Engineering Structures 32 (2010) 2229–2236

Contents lists available at ScienceDirect

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

Acceleration trajectory tracking control for earthquake simulators

Narutoshi Nakata ∗
Department of Civil Engineering, Johns Hopkins University, 210 Latrobe Hall, 3400 N. Charles St., Baltimore, MD 21218, United States

article info abstract

Article history: The purpose of earthquake simulators is to reproduce reference accelerations on a shake table. However,
Received 14 July 2009 acceleration control is extremely difficult in earthquake simulators due to an inability to directly
Received in revised form control acceleration with feedback, inherent nonlinearities in servo hydraulic systems, control-structure
4 September 2009
interactions, etc. This study presents a control method called acceleration trajectory tracking control
Accepted 25 March 2010
Available online 24 April 2010
(ATTC) that improves the acceleration control performance of earthquake simulators. The ATTC method
consists of an acceleration feed-forward controller, a system dynamics command shaping, an intentional
Keywords:
time-delay, a Kalman filter for displacement measurement, and an actuator displacement feedback
Earthquake simulators controller. The ATTC method not only provides acceleration tracking capability, but also ensures stability
Shaking tables of the system. Following the theoretical description of the ATTC method, an experimental investigation
Hydraulic actuators is presented. The ATTC method is successfully implemented in the control system for the uniaxial
Acceleration control earthquake simulator at Johns Hopkins University, and the experimental results show the superior
Feed-forward control performance of the ATTC method over the conventional displacement feedback with command shaping.
Furthermore, repeatability of the ATTC method is experimentally verified.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction in nature unstable. A closed-loop feedback control is required

