Large-Scale Real-Time Hybrid Simulation for Evaluation

of Advanced Damping System Performance

Anthony Friedman 1; Shirley J. Dyke, A.M.ASCE 2; Brian Phillips, A.M.ASCE 3; Ryan Ahn 4; Baiping Dong 5;
Yunbyeong Chae 6; Nestor Castaneda 7; Zhaoshuo Jiang, A.M.ASCE 8; Jianqiu Zhang 9; Youngjin Cha 10;
Ali Irmak Ozdagli 11; B. F. Spencer 12; James Ricles 13; Richard Christenson 14; Anil Agrawal, M.ASCE 15;
and Richard Sause, M.ASCE 16
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

Abstract: As magnetorheological (MR) control devices increase in scale for use in real-world civil engineering applications, sophisticated
modeling and control techniques may be needed to exploit their unique characteristics. Here, a control algorithm that utilizes overdriving and
backdriving current control to increase the efficacy of the control device is experimentally verified and evaluated at large scale. Real-time
hybrid simulation (RTHS) is conducted to perform the verification experiments using the nees@Lehigh facility. The physical substructure of
the RTHS is a 10-m tall planar steel frame equipped with a large-scale MR damper. Through RTHS, the test configuration is used to represent
two code-compliant structures, and is evaluated under seismic excitation. The results from numerical simulation and RTHS are compared to
verify the RTHS methodology. The global responses of the full system are used to assess the performance of each control algorithm. In each
case, the reduction in peak and root mean square (RMS) responses (displacement, drift, acceleration, damper force, etc.) is examined. Beyond
the verification tests, the robust performance of the damper controllers is also demonstrated using RTHS. DOI: 10.1061/(ASCE)ST.1943-
541X.0001093. © 2014 American Society of Civil Engineers.

Introduction must be sought and developed. Validation of these new concepts

requires sophisticated experimental techniques. Traditional struc-
As the existing infrastructure evolves to meet the needs of future tural testing methodologies encompass many different approaches,
generations, innovative approaches to design and hazard mitigation from quasistatic testing to shake-table testing to conventional hy-
brid testing. Ideally, the full-scale structure should be the test speci-
1
Ph.D. Recipient, Purdue Univ., 550 Stadium Mall Dr., West Lafayette, men, as this eliminates errors associated with modeling and scaling.
IN 47907 (corresponding author). E-mail: [email protected] However, this type of testing would be extremely time consuming
2
Professor, Purdue Univ., 550 Stadium Mall Dr., West Lafayette, IN 47907. and cost prohibitive in virtually all situations. Shake-table testing
3
Assistant Professor, Univ. of Maryland, 1145 Glenn L. Martin Hall, allows the test structure to undergo inertial forces due to ground
College Park, MD 20742. motion, but very few shake tables in the world are capable of testing
4
Master’s Degree Recipient, Lehigh Univ., 117 Atlss Dr., Bethlehem,
full- and large-scale specimens. Conventional hybrid simulation
PA 18015.
5
Ph.D. Candidate, Lehigh Univ., 117 Atlss Dr., Bethlehem, PA 18015.
(Dermitzakis and Mahin 1985), or pseudodynamic testing, com-
6
Ph.D. Recipient, Lehigh Univ., 117 Atlss Dr., Bethlehem, PA 18015. bines physical and analytical substructures to perform realistic test-
7
Ph.D. Recipient, Purdue Univ., 550 Stadium Mall Dr., West Lafayette, ing over an extended period (e.g., a 30-s earthquake may require
IN 47907. hours or days to complete). This type of testing captures the inertial
8
Ph.D. Recipient, Univ. of Connecticut, 261 Glenbrook Rd., Unit 3037, effects in either structure due to ground excitation by including
Storrs, CT 06269. them in the numerical substructure.
9
Ph.D. Recipient, City College of New York, Steinman Hall T-121, However, the need to perform structural testing in real time is
Convent Ave., 140th St., New York, NY 10031. paramount when rate-dependent effects play a significant role in the
10
Postdoctoral Researcher, City College of New York, Steinman Hall system’s behavior, as is the case with most existing advanced
T-121, Convent Ave., 140th St., New York, NY 10031.
11 damping devices. Thus, an alternative to the traditional test meth-
Ph.D. Candidate, Purdue Univ., 550 Stadium Mall Dr., West
Lafayette, IN 47907.
ods that is gaining traction in the earthquake engineering commu-
12
Newmark Endowed Chair in Civil Engineering, Univ. of Illinois at nity is real-time hybrid simulation (RTHS; Fig. 1). The RTHS
Urbana-Champaign, 205 N. Mathews Ave., Urbana, IL 61801. enforces the real-time constraint for command displacements to
13
Bruce G. Johnston Professor, Lehigh Univ., 117 Atlss Dr., Bethlehem, be applied to the experimental specimen to preserve rate depend-
PA 18015. ence when examining the behavior of any physical substructure.
14
Associate Professor, Univ. of Connecticut, 261 Glenbrook Rd., Unit There are several advantages to this type of testing, including in-
3037, Storrs, CT 06269. creased safety and reduced costs as a result of not requiring complete
15
Professor, City College of New York, Steinman Hall T-121, Convent full-scale physical models, not requiring high-fidelity analytical
Ave., 140th St., New York, NY 10031. models for components that may be quite difficult to effectively
16
Joseph T. Stuart Professor, Lehigh Univ., 117 Atlss Dr., Bethlehem,
model (i.e., highly nonlinear devices), and the ability to effectively
PA 18015.
Note. This manuscript was submitted on April 23, 2013; approved on capture the behavior of rate-dependent devices. Thus, advances in
April 3, 2014; published online on July 24, 2014. Discussion period open the field of RTHS are enabling more cost-effective and efficient
until December 24, 2014; separate discussions must be submitted for in- evaluations of new structural systems under realistic conditions.
dividual papers. This paper is part of the Journal of Structural Engineer- In RTHS, the integration algorithm must generate smooth, con-
ing, © ASCE, ISSN 0733-9445/04014150(13)/$25.00. tinuous, and stable displacement commands that can be realized by

© ASCE 04014150-1 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 dampers is used to represent a portion of two code-compliant,
general-purpose structures equipped with advanced damping sys-
tems (the rest of the structure is modeled analytically). Compari-
sons of the RTHS results with those from numerical simulation
are first performed to verify the RTHS methodology. The availabil-
ity of verified RTHS methods will enable a much wider variety
of real-time studies than has been attempted previously. To assess
the performance of each damper control algorithm, the global
responses of the structure under general earthquake loading are ex-
amined and compared using both numerical simulation and RTHS.
Furthermore, to examine the robust performance of the controllers
when facing uncertain conditions, changes in the structural mass
are considered using RTHS and the results are compared with
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

the results with the nominal structure.

