr. Frank McKenna and Dr. Selim Günay of University of California. The
findings and conclusions are those of the authors alone and may not reflect those of NSF or other sponsors.
REFERENCES
1. Ziyaeifar M, Noguchi H. Partial mass isolation in tall buildings. Earthquake Engineering and Structural Dynamics 1998;
27(1):49–65.
2. Ogura K, Takayama M, Tsujita O, Kimura Y, Wada A. Seismic response of mid-story isolated buildings. Journal of Struc-
tural and Construction Engineering, Architectural Institute of Japan 1999; 516:99–104.
3. Villaverde R, Mosqueda G. Aseismic roof isolation system: analytical and shake table studies. Earthquake Engineering and
Structural Dynamics 1999; 28(3):217–234.
4. Kobayashi M, Koh T. Modal coupling effects of midstory isolated buildings. Proceedings of the 14
th
World Conference on
Earthquake Engineering, Beijing, China, 2008.
5. Wang SJ, Chang KC, Hwang JS, Lee BH. Simplified analysis of mid-story seismically isolated buildings. Earthquake
Engineering and Structural Dynamics 2011; 40:119–133.
6. Murakami K, Kitamura H, Ozaki H, Yamanashi T. Design of a building with seismic isolation system at the mid-story.
Journal of Technology and Design, Architectural Institute of Japan 1999; 7:51–56.
7. Sueoka T, Torii S, Tsuneki Y. The Application of Response Control Design Using Middle-Story Isolation System to High-
Rise Building, Proceedings of 13WCEE. Vancouver: B.C., Canada, 2004.
8. Dutta A, Sumnicht J, Mayes R, Hamburger R, and Citipitioglu A. An innovative application of base isolation technology.
Proceedings of the 2009 ATC SEI Conference on Improving the Seismic Performance of Existing Buildings and Other
Structures, 2009.
9. Takanashi K, Udagawa K, Seki M, Okada T, Tanaka H. Non-linear earthquake response analysis of structures by a
computer-actuator on-line system (details of the system). Transaction of the Architectural Institute of Japan 1975;
229:77–83.
10. Nakashima M, Kato M, Takaoka E. Development of real-time pseudo-dynamic testing. Earthquake Engineering and Struc-
tural Dynamics 1992; 21(1):79–92.
DOI: 10.1002/stc

11. Horiuchi T, Konno T. A new method for compensating actuator delay in real-time hybrid experiments. Philosophical Trans-
actions of the Royal Society, Mathematical, Physical and Engineering Sciences 2001; 359:1893–1909.
12. Wallace MI, Wagg DJ, Neild SA. An adaptive polynomial based forward prediction algorithm for multi-actuator real-time
dynamic substructuring. Proceedings of the Royal Society A 2005; 461:3807–3826.
13. Ahmadizadeh M, Mosqueda G, Reinhorn AM. Compensation of actuator delay and dynamics for real-time hybrid structural
simulation. Earthquake Engineering and Structural Dynamics 2008; 37:21–42.
14. Chae Y, Kazemibidokhti K, Ricles JM. Adaptive time series compensator for delay compensation of servo-hydraulic actu-
ator systems for real-time hybrid simulation. Earthquake Engineering and Structural Dynamics 2013; 42:1697–1715.
15. Bursi OS, Shing B. Evaluation of some implicit time-stepping algorithms for pseudodynamic tests. Earthquake Engineering
and Structural Dynamics 1996; 25(4):333–355.
16. Bonelli A, Bursi OS. Generalized-alpha methods for seismic structural testing. Earthquake Engineering and Structural Dy-
namics 2004; 33(10):1067–1102.
17. Bonnet PA, Williams MS, Blakeborough A. Evaluation of numerical time-integration schemes for real time hybrid testing.
Earthquake Engineering and Structural Dynamics 2008; 37(13):1467–1490.
18. Neild S, Stoten D, Drury D, Wagg D. Control issues relating to real-time substructuring experiments using a shaking table.
19. Dorka UE, Queval JC, Nguyen VT, Maoult A.L. Real-time sub-structure testing on distributed shaking tables in CEA
Saclay. Proceedings of the 4th World Conference on Structural Control and Monitoring, San Diego, USA, 2006.
20. Lee S, Parka E, Mina K, Park J. Real-time substructuring technique for the shaking table test of upper substructures.
Engineering Structures 2007; 29(9):2219–2232.
21. Ji X, Kajiwara K, Nagae T, Enokida R, Nakashima M. A substructure shaking table test for reproduction of earthquake re-
sponses of high-rise buildings. Earthquake Engineering and Structural Dynamics 2009; 38(12):1381–1399.
22. Igarashi A, Iemura H, Suwa T. Development of substructured shaking table test method. 12th World Conference on Earth-
quake Engineering, Auckland, New Zealand, 2000.
23. Nakata N, Stehman M. Substructure shake table test method using a controlled mass: formulation and numerical simulation.
24. Shao X, Reinhorn AM, Sivaselvan MV. Real-time hybrid simulation using shake tables and dynamic actuators. ASCE Jour-
nal of Structural Engineering 2011; 137(7):748–760.
25. Magonette G. Development and application of large-scale continuous pseudo-dynamic testing techniques. Philosophical
Transactions of the Royal Society: Mathematical, Physical and Engineering Sciences 2001; 359(1786):1771–1799.
26. Schellenberg AH, Mahin SA, Fenves GL. Advanced implementation of hybrid simulation. Report No: PEER 2009/?104.
Paciﬁc Earthquake Engineering Research Center, University of California, Berkeley, 2009.
27. Nakata N. Acceleration trajectory tracking control for earthquake simulators. Engineering Structures 2010; 32(8):
2229–2236.
28. Thoen BK. 469D Seismic Digital Control Software. MTS Systems Corporation 2010, Eden Prairie, MN.
29. McKenna F, Fenves GL, Filippou FC. 2010, OpenSees. http://opensees.berkeley.edu.
30. Schellenberg AH, Mahin SA, Fenves GL. 2013, OpenFresco. http://openfresco.berkeley.edu.
31. Mercan O, Ricles JM. Experimental studies on real-time testing of structures with elastomeric dampers. ASCE Journal of
Structural Engineering 2009; 135:1124–1133.
DOI: 10.1002/stc

Experimental Assessment of the Seismic Response of a Base-Isolated Building Through a Hybrid Simulation Technique

More Related Content

Similar to Experimental Assessment of the Seismic Response of a Base-Isolated Building Through a Hybrid Simulation Technique (20)

Recently uploaded (20)

Experimental Assessment of the Seismic Response of a Base-Isolated Building Through a Hybrid Simulation Technique