Earthquake simulators, also known as shake tables, provide
the most direct experimental means for the performance as-
sessment of structures subject to ground motion. At present, a
large number and a wide variety of earthquake simulators are
in active use around the world (for example, a 1,200 ton pay-
load extremely large shake table [1], an outdoor shake table [2],
6-degrees-of-freedom shake tables [3]). The purpose of the con-
trol systems in earthquake simulators is to reproduce reference ac-
celerations at the table. In general, the reference accelerations are
either recorded accelerations during earthquake, synthetic accel-
erations from attenuation and seismological study, or some sort
of waveforms such as sinusoidal and random waves. However, ac-
celeration control of shake tables is extremely difficult because of
inherent nonlinearities in servo hydraulic systems (i.e., valve dy-
namics, oil flow, etc.), control-structure interactions, dynamics of
the base support, etc. Although measured accelerations at the ta-
ble can be used as the input acceleration in the performance as-
sessment, it is essential to have an acceptable acceleration control
capability to assess the true impact of the reference accelerations
on structures.
Most earthquake simulators are driven by servo hydraulic
actuators to meet the large force and high velocity requirements
for shaking the table with payload. Servo hydraulic actuators are
∗ Tel.: +1 410 516 6482; fax: +1 410 516 7473.
E-mail address: nakata@jhu.edu.
URL: http://www.ce.jhu.edu/nakata.
0141-0296/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.03.025
to stabilize the motion of the actuator piston. Due to the
existence of an unobservable and marginally unstable mode in
the acceleration measurement (e.g., constant velocity motions
cannot be detected from the acceleration measurement; see the
theoretical proof in the next section), acceleration feedback control
of the hydraulic actuators is unstable, and cannot be adopted for
the earthquake simulators. In practice, actuator displacement is
used as the feedback, and thus the reference input to the hydraulic
actuators is also the displacement. This inability to directly control
the acceleration with feedback makes the control systems for
earthquake simulators insensitive to unpredictable disturbances in
acceleration.
In earthquake simulators, the reference displacement inputs
to hydraulic actuators are calculated from a double integration
of the reference accelerations [4,5]. Due to the aforementioned
inherent nonlinearities in servo hydraulic systems, displacement
feedback control does not ensure acceptable displacement con-
trol performance even if boundary conditions are ideal. Several
researchers proposed control methods that compensate the actua-
tor/system dynamics to improve the control performance. Spencer
and Yang [4] presented the transfer function iteration method em-
ployed in many commercial earthquake simulators. The method
is based on the iterative command shaping using the inverse of
the transfer function from the actuator reference displacement to
the measured acceleration. Twitchell and Symans [5] proposed a
simplified approach without iterations using an inverse of the
transfer function from the reference displacement to the measured
displacements. Those methods are basically command shaping [6]
2230 N. Nakata / Engineering Structures 32 (2010) 2229–2236
(i.e., the performance improvement is achieved by shaping the
command without making changes in the displacement feedback
loop), and can improve displacement control performance in both
the time and frequency domains. However, improvement in the
acceleration control performance is rather limited; while the fre-
quency domain distortion (particularly magnitude of transfer func-
tion) can be reduced, acceleration tracking in the time domain is
either not achieved or limited within the low frequency range.
In addition to the displacement feedback with command shap-
ing, feed-forward and the other feedback controls have been in-
vestigated for earthquake simulators. Kuehn et al. [7] developed
a feedback/feed-forward control strategy based on the receding
horizon control (RHC). Experimental results showed that the RHC
based controller had better phase characteristics in the acceler-
ation transfer function than a feedback control using the linear
quadratic regulator (LQR) control. Trombetti and Conte [8] em-
ployed a control method combining the displacement feedback,
velocity feed-forward and differential pressure feedback in the
test-analysis comparison study. Their sensitivity analysis showed
the effectiveness of the velocity feed-forward in the magnitude
characteristics of the displacement transfer function. While those