Fig. 1. RTHS concept Problem Formulation

Consider a seismically excited structure being controlled with

n devices placed between the floors of the structure. In a RTHS,
the structure in question is partitioned into two components,
servohydraulic actuators. Several integration algorithms have been namely, an analytical substructure and a physical substructure
suggested. Chen and Ricles (2008) proposed an explicit integration (shown in Fig. 2). The physical substructure is tested in the
algorithm, which utilizes the current measured state of a system to laboratory, whereas the analytical substructure is evaluated compu-
calculate the state at a future time. By contrast, Wu et al. (2007) tationally. Displacements calculated from the analytical substruc-
proposed an implicit method, which solves equations using both ture are applied to the physical substructure using high-speed
current and future systems states. Another critical component of hydraulic actuators, and the resulting measured forces from the
RTHS is achieving accurate tracking of a desired displacement physical substructure are fed back into the analytical substructure
command signal using servohydraulic actuators. Because actuators for the next set of calculations. This feedback relationship is out-
are dynamic systems, time lags may result due to the physical dy- lined in the following equations. Consider a reference multidegree
namics/limitations of the servohydraulic actuators and can vary of freedom (MDOF) mass-spring-dashpot system with its constitu-
with both the excitation and specimen (Dyke 1996). To overcome ent components.
this, several control methods have been proposed. Horiuchi et al. The equations of motion for the total MDOF system under a
(1996) proposed the polynomial extrapolation delay compensation seismic excitation are
method. Modern approaches (Chen and Ricles 2010) are creating
inverse transfer function of low-order models that represent the
Fẍg ¼ m ẍ þ c ẋ þk x ð1Þ
dynamics of the actuator. Model-based controllers (Carrion and
Spencer 2007; Carrion et al. 2009) address the tracking issue
using feedforward–feedback combination controllers. While where m = the mass matrix; c = the damping matrix; m = the
displacement-based command and tracking is accepted and stiffness matrix; Fẍg ¼ ½Fẍg ; : : : ; Fẍgn
T is the force vector due
1
widely implemented, other methods, including force control to the ground excitation; ẍ ¼ ½ẍ1 ; : : : ; ẍn
T is the vector of floor
(Chantranuwathana and Peng 1999) and acceleration control accelerations; ẋ ¼ ½ẋ1 ; : : : ; ẋn
T is the vector of floor velocities
(Nakata 2010) may also prove useful in future endeavors. (relative to the ground); and x ¼ ½x1 ; : : : ; xn
T = vector of floor
Although conceptually very attractive, further research is displacements (relative to the ground), respectively. The mass,
needed to identify and meet the challenges associated with the stiffness, and damping coefficients of each floor can be divided into
use of RTHS in large-scale experiments to assess these complicated analytical and physical components
physical phenomena. Progress in this cutting-edge research field
requires expertise from interdisciplinary research teams. Moreover,
numerous practical challenges, including those on the physical side
(e.g., time delays/lags, sensor noise, actuator dynamics, control–
structure interaction) and on the computational side [e.g., computa-
tional support (Saouma et al. 2012) real-time execution, numerical
modeling errors, robust actuator control, integration algorithms]
need to be considered and understood. Although tests using several
types of isolated physical substructures have been conducted
[e.g., magnetorheological (MR) dampers], very few have utilized
a significant structural component, such as a frame (Gao et al.
2013; Nakata and Stehman 2012; Cha et al. 2012, 2014; Saouma
et al. 2013).
The focus of this paper is twofold: (1) performance verification
of large-scale RTHS through comparisons with numerical simula-
tion, and (2) performance evaluation of control algorithms devel-
oped for use with large-scale MR dampers. In each RTHS test, a
Fig. 2. Example of substructuring
multistory, large-scale steel frame equipped with large-scale MR

© ASCE 04014150-2 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 mi ¼ mai þ mpi ; ci ¼ cai þ cpi ; ki ¼ kai þ kpi ð2Þ

where i ¼ 1; : : : ; n. In addition, the measured forces from the

physical substructure (which incorporate the internal forces due
to any control devices present within the physical substructure)
are defined as

FM ¼ mp ẍ þ cp ẋ þ kp x ð3Þ

where FM , mp , cp , and kp = vector of measured forces, and the

mass, damping and stiffness matrices for the physical substructure,
respectively. Substituting Eqs. (2) and (3) into Eq. (1) yields the
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

equations of motion for the analytical substructure

Fẍg ¼ ma ẍ þ ca ẋ þ ka x þ FM ð4Þ

Fig. 3. RTHS large-scale frame testing setup at Lehigh University

where ma , ca , and ka = mass, damping, and stiffness matrices, re-
spectively, of the analytical substructure.
In the implementation of the RTHS, the measured force is ob-
tained using load cells attached to the actuators driving the physical
substructure. This quantity is necessary to compute the displace- Experiments are conducted at the nees@Lehigh real-time multi-
ment in the analytical substructure (using a numerical solver) directional (RTMD) large-scale testing facility. For these simula-
for the next time step. The resulting calculated displacement is then tions, three actuators are used to apply displacements in the
imposed on the physical substructure, and the measured forces are DBF (Fig. 3). The actuators and specimen are mounted to a con-
acquired again. This cycle repeats at every time step in real time for crete L-shaped reaction wall designed to minimize elastic and out-
the duration of the experiment. of-plane deformation under any high forces experienced during
testing procedures. A braced frame system, consisting of two rel-
atively stiff steel frames placed parallel to each other, are securely
Experimental Setup and Procedure bolted to the concrete reaction wall to minimize out-of-plane mo-
tion in the test specimen. To minimize friction between the speci-
To perform RTHS, appropriate division of the specimen into men and the braced frame, Teflon pads are installed at several
physical and analytical substructures is needed. Furthermore, an contact points along each floor. The specimen is restrained against
in-depth understanding of the behavior of the physical substructure, rocking with bolted connections to rigid steel base plates mounted
including the actuators driving the specimen, will facilitate effec- to the strong floor.
tive RTHS.
Two full-scale structures are used as the basis for the RTHS
performed at the nees@Lehigh facility, namely, a 3-story proto- System Identification
type structure designed for this study and a 9-story benchmark
To conduct a RTHS, the global stiffness matrix for the physical
structure. The 3-story prototype structure used in this study
DBF constructed in the laboratory must be identified (Friedman
was designed by the authors from Lehigh University (Friedman
2012). Quasistatic testing, in which a force is applied cyclically
2012), whereas the 9-story structure was designed by
Brandow & Johnston Associates for the SAC Phase II Steel to a single floor through a hydraulic actuator and the resulting floor
Project (Ohtori et al. 2004). Although not actually constructed, displacement at each level is measured, is used to determine the
both structures adhere to regulations in the seismic code and global stiffness matrix representing the frame. In this approach,
are meant to exemplify typical low-rise and midrise building office each floor is loaded individually while the two remaining floors
buildings (for general commercial use), respectively, designed for are allowed to displace with no external force as measured by their
the Los Angeles, California region. respective actuators. The sampling rate is 128 Hz.
An individual RTHS is performed to analyze each structure. In Each floor of the experimental DBF is treated as a horizontal
both cases, a large-scale, 3-story steel damped-braced frame DOF, yielding a 3 × 3 stiffness matrix. Each floor of the analytical
(DBF) is used as the physical substructure, whereas the rest of substructure is assumed to be rigid, whereas the actual stiffness is
the structure is modeled analytically using the real-time capabil- used for the physical substructure. Thus, an effective stiffness is
ities provided by an xPC Target system (MATLAB 2011). The obtained using this testing procedure. By plotting the displacement
single-bay frame is 8.38 m (27.5 ft) tall and 4.57 m (15 ft) wide. of each floor versus the known applied load from an actuator, the
It is composed of two W8 × 67 steel columns, three W10 × 30 flexibility coefficient f ij (i.e., the displacement at the ith DOF due
steel beams, and six HSS8 × 6 × 3=8 steel braces. The frame to a unit force at the jth DOF and corresponding matrix element)
has natural frequencies of 1.2, 4.0, and 8.2 Hz. The large-scale can be determined. Each test produces three flexibility coefficients,
DBF is chosen as the physical substructure for the RTHS so that and for any test that exhibits hysteresis, an average slope is taken
a substantial portion of the stiffness of the total structures is tested for both the loading and unloading curves, as shown in Fig. 4.
physically in the laboratory. The constructed DBF is 60% scale, as Using a set of three tests, nine flexibility coefficients are deter-
compared with the full-scale 3-story prototype structure, due to mined and the flexibility matrix is formed. Because the flexibility
the physical limitations of the nees@Lehigh facility. Thus, time matrix includes the external constraints and represents the re-
is scaled by the square root of 0.6. strained condition, the symmetric stiffness matrix representing