methods reported to date improve control performance of earth-
quake simulators, improvements in acceleration tracking is not
addressed.
This paper presents a method to improve the acceleration
control performance for earthquake simulators. The proposed
method, called the acceleration trajectory tracking control (ATTC)
method, combines an acceleration feed-forward controller, a
displacement feedback, a command shaping, an intentional
time delay component, a Kalman filter for the displacement
measurement, and a displacement feedback controller. First,
analytical open-loop transfer functions are developed to discuss
the stability, controllability, and observability of servo hydraulic
actuators. Then, the ATTC method is developed based on the
system dynamics of an earthquake simulator with focus on the
performance and the stability. The ATTC method is implemented
in the uniaxial earthquake simulator at Johns Hopkins University,
and an experimental investigation is conducted to verify and
understand the performance of the ATTC method in acceleration
control. The experimental results and their implications are
discussed in this paper.
2. System dynamics of earthquake simulators
Earthquake simulators are comprised of electrical and mechan-
ical components that have unique dynamic characteristics. System
dynamics of the earthquake simulators are subject to the serial,
parallel and feedback connections of those components, and are
complex and nonlinear. In addition, due to the control-structural
interactions [9], the system dynamics are influenced by the dy-
namics of test specimens that are placed on the shake table. While
the displacement feedback with the conventional proportional-
integral-differential (PID) controller provides a reasonable dis-
placement tracking, acceleration tracking is still poor due to the
slowness and time lag associated with the displacement feed-
back [5]. To facilitate the acceleration control design and stability
assessment, this section briefly derives the dynamics and interac-
tions of the governing components in earthquake simulators.
2.1. Servo valve and oil flow in hydraulic actuator
First, we consider the relationship between an electrical
command to a servo valve and oil flow in the actuator chambers.
The electrical command changes the position of the valve spool
that regulates the oil flow into the actuator chambers. The oil
flow is often modeled proportional to the valve command with a
Fig. 1. A schematic of a uniaxial earthquake simulator.
constant time-delay [9,10]. The first-order approximation model
of the transfer function from the valve command u to the oil flow
q can be expressed as:
q (s) kv
Hqu (s) = = (1)
u (s) 1 + τs
where kv is the valve gain; τ is the time delay of the servo valve;
and s is the Laplace variable.
Oil flow in the actuator chambers is the driving source of hy-
draulic actuators. However, flow rate is also affected by the dynam-
ics of the actuator piston. From the equilibrium in flow rate q, the
first-order governing oil flow equation can be obtained as:
V
q (t ) = Aẋa (t ) + ke Fa (t ) + Ḟa (t ) (2)
4β A
where A is the piston area; xa is the actuator displacement; ke is
the flow-force coefficient; Fa is the actuator force; V is the volume
of the chamber; and β is the bulk modulus of the fluid. For more
details, see [10]. The first, second, and third terms are those from
piston movement, oil leakage, and change in chamber volume, re-
spectively. Taking the Laplace transform of Eq. (2), the transfer
function from the oil flow q to the actuator displacement xa is
obtained as follows:
xa (s) H xa Fa
Hxa q (s) = = (3)
q (s) AsHxa Fa + ke + k1 s
where k1 = V /4β A, and Hxa Fa is the transfer function from the ac-
tuator force Fa to the actuator displacement xa .
2.2. Equations of motion for earthquake simulators
To establish the actuator transfer function, we consider the
equations of motion for earthquake simulators. Fig. 1 shows a
schematic of a uniaxial earthquake simulator including a shake
table, a hydraulic actuator, a base support, and a test specimen.
Note that the base support is not rigidly connected to a fixed end
to prevent propagation of vibration from the earthquake simulator.
Force resistance of the base support is modeled as a combination
of a linear spring kb and a dashpot cb for the sake of simplicity.
The equations of motion for the shake table and the base
support can be expressed as:
Shake Table: mt ẍt (t ) − Fa (t ) − Fs (t ) = 0 (4)
Base Support: mb ẍb (t ) + cb ẋb (t ) + kb xb (t ) + Fa (t ) = 0 (5)
where mt and mb are the mass of the table and the base support,
respectively; xt , xa and xb are the displacement of the shake table,
the actuator, and the base support, respectively; and Fs is the base
 N. Nakata / Engineering Structures 32 (2010) 2229–2236 2231