© ASCE 04014150-3 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 the experimental DBF is calculated by inverting the flexibility ma-
trix and averaging the off-diagonal terms.
In addition to the experimentally identified interstory stiffness
for the physical DBF, it is necessary to identify the effective stiff-
ness of the bracing system that transfers the force of the MR
damper to the DBF (Fig. 5). Because the MR damper piston dis-
places relative to the floor of the DBF where it is attached, the brac-
ing system is treated as an additional horizontal DOF at each floor
where a MR damper is installed. This differential can be attributed
to several sources, including compliance within the clevis plate at-
tachment to the MR damper and imperfect rigidity with the result-
ing displacement/rotation of the beam representing the ground
floor. For effective control design, it is particularly important to
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

identify and include the effective stiffness at this DOF, as an inac-

curate model may cause controller performance to suffer.
To identify the effective stiffness of the bracing DOF, data from
several dynamic actuator tests are used. The effective bracing stiff-
ness is defined as Fig. 4. Example of a flexibility coefficient determination from quasi-
static test data
kbr ¼ FMR =ðxi;act − xi−1;act Þ − xmr piston ð5Þ

where FMR = measured damper force; xi;act = measured displace-

ment of the actuator connected to the ith floor of the structure brac-
ing being identified; xi−1;act = measured displacement of the
actuator connected to the floor below the ith floor of the structure;
and xmr piston = displacement of the MR damper piston. If i ¼ 1,
xi−1;act becomes the displacement of the ground floor. Because
FMR and the displacements are sampled at each time interval,
the computed stiffness estimate is also a function of time. Because
noise and numerical errors pollute this estimate, a histogram is used
to find the value with the highest occurrence, which is taken as the
identified bracing stiffness. From the data, the effective stiffness of
the bracing DOFs at both the first and second floors of the DBF are
identified as 48,600 and 42,135 kN=m, respectively.
With the measured mass of the DBF, and an assumed damping
ratio of 5% for each DOF of the DBF, the identified characteristic
matrices for the 5DOF (horizontal) DBF are summarized as follows
(the mass matrix is given as a row vector of the diagonal terms):


N × s2
M DBF ¼ ½ 3.15 3.15 3.15 0.3 0.3

m (a)
2 5 4
3
1.1e −3.9e 9.4e −4.9e4
3 0
6 7 6000
6 −3.9e4 4.75e4 −2e4 0 −4.2e4 7 Data Set 1
6 7 Data Set 2
6 7 Data Set 3
K DBF ¼ 6 9.4e3 −2e4 1.12 0 0 7ðNÞ Data Set 4
6 7 5000
6 −4.9e4 0 0 4.9e4 0 7
4 5
0 −4.21e4 0 0 4.2e4 4000
2 3
# of Occurrences

161 −59.3 3.0 −21.9 0

6 7
6 −59.3 152 −53.2 0 −21.9 7
3000
6 7 N × s
6 7
CDBF ¼ 6 3.0 −53.2 80.5 0 0 7
6 7 m
6 −21.9 0 0 57.8 0 7
2000
4 5
0 −21.9 0 0 57.8
1000

Large-Scale MR Damper Identification 0

2 3 4 5 6 7
Many nonlinear models have been developed to capture the behav- 10 10 10 10 10 10
Stiffness (kN)
ior of MR devices, including several nonparametric models
(Metered et al. 2009). Bass and Christenson (2007) adapted the (b)
hyperbolic tangent model for use with a 200-kN MR damper.
Fig. 5. (a) Bracing system in the DBF; (b) identification of the bracing
Another approach was adopted by Spencer et al. (1997) to create
stiffness
the phenomenological Bouc–Wen model. This model has been

© ASCE 04014150-4 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 shown to capture the behavior of MR dampers very well and can be Table 1. Identified MR Damper Model Parameters
simulated with little computational effort. As such, this model is Parameter Value
chosen for this experimental study.
αA 688 kN
To accurately reproduce the behavior of the device in the oper-
αB 0.006 kN
ating range, parameter identification was performed for the damp- αC −698 kN
ers subjected in these experiments. Certain parameters of the model αD −0.9 kN
are clearly related to the current in the electromagnet. Given the c0A 201 kN · s=m
expanded operating range of the devices for this study (7.5 A, c0B 0.1 kN · s=m
as opposed to 0–2.5 A in previous experiments), three parameters c0C −109 kN · s=m
are identified as functions of current input, as given in the following c0D −2.1 kN · s=m
equations: c1A 100 kN · s=m
c1B 15,000 kN · s=m
α ¼ αA eðαB iÞ þ αC eðαD iÞ ð6Þ k0 0.06 kN=m
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

k1 0.06 kN=m
x0 0.01 m
c0 ¼ c0A eðc0B iÞ þ c0C eðc0D iÞ ð7Þ β 3,000 m−1
γ 3,000 m−1
pffiffiffiffiffi A 337
c1 ¼ c1A jij þ c1B ð8Þ n 2

where i = current applied to damper from the current driver. In total,

optimal values of 17 parameters (αA , αB , αC , αD , c0A , c0B , c0C ,
c0D , c1A , c1B , k0 , k1 , x0 , β, γ, η, and A) must be determined to
model the MR damper. represent the behavior of both MR dampers. The identified param-
In addition, the dynamics associated with commanding current eters are summarized in Table 1.
and achieving rheological equilibrium have a large impact on the Because MR dampers are intrinsically nonlinear and rate depen-
performance of the MR damper and must be modeled (Jiang et al. dent, it is challenging to develop accurate models that describe their
2010; Jiang and Christenson 2012). The pulse-width modulation behavior for control design and evaluation purposes. However, it is
(PWM) current driver is represented by a first-order filter conceivable that during a RTHS, feedback and dynamic interaction
of physical and analytical substructures could result in potentially
I PWM 1 unstable behavior. Thus, it is desirable to conduct pretest simula-
HðsÞ ¼ ¼ ð9Þ
U 1 þ τs tions where the physical specimen is replaced with an appropriate
numerical model with the numerical RTHS component left un-
where I PWM = Laplace transform of the current output from the changed. These pretest simulations require a MR damper model
PWM; U = Laplace transform of the command current sent to that can exhibit stability and convergence at larger fixed integration
the PWM; and τ = time constant. time steps, and provide computational efficiency, speed of calcu-
The dynamics associated with the damper reaching rheological lation, and accuracy during pretest verification of the experimental
equilibrium as a result of the application of a magnetic field are setup.
related to the current within the electromagnetic coil inside the The addition of the PWM current driver and the inductance
damper. This current will experience dynamics due to the induct- models add accuracy to the model in the more commonly exam-
ance within this coil, and is modeled as ined higher velocity ranges in seismic applications. However, the
behavior of the MR damper can be difficult to model under low-
I 1
GðsÞ ¼ ¼ ð10Þ amplitude and low-velocity range. Under such conditions, while
I PWM 1 þ RL s the low-velocity force is still controllable, the yielding behavior is
less pronounced for the higher damper currents (this behavior, as
where R = resistance of the damper and L = inductance in the coil
the authors speculate, is because the MR damper fluid in the gap
which, as a function of current, is defined as the following:
between the electromagnetic coil and the housing can remain ener-
R×i gized and present in the gap even at zero current). This results in
LðiÞ ¼ þ L0 ð11Þ uneven and offset forces with zero current input observed. For this
di=dt
reason, RTHS is helpful in evaluating the performance of the MR
where L0 = inductance constant and i = time-varying current. dampers (Jiang et al. 2013).
A set of experimental 200-kN MR dampers (Christenson et al.
2008; Jiang and Christenson 2011) was selected as damping devi-
Actuator Controller Design
ces for these experiments. Because the dampers have been used in
prior applications, new model parameters for numerical simulation The design of an actuator control strategy is an important step
must be identified to capture any changes in behavior/performance toward successful RTHS due to the real-time experimental con-
that may have occurred. Testing for parameter identification was straints. Actuator control is necessary to ensure accurate tracking
conducted at the Smart Structures Technology Laboratory at the of the desired displacements and the compatibility of the analytical
University of Illinois (Urbana-Champaign) and at the RTMD fa- and physical substructures. Furthermore, when multiple actuators
cility at the nees@Lehigh laboratory. The dampers were subjected are connected to the same specimen, the dynamics of the actuators
to sinusoidal excitation, with fixed amplitude, frequency, and con- become coupled through the specimen, leading to a complex con-
stant voltage input. Utilizing a wide range of frequencies and volt- trol challenge (i.e., when an actuator applies a force to the structure,
age values, the resulting data are used to identify the parameters the other actuators will also experience this force; Phillips and
of the model with a constrained nonlinear optimization curve fit Spencer 2012). Consider the multiinput multioutput transfer func-
in MATLAB (2011). The performance of the dampers was similar, tion model Gdu ðsÞ of the linearized servohydraulic system, includ-
such that a single set of model parameters was sufficient to ing the actuators, servovalves, servocontrollers, and specimen as