shear of the test specimen. The actuator, table and base support kv (mt + mb ) s2 + cb s + kb
displacements hold the following relationship. Hxa u (s) = · (12)
s (1 + τ s) D̄ (s)
xt (t ) = xa (t ) + xb (t ) . (6)
kv s mb s2 + cb s + kb
From Eqs. (4)–(6) and further derivation, the transfer function Hat u (s) = · (13)
( 1 + τ s) D̄ (s)
from the actuator force Fa to the displacement xa can be obtained
D̄ (s) = mt s (ke + k1 s) mb s2 + cb s + kb


as:
+ A (mt + mb ) s2 + cb s + kb .

xa (s) 1 1 (14)
Hxa Fa (s) = = + (7)
Fa (s)


mt + HFs at s2 mb s2 + cb s + kb Based on the linear systems theory [11], the dynamics of
the hydraulic actuator in earthquake simulators described in
where HFs at is the transfer function from the table acceleration at Eq. (12) can be expressed using six state variables; the order
to the base shear of the payload Fs . of the denominator in the transfer function is six. If all of the
six poles (roots of the denominator) lie in the left side of the
2.3. Open-loop transfer functions complex plane, systems are guaranteed to be stable [12]. As seen in
Eq. (12), the actuator displacement open-loop transfer function
An open-loop transfer function from the electrical valve com- has a marginally stable pole at the origin (s = 0). Note that the
mand u to the actuator displacement xa can be obtained by substi- rest of the poles (−τ and roots of D̄ (s)) are stable. Practically, the
tuting Eq. (7) into Eq. (3) and multiplying Eqs. (1) and (3) pole at the origin is unstable; as s gets close to zero, the output
xa (s) goes to infinity (unbounded instability). However, the pole at the
Hxa u (s) = = Hxa q (s) Hqu (s) origin is observable in the displacement measurement and more
u ( s)
importantly controllable from the valve command. Therefore, this
kv mb + mt + HFs at s2 + cb s + kb practically unstable pole can be stabilized by introducing a closed-
= · (8)
s ( 1 + τ s) D loop feedback with the displacement measurement.
On the other hand, the pole at the origin is cancelled out in the
D = s (ke + k1 s) mt + HFs at m s2 + cb s + kb




2 b table acceleration open-loop transfer function, Eq. (13), due to the
mt + mb + HFs at s + cb s + kb .

+A (9) zeros (roots of the numerator) from the double integration of s. This
In the same manner, the open-loop transfer function from the pole-zero cancellation makes the pole at the origin unobservable in
electrical valve command to the shake table acceleration can be the acceleration measurement. In other words, the pole at the ori-
given by gin cannot be controlled and stabilized based on the acceleration
measurement. Thus, acceleration feedback control for earthquake
at (s) xt (s) simulators is unstable and practically not feasible. Therefore, this
Hat u (s) = = s2
u (s) u ( s) study explores acceleration control methods without an accelera-

tion feedback control.
kv s mb + HFs at s2 + cb s + kb
= · . (10)
1 + τs D 3.2. Acceleration trajectory tracking control
It should be noted from Eqs. (8) and (10) that because of the
dynamics of the base support, the double differentiation of the The goal of control design is to obtain the configuration, speci-
actuator displacement does not yield the table acceleration, that fication, and identification of key parameters of a proposed system
is, to produce the desired output. With the focus placed on the con-
trol performance and the stability, this paper proposes a control
Hat u (s) 6= s2 Hxa u (s) . (11) method called acceleration trajectory tracking control (ATTC) that
The above relationship implies that the actuator displacement consists of both acceleration feedforward and displacement feed-
tracking does not necessarily ensure the reference acceleration back control loops. The displacement feedback loop incorporates a
tracking at the shake table. system dynamics command shaping, an intentional systems time
delay, and a Kalman filter. Fig. 2 shows a schematic of all of the
components in the proposed control method. The details and roles
3. Acceleration trajectory tracking control
of each component are given below.
As mentioned earlier, displacement feedback controls are lim-
ited in term of acceleration tracking in earthquake simulators. A 3.2.1. Acceleration feed-forward using the pseudo inverse of the table
discrepancy between the reference and the measured accelera- acceleration transfer function
tion is particularly prominent at high frequency because of the An acceleration feed-forward controller herein is adopted to
poor phase characteristics in the displacement control. This section generate the primary driving command for the reference acceler-
investigates the open- and closed-loop stabilities of earthquake ation, compensating the dynamics from the valve to the table ac-
simulators using the transfer functions obtained in the previous celeration. The feed-forward controller consists of a lowpass filter
section, and then introduces a possible control method for accel- Glf and the inverse of the approximated table acceleration transfer
eration tracking control for earthquake simulators. function H̄at u . The roles of the lowpass filter are (i) to make Glf H̄a−t u1
proper so that it can be realized in a state space model to calculate
3.1. Stability, controllability and observability of earthquake simula- the valve command uff from the reference acceleration ãt ; and (ii)
tors to reduce the frequency content that is higher than the frequency
range of interest. The relationship between the reference acceler-
Prior to the development of control methods, we consider the ation and the valve command from the feed-forward controller is
stability of the open-loop transfer functions. For the sake of sim- given as:
plicity, we consider an earthquake simulator without test speci-
uff = Glf H̄a−t u1 ãt . (15)
mens (i.e., HFs xt = 0). Without losing generality, the open-loop
transfer functions from the electrical command to the actuator dis- Note that the computation of the feed-forward command can
placement and to the table acceleration are simplified as be performed off-line.
2232 N. Nakata / Engineering Structures 32 (2010) 2229–2236

Fig. 2. A block diagram of the acceleration trajectory tracking controller.