© ASCE 04014150-5 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 Servo-Hydraulic System Gdu(s)
The feedforward controller is created as a model inverse to can-
cel the modeled dynamics of the servohydraulic system. The feed-
u + f d
G s (s ) G a (s ) G df ( s ) back controller is added to further reduce the tracking error in the
+
− − presence of changing specimen conditions, modeling errors, non-
Servo-Controller Actuator Specimen linearities, and disturbances. Invoking the separation principal,
and Servo-Valve
As a linear quadratic Gaussian (LQG) controller is designed from in-
Natural Velocity Feedback
dependent linear quadratic regulator and Kalman filter designs
(Stengel 1986) using a state-space representation of the identified
model. The details of the model-based control strategy can be
Fig. 6. Servohydraulic system model found in Phillips and Spencer (2011), which is extended to multi-
actuator systems in Phillips and Spencer (2012).
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

represented in Fig. 6. In this block diagram, u is the command Experimental Results

vector, f is the force imparted by the actuators, and d is the dis-
placement vector as realized by the actuators and specimen. The Two series of experimental tests are conducted, namely, (1) verifi-
servohydraulic system contains inner-loop feedback control, which cation of large-scale RTHS and (2) evaluation of MR damper con-
provides nominal tracking of u as measured by d, however this troller performance using RTHS. To verify the performance of the
alone is inadequate for accurate and stable RTHS (Horiuchi et al. RTHS, the following two approaches are used: (1) comparison of
1996; Gao et al. 2013). results of a RTHS with results of an equivalent numerical simula-
The dynamics of the servocontroller and servovalve, actuator, tion, and (2) comparison of results from multiple repetitions of a
and specimen have been condensed into transfer functions single RTHS.
Gs ðsÞ, Ga ðsÞ, and Gdf ðsÞ, respectively. The parameter A repre- To demonstrate the MR damper controller performance, the
sents the effective cross-sectional areas of the actuator pistons. MR damper is used to control the response of two structures:
The input–output transfer function can be written in the Laplace (1) a 3-story prototype structure equipped with a single damper
domain as follows: on the first floor (Case I: 4DOF), and (2) a 9-story benchmark struc-
ture equipped with two MR dampers on the first and second floors
Gs ðsÞGa ðsÞGdf ðsÞ (Case II: 11DOF). In both cases, a large-scale steel frame, as de-
DðsÞ
Gdu ðsÞ ¼ ¼ ð12Þ fined earlier and equipped with MR damper(s) on the specified
UðsÞ I þ ½Gs ðsÞ þ As
Ga ðsÞGdf ðsÞ floors, is used as the physical substructure. The analytical substruc-
tures for both cases, summarized in the following sections, are
In multiactuator systems, the servohydraulic components modeled as linear, lumped parameter structural systems, reorgan-
operate independently, thus Gs ðsÞ, Ga ðsÞ, and A are diagonal. ized into state-space form, and executed using Simulink. Each
The source of actuator coupling is the specimen, namely, the off- RTHS is conducted using a 1=1,024-s time step, while utilizing
diagonals of Gdf ðsÞ. the Runge–Kutta numerical solver to evaluate the analytical
In model-based actuator control, an outer-loop controller is cre- substructure at each time step.
ated to compensate for the dynamics of the servohydraulic system, The 4DOF characteristic matrices (the mass matrix is given as
including specimen dependency and actuator coupling. For the a row vector of the diagonal terms) comprising the analytical
controller design, the tracking error e is defined as the difference substructure in Case I are defined as follows:
between the desired displacement vector r and measured displace-
ment vector d. Through regulator redesign, the tracking problem is

defined as a regulator problem about a set point (Phillips and N × s2
M ASCI ¼ ½ 98.1 98.1 70.8 0

Spencer 2011). The resulting model-based controller, incorporating m
both feedforward and feedback links, is represented in Fig. 7. 2 3
8e5 −5.6e 4 9.4e 3 0
The control strategy used in this study is based on a linearized 6 7
model of the servohydraulic system. Nonparametric system iden- 6 −5.6e4 8.2e4 −3.1e4 0 7
K ASCI ¼6
6
7ðNÞ;
7
tification is used to identify the poles and zeros of the multiinput 4 9.4e3 −3.1e4 1.9e4 0 5
multioutput system model following the procedure outlined in
Phillips and Spencer (2012). This identified model is used to design 0 0 0 0
2 3
both the feedforward and feedback controllers of Fig. 7. 225 −100 3.8 0
6 7

6 −100 176 −81 07 N × s
CAS ¼6
6 3.8
7
83.5 0 7
CI
4 −81 5 m
uFF
GFF(s) 0 0 0 0
Feedforward Controller
+
r + e uFB u d The fourth row and column of each matrix correspond to the
LQG Gdu(s) first-floor bracing DOF. Because the bracing is entirely physical,
– +
Feedback Controller Servo-Hydraulic
the row/column representing this DOF consists of zeros. These
System rows and columns are eliminated before constructing the state-
space model of the numerical substructure.
Similarly, the 11DOF characteristic matrices (the mass matrix is
Fig. 7. Model-based actuator control with feedforward and feedback
given as a row vector of the diagonal terms) comprising the ana-
links
lytical substructure in Case II are defined as follows:

© ASCE 04014150-6 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150



N × s2
M ASCII ¼ ½ 109 107 107 110 110 110 110 110 118 0 0

m
2 3
8.7e3 −9.6e3 −9.4e3 0 0
6 −9.6e3 4.8e4 −2.7e4 7
6 7
6 7
6 −9.4e3 −2.7e4 7.4e4 −3.9e4 7
6 7
6 7
6 0 −3.9e 4
7.8e4 −3.8e4 7
6 7
6 7
6 −3.8e4 6.9e4 −3.1e4 7
6 7
6 7
K ASCII ¼6 −3.1e4 5.6e4 −2.5e4 7ðNÞ
6 7
6 −2.5e4 4.5e4 −2.0e4 7
6 7
6 7
6 −2.0e4 3.9e4 −1.9e4 7
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

6 7
6 7
6 −1.9e4 1.9e4 7
6 7
6 7
4 0 5
0 0
2 3
92
6 7
6 −37 132 7
6 7
6 7
6 −17 −40 162 7
6 7
6 −5 −13 −60 184 7
6 7
6 7
6 −3 −5 −12 −62 173 7
6 7

6 7 N×s
CAS CII ¼6
6 −2 −3 −4 −11 −54 156 7
7 m
6 7
6 −1 −2 −3 −4 −10 −49 141 7
6 7
6 7
6 −0.7 −1 −2 −3 −5 −10 −45 128 7
6 7
6 −0.5 −1 −1 −2 −3 −5 −11 −50 89 7
6 7
6 7
6 0 0 7
4 ··· 5
0 ··· 0

As before, the tenth and eleventh rows and columns of each ma- numerical simulation results. An additional investigation is per-
trix correspond to the first-and second-floor bracing DOF, respec- formed to assess repeatability. Three RTHS runs using the same
tively. Because the bracing is entirely represented in the physical input are compared to demonstrate that this method generates
substructure, the row/column representing these DOFs consists of consistent results. The differences observed in the three tests are
zeros and eliminated before constructing the state-space model of observed to be negligible (Friedman 2012).
the numerical substructure.

MR Damper Controller Evaluation

RTHS Verification
Control strategies that can accommodate the dynamics of a particu-
To verify the RTHS method at large scale, comparisons are first lar device should be more effective in decreasing a structure’s re-
made between the results from a RTHS and those of an equivalent sponses and minimizing damage that may result from a seismic
numerical simulation. In this case, the 9-story structure from event. Control strategies can range from model-based controllers
Case II is used as the benchmark, given that over 90% of the total (bang bang, sliding mode, etc.) to soft-computing-based control
stiffness of the first three floors is present in the physical substruc- (fuzzy logic, neural networks, etc.).
ture. The numerical simulation of the total structure is performed Several control strategies are evaluated as a part of this inves-
before the experimental testing, with all components modeled in tigation. Two previously established control algorithms for MR
Simulink. The ground motion record is the 1940 El Centro earth- dampers are utilized, namely, (1) passive on (PON) and (2) clipped
quake (0.95-Hz natural frequency). For the first assessment, Fig. 8 optimal control (COC). The PON control sends a constant com-
shows a comparison of the ninth floor displacement, velocity, and mand signal of 2.5 V (the maximum operational voltage value)
acceleration responses, for the numerical simulation and the RTHS, to the damper. The COC (Dyke 1996) combines a LQG regulator
respectively. In this case, because it was not possible to measure the with a control law to select the voltage command level based on a
velocity in the laboratory and the accelerations contained high- desired restoring force. The third control algorithm, (3) overdriven
frequency content due to the data-acquisition system, both COC (ODCOC; Friedman 2012; Friedman and Dyke 2013; Yang
responses in the RTHS are taken from the numerical substructure 2001), is formulated specifically to take advantage of the unique
output. For each response, the RTHS results compare well with the features of large-scale MR dampers. The ODCOC is similar in

© ASCE 04014150-7 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 0.05 0.01
RTHS
Disp (m) 0.005

Disp (m)
SIM
0 0
−0.005
−0.05 −0.01
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50

0.2 0.02
0.1 0.01
Vel (m/s)

Vel (m/s)
0 0
−0.1 − 0.01

−0.2 − 0.02
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50

1 0.04
RTHS 1
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

Acc (m/s2)

Acc (m/s2)
0.5 RTHS 2
0.02
RTHS 3
0
0
−0.5

−1 − 0.02
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time (s) Time (s)
(a) (b)

Fig. 8. Comparison of (a) ninth floor responses; (b) repeatability

concept to the COC, with the added effects of overdriving and earthquake, and (7) SAC large earthquake. The SAC ground
backdriving incorporated into the control law. Previous studies motions are generated time histories for use in topical investiga-
have demonstrated that these features led to an increase in the abil- tions, case studies, and trial applications in the SAC Phase 2 Steel
ity to track a desired force in large-scale devices (Friedman Project (Los Angeles area, corresponding to seismic Zone 4).
et al. 2013f). Because of safety limitations for the frame, the maximum applied
To design the semiactive optimal controllers, a 20-Hz band- interstory drift is limited to 1% of the story height, or 23 mm. Thus,
limited white-noise signal is used as the excitation for the structure. each earthquake is scaled to 50% and 15% intensity for the 3-story
For control design purposes, the R matrix is selected to be an iden- prototype structure and 9-story benchmark structure, respectively,
tity matrix of proper order, and a wide variety of Q matrices are to maintain this limitation (the peak ground acceleration values
chosen and tested using a band-limited white-noise input. Based shown in Tables 2–4 are for the nominal earthquakes). In addition,
on the performance of these controllers, the weighting values for the 3-story structure, each SAC ground motion is scaled appro-
pffiffiffiffiffiffi
are selected and implemented. Ultimately, in both RTHS test cases, priately in time, by a factor of 0.6.
the optimal controllers are designed to minimize the absolute ac- In addition to the MR damper controller effectiveness under
celeration, with equal weighting on all floors of the structure. In the general earthquake loading, it is important for the controller to
first case, for the 3-story prototype structure, a weighting value of be robust to modeling errors (Spencer et al. 1998). To assess the
100,000 m2 is selected for both the COC and the ODCOC control- robust performance of the MR damper controller, the mass of
lers. In the second case, for the 9-story benchmark structure, the structure on which the controller is implemented is varied in
weighting values of 70,000 and 50,000 m2 are selected for both increments from 80% to 120% of the original design, in an effort
the COC and the ODCOC controllers, respectively. to gage the impact of uncertainties/modeling errors on control per-
To evaluate the efficiency of the proposed controllers, a selec- formance. The original controller is used in these tests, simulating
tion of several input excitations, ẍg , are used in simulation, includ- the modeling errors. The SAC small I earthquake is used as the
ing: (1) SAC small I earthquake, (2) SAC II small earthquake, input excitation for the structure, and all changes in mass are
(3) the north–south (NS) component of the 1940 El Centro earth- incorporated in the analytical substructure.
quake, (4) SAC medium I earthquake, (5) SAC medium I earth- Focusing on global structural performance, the controller
quake, (6) the NS component of the 1987 Superstition Hills evaluation criteria are divided into two categories, namely, peak

Table 2. Comparison of RTHS Results to Evaluate Robust Controller Performance, Case I

Evaluation criteria
Ground
motion % Mass Controller J1 J2 J3 J 4 (kN) J5 J6 J7 J 8 (kN)
SAC small 100 PON 0.97 1.38 1.5 136 1.09 1.31 1.48 21
EQ I PGA, 0.15 g ODCOC 0.88 0.93 0.95 78 0.88 0.95 1.29 11
105 PON 1.02 1.4 1.63 138 0.64 1.09 0.99 20
ODCOC 0.93 0.96 1.07 86 0.58 0.89 1.19 11
110 PON 1.06 1.35 1.69 142 0.61 1.04 1.25 20
ODCOC 0.94 0.94 1.13 85 0.58 0.88 1.18 11
120 PON 1.12 1.24 1.46 145 0.53 0.9 1.18 18
ODCOC 0.98 0.91 1.04 83 0.56 0.84 1.12 11
Note: EQ = earthquake; PGA = peak ground acceleration.

© ASCE 04014150-8 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 Table 3. Comparison of RTHS Results for General Earthquake Input, Case II
Evaluation criteria
Ground motion Controller J1 J2 J3 J 4 (kN) J5 J6 J7 J 8 (kN)
SAC small EQ I PGA, 0.15 g PON 0.85 0.58 1.37 115 0.76 0.48 1.15 18
COC 0.85 0.79 1.28 84 0.7 0.6 1 13
ODCOC 0.87 0.81 0.86 79 0.71 0.64 0.95 11
SAC small EQ II PGA, 0.325 g PON 0.51 0.35 0.84 71 0.61 0.39 0.85 22
COC 0.57 0.59 0.73 57 0.63 0.6 0.78 17
ODCOC 0.64 0.55 0.7 54 0.68 0.67 0.77 15
El Centro PGA, 0.3 g PON 0.66 0.33 0.98 134 0.62 0.4 0.89 28
COC 0.76 0.61 0.84 81 0.64 0.61 0.75 18
ODCOC 0.83 0.6 0.74 72 0.69 0.63 0.73 15
SAC medium EQ I PGA, 0.5 g PON 0.51 0.41 1.68 105 0.47 0.34 0.75 26
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

COC 0.58 0.69 1.44 59 0.52 0.54 0.7 18

ODCOC 0.6 0.73 1.22 54 0.56 0.59 0.68 16
SAC medium EQ II PGA, 0.55 g PON 0.97 0.54 0.97 142 0.89 0.54 0.95 32
COC 0.93 0.58 0.8 84 0.75 0.65 0.73 20
ODCOC 0.93 0.82 0.75 57 0.77 0.73 0.76 16
SAC large EQ I PGA, 1 g PON 0.79 0.83 1.87 152 0.5 0.39 1.19 29
COC 0.8 0.83 1.2 80 0.53 0.55 0.86 18
ODCOC 0.85 0.69 1.12 66 0.57 0.56 0.83 16
Note: EQ = earthquake; PGA = peak ground acceleration.

Table 4. Comparison of RTHS Results for Robust Performance Testing, Case II

Evaluation criteria
Design
Ground motion mass case Controller J1 J2 J3 J 4 (kN) J5 J6 J7 J 8 (kN)
SAC small EQ I PGA, 0.15g 80 PON 0.84 0.59 1.32 118 0.87 0.58 0.91 20
ODCOC 0.81 0.8 0.97 76 0.82 0.72 0.86 13
90 PON 0.82 0.58 1.49 116 0.87 0.57 0.94 18
ODCOC 0.83 0.81 0.96 79 0.81 0.73 0.87 12
95 PON 0.83 0.59 1.58 113 0.82 0.5 0.9 18
ODCOC 0.84 0.82 0.96 77 0.77 0.7 0.84 11
100 PON 0.85 0.58 1.37 115 0.76 0.48 1.15 18
ODCOC 0.87 0.81 0.86 79 0.71 0.64 0.95 11
105 PON 0.87 0.57 1.15 107 0.69 0.45 0.79 17
ODCOC 0.88 0.79 0.76 76 0.67 0.61 0.74 11
110 PON 0.89 0.56 1.14 119 0.65 0.41 0.75 17
ODCOC 0.92 0.8 0.82 76 0.63 0.6 0.71 10
120 PON 0.9 0.56 1.35 121 0.59 0.39 0.68 16
ODCOC 0.93 0.84 0.95 70 0.6 0.58 0.67 10
Note: EQ = earthquake; PGA = peak ground acceleration.