3.2.2. System dynamics command compensator stabilize the pole at the origin, but not to affect the tracking of
A displacement feedback is used to stabilize the earthquake displacement and acceleration. In other words, the displacement
simulator. The reference actuator displacement is computed from feedback is just to prevent the drift of the earthquake simulator.
the reference table acceleration based on the system dynamics The displacement feedback loop with a proportional and a
compensator instead of the double integration. The computed, derivative controller can be expressed as:
reference actuator displacement should provide better prediction


of the actuator displacement for the reference acceleration, and the ufb = Kp + Kd s x̂a − x̄a (19)
relationship is given as: where Kp and Kd are the proportional and the derivative gains,
respectively; and ufb is the valve command from the displacement
x̃a = H̄xa u H̄a−t u1 ãt (16) feedback. In the implementation, the proportional gain needs to
where x̃a is the reference actuator displacement; and H̄xa u is be tuned relatively low to reduce the influence on the system
the approximated open-loop transfer function from the valve dynamics at the high frequency.
command to the actuator displacement.
3.2.6. Overall transfer functions
3.2.3. Intentional system delay The command to the servo valve in the ATTC method is a sum
In the ATTC method, an intentional time delay is introduced of the feed-forward and feedback terms as;
between the reference table acceleration and the reference u = uff + ufb . (20)
actuator displacement. The purpose of the intentional system delay
is to ensure that the acceleration feed-forward loop is the driving The overall system transfer functions from the reference table
source for the reference acceleration and that the displacement acceleration ãt to the measured table acceleration at , and from
feedback loop serves only to provide the stability (prevent drift) the reference actuator displacement x̃a to the measured actuator
of the table. The intentional system time delay is given as: displacement xa can be obtained using Eqs. (15)–(20) and further
derivation:
1 − τs
x̂a = Gt (s) x̃a = x̃a (17) at
1 + τs Hat ãt (s) =
ãt
where x̂a is the delayed, reference actuator displacement.



Glf Hat u H̄a−t u1 1 + Gt Kp + Kd s H̄xa u
=

(21)
3.2.4. Kalman filter 1 + Kp + Kd s Kxa u + Kxa xa Hxa u
A Kalman filter [13] is employed in the ATTC method to Hxa u H̄x−a 1u G− 1
+ Kp + Kd s H̄xa u



xa t
reduce noise in the actuator displacement without time delay. The Hxa x̃a (s) = =

. (22)
model-based filtering effectively reduces noise in the displacement x̃a 1 + Kp + Kd s Kxa u + Kxa xa Hxa u
feedback loop so that the displacement loop does not introduce a If the proportional gain becomes small (i.e., Kp → 0), the
high frequency disturbance in the valve command. Note that the transfer functions get close to simply the inverse of the open-loop
valve command for high frequency acceleration is generated from transfer functions
the acceleration feed-forward controller.
The relationship between the valve command u, the measured limKp →0 Hat ãt (s) = Glf H̄a−t u1 Hat u (23)
actuator displacement xa , and the estimated actuator displacement
x̄a are given by: limKp →0 Hxa x̃a (s) = Glf H̄xa u Hxa u .
−1
(24)

x̄a = Kxa u (s) u + Kxa xa (s) xa (18) On the other hand, if the proportional gain is relatively large,
the transfer functions get close to the conventional displacement
where Kxa u and Kxa xa are the Kalman gains from the valve command
feedback control with command shaping as:
and the measured actuator displacement to the estimated actuator

displacement, respectively. The Kalman filter model and gains Glf Gt Kp + Kd s H̄xa u H̄a−t u1 Hat u
are determined from the open-loop transfer function from the limKp →∞ Hat ãt (s) =

(25)
1 + Kp + Kd s Kxa u + Kxa xa Hxa u
valve command to the actuator displacement with measured
displacement noise.