responses and RMS responses. Because most criteria are a For the RMS response categories, ODCOC and COC achieve
ratio of the controlled value to the uncontrolled value, in general, similar performance overall.
smaller values for the evaluation measures are indicative of For the medium earthquakes cases, as before, the ODCOC
better performance. A summary of these criteria is presented in algorithm performs significantly better than both PON (an average
Table 5. of 15% improvement) and COC (an average of 4% improvement)
in peak acceleration reduction. The ODCOC and PON each achieve
good reductions in drift and displacement reduction for both earth-
Case I Experimental Results quakes. However, the COC and ODCOC achieve these results with
In Case I (Fig. 9, Tables 2 and 6), for the smaller earthquakes, the significantly smaller peak and RMS control forces associated
PON and ODCOC controllers split achieving greater reduction in with them.
peak displacement and drift. However, the ODCOC results in lower For the large earthquake cases, the optimal controllers achieve
responses than the COC for these categories, by an average of 7% significantly better results in terms of acceleration reduction
for displacement, and 5% for drift. For peak absolute acceleration, compared with the passive control approach (approximately
the COC and ODCOC approaches yield lower responses than the 15% improvement), and ODCOC is only marginally better than
PON controller, by an average of approximately 25%; however, the COC (2% improvement). The ODCOC and PON again split the
ODCOC is typically superior to COC by 5–6%. The COC does best peak-displacement reduction, whereas the best drift-reduction
have the best performance in terms of peak and RMS force (indi- varies between all controllers. COC is the best control in terms of
cating that this approach used the least amount of force to achieve peak and RMS damper force.
response reduction, which can result in economic savings in real- Overall, the ODCOC seems to be the superior controller
world applications), compared with the other two controllers. in terms of peak acceleration response reduction (for which it

© ASCE 04014150-9 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 Table 5. Evaluation Criteria Definitions
J value Equation Description
J1 maxEQ f½maxt;i jxi ðtÞj
=xmax g Peak floor displacement
Ratio of controlled maximum relative displacement to the uncontrolled value
J2 maxEQ f½maxt;i jdi ðtÞj=hi
=dmax
n g Peak interstory drift
Ratio of controlled maximum interstory drift to the uncontrolled value
J3 maxEQ f½maxt;i jẍai ðtÞj
=ẍmax
a g Peak floor acceleration
Ratio of controlled maximum absolute acceleration to the uncontrolled value
J4 maxEQ fmaxt;i jf i ðtÞjg Peak control force
Ratio of the maximum device output force to the weight of the structure
J5 maxEQ f½maxt;i jxRMSi ðtÞj
=xRMSmax g RMS floor displacement
Ratio of controlled root-mean-square relative displacement to the uncontrolled value
J6 maxEQ f½maxt;i jdRMSi ðtÞj=hi
=dRMSmax
n
g RMS interstory drift
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

Ratio of controlled root-mean-square interstory drift to the uncontrolled value

J7 maxEQ f½maxt;i jẍRMSai ðtÞj
=ẍRMSmax
a
g RMS floor acceleration
Ratio of controlled maximum absolute acceleration to the uncontrolled value
J8 maxEQ fmaxt;i jf RMSi ðtÞjg RMS control force
Ratio of the maximum device output force to the weight of the structure

3 3
was designed), with an average improvement of 20% versus PON
control, and 4–5% versus COC. The ODCOC and PON are similar
2.5 2.5 in terms of displacement and drift reduction, with no clear trend
across all earthquakes. The COC and ODCOC also typically use
approximately 25% less force than PON control.
2 2 In addition, for the robust performance testing, the ODCOC
controller outperforms the PON controller in nearly every category
for every mass level. The results clearly demonstrate that the
Story #

1.5 1.5 ODCOC controller is robust in terms of performance. The peak

displacement is reduced by an average of 15%, as compared with
the PON case, and peak drift is reduced a further 30%. The peak
1 1 acceleration is reduced by an average of 40% in comparison to the
PON reduction, while using approximately 70% of the peak force.
All of the RMS categories favor the ODCOC as well.
0.5 0.5 UNCTRL
PON
COC
ODCOC Case II Experimental Results
0 0
0 0.2 0.4 0.6 0.8 1 0 2 4 6 In Case II (Fig. 10, Tables 3 and 4), the 9-story benchmark building,
Peak Drift (% Story Height) Peak Acceleration (m/s2)
several important trends emerge in the results. For the smaller earth-
Fig. 9. Response profiles for SAC Large EQ I, Case I quake cases, the ODCOC controller consistently achieves a greater
reduction in peak absolute acceleration, with an average 25%

Table 6. Comparison of RTHS Results for Various Ground Motions, Case I

Evaluation criteria
Ground motion Controller J1 J2 J3 J 4 (kN) J5 J6 J7 J 8 (kN)
SAC small EQ I PGA, 0.15 g PON 0.98 1.388 1.5 136 1.09 1.31 1.48 21
COC 0.93 0.98 0.98 77 0.89 0.97 1.36 12
ODCOC 0.88 0.93 0.95 78 0.88 0.95 1.29 11
SAC small EQ II PGA, 0.325 g PON 0.64 0.73 0.87 162 0.6 0.72 0.81 41
COC 0.71 0.76 0.74 145 0.59 0.63 0.66 26
ODCOC 0.72 0.76 0.74 146 0.61 0.64 0.67 26
El Centro PGA, 0.3 g PON 0.72 0.82 0.93 182 0.62 0.74 0.82 39
COC 0.74 0.7 0.73 154 0.62 0.69 0.7 27
ODCOC 0.74 0.7 0.68 180 0.62 0.68 0.69 27
SAC medium EQ I PGA, 0.5 g PON 0.65 0.72 1.07 179 0.59 0.7 0.77 34
COC 0.68 0.75 0.97 156 0.62 0.66 0.67 22
ODCOC 0.69 0.75 0.93 155 0.63 0.67 0.69 22
SSH (DBE) PGA, 0.75 g PON 0.52 0.62 0.86 185 0.5 0.61 0.69 50
COC 0.68 0.8 0.86 182 0.65 0.7 0.71 35
ODCOC 0.68 0.79 0.84 211 0.64 0.71 0.71 37
SAC large EQ I PGA, 1 g PON 0.91 1.11 1.12 198 0.79 0.95 1.02 43
COC 0.85 0.96 0.89 152 0.72 0.81 0.85 25
ODCOC 0.84 0.99 0.87 176 0.73 0.82 0.86 25
Note: Bold values indicate the best performance of the MR control strategies. If a controller has the most bold entries for a given earthquake, that controller had
the best performance, in terms of response reduction. EQ = earthquake; PGA = peak ground acceleration.

© ASCE 04014150-10 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 9 9
the optimal controllers actually outperform the passive controller
for both displacement and interstory drift, in general.
8 8

7 7
Conclusions
6 6
RTHS is a new experimental approach that can facilitate dynamic
5 5
testing of rate-dependent specimens at large scale without the pro-
Story #

hibitive expenses associated with alternative approaches. Further-

4 4
more, as demonstrated in this study, RTHS enables a variety of tests
using a single specimen. The availability of the RTHS method will
3 3
enable a much wider variety of dynamic tests than has been at-
tempted previously. Using a significant structural component
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

2 2
(a large-scale steel frame equipped with multiple large-scale MR
dampers) at the nees@Lehigh facility, scenarios involving two
UNCTRL
1 1 PON
different structures are considered using RTHS. To verify the RTHS
COC approach, comparisons are made between the results of a RTHS
ODCOC
0 0 with a pure numerical simulation. The global results demonstrate
0 0.05 0.1 0.15 0.2 0 0.2 0.4 0.6 0.8
that RTHS is a viable experimental approach for large-scale,
Peak Drift (% Story Height) Peak Acceleration (m/s2)
dynamic testing.
Fig. 10. Response profile for SAC Small EQ I, Case II Subsequently, RTHS is used to evaluate MR damper controller
approaches. Two scenarios involving different structures and multi-
ple earthquake excitations are tested. The first case considers a
3-story office building equipped with a single MR damper on the
improvement as compared with the PON controller, and an average first floor. The second case involves a 9-story office building
10% improvement when compared with the COC controller. In equipped with MR dampers on the first and second floors. In gen-
addition, the semiactive control methods use significantly less peak eral, the semiactive control approaches are superior to the passive
and RMS damper forces to achieve their respective performances, control approach for reduction in peak absolute acceleration. The
as compared with the passive approach. However, using PON con- ODCOC is the best option for each earthquake, with an average
trol did yield greater reductions in peak relative displacement and improvement of 25% versus the PON cases and 5% versus the
drift. It should be noted that while the semiactive methods could not COC case. Further, PON and ODCOC both perform well in terms
match the PON performance in tests using small ground motions, of drift and displacement reduction. The COC approach is superior
the ODCOC results in lower responses than COC for these at utilizing the smallest damper force to achieve good results.
categories. The proposed controller is also found to be robust in terms of
For the medium earthquakes cases, the ODCOC algorithm again its performance under model mass variations.
performs much better than both PON (an average 25% improve- All data pertaining to this study are available in the NEES
ment) and COC (an average 13% improvement) in peak accelera- Project Warehouse (http://nees.org), including data sets related
tion reduction. Using a larger force, the PON controller yields to the identification of the large-scale MR damper (Chae et al.
lower interstory drifts in both cases, by an average of 10%, 2013a, b; Phillips et al. 2013), identification of the large-scale steel
although COC and ODCOC require significantly smaller forces test frame (Ozdagli et al. 2013a, b, c, d, e), and the RTHS data
(peak and RMS) to achieve the associated reductions. (Friedman et al. 2013a, b, c, d, e).
For the SAC large earthquake case, the largest earthquake
tested, the ODCOC is clearly the superior controller, with improved
reductions in peak drift, peak acceleration, and damper force (both Acknowledgments
peak and RMS) compared with both PON and COC. Both semi-
active controllers have smaller peak forces and RMS forces, as The authors acknowledge the financial support from the National
compared with PON. Science Foundation (NSF-CMMI Grant No. 1011534). In addition,
These trends continue in the results of the tests to evaluate ro- the use of the RTMD facility at the nees@Lehigh facility to conduct
bustness (Table 4). In each test, ODCOC is the superior controller the RTHS (supported by the National Science Foundation CMMI
in terms of peak and RMS acceleration as well as peak and RMS Directorate under Cooperative Agreement Number CMMI-
damper force, by a wide margin. The PON control is superior for 0927178) and the utilization of the large-scale MR dampers owned
peak and RMS drift in all cases, and the best controller for peak and by Dr. Richard Christenson, are greatly appreciated. Lastly, the sup-
RMS displacement varies as the mass increases. port of Gary Novak, Thomas Marullo, Peter Bryan, and all of the
When considering the performance of the controller, the inter- staff at the nees@Lehigh facility were instrumental in conducting
story drift performance should consider the characteristics of the these tests.
structure. This structure has a soft story at the first floor, due to
the increased height at this level. Thus, for each excitation, the peak
interstory drift occurred at the first floor, where the damper is References
located. For the PON case, the damper uses greater force and es-
Bass, B., and Christenson, R. E. (2007). “System identification of 200 kN
sentially locks up the floor. Because the optimal controllers are magnetorheological fluid damper for structural control in large-scale
designed to generate the best acceleration response for the overall smart structures.” American Control Conf., IEEE, 2690–2695.
structure, they generally allow a higher degree of flexibility Carrion, J. E., and Spencer, B. F. (2007). “Model-based strategies for real-
in the first floor, and so the global response of the structure is time hybrid testing.” Newmark Structural Engineering Laboratory Rep.
improved. If this soft story is eliminated from consideration, Series No. 6, Univ. of Illinois at Urbana-Champaign, Urbana, IL.

© ASCE 04014150-11 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 Carrion, J. E., Spencer, B. F., and Phillips, B. M. (2009). “Real-time hybrid large-scale MR dampers using real-time hybrid simulation.” IISL
simulation for structural control performance assessment.” Earthquake Rep. Series 03, 〈http://nees.org/resources/6622〉 (Jun. 17, 2013).
Eng. Eng. Vib., 8(4), 481–492. Friedman, A. J., Dyke, S. J., and Phillips, B. M. (2013f). “Over-driven con-
Cha, Y.-J., et al. (2012). “Comparative studies of semi-active control trol for large-scale MR dampers.” Smart Mater. Struct., 22(4), 045001.
strategies for MR dampers: Pure simulation and real-time hybrid Gao, X., Castaneda, N., and Dyke, S. J. (2013). “Real-time hybrid simu-
tests.” J. Struct. Eng., 10.1061/(ASCE)ST.1943-541X.0000639, lation: From dynamic system, motion control to experimental error.”
1237–1248. Earthquake Eng. Struct. Dyn., 42(6), 815–832.
Cha, Y.-J., et al. (2014). “Performance validations of semiactive controllers Gao, X., Castaneda, N., and Dyke, S. J. (2013). “Experimental validation of
on large-scale moment resisting frame equipped with 200-kN MR a generalized procedure for MDOF real-time hybrid simulation.” J.
damper using real-time hybrid simulations.” J. Struct. Eng., 10.1061/ Eng. Mech., 10.1061/(ASCE)EM.1943-7889.0000696, 04013006.
(ASCE)ST.1943-541X.0000982, 04014066. Horiuchi, T., Nakagawa, M., Sugano, M., and Konno, T. (1996). “Develop-
Chae, Y., Jiang, Z., Ricles, J., Christenson, R., and Dyke, S. J. (2013a). ment of a real-time hybrid experimental system with actuator delay
“MR damper characterization—Lehigh—damper 1.” Network for compensation.” Proc., 11th World Conf. Earthquake Engineering,
Earthquake Engineering Simulation (distributor). Purdue Univ., West IAEE, Tokyo, Japan.
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