Glf Gt Kp + Kd s Hxa u
limKp →∞ Hxa x̃a (s) =

. (26)
3.2.5. Displacement feedback with a PD controller 1 + Kp + Kd s Kxa u + Kxa xa Hxa u
The displacement feedback loop herein is included to stabilize Thus, the proposed ATTC method implicitly enables flexi-
the earthquake simulator. More specifically, it is intended to ble setting, ranging from a true open-loop table acceleration
 N. Nakata / Engineering Structures 32 (2010) 2229–2236
Fig. 3. A photo of the uniaxial earthquake simulator at Johns Hopkins University.
feed-forward control to a conventional closed-loop actuator dis-
placement feedback control, by changing the controller
parameters. The control parameters have to be tuned based on the
open-loop transfer functions for each earthquake simulator.
4. Experimental setup
The proposed acceleration trajectory tracking control method
is implemented in the control system for the recently constructed
uniaxial earthquake simulator at Johns Hopkins University.
(See Fig. 3.)
4.1. Uniaxial earthquake simulator
The uniaxial earthquake simulator shown in Fig. 3 consists of
a 1.2 m × 1.2 m aluminum sliding table mounted on two linear
guides with high-precision, low-friction, linear ball-bearings. The
shake table is driven by a 27 kN hydraulic actuator manufactured
by Shore Western, Inc. The specifications of the earthquake
simulator are: maximum displacement of ±7.6 cm, maximum
velocity of ±5.1 cm/s, maximum acceleration of 3.8 g; and
maximum payload of 0.5 ton. The operational frequency range of
the simulator is 0.1–50 Hz.
4.2. Control and data acquisition systems
The control hardware for the earthquake simulator includes
a National Instruments 8-axis high-performance servo motion
controller board (PXI-7358), a 2.3 GHz high-bandwidth dual-core
PXI express controller (PXIe-8130), a windows-based host PC and
other accessories. The NI PXI-7358 is capable of running a PID
loop rate at 62.5 µs. The data acquisition system consists of a 16-
bit high-speed multifunction data acquisition board (PXI-6251), a
signal conditioner (SCXI-1000), and various analog input modules.
Programs for the control and data acquisition are written in
NI LabVIEW, and are deployed in a real-time operating system
on the PXIe-8130. The integrated control and data acquisition
system enables simultaneous sampling of all of the input and
output signals, and user-defined control and signal processes. In
the following experiments, both control and data sampling rates
of 2 kHz are employed.
5. Experimental verification
To verify the performance of the ATTC method in acceleration
control, an experimental investigation was conducted using the
2233
uniaxial earthquake simulator at Johns Hopkins University. This
paper presents the initial experimental work, which did not
include payloads.
As described earlier, the ATTC method requires system dynam-
ics of the earthquake simulator for proper selection of the design
parameters. Using a band-limited random signal with a frequency
range between 0.1 and 30 Hz, the system dynamics of the uniax-
ial earthquake simulator are experimentally obtained. Fig. 4 shows
the open-loop transfer functions and coherence from the valve
command to the actuator displacement and the table acceleration.
The coherence plots depict that adequate correlation between the
valve signal and the earthquake simulator dynamics was obtained
in the system identification tests.
The experimentally obtained open-loop transfer functions
were approximated in the form of rational polynomial functions,
employing the least square curve fitting method. The polynomial
orders used in the approximation functions are the same as the
ones in the analytical models: that is, orders for the numerator
and denominator for the transfer function to the table acceleration
are 3 and 5, respectively. The orders for the numerator and
denominator for the transfer function to the actuator displacement
are 2 and 6, respectively. The approximated analytical models for
the table acceleration and the actuator displacement shown in
Fig. 4 exhibit excellent agreement with the experimental results in
both the amplitude and the phase characteristics. It can be verified
that the analytical models for the open-loop transfer functions in
Eqs. (12) and (13) are sufficient in the frequency range of interest.
With the experimental and approximated analytical transfer
functions and the experimentally identified system delay of
0.01 s in valve dynamics, the shake table acceleration transfer
function (from the reference acceleration to the measured table
acceleration) in Eq. (21) was established. The Kalman filter gains
are determined based on the high frequency noise reduction
and the time delay, accounting for the experimental transfer
function of the shake table and the noise level in the displacement
measurement. Then the PD gains in the displacement feedback
loop are selected based on the transfer function and a numerical
time-history analysis such that the displacement loop provides
stability, but does not influence the high frequency response of the
shake table. The shake table acceleration transfer function in the
ATTC method is shown in Fig. 5. To evaluate the performance of the
ATTC method in comparison with the conventional displacement
feedback with command shaping (DFCS), a shake table acceleration
transfer function of the DFCS was also plotted in Fig. 5. The ATTC
method shows better performance than the DCFS in the transfer
function over a wide range of frequency. In particular, the ATTC
method is superior to the DFCS method in the phase characteristic.
It should be mentioned that the scatter in the low frequency range
of the acceleration transfer functions are due to the low coherence
in acceleration in that range, and that the stability was ensured in
both acceleration and displacement.
Acceleration tracking performance of the ATTC method was
experimentally investigated for earthquake records. Fig. 6 shows
the comparison between the ATTC method and DFCS method in
the acceleration time histories and the acceleration power spectral
density for scaled Loma Prieta and Kobe earthquake records. The
ATTC method exhibits smooth trajectory tracking to the reference
earthquake not only in a long time scale (see Figs. 6(a) and (b))
but also in a short time scale (see Figs. 6(c) and (d)). On the
other hand, while the DFCS method shows reasonable trajectory
tracking in a long time scale, it shows poor performance in a
short time scale. This high-frequency pitching is associated with
the displacement feedback, and the results agree with studies
in literature (e.g., [5]). The ATTC method also shows better
performance than the DFCS method in the frequency domain,
particularly in the high frequency range (see Figs. 6(e) and (f)).
2234 N. Nakata / Engineering Structures 32 (2010) 2229–2236

a b

c d

e f
Fig. 4. System dynamics of the uniaxial earthquake simulator: (a), (c), and (e) are the magnitude, phase, and coherence of the open-loop transfer function from the valve
command to the actuator displacement, respectively; and (b), (d), and (f) are the magnitude, phase, and coherence of the open-loop transfer function from the valve command
to the table acceleration, respectively.

a b

c d
Fig. 5. Closed-loop transfer functions: (a) and (c) are the magnitude and phase plots of the transfer function from the reference table acceleration to the measured table
acceleration, respectively; and (b) and (d) are the magnitude and phase plots of the transfer function from the reference actuator displacement to the measured actuator
displacement, respectively.
 N. Nakata / Engineering Structures 32 (2010) 2229–2236 2235

a b

d
c

e f

Fig. 6. Comparison of the acceleration among the ATTC, DFCS, and reference: (a), (c), and (e) are wide and a narrow views of the acceleration tracking, and the acceleration
power spectral density for the Kobe earthquake, respectively; (b), (d), and (f) are wide and a narrow views of the acceleration tracking, and the acceleration power spectral
density for the Loma Prieta earthquake, respectively.

Table 1 ATTC method was experimentally investigated. Fig. 7 includes four

RMS of the error between the reference and measured accelerations in the time acceleration time histories using the same input, but at different
domain and the frequency domain. times, for the ATTC and DFCS methods, respectively. The results
Time domain RMS (g) Frequency in the ATTC method are virtually identical. On the other hand,
domain RMS the results in the DFCS method show variations in the amplitude
(g2 /Hz)
and phase. These experimental results demonstrate the superior
ATTC DFCS ATTC DFCS repeatability of the ATTC method over the DFCS method. However,
ChiChi (1999) 0.141 0.190 0.198 0.290 it should be noted that these results do not necessarily ensure the
Coalinga (1983) 0.239 0.297 0.396 0.490 repeatability of the ATTC method with payload.
Duzce (1999) 0.447 0.528 0.723 0.943
Imperial Valley (1979) 0.213 0.363 1.135 1.446
Kobe (1995) 0.263 0.309 0.861 0.816 6. Conclusions
Loma Prieta (1989) 0.152 0.214 0.373 0.442
Morgan Hill (1984) 0.181 0.246 0.842 0.887
This paper introduced an acceleration trajectory tracking con-
Taiwan SMART (1986) 0.245 0.278 0.109 0.181
trol (ATTC) method for earthquake simulators. The ATTC method
incorporates an acceleration feed-forward, a system dynamic com-
Further comparison between the ATTC and the DFCS methods mand shaping, an intentional time-delay, a Kalman filter for the
was made for more recorded earthquakes. Table 1 lists the root displacement measurement, and a displacement feedback loop.
mean square (RMS) of the error between the reference and The theory behind the ATTC method, including the stability, were
measured acceleration in the time domain and the frequency discussed and the table acceleration transfer function was analyt-
domain, respectively. In all of the cases, the ATTC method ically derived.
outperforms the DFCS method in the trajectory tracking (i.e., the The ATTC method was successfully implemented in the
time domain RMS). While exceptions exist, the ATTC method also control system for the uniaxial earthquake simulator at Johns
shows better performance than the DFCS method in the frequency Hopkins University. Experimental investigation demonstrated the
domain RMS. These comprehensive experimental results clearly superior performance of the ATTC method over the conventional
show that better acceleration trajectory tracking control can be displacement feedback with command shaping (DFCS) method
achieved with the ATTC method than the conventional DFCS that is used in commercial earthquake simulator control. The ATTC
method. method shows excellent performance in acceleration trajectory
Repeatability is one of the most important features of an tracking in a wide range of frequencies without the pitching that
earthquake simulation test to identically test more than one can be found in the DFCS method. Furthermore, the repeatability
specimen and to compare the results [14]. The repeatability of the of the ATTC method was experimentally investigated. The results
2236 N. Nakata / Engineering Structures 32 (2010) 2229–2236
a 2003.
b comparison study. J Earthq Eng 2002;(4):513–51.
Fig. 7. Comparison among multiple trials: (a) ATTC and (b) DFCS.
showed that the acceleration time-histories in the ATTC method
were virtually identical.
This study indicated the performance and potential of the ATTC
method. Further studies are needed to understand and improve the
performance of the ATTC method. In particular, the ATTC method
has to be verified with payloads and also for multi-axial earthquake
simulators.
References
[1] Ohtani K, Ogawa N, Katayama T, Shibata H. Project E-defense: 3D full-scale
earthquake testing facility. In: Proceedings of the joint NCREE/JRC workshop.
[2] Van Den Einde L. et al. Development of the George E. Brown Jr. network for
earthquake engineering simulation (NEES) large high performance outdoor
shake table at the Univ. of California, San Diego. In: Proc of the 13th world
conference on earthquake engineering. 2004.
[3] Bruneau M. et al. Versatile shake tables and large-scale high-performance
testing facility towards real-time hybrid seismic testing. In: Proc of ASCE
structures congress. 2002.
[4] Spencer BF, Yang G. Earthquake simulator control by transfer function
iteration. In: Proc of the 12th ASCE engineering mechanics conference. 1998.
[5] Twitchell BS, Symans MD. Analytical modeling, system identification, and
tracking performance of uniaxial seismic simulators. J Eng Mech 2003;(12):
1485–8.
[6] Daley S, Hatonen J, Owens DH. Hydraulic servo system command shaping
using iterative learning control. In: Proc of the control. 2004.
[7] Kuehn J, Epp D. Patten WN. High-fidelity control of a seismic shake table.
Earthq Eng Struct Dyn 1999;(11):1235–54.
[8] Trombetti TL. Conte JP. Shaking table dynamics: results from a test-analysis
[9] Dyke SJ, Spencer BF, Quast P. Sain MK. Role of control-structure interaction in
protective system design. J Eng Mech 1995;(2):322–38.
[10] Conte JP, Trombetti TL. Linear dynamic modeling of a uniaxial servo-hydraulic
shaking table system. Earthq Eng Struct Dyn 2000;(9):1375–404.
[11] Antsaklis PJ, Michel AN. Linear systems. McGraw-Hill; 1999.
[12] Dorf RC, Bishop RH. Modern control systems. Prentice Hall; 2000.
[13] Grewal MS, Andrews AP. Kalman filtering. Prentice Hall; 1993.
[14] Dimig J, Shield C, French C, Bailey F, Clark A. Effective force testing: a method
of seismic simulation for structural testing. J Struct Eng 1999;(9):1028–37.