Lafayette, IN. 〈https://nees.org/warehouse/experiment/3031/project/ Jiang, Z., et al. (2010). “Comparison of 200 KN MR damper models for
648〉 (Sep. 02, 2011). use in real-time hybrid simulation.” 〈https://nees.org/resources/670〉
Chae, Y., Jiang, Z., Ricles, J., Christenson, R., and Dyke, S. J. (2013b). (Aug. 16, 2010).
“MR damper characterization—Lehigh—damper 2.” Network for Jiang, Z., and Christenson, R. (2011). “A comparison of 200 kN magneto-
Earthquake Engineering Simulation (distributor). Purdue Univ., West rheological damper models for use in real-time hybrid simulation
Lafayette, IN. 〈https://nees.org/warehouse/experiment/3032/project/ pretesting.” J. Smart Mater. Struct., 20, 065011.
648〉 (Sep. 02, 2011). Jiang, Z., and Christenson, R. E. (2012). “A fully dynamic magneto-
Chantranuwathana, S., and Peng, H. (1999). “Force tracking control for rheological fluid damper model.” Smart Mater. Struct., 21(6), 065002.
active suspensions-theory and experiments.” Control Applications Jiang, Z., Kim, S. J., Plude, S., and Christenson, R. E. (2013). “Real-time
Proc., 1999 IEEE Int. Conf., Vol. 1, IEEE, New Brunswick, NJ. hybrid simulation of a complex bridge model with MR dampers using
Chen, C., and Ricles, J. M. (2008). “Development of direct integra- the convolution integral method.” Smart Mater. Struct., 22(10), 105008.
tion algorithms for structural dynamics using discrete control MATLAB [Computer software]. (2011). Natick, MA, Mathworks.
theory.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(2008)134:8(676), Metered, H., Bonello, P., and Oyadiji, S. (2009). “Nonparametric identifi-
676–683. cation modeling of magnetorheological damper using Chebyshev poly-
Chen, C., and Ricles, J. M. (2010). “Tracking error-based servohydraulic nomials fits.” SAE Int. J. Passenger Cars Mech. Syst., 2(1), 1125–1135.
actuator adaptive compensation for real-time hybrid simulation.” Nakata, N. (2010). “Acceleration trajectory tracking control for earthquake
J. Struct. Eng., 10.1061/(ASCE)ST.1943-541X.0000124, 432–440. simulators.” Eng. Struct., 32(8), 2229–2236.
Christenson, R., Lin, Y. Z., Emmons, A., and Bass, B. (2008). “Large-scale Nakata, N., and Stehman, M. (2012). “Substructure shake table test method
experimental verification of semiactive control through real-time hybrid using a controlled mass: Formulation and numerical simulation.” Earth-
simulation.” J. Struct. Eng., 10.1061/(ASCE)0733-9445(2008)134: quake Eng. Struct. Dyn., 41(14), 1977–1988.
4(522), 522–534. Ohtori, Y., Christenson, R., Spencer, B. F., and Dyke, S. J. (2004).
Dermitzakis, S. N., and Mahin, S. A. (1985). “Development of substructur- “Benchmark control problems for seismically excited nonlinear build-
ing techniques for on-line computer controlled seismic performance ings.” J. Eng. Mech., 10.1061/(ASCE)0733-9399(2004)130:4(366),
366–385.
testing.” Rep. No. UBC/EERC-85/04, Earthquake Engineering Research
Center, Univ. of California, Berkeley, CA. Ozdagli, A., Friedman, A., Ahn, R., Dong, B., Dyke, S. J., and Ricles, J. M.
(2013a). “Dynamic system identification on DBF—rigid links ON—
Dyke, S. J. (1996). “Acceleration feedback control strategies for active
ground links ON—loading beams OFF.” Network for Earthquake
and semi-active control systems: Modeling, algorithm development,
Engineering Simulation (distributor). Purdue Univ., West Lafayette, IN.
and experimental verification.” Ph.D. dissertation, Univ. of Notre
〈https://nees.org/warehouse/experiment/3070/project/648〉 (Sep. 09, 2011).
Dame, Notre Dame, IN.
Ozdagli, A., Friedman, A., Ahn, R., Dong, B., Dyke, S. J., and Ricles, J. M.
Friedman, A. J. (2012). “Development and validation of a new control strat-
(2013b). “Dynamic system identification on DBF—rigid links OFF—
egy considering device dynamics for large-scale MR dampers using
ground links ON—loading beams OFF.” Network for Earthquake
real-time hybrid simulation.” Ph.D. dissertation, Purdue Univ., West
Engineering Simulation (distributor). Purdue Univ., West Lafayette,
Lafayette, IN.
IN. 〈https://nees.org/warehouse/experiment/3071/project/648〉 (Sep.
Friedman, A., et al. (2013a). “RTHS (damper)—3StoryPS—single MR 09, 2011).
damper (floor 1).” Network for Earthquake Engineering Simulation Ozdagli, A., Friedman, A., Ahn, R., Dong, B., Dyke, S. J., and Ricles, J. M.
(distributor). Purdue Univ., West Lafayette, IN. 〈https://nees.org/ (2013c). “System identification on DBF—rigid links ON/OFF—ground
warehouse/hybrid/3785/project/648〉 (Oct. 28, 2013). links ON—loading beams ON—actuator @ floor 1.” Network for
Friedman, A., et al. (2013b). “RTHS (damper)—3StoryPS—single MR Earthquake Engineering Simulation (distributor). Purdue Univ., West
damper (floor 1).” Network for Earthquake Engineering Simulation Lafayette, IN. 〈https://nees.org/warehouse/experiment/3075/project/
(distributor). Purdue Univ., West Lafayette, IN. 〈https://nees.org/ 648〉 (Sep. 17, 2011).
warehouse/hybrid/3792/project/648〉 (Oct. 28, 2013). Ozdagli, A., Friedman, A., Ahn, R., Dong, B., Dyke, S. J., and Ricles, J. M.
Friedman, A., et al. (2013c). “RTHS (frame + damper)—3StoryPS—single (2013d). “System identification on DBF—rigid links ON/OFF—
MR damper (floor 1).” Network for Earthquake Engineering Simulation ground links ON—loading beams ON—actuator @ floor 2.” Network
(distributor). Purdue Univ., West Lafayette, IN. 〈https://nees.org/ for Earthquake Engineering Simulation (distributor). Purdue Univ.,
warehouse/hybrid/3807/project/648〉 (Oct. 28, 2013). West Lafayette, IN. 〈https://nees.org/warehouse/experiment/3206/
Friedman, A., et al. (2013d). “RTHS (frame + damper)—9StoryBM— project/648〉 (Sep. 17, 2011).
single MR damper (floor 1).” Network for Earthquake Engineering Ozdagli, A., Friedman, A., Ahn, R., Dong, B., Dyke, S. J., and Ricles, J. M.
Simulation (distributor). Purdue Univ., West Lafayette, IN. 〈https:// (2013e). “System identification on DBF—rigid links ON/OFF—ground
nees.org/warehouse/hybrid/3809/project/648〉 (Oct. 28, 2013). links ON—loading beams ON—actuator @ floor 3.” Network for
Friedman, A., et al. (2013e). “RTHS (frame + damper)—9StoryBM— Earthquake Engineering Simulation (distributor). Purdue Univ., West
two MR dampers (floor 1 + 2).” Network for Earthquake Engineering Lafayette, IN. 〈https://nees.org/warehouse/experiment/3207/project/
Simulation (distributor). Purdue Univ., West Lafayette, IN. 〈https://nees 648〉 (Sep. 17, 2011).
.org/warehouse/hybrid/3811/project/648〉 (Oct. 28, 2013). Phillips, B., Jiang, B., Spencer, B. F., Christenson, R., and Dyke, S. J.
Friedman, A., and Dyke, S. J. (2013). “Development and experimental (2013). “MR damper characterization—UIUC—damper 3.” Network
validation of a new control strategy considering device dynamics for for Earthquake Engineering Simulation (distributor). Purdue Univ.,

© ASCE 04014150-12 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150

 West Lafayette, IN. 〈https://nees.org/warehouse/experiment/2796/ Simulink [Computer software]. (2011). Natick, MA, Mathworks.
project/648〉 (Aug. 31, 2011). Spencer, B. F., Christenson, R. E., and Dyke, S. J. (1998). “Next generation
Phillips, B. M., and Spencer, B. F., Jr. (2011). “Model-based feedforward- benchmark control problem for seismically excited buildings.” Proc.,
feedback tracking control for real-time hybrid simulation.” Newmark 2nd World Conf. on Structural Control, Wiley, Chichester, U.K., Vol. 2.
Structural Engineering Laboratory Rep. Series, Univ. of Illinois at Spencer, B. F., Dyke, S. J., Sain, M. K., and Carlson, J. D. (1997).
Urbana-Champaign, Urbana, IL. “Phenomenological model of a magnetorheological damper.” J. Eng.
Phillips, B. M., and Spencer, B. F., Jr. (2012). “Model-based multiactuator Mech., 10.1061/(ASCE)0733-9399(1997)123:3(230), 230–238.
control for real-time hybrid simulation.” J. Eng. Mech., 10.1061/ Stengel, R. F. (1986). Stochastic optimal control: Theory and application,
(ASCE)EM.1943-7889.0000493, 219–228. Wiley, New York.
Saouma, V., Haussmann, G., Kang, D., and Ghannoum, W. (2013). “Real Wu, B., Wang, Q., Shing, P. B., and Ou, J. (2007). “Equivalent force control
time hybrid simulation of a non-ductile reinforced concrete frame.” method for generalized real-time substructure testing with implicit in-
J. Struct. Eng., 10.1061/(ASCE)ST.1943-541X.0000813, 04013059. tegration.” Earthquake Eng. Struct. Dyn., 36(9), 1127–1149.
Saouma, V., Kang, D.-H., and Haussmann, G. (2012). “A computational Yang, G. (2001). “Large-scale magnetorheological fluid damper for vibra-
finite-element program for hybrid simulation.” Earthquake Eng. Struct. tion mitigation: Modeling, testing and control.” Ph.D. dissertation,
Downloaded from ascelibrary.org by Purdue University Libraries on 05/14/23. Copyright ASCE. For personal use only; all rights reserved.

Dyn., 41(3), 375–389. Univ. of Notre Dame, Notre Dame, IN.

© ASCE 04014150-13 J. Struct. Eng.

J. Struct. Eng., 2015, 141(6): 04